The Impact of Forced Displacement on Host Communities A Review of the Empirical Literature in Economics 1 © 2019 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522- 2625; e-mail: pubrights@worldbank.org. 2 Table of Contents Acknowledgments......................................................................................................................................... 4 Abstract ......................................................................................................................................................... 5 Executive Summary....................................................................................................................................... 6 1. Introduction ......................................................................................................................................... 13 2. Basic theory ........................................................................................................................................ 15 3. Empirical modelling and identification strategies............................................................................... 17 4. Meta-analysis of empirical results ...................................................................................................... 21 4.1 Data ............................................................................................................................................. 21 4.2 Results ......................................................................................................................................... 23 5. Conclusion .......................................................................................................................................... 26 References ................................................................................................................................................... 29 Annex 1 – Review of Empirical Models ....................................................................................................... 47 1.1 Well-being and prices ................................................................................................................. 47 1.2 Employment and wages .............................................................................................................. 50 Annex 2 – Empirical Results by Crisis .......................................................................................................... 60 2.1 High Income countries ................................................................................................................ 60 2.2 Middle-income countries ............................................................................................................ 66 2.3 Low income countries ................................................................................................................. 69 3 Acknowledgments The study has been prepared by Paolo Verme (FCV) and Kirsten Schuettler (SPJ). The authors are grateful for comments received from Xavier Devictor (Program Manager, FTFOS), William Wiseman (Program Leader, LCC1C), Caglar Ozden (Lead Economist, DECTI), Christian Eigen-Zucchi (Program Leader, SACBN), and Harun Onder (Senior Economist, GMTMN). This work is part of the program ``Building the Evidence on Protracted Forced Displacement: A Multi-Stakeholder Partnership''. The program is funded by UK aid from the United Kingdom's Department for International Development (DFID), it is managed by the World Bank Group (WBG) and was established in partnership with the United Nations High Commissioner for Refugees (UNHCR). The scope of the program is to expand the global knowledge on forced displacement by funding quality research and disseminating results for the use of practitioners and policy makers. This work does not necessarily reflect the views of DFID, the WBG or UNHCR. The authors are also grateful to participants to the conference on the impacts of refugees in hosting economies held at the University of Southern California, Los Angeles, September 14-15, 2018. All remaining errors are responsibility of the authors. 4 Abstract The paper reviews 49 empirical studies that estimate the impact of forced displacement on host communities. A review of the empirical models used by these studies and a meta-analysis of 762 separate results collected from these studies are the main contributions of the paper. Coverage extends to 17 major forced displacement crises occurred between 1922 and 2015, to host countries at different level of economic development and to different types of forced migrants. The focus is on outcomes related to household well- being, prices, employment and wages. All studies can be classified as ex-post quasi natural experiments. The analysis on empirical modelling shows a preference for partial equilibrium modelling, differences-in- difference evaluation methods and cross-section econometrics with all these choices largely dependent on the type of data available. The meta-analysis on household well-being shows that between 45% and 52% of results are positive and significant indicating a net improvement in household well-being. An additional 34-42% of results are found to be non-significant and between 6% and 20% show a decrease in household well-being. The analyses on employment and wages show positive and significant improvements for 12- 20% of results, non-significant results in 63% of the cases and negative and significant results for 22-25% of results. Negative results on employment and wages relate to young and informal workers in middle- income countries. Results on prices show asymmetric behavior across types of products. Overall, the probability of having a negative outcome for host communities in the consumer and labor markets is below 20%. 5 Executive Summary The question of whether forced displacement is beneficial or detrimental to host communities1 has become a hotly debated issue in policy, political and media circles since the start of the Syrian refugee crisis in 2011 and the EU migration crisis in 2015. Economics has for long paid little attention to this phenomenon with only occasional studies of mostly historical interest until this issue became popular among media outlets. The economics discipline recently responded to this growth in public interest by multiplying research efforts and producing a string of studies published in major journals. As shown in Figure 1, the first study of this kind dates back to 1990 and between 1990 and 2011 an average of only one study per year reached publication. This changes after 2011 when the average number of studies per year increases ten folds. Thanks to these recent efforts, we now have a much more solid body of evidence addressing this simple question: What is the impact of forced displacement on host communities? Figure 1 - Number of Published Studies (1990-2018) 10 Syrian crisis num 5 0 1990 2000 2010 2020 year This is not, of course, an easy question to answer. The impact of forced displacement on host communities requires qualifying the nature of forced displacement and the outcome considered, defining other factors influencing these outcomes, and overcoming several technical difficulties. Among forcibly displaced populations we find refugees, Internally Displaced Persons (IDPs), returnees, escapees, repatriates, and other groups that flee various types of situations such as wars, political persecution, domestic or ethnic violence or natural disasters. The economics literature has covered most of these groups but focused essentially on forced displacement generated by some form of violence or high levels of insecurity or uncertainty. The outcome observed can also be wide ranging. Host communities may be affected by changes in the financial, labor and consumer markets, provision of services, psychological effects, cultural differences, environmental degradation, traffic congestions, and other effects that a mass inflow of people 1 Host communities are defined as natives or existing residents who are affected by a sudden influx of forcibly displaced persons. For measurement purposes, these communities are generally identified by the literature in terms of administrative areas, but it is evident that these areas may include or exclude persons who are or are not affected by the displacement shock, and that the size of the impact might differ between them. 6 may generate. The economics profession has clearly been narrow in this respect and largely focused on the labor and consumer markets. There are also several technical challenges that make these evaluations complex and difficult to compare. The very nature of forced displacement implies that crises are unpredictable, and this excludes any form of planned experimental evaluation. Evaluations are ex-post and need to rely on existing data collected in situations that are typically challenging. Data are often scarce and of poor quality and determining correlations and causality is extremely hard. Separating the impacts of forced displacement from the impacts of conflict itself, global trends or other events happening at the same time is challenging. Comparing studies across countries and time is also arduous. The displacement crises observed can be very different, the population studied are very diverse, and impacts might change over time. Several factors influence results that vary between the different settings, including the economic conditions in the host country before the inflow and the policy response to the inflow. The models used by researchers are similar but never the same. Overall, none of the papers reviewed in this study can claim to have the ultimate answer to this question. This is the first comprehensive review of its kind, a fact that is largely explained by the scarcity of studies that characterized this discipline until very recently. The paper reviews 49 empirical studies that estimate the impact of forced displacement on host communities and provides a meta-analysis of 762 separate results collected from these studies. This literature covers a total of 17 major forced displacement crises occurred between 1922 and 2015. It includes host countries at different level of economic development, different types of forced migrants and situations where the scale of the phenomenon and the policy response has been very different. The focus is on the consumer and labor market and, more precisely, on outcomes related to household well-being, prices, employment and wages. In order of coverage of these outcomes, employment comes first (334 results) followed by wages (244), prices (120) and well-being (64).2 Among these four outcomes, household well-being is the only one that can be considered a comprehensive indicator of the impact of forced displacement on host communities. The review has two objectives. First, we wish to provide a review of the specific modelling and econometric challenges that this type of work entails for the benefit of social scientists who wish to work in this area. For those mostly interested in outcomes, this may not seem a critical contribution of this study and such analysis may be hard to follow at times. But for those social scientists engaged in this type of research, this is a real asset as they will find one place where most of the available models and methodologies designed to address these types of questions can be found. We hope that this will contribute to assure quality and rigor in this growing field of research. Second, we review the empirical results emerged from this literature and provide a meta-analysis with the objective of summarizing results by selected outcome and provide initials leads on some of the factors that may drive these outcomes. By doing so, we wish to bring some clarity to a very complex and controversial topic. As this is the first review of this kind, the meta-analysis is also the first attempt to provide an answer to the question of whether forced displacement affects host populations overall and across countries. What is the gold standard to estimate the impact of forced displacement on host populations’ outcomes? There are no magic tricks here and, as already mentioned, these evaluations are carried out ex-post, under data scarcity and difficult environments. The classic gold standard of impact evaluations – Randomized Controlled Trials (RCT) – cannot be applied in the context of forced displacement by definition. The shock (conflict) cannot be simulated and people cannot be forced to flee while others are forced to stay. The second best are natural experiments. Most authors claim that the random selection and allocation of the displaced due to an unpredictable event represents a natural experiment. In reality, all authors accept the fact that the selection and allocation of the displaced may not be so random and that many unobserved factors are at play. In line with best standards under these conditions, these authors use various techniques 2 For household well-being, it is intended the household per capita level of income, consumption or expenditure. 7 to approximate a controlled experiment including Differences-in-Difference, longitudinal, panel or Instrumental Variables (IV) models accompanied by matching, placebo controls and other techniques designed to approximate random conditions. As many of these studies have been published in top economics journals, the results analyzed in this review have gone through close scrutiny and should be regarded as the gold standard in this particular research area. The meta-analysis of these results provides, therefore, some initial and general answers to the questions that this literature has addressed. Does forced displacement affect household well-being of hosts populations? The answer is largely “yes” and, according to 45-52%3 of the results analyzed, the impact is positive meaning that household well-being of the hosts increases as a result of forced displacement. An additional, 34-42% of results are non-significant meaning that a clear correlation between forced displacement and household-wellbeing could not be established. For the remaining 6-20% of results the impact is negative, meaning that household well-being for the hosts has decreased as a result of displacement. Overall, less than 1 in 5 results is negative. Unfortunately, the limited number of results analyzed is not sufficient to breakdown results by type of household, type of crisis or type of host country. This is therefore one area where much more research is needed to provide some insights into the drivers of changes in well-being. However, the papers analyzed seem to support the hypothesis that positive results are induced by increases in consumer demand that translates into increased production, productivity and service provision on the part of hosts. In some cases, evidence of skills upgrading for host workers and technology changes for host producers has also been reported. Negative results are too few to derive any lesson. Does forced displacement affect prices in hosts countries? Yes, prices in host countries are affected by forced displacement. Between 37 and 46% of results show that prices increase as a result of forced displacement and between 36 and 39% of results show that they decrease. Only between 18 and 23% of results show no correlation between forced displacement and prices. There is no good or bad interpretation of these results. Increases in prices benefit producers and home owners while damaging consumers and tenants. The net effect on household well-being is difficult to predict when prices change. However, an analysis of the positive results (price increases) shows that these are almost entirely represented by food and rents prices (for certain types of housing) whereas an analysis of the negative values (prices decreases) shows that these are represented mostly by overall inflation, luxury goods, services, and labor-intensive products. These results are very preliminary and not necessarily intuitive. Differences in supply elasticity seem to be an important factor shaping results. Future research on this topic will need to focus on changes by type or class of products to start discerning the relation between changes in prices and household well- being. A better understanding of the channels through which the forced displacement inflow affects prices besides an increase in demand would also be important to inform policy. Does forced displacement affect employment among host populations? Only in a minority of cases and only for selected groups of the population. The meta-analysis shows that between 12 and 14% of results are positive and significant meaning that employment for the hosts increases after a forced displacement crisis, whereas between 19 and 24% are negative and significant meaning that employment decreases. Therefore, in this case, negative results are more frequent than positive results. However, for the great majority of results (between 65 and 67%) the impact on employment is non-significant, meaning that authors could not identify any correlation between forced displacement and employment. Overall, the likelihood of a forced displacement crisis to have a negative impact on employment for the host population is below 25%. It is also found that negative outcomes are associated with young and informal workers whereas the analysis of results on gender and education remains inconclusive. Overall, employment is the most researched of the four outcomes considered by this study but results on population sub-groups remain scarce and future 3These ranges derive from the analysis of results with and without weights where weights are the journals’ impact factor, a measure of the quality of journals. 8 research will need to address this gap. Even if forced displacement contributes to decrease overall employment in a minority of cases, the number of workers affected may be large and concentrated among few groups. From a policy perspective, it is essential to better understand the distributional effects on employment, a question that this review could not address because of lack of data. Does forced displacement affect wages among host populations? The majority of results show no correlation between these two factors, negative results are more frequent than positive results, but these tend to be short lived effects concentrated in Middle Income Countries (MICs). Between 13 and 22% of results are positive and significant (wages increase), between 55 and 59% are non-significant (no correlation found) and between 22 and 27% are negative and significant (wages decrease). The overall evidence shows that only about one in four results is negative. There is also some initial evidence that negative results turn into positive results in the long-term and that wages for hosts in High Income Countries (HICs) perform better than wages for hosts in MICs. This may be evidence of skills’ complementarities in HICs and skills substitutability in MICs. The meta-analysis was unable to uncover any robust result related to population characteristics such as gender, age, or education. Again, this is an area that will require increased research in the future. Figure 2 - Share of Results by Outcome Well-being Well-being (weighted) Prices Prices (weighted) Employment Employment (weighted) Wages Wages (weighted) 0 20 40 60 80 100 120 Positive (increase) Nonsignificant (no correlation) Negative (decrease) What can we derive from these results? Despite recent efforts in understanding the relation between forced displacement and host populations, research in this area remains in its infancy. Studies have focused on selected markets, first round and short and medium-term effects. Very little work is available on second- round and long-term and dynamic effects, on the production side of the economy and on the impact of forced displacement on primary services such as water, electricity, education or health. Results on household well-being, which should be the most important outcome to study, remain few. These are all areas that will require greater research efforts to complement the existing literature. Results are also derived from a multitude of models and case-studies and their comparability remain a challenge. We could not 9 provide, for example, reliable summary figures on the size of the measured effects forcibly limiting our analysis to the significance level of the econometric estimates. As it is often the case in newly researched topics, the first conclusion is that more research is needed and the first recommendation is that results should be treated with caution. However, the study has challenged with evidence based on microdata several assumptions based on anecdotal or selected evidence. The first assumption is that forced displacement is invariably negative for host countries. The study has shown that this is not the case. If one really wants to generalize, the likelihood of a positive result is much higher than the likelihood of a negative result. Results are also dynamic and negative impacts tend to disappear over time. A second assumption is that forced displacement is bad for all hosts. The study has provided some initial evidence that negative effects are concentrated among selected population groups and certain type of countries, while other parts of the population and some countries benefit from positive effects. A third assumption is that increases in prices due to forced displacement damage invariably the host population. The study has shown that increases in prices benefit some people and damage others and that household well-being and prices can increase at the same time. A fourth assumption is that displaced people are forcibly a net cost for the host population. The study has shown that household well-being of the hosts - a measure of the overall economic impact on hosts - is more likely to increase than decrease in the aftermath of a crisis. A fifth assumption is that displaced people are idle and non-productive. The study has shown that the increase in consumer demand is one of the explanations for a positive outcome on household well-being of the hosts and that many displaced people are occupied in either the informal or formal markets despite restrictions in work permits. A sixth assumption is that displaced people steal jobs. The study has found that there can be substitutability as well as complementarity between displaced people and hosts and that the net effect on household well-being is more likely to be positive than negative for the hosts. Given the results described, it is important to ask what drives popular assumptions, a question that this study has not addressed. There can be noticeable differences between the measured impacts on host communities and perceptions of these impacts from these same hosts. The impact of displacement on the subjective well-being of hosts does not necessarily correspond to the impact on their objective well-being. To our knowledge, only one study looked at these differences (Kreibaum, 2015) and found this difference to be sizable. Subjective well-being can be a powerful driver for change and understanding its relation with objective well-being is key from a policy perspective. New data collection and research efforts should take this aspect into account. Political upheaval is largely driven by perceptions and politicians are evidently and rightly concerned about perceptions. The historical cases studied show how the forced migrants are defined and how they are perceived influences the willingness to integrate them by the host society, the possibility of the state to pursue certain policies to integrate them and thus the actual outcomes, especially in the long-term. Contributing to align perceptions with facts and understanding the drivers of the mismatch between perceptions and reality is one of the important tasks that policy-oriented research is called to uncover. What does this mean for policy? None of the studies covered by this review explicitly measured the effects of policy changes on host communities. This is an important issue of course not only from a policy perspective but also from a research perspective. Policies affect outcomes and the different policies administered cross-country are not easily comparable. Unfortunately, only a few studies analyzed the channels through which the influx of forcibly displaced people influences outcomes for the host communities and we know very little on which policies are effective and which are not. A common thread to most papers that find positive impacts on the well-being of host communities is that the growth in demand determined by the population and expenditure shocks and the simultaneous increase in prices and decline 10 in wages are very beneficial to local producers and owners of assets boosting production, encouraging investments and eventually outweighing the negative substitution effects for unskilled labor. None of these effects directly derive from policy changes but it is clear that a reduction in entry barriers to the labor market can only accelerate such process. Some papers have found evidence of the importance of investments and regulations for improving living conditions in areas affected by displacement. Prices increase because supply might be non-elastic, at least in the short-term, in poor and isolated areas and for non-tradables, like housing. To increase the price elasticity of supply for food items in poor and isolated areas, investments by the government, donors and humanitarians can help connect these places to markets. The improved road network seems to have a positive impact on household welfare even after the forced migrants return (Maystadt and Verwimp 2018). An improved business and investment climate will also speed up the reaction of the private sector to an increase in demand. An increase in the issuance of construction permits, notably for social housing, can help buffer effects on the housing market, at least in the medium term. If construction permits for high- income housing crowd out construction permits for social housing instead, the negative incomes on lower income hosts are reinforced (Depetris-Chauvin and Santos 2018). Relatively simple labor market measures can also contribute to normalize labor markets that have been upset by an inflow of displaced persons. There is some evidence that negative impacts on employment of hosts might be stronger in countries with more rigid labor markets (Angrist and Kugeler 2003). Restrictions on the right to work usually mean that refugees of all skills are limited to compete with low-skilled workers in the informal sector, potentially increasing negative impacts on already vulnerable groups (as the studies on Turkey and Jordan show). Allowing refugees to work will disperse the impacts across different sectors and skill levels. Policy makers should also support additional job creation. As two papers on Turkey exemplify, enterprises created by refugees themselves can contribute to these efforts, if policies and regulations allow them to (Akgündüz, van den Berg, and Hassink 2018; Altindag, Bakis, and Rozo, 2018). The papers we reviewed that looked at tasks complexities and the question of substitution versus complementarities between refugees and natives found occupational upgrading among natives as a result of the refugee inflow (Akgündüz, van den Berg, and Hassink 2018, Akgündüz and Torun 2018, and Foged and Peri 2015). Policies can reinforce these complementarities between forced migrants and native workers and increase the productivity of native workers by providing incentives to upgrade their skills. As a number of studies showed, internal migration helps dissolve some of the impacts on the labor market and could be incentivized by policy makers. Capital flows can help re-equalize capital/labor ratios within the country, if allowed to do so. In general, policies are needed to counterbalance the distributional impacts of a forced displacement inflow on the labor and consumer markets. There are also some recurrent findings once we consider small and large crises, poor and rich host countries and short and long-term effects. Studies on larger crises in richer countries (HICs and MICs) with more developed labor markets provided evidence that low-skilled workers are the main losers during a displacement crisis. In line with the literature on economic migration, studies that compared short and long- term effects found that negative effects tend to disappear in the long-run. Studies on the overall well-being of hosts in the aftermath of a displacement crisis in low-income countries show rather consistent positive results also explained by the inflow of international aid and its relative importance in a poor country. These are preliminary but consistent findings that emerged from the review. Policy makers need to be aware of a potential trade-off between containing negative short-term impacts and ameliorating the long-term prospects. If refugees are not able to work, this might limit the negative impacts on the labor markets of the host community in the short term. But it also means that potential positive contributions of the refugees remain limited to financial flows (i.e. aid/government spending and 11 remittances). Limitations on the freedom of movement of forced migrants also shape impacts. Encampment reduces the aggregate effects but concentrates them around the camp. Integrating the refugees into society increases their economic effects in aggregate terms but diffuses it throughout the whole country (Artuc et al. 2017). Placement policies which distribute refugees across the country based on non-economic criteria diffuse impacts but might also hinder refugee labor market integration and their ability to contribute to the economy in the medium to long-term. Investments in health and education services as well as labor market integration are costly in the short-term but will pay off in the medium to longer term. Low- and medium- income countries need support from the international community to fund these investments. Policy makers need to solve challenges that pre-date the influx of forced migrants. The influx of forced migrants often only exacerbates policy issues that existed among host communities beforehand, like a too rigid labor market, an important informal sector, inefficient subsidies or non-existent services. These initial conditions contribute to shape the effects of forced displacement on host communities. The influx can create the pressure needed to finally tackle these challenges. To better inform policy making, future studies should not only ask what the impact of forced displacement is - as we did in this paper - but provide evidence on the factors that drive these impacts. 12 1. Introduction The question of whether Forced Displacement4 is beneficial or detrimental to host communities5 has become a hotly debated issue in policy, political and media circles since the start of the Syrian refugee crisis in 2011 and the EU migration crisis in 2015. Economics has for long paid little attention to this phenomenon with only occasional studies of mostly historical interest until this issue became popular among media outlets. The first study of this kind dates back to 1990 and between 1990 and 2011 an average of only one study per year reached publication (Figure 1). This changes after 2011 when the average number of studies per year increases ten folds. Thanks to these recent efforts, we now have a much more solid body of evidence addressing this simple question: What is the impact of forced displacement on host communities? This paper provides a review of the economics literature that addressed this question. Forced displacement is a different phenomenon from economic migration justifying a separate review.6 By definition, FD is less of a choice and less voluntary than EM, although there is ultimately always a choice behind most (but not all) migration decision. FD is a decision that is taken quickly following a sudden shock as opposed to EM, which is more often a carefully planned move. Forced migrants typically carry some small savings with them but little else because of the sudden nature of the shock whereas economic migrants tend to carry savings and assets or transfer these in advance of the move. Economic migrants tend to rely on extended networks in the place of origin and destination and plan their move in accordance with these networks. Forced migrants tend to move to destinations based on proximity and security criteria rather than personal networks, although networks can occasionally play a role. EM is a regular phenomenon with increasing and decreasing trends whereas FM happens in sudden and unexpected burst of population movements which can be massive in nature.7 Behavioral characteristics can be very different between economic and forced migrants. FD is therefore a rather different phenomenon from EM calling for different types of theoretical and empirical instruments (Verme 2016, Ceriani and Verme 2018). The objective of this review is two-fold. First, we wish to provide a review of the specific modelling and econometric challenges that this type of work entails for the benefit of social scientists who wish to work in this area. Second, we review the empirical results emerged from this particular literature and provide a meta-analysis with the objective of summarizing results by selected outcome and provide initials leads on some of the factors that may drive these outcomes. By doing so we wish to bring some clarity to a very complex and controversial topic. The focus of the review is dictated by the existing literature which concentrated its efforts in understanding the impact on the consumer and labor markets and on four outcomes: household well-being, prices, employment and wages.8 Household well-being measured in terms of income, consumption or wealth is the main outcome of interest to understand whether the net effect of a FD crisis is positive or negative for the 4 Under the term FD we include refugees, returnees, expellees, escapees and Internally Displaced Persons (IDPs). These populations may have different characteristics, but they represent groups that have been subjected to FD due to some form of conflict, violence, persecution, human rights violations or high levels of insecurity or uncertainty resulting in a sudden and massive movement of people. We exclude episodes of Forced Migration (FM) due to environmental or other types of disasters and occasional or small episodes of forced displacement. Forced displacement is sometimes referred to as forced migration. We use these terms interchangeably in this paper. 5 Host communities are defined as natives or existing residents who are affected by a sudden influx of forcibly displaced persons. For measurement purposes, these communities are generally identified by the literature in terms of administrative areas, but it is evident that these areas may include or exclude persons who are or are not affected by the displacement shock. 6 We are aware that the lines between forced migrants and economic migrants can be blurred, and that over time forced migrants might become more similar to economic migrants, notably in the case of secondary movements. 7 Sudden and massive movements are much more frequent in the context of forced displacement, but there are cases of sudden and massive inflows of migrant workers. One example is a new commuting policy that led to a sharp and unexpected inflow of Czech workers to areas along the German-Czech border (Dustmann, Schoenberg, and Stuhler 2017). 8 Few papers look at the impacts of refugees and IDPs on education (Semrad 2015; Assad, Ginn and Saleh 2018; Tumen 2018), health (Baez 2011), the environment (Martin et al. 2017) or at the impacts on crime and social cohesion in the host communities (Amuedo-Dorantes, Bansak and Pozo 2018; Depetris-Chauvin and Santos 2018; Masterson and Yasenov 2018). 13 host population. Negative changes in markets outcomes such as increases in consumer prices or decreases in wages damage consumers and workers but benefit producers and owners of assets. The net effect on household well-being is not obvious when wages and prices change. Besides increasing the labor supply and creating a demand stimulus on consumer markets, refugees can also have an impact on productivity and structural change (Paserman 2013; Hornung 2014; Braun and Kvasnicka 2014; Savimaki 2011; Peters 2017), innovation and new patents (Moser, Voena, and Waldinger 2014), create new enterprises (Akgündüz, van den Berg, and Hassink 2018; Altindag, Bakis, and Rozo) or increase FDI (Mayda, Parsons, and Vézina 2017) and trade with their countries of origin (Parsons and Vezina 2018; Ghosha and Enamib 2015; Mayda, Parsons, and Steingress 2017). The overall impact on household well-being is evidently the product of a combination of multiple factors and labor market analyses capture only some of these factors. In our knowledge, this is the first comprehensive review of its kind. Ruiz and Vargas-Silva (2013) carried out a literature review of the impact of forced displacement on the displaced and on host communities but the review on host communities provides a brief overview of only eight papers, as most of the available literature is more recent. Related reviews on migration or the impacts of war and violence are broader in scope and have only occasional references to papers covering the impact of forced migrants on host communities. The recent reviews by Özden and Wagner (2018) and Dustmann, Schönberg and Stuhler (2016) focus on the labor market impacts of migration and only cover some of the natural experiments included in this paper. Other reviews only cover one country or region (e.g. Ogude 2017; Verwimp and Maystadt 2015; Mabiso et al. 2014). Some of the empirical papers we review offer an overview of the literature, such as Borjas and Monras (2017) and Clemens and Hunt (2017) who revisit several cases of large and sudden inflows in high-income countries, but none of these papers covers the range or scope of this work. The literature on the impacts of forced displacement on the displaced themselves was very limited until recently (see the reviews by Kondylis and Mueller 2014 and Ruiz and Vargas-Silva 2013) but has also started to grow in the last years (e.g. Gimenez-Nadal, Jose Ignacio, José Alberto Molina, and Edgar Silva- Quintero 2018; Fransen, Vargas-Silva, and Siegel 2018). The results of this literature are linked to our review as the impacts on the financial assets, human capital and psychological well-being of forced migrants will in return influence how they impact the consumer and labor markets of the host community. Most of the papers covered are published in peer-reviewed international journals and most of these journals are top ranked journals in their respective disciplines. This set of papers is complemented by papers published as working papers in reputable series by known authors using standard modelling techniques. The oldest paper covered is dated 1990 and the newest 2018. No papers prior to 1990 were found. The episodes of FD included in this literature span from 1922 to 2015 and cover 17 of the major FD crises of this period, those that received the most attention from scholars. These are distributed between high, medium and low-income host countries and include episodes of FD in the US and Europe, Middle East, North-Africa, Sub-Saharan Africa and Latin America. Only empirical papers in economics with original results are covered by the review. Almost all studies are described as natural experiments by the authors but face major measurement and identification challenges. Availability of micro data is one of the challenges and the first explanation of why these types of studies have emerged only very recently.9 All studies have been undertaken ex-post, after the displacement crisis has taken place. The unexpected nature of the crisis and the randomness of the allocation of displaced persons are two elements used to defend the natural experiment assumption. However, all papers address the central question of endogeneity and unobserved heterogeneity. How random is the decision to leave, the choice of destination or the type of people who flee? What other 9 National household surveys do not normally cover displaced populations and humanitarian agencies in charge of displaced populations do not normally cover host populations in their surveys. These latter surveys also rarely contain socio-economic information of sufficient quality to be used in econometric studies, not least because issues such as sampling and questionnaire design are extremely difficult with mobile populations such as refugees and IDPs. Registration data do not always capture the whole displaced population, might be outdated and focus on the displaced rather than host communities. Displaced people are usually hosted in marginal areas where data are scarce or of poor quality. 14 unobserved concomitants factors such as growth, natural disasters or international aid have contributed to the observed outcomes? These studies are therefore better described as quasi-natural experiments. There are several factors affecting results that should be considered when comparing these results across countries and across FD episodes. The income per capita of the host country is an obvious factor which has also implications on the economic structure of the labor and consumer markets’ institutions and the degree of formality of these markets. It also determines whether international aid or increased government spending accompany these crises or not. Host countries may be big or small, some may be going through periods of growth and others through periods of recession. The legal framework and policies in place (right to work, freedom of movement) are different across countries and sometimes different within countries along space or time. Some studies focus on displaced populations hosted in camps and others on those outside camps, some of the displaced live in urban areas and others in rural areas. Some refugees move to countries with similar cultures, profiles and languages, others do not. Some of the FD episodes studied are massive in size while others are relatively small and the size relative to the host population can vary significantly across studies. Most inflows are sudden, but some are spread over a long period of time. Also, very few studies consider the role of international aid, which is a confounding factor to the displacement shock (Alix-Garcia and Saah, 2010). The comparative analysis of the empirical models used by this literature shows a certain homogeneity in the choice of identification and modelling strategy. Double difference and linear elasticity models are the dominant choice. The key independent variable (FD shock) is generally used in both its natural form and its instrumented version where variables such as geographical distances and (forced) migrants’ location prior to the shock (such as occupation) are used to construct the instrumental variable. Matching and placebo counterfactuals often support these choices. Cross-section econometrics is the predominant approach (largely dictated by the type of data available), few papers use time-series models and Computable General Equilibrium (CGE) models whereas panel data models are very rare. The main results of the meta-analysis can be summarized as follows. Between 45 and 52% of results show that household well-being for the hosts increases as a result of forced displacement.10 An additional 34- 42% of results are found to be non-significant and only in between 6 and 20% of results are found to be negative and significant. The review of studies on employment and wages shows that between 12 and 20% of results are positive and significant, around 63% are non-significant and between 22 and 25% are negative and significant. Results on prices show asymmetric behavior across types of products. Overall and across the four outcomes of interest, less than one in five results can be considered as negative for host communities. An assessment of these negative results shows that they mostly relate to young and informal workers in middle-income countries (MICs) whereas results on gender, skills and informality are inconclusive. The paper is organized as follows. The next section illustrates what basic economic theory would predict as an outcome of mass displacement in the consumer and labor markets. . Section 3 reviews the empirical models and identification strategies used to address the question of impact on the various outcomes considered. Section 4 provides a statistical overview of the literature covered and a meta-analysis of all results from all papers considered and Section 5 concludes. 2. Basic theory This section outlines in broad terms what standard economic theory would predict in terms of household well-being and labor market outcomes in the aftermath of a mass population displacement shock. This will set the benchmark against which the results of the review can be discussed. 10These boundaries are determined by estimations made with or without weights where weights are the journals’ impact factor. See section 6 for more details. 15 A forced displacement crisis typically results in two types of economic shocks. The first is a population shock with a sudden increase in population generated by an inflow in a particular geographical area. The second is a expenditure shock determined by the increased financial flows that a forced displacement crisis may attract, including aid from international donors and/or increased government spending on the part the host government. How these two components of the expenditure shock play out largely depends on the level of income per capita of host countries. In low income countries, displacement crises are typically accompanied by an almost simultaneous inflow of international aid. In middle income countries, international aid is usually accompanied by an increase in public spending on the part of the host government. And in high income countries international aid is typically absent whereas an increase in public spending would be the norm (social transfers to refugees or asylum seekers and subsidies to access education, health and other public services). In all these cases, we should think in terms of an expenditure shock channeled through an increase in welfare programs and the provision of public services targeting areas hosting refugees. We consider the population and expenditure shocks as quasi simultaneous short-term shocks. International aid and public welfare programs would generally be established later than the first inflow of refugees or displacement of IDPs but international aid can flow in within a few weeks, sometimes days, and the increased use of national public services on the part of forcibly displaced people is often immediate where services exist. However, the impacts of these shocks on the consumer and labor market are not necessarily immediate. While consumer demand responds promptly to these shocks, consumer supply may be slower to adapt. Similarly, while labor supply may increase rather quickly there are several constraints that may slow down labor demand adjustments. Refugees require some time to adapt to the new labor market opportunities, if any, international aid organizations and government services will take some time to be established, hire new local workers and have an impact on the local economy. Local firms will take additional time to react to the increased demand for goods and services and the increased labor supply by increasing production and hiring new workers and the degree of supply elasticity of goods and services varies. Local workers need time to reassess their situation and take decisions such as accept lower wages, drop out of the workforce or move out of the affected area. Population and expenditure shocks should also be expected to operate differently in the consumer and labor markets. A population shock results first in a shock to consumer demand and labor supply whereas a expenditure shock results first in a shock to consumer demand. In the absence of any other information regarding elasticities and the shape of the demand and supply curves, we take the classic textbook approach − described as () = − and () = + ⅆ respectively with equilibrium price ∗ = and + + equilibrium quantity ∗ = where D is demand, S is supply and p is price. The basic mechanics of the + shocks to the consumer and labor markets are described as follows. Shock to the consumer market. The first shock to the system occurs via an increase in consumer demand induced by savings, aid, and public spending (Figure 1, left-hand panel). Forced migrants usually carry a minimum amount of savings in kind or cash and these savings are typically spent on primary goods and services such as food, health services and shelter. Concomitantly, international aid or government spending boosts the spending capacity of the forcibly displaced via social transfers that are or can be monetized and via increased public spending that reduces living costs for the forcibly displaced.11 These factors are 11Support for refugees and IDPs usually take the form of cash, food vouchers, food in-kind, shelter, health and education services. Cash, food vouchers and food in-kind should be expected to have similar effects on consumer demand. Refugees are known to market food vouchers and when the vouchers are used to buy food, they tend to increase the demand for locally produced food just as cash would do. Humanitarian agencies tend to facilitate the availability of locally produced goods in stores that accept food vouchers and even when the food is delivered in-kind there is an effort to buy stocks from local producers. Moreover, humanitarian agencies have progressively shifted towards cash and food vouchers over the years as opposed to food in-kind. Free services such as health and education also increase the spending capacity of refugees by not diverting savings towards these expenditures. 16 expected to push the consumer demand towards the right with a subsequent increase in prices and consumption.12 In a second round, local producers are then expected to expand production encouraged by higher prices and cheap labor available, with a consequent increase in consumer supply.13 The net demand- supply effect is expected to result in higher prices and higher consumption for hosts. During this process, we should expect to have winners and losers with winners concentrated among net producers in rural areas and assets owners in urban areas and losers to be concentrated among manual labor in rural areas and net consumers in urban areas. Shock to the labor market. An influx of forcibly displaced would generally increase labor supply with this effect varying significantly depending on where the displaced are hosted (in camps or outside camps, urban or rural areas), on the host country legislation in relation to work status as well as on the socio-economic characteristics of the displaced (Figure 1, right-hand panel). Whether employment of the hosts decreases or not, this will depend on the degree of substitutability between local and displaced workers, on the opportunity wage available to low skilled locals, and on whether the influx ultimately results in outmigration of locals from these areas. Firms might also adapt their technologies to the increased labor supply and substitute capital for labor. As the returns to capital go up, investments will increase, and in the long term, labor-capital ratios can be expected to equalize. These elasticities are all largely unknown, but we can reasonably assume a displacement effect for some local workers, particularly workers with similar skills to the ones of immigrants, at least in the short- to medium-term. On the other hand, the increase in consumer demand generates a second-round effect of an increase in production which generates new employment opportunities for locals. The net effect of these two forces is hard to predict of course. Concomitantly, an influx of aid, an increase in government spending, and an increase in public services increases the demand for skilled and unskilled labor. Humanitarian agencies typically recruit local staff for registering refugees and IDPs, distributing food, setting up camps, driving vehicles and various other skilled and non-skilled activities. International aid workers also generate a demand for domestic unskilled labor. National agencies would also need to recruit more staff to scale-up programs. This shock shifts the labor demand curve outwards with a subsequent increase in wages, particularly for skilled labor, and an increase in employment (Figure 1, right-hand panel). As before, net effects are difficult to predict but we should expect to see winners and losers in this process with winners concentrated among high skilled formal workers and losers concentrated among low skilled informal workers. [Figure 1] Overall, the most important question is whether average household income for the host population increases or decreases. The growth of the consumer market and the arrival of aid and/or increase in government spending, and the subsequent growth of local production and employment drive household income upwards but the displacement effects and the decrease in employment and wages for some workers drive household income down. The net effect is difficult to predict and is likely to vary depending on the host country income per capita level and the substitutability of local workers with foreign workers. What is certain is that changes in relative prices and wages have distributional effects resulting in some low skilled/net consumer households to be worse off overall, at least in the short- to medium-term. Table 1 summarizes the predictions stated above. [Table 1] 3. Empirical modelling and identification strategies The purpose of this section is to provide some guidance to practitioners on the main models and identification strategies used by empirical economists. Annex 1 provides a detailed discussion of each 12 When subsidies are in place for certain products, prices for these goods would not increase but fiscal costs would. 13 Supply might be non-elastic, at least in the short-term, in very poor and isolated areas and notably for non-tradables, like housing. 17 model divided into a section on prices and consumption models and a section on wages and employment models. Table 2 provides a summary of the main equations used by each paper using comparable notations. As displacement crises are largely unpredictable, all the studies surveyed in this paper are evaluations conducted ex-post. In theory, a few of the crises studied could have been predicted but it would not be possible to allocate individuals to treated and non-treated groups randomly given that, by the definition of forced displacement we provided, people are fleeing violence, persecution or high levels of insecurity or uncertainty. Consequently, none of the papers reviewed is based on a Randomized Controlled Trial (RCT). Due to the randomness of the decision to leave (because of conflict, violence, insecurity or major political events) and/or the random allocation of displaced people in the country of destination (by policy or by default), some authors argue that they are in the presence of natural experiments. All authors do, however, address the question of endogeneity and, if one searches for a common thread, these evaluations would be better described as quasi-natural experiments. The basic model used by the literature is a model of the following form: = + + + Where i is the unit of observation, y is one of the four outcomes described, FD is the forced displacement shock and FE are fixed effects. Most papers with few exceptions use standard OLS estimators or some of its variants (Table 2). Two papers use general equilibrium models (Bodvarsson,Van den Berg, and Lewer 2008; Hercowitz and Yashiv 2002) and two papers simply compare means between treated and non-treated groups resulting in simple difference estimations (Card, 1990 and Alix-Garcia and Bartlett, 2015). [Table 2] The unit of observation varies depending on the data at hand. Most studies rely on household survey data where individuals or households are the unit of observations and most studies include some regional dimension (more frequently administrative areas). Where longitudinal or panel data are available time is also included. Other choices for unit of observations include skills or education level, various types of population groups (based on gender, age etc.), and, in a few cases, economic sectors, industry or labor market segments. The use of fixed effects varies. Some papers use the full set of parameters depicting units of observations (for example, household, region and time fixed effects in equations where the unit of observation is constructed using household, region and time). Other papers use subsets of these parameters whereas other papers introduce variables that are not used to identify the unit of observation. Very few papers provide explanations for these choices and there is no clear common approach to this choice. There are also only a handful of paper that discuss estimations of the error term and choices made in this regard. The two prevalent evaluation methods used by these studies are Differences-in Difference (DD) methods and linear elasticities models. In the first case, the variable of interest (FD) is a discrete status variable (generally a pre/post- treated/non-treated interaction term) and the coefficient of interest measures the impact on outcomes in the presence or absence of displaced people. In the second case, the model is typically in log form and is based on a shock variable that measures the intensity of the shock such as the number or share of refugees per geographical unit. In this case, the coefficient measures the elasticity of outcomes to the intensity of displacement. A few papers conduct simple differences illustrating results graphically or in tabular form. A few papers use ordinary matching methods (Alix-Garcia and Bartlett 2015, Aydemir and Kirdar 2018, Murard and Sakalli 2017; Mayda et al. 2017) and three papers use Synthetic Matching Methods (Peri and Yasenov 2017; Borjas 2017; Makela 2017). We could not find any paper using a discontinuity design.14 The essential ingredients used to measure the population shock are the number or presence of forcibly displaced persons, the size of the host population and the distance of the displaced from host communities 14 Schumann (2014) is an exception, but only looks at the impacts on municipality size. 18 if the displaced are clustered in camps or other forms of independent settlements. The literature covering high-income countries tends to focus on labor markets and the host population is often defined in terms of labor force whereas the literature covering middle and low-income countries often expands the work to household well-being and considers as host the entire population in a certain geographical area. Papers looking at labor market impacts either measure refugees or IDPs as a percentage of the population or labor force in a certain geographical area or as a percentage of the labor force in a certain education-experience group or both. The latter is used for the so-called skill-cell approach, which is prevalent in the economic migration literature and measures the impact of refugees or IDPs for specific population groups defined along education, skills or experience characteristics (see the recent reviews by Dustmann, Schönberg, and Stuhler 2016, and Özden and Wagner 2018). The outcome variables are usually measured on the sub-national level, but in a few cases nation-wide or across countries. They are measured across all sectors or, in three of the papers reviewed, for specific sectors of the economy (i.e. the construction sector in Portugal as in Carrington and de Lima 1996 and Makela 2017, or the retail sector in Miami as in Bodvarsson, Van den Berg and Lewer 2008). The authors aggregate results across all workers and types of employment or disaggregate them for specific groups of workers (based on their age, gender or experience and education level) and types of employment (formal or informal; as employee, employer, self-employed; full-time or part-time). Results either measure absolute effects or relative effects for certain groups compared to other groups. The studies also vary in terms of the time frame studied, with most of the studies looking at short- and, medium-term impacts and only few studies at long- term or dynamic impacts. The question of endogeneity is central to all papers irrespective of claims related to natural experiments and the main approach to address this issue is the instrumental variable approach. The choice of instruments varies across contexts. The distance from the shock, such as the distance to the border with the country of origin of the refugees (Ruiz and Vargas-Silva 2016) or the distance from the capital or the nearest larger city in the country of origin (Angrist and Kugler 2003) are popular choices. Fallah et al. (2018) instrument for the locality share of refugees based on the distance from the main refugee camp (Zaatari). Ruiz and Vargas-Silva (2015) measure the distance between each host community and 13 refugee camps and their population over time. Distance is often combined with (proxies for) outflow numbers. Rozo and Sviatschi (2018) use the inverse distance of each geographic unit to each of the three main refugee camps and the number of individuals fleeing Syria each year. Del Carpio and Wagner (2015) combine the distance to the different governorates in Syria with the number of registered refugees from these governorates in Turkey. Depetris-Chauvin and Santos (2018) use the weighed sum of IDP outflows from all municipalities (except the receiving host city), where the weights are the inverse of the road distance between the host city and each municipality of origin. Ibanez and Calderon-Mejia (2016) use the number of deaths due to civil violence in the previous year, weighted by the distance between the urban labor market and the site of the violence. IV modele using some form of distance to the border need to be cautious of potential correlations between distance to the shock and economic conditions. This is notably the case when border regions are very remote or are affected by the conflict in the neighboring country through a decline in trade and an increase in insecurity. Within a country, there might be spill-overs from violence in affected municipalities to municipalities nearby. The other frequent approach to instruments is the prior refugee or migration stock in the area, based on Altonji and Card (1991) and the idea that previous migrants attract new migrants (network effect). Borjas and Monras (2017), for example, instrument for the refugee shock with prior migration to that region. Rozo and Sviatschi (2018) use the settlements of Syrians in Jordan before the start of the war in Syria in 2011. Hunt (1992) uses the share of early (1954-1962) repatriates as a share of the 1962 population to instrument the 1962-1968 repatriates as a share of the labor force. Like in the case of distance, this is often combined with (proxies for) outflow numbers. Morales (2018) uses an instrument for inflows of IDPs in municipalities that combines outflows with immigrant stock. A common criticism of the migrant stock instrument is that the settlement of previous immigrants or refugees may be correlated with economic conditions across these 19 locations that may persist until today. To confront this criticism authors either use migrant stock data from a number of years before the influx they study or argue that the settlement of previous (forced) migrants was independent of economic conditions. Aydemir and Kirdar (2017), for example, use the share of earlier repatriates and show that the Turkish state took the decision where to settle them independent of economic conditions. Another criticism of this shift-share type of instrument is that if (forced) migrant inflows are stable over time, it conflates the short-run impacts of a new inflow with the long-run impacts of previous inflows (Jaeger, Ruist, and Stuhler 2018). The previous occupational distribution of the refugees in their country of origin (Friedberg 2001) or the occupational distribution of previous immigrants or refugees in the country of destination (Borjas and Monras 2017) is also used. Authors who prefer the latter argue that refugees might experience occupational downgrading upon arrival and their previous occupation might only be a weak instrument for their current occupation. Braun and Mahmoud (2014) combine previous occupational distribution and distance when they instrument the share of male expellees in the total male labor force in state-occupation cell exploiting regional variations in pre-war distribution of occupations and the distance of the expellees’ origin from West Germany. Hunt (1992) proposed the annual average temperature in each department in France, as repatriates from Algeria had a tendency to settle in areas in the South of France with higher annual average temperature. Sarvimaki (2011) uses the elements of the government’s placement policy as instruments (i.e. the proportion of a municipality’s population speaking Swedish and the hectares of potential agricultural land). Other authors focus instead on the counterfactual group testing alternative designs of the control group, sometimes including placebo groups and other times recurring to matching methods. The choice of matching methods varies from ordinary methods such as nearest neighbor to more recent advances such as Synthetic Control Methods (Abadie and Gardeazabal, 2003). The inclusion of fixed effects is common to almost all papers although the choice of fixed effects can be very different, as described above. Only one paper uses Fixed Effects (FE) and Random Effects (RE) formal models in conjunction and tests for differences (Esen and Binatli 2017). Cross-section econometrics is, by far, the method of choice even if time is included into the equations but we also found three papers employing time-series models (Carrington and de Lima 1996, Makela 2017, Fakih and Ibrahim 2015). Only few papers are able to exploit panel data (Foged and Peri 2015, Depetris-Chauvin and Santos 2017) and several of them use the same dataset (Maystadt and Duranton 2018, Maystadt and Verwimp 2014; Ruiz and Vargas-Silva 2015, 2016, 2017). Not all cross-sectional studies have multiple rounds of comparable data, covering the period before and after the crisis. When comparing impacts between locations within a country, cross-sectional data also usually does not allow to capture impacts on those who moved out and to differentiate impacts between those who were already there before the shock and those who moved in afterwards. Some of the models based on administrative areas qualify as spatial econometrics models in that they use estimation methods that derive from this literature and are published in spatial econometrics journals. Studies that compare different areas within a country are not only confronted with the potential endogeneity of the size and skill composition of the inflow and the choice of destination, but also with the endogenous reactions of the host community. Local workers might respond to the labor supply shock by dropping out of the labor force, investing in education, occupational upgrading or moving to other areas and diffusing the impact of the inflow. Even if local workers do not respond to wage variations, capital flows may equalize capital/labor ratios within the country, labor-intensive industries might move towards the regions with a high refugee or IDP influx or firms might use more labor-intensive production technologies. The reactions of the host country workers, investors and firms are medium-to long-term in nature and will play less of a role in the short-term if there are large, sudden and geographically concentrated inflows. Some of the papers explicitly analyze these potential channels, notably migration of local workers, and, to a lesser extent, occupational upgrading. Outmigration of hosts is a critical complement to the labor market analysis and excluding this outcome can lead to an underestimation of the impacts of forced displacement on the labor market outcomes of natives. The papers we reviewed that looked at tasks complexities and the question of 20 substitution vs complementarities between refugees and natives found occupational upgrading among natives as a result of the refugee inflow (Akgündüz, van den Berg, and Hassink 2018, Akgündüz and Torun 2018, and Foged and Peri 2015). Measurement challenges also arise due to the phenomenon of skill downgrading (Ozden and Wagner 2018). Refugees are often not able to find jobs that correspond to their education level and previous work experience. This has an impact on the degree of substitution between refugees and natives with the same observable education and work experience. Papers using the skill-cell approach face these measurement challenges, while papers which look at the impacts of refugees or IDPs across all skills and experience levels do not have to address this issue. There are two important questions related to endogeneity and spurious correlation that have been raised and addressed in two separate papers but are relevant for and have been largely ignored by the rest of the literature. The first question related to endogeneity was raised by Borjas and Monras (2017). The displacement shock has an impact on local wages and this affects native labor supply at the intensive margin (by affecting the amount of labor that working natives provide) and at the extensive margin (by affecting the number of natives who participate in the labor market). In order to address this issue, one has to consider a labor supply model that is able to measure both effects separately whereas most papers confound these two effects into one. Foged and Peri (2015) is one of the exceptions, as their paper looks at the intensive margin (fraction of year worked). Rozo and Sviastchi (2018) include the number of hours worked, and Ruiz and Vargas-Silva (2017) look at the changes in number of hours dedicated to a task (including employment outside the household). The second question relates to possible spurious correlations generated by how variables are combined in models. Linear models that use ratios of two variables as dependent variable (think of average prices or wages, employment rates or consumption per capita) and the denominator of this ratio as independent variables (think of the share of refugees on host communities or household size) can produce spurious correlations (Kronmal 1993). This is noted and addressed in Clemens and Hunt (2017) who show how addressing this issue change results for several studies in the literature covered here. Indeed, almost all models reviewed use the same population or household size on both sides of the equations. Finally, the expenditure shock which we discussed in the theory section (international aid or an increase in public spending associated with the forced displacement crisis) is considered by only a handful of papers. This is a possible confounding factor of the impact of forced displacement on host communities and one that is not easily addressed with the use of fixed effects. This is clearly a shortcoming of this literature that will require increased attention in the future. 4. Meta-analysis of empirical results 4.1 Data The literature review covers 49 papers spanning over a period of 29 years. We were not able to find published papers prior to the work by Card in 1990, which effectively started this literature, and there is a relatively low interest in this topic between 1990 and 2011 with only one or two papers published per year. With the Syrian crisis starting in 2011 and the EU crisis in 2015 the number of papers per year increased by several folds. Most of the papers and results considered in this review are therefore very recent (Figure 2). We used academic databases and search engines (EconLit, Social Science Research Network, JSTOR, Google Scholar) and searched websites of institutions with relevant working paper series (NBER, IZA, ERF and others). Relevant unpublished papers were included by searching agendas of workshops and conferences organized during the past few years. [Figure 2] From the papers reviewed, we selected a total of 762 results summarized in Table 3. The results database was compiled as follows. For each paper we focused on the results that the authors considered the main and 21 most reliable findings.15 For the same paper, results are considered different if the dependent, the key independent variable or the population group considered change. For each of these variations, we include two results, a minimum and a maximum value, derived from variations in estimators, set of regressors or modalities for the estimation of the standard error.16 The sample is therefore unbalanced with respect to papers and authors. Employment is the outcome most studied followed by wages, prices and well-being in this order. Considering that well-being is the only indicator that captures the overall impact on host households, the relatively low number of outcomes is clearly a shortcoming of this literature. Overall, there is a good spread of results across papers and outcomes. Most authors consider more than one outcome and all outcomes are covered by a significant number of authors. [Table 3] Table 4 provides the number of papers and results by outcome as well as the average results per paper and the Gini index. The latter is calculated using number of papers (as “individuals”) and number of results (as “income”). In this case, a lower Gini is a good property as it indicates more spread of results across papers. From Table 2, we can see that employment and wages have a considerable number of papers and results but not necessarily a good Gini whereas prices have much fewer papers and results but a good Gini. Well- being is the outcome with the least number of papers and results and a comparable Gini to employment and wages. Overall, all four outcomes have a sufficient number of observations and a fair diversity across papers and authors to allow for a meta-analysis. [Table 4] The literature covered includes 17 displacement crises well distributed across high, medium and low- income hosting countries (Table 5). There is a good coverage of all three groups of countries and there is a good coverage of most crises with a few exceptions. The single crisis that dominates the literature is the Syrian crisis with over a third of all results covering this crisis alone. Other well studied crises are Burundian and Rwandan refugees in Tanzania, Cuban refugees in Miami, Former Soviet Union (FSU) escapees to Israel and IDPs in Colombia. One case (refugees in Denmark) has many results but they derive from a single paper (Foged and Peri, 2015). The incidence of the number of refugees on the host population varies across crisis and averages around 14.6%.17 All crises show a significant incidence of displacement versus the local population reaching 50% in the case of Burundian and Rwandan refugees in the Kagera region of Tanzania. There is also a certain variability in the time gap between the beginning of the crisis and the year when the impacts are measured by the studies. This is 13.5 years on average with a range between 1 and 58 years. [Table 5] The coverage in terms of journals is of high quality (Table 6). The average recursive impact factor for the last ten years is around 1 and journals include top journals such as the Quarterly Journal of Economics, the Journal of Political Economy, the Review of Economics and Statistics, the Journal of Labor Economics, the Journal of International Economics, the AEA: Applied Economics, the Journal of Development Economics, the Journal of Economic Geography and the World Bank Economic Review.18 The number of 15 When OLS and IV estimations are reported, for example, IV estimations are almost invariably preferred by the authors. Robustness checks are excluded from the list of results. 16 Only in a few cases, we considered a change in estimator a separate result. That is when the two estimators convey clearly different information. 17 The incidence of refugees and IDPs is estimated based on the peak stock value of refugees or IDPs divided by the host population, which can be a country or a smaller geographical area affected by refugees or IDPs. These data are mostly provided by the papers that cover these crises. 18 The recursive impact factor for the last ten years is taken from the IDEAS/Repec repository as for September 10, 2018. The same listing includes journals and working papers. Working papers not included in the list were attributed an impact factor of 0.01. 22 papers is well distributed across journals. The average number of papers per journal is 1.9. Industrial and Labor Relations Review, the review that published the first paper by Card, is the journal with more papers with four articles followed by the Journal of Development Economics and the Journal of Economic Geography with three articles each. [Table 6] 4.2 Results This section discusses the overall results of the four outcomes reviewed by the study using the database of 762 results. Table 7 provides the full results for the four outcomes classified into positive, non-significant and negative values where positive and negative values are intended as significant. Figure 3 provides an overview of the share of these results by outcome. Table 8 provides the results of the multivariate logit regressions where the dependent variables are dummies for negative and positive significant values and the independent variables are population characteristics. Table 9 provides the same results for the bivariate regressions with the addition of a number of dummies including: the time lag between shock and evaluation, the average share of refugees over the host population for the period considered, dummies for HICs, MICs and LICs and each of the crisis studied.19 All regressions results in Tables 8 and 9 include weighted and unweighted results where weights are the journals’ impact factors. Results in Tables 8 and 9 are presented for employment and wages only. That is because these are the only two outcomes with sufficient number of observations to run the regressions as the literature rarely provides results by population groups for well- being and prices. Below we comment all results by outcome. For readers interested in selected forced displacement crises, Annex 2 provides a full discussion of results by crisis. The overview of results organized by crisis and income level of the host countries is useful in as far as this is the most defining feature that could explain differences across results. In particular, the level of economic development of host countries, the absolute and relative scale of the crises and the timing of the inflow are very important factors in determining outcomes. 4.2.1 Well-being and Prices Well-being. Among all the papers reviewed, 13 papers explicitly measure the impact of displacement on the economic well-being of host communities for a total of 64 distinct results. Of these, 34 are on income, consumption or output, 20 are on housing or assets, 6 are on night luminosity and 4 are on poverty. Only two of these results are on HICs, 26 are on MICs and 36 on LICs. In all these cases, a positive result is considered a good outcome. 20 Table 7 shows that between 45 and 52% of results are positive and significant depending on whether results are weighted for the journals’ impact factor or not. This ind icates a net improvement in household well- being according to about half of results. An additional, 34-42% of results are non-significant and the remaining 6-20% of results are negative and significant. Looking more in details at the 13 negative results, these are equally split between income and consumption and assets and housing indicators. Results on assets and housing indicators refer to individual items such as construction materials of dwellings and are therefore less representative of household well-being as compared to aggregate income or expenditure. Overall, between 80 and 94% of results are either positive or non-significant indicating that the likelihood of a forced displacement shock resulting in a negative outcome for host communities is less than 20%. The basic theory illustrated at the outset of the paper could not predict these outcomes whereas such outcomes 19These dummies could not be used in the multivariate analysis because of collinearity with other variables. 20Note that when poverty was used as an indicator of well-being, the sign of the coefficient was reversed to make it consistent with the other indicators of well-being where a positive sign indicates an improvement in well-being. 23 are clearly in contrast with the popular view that forced displacement is detrimental to host communities. These results are encouraging but much more research on household well-being is needed. Future research will need to provide more hard evidence based on income, consumption or expenditure indicators and expand its coverage to many more countries and situations and possibly disaggregating results by type of households. Prices. The database includes 120 results on prices where prices refer to all types of prices from food and non-food items, to rents and services. Results in Table 7 show that 37-46% of results are positive and significant, between 18 and 23% are non-significant and between 36 and 39% are negative and significant. As mentioned, there is no good or bad interpretation of these results as changes in prices benefit some hosts and damage others. An analysis of the positive results shows that these are almost entirely represented by food and rents prices whereas an analysis of the negative values shows that these are represented mostly by overall inflation, luxury goods, services, and labor-intensive products. Overall, a forced displacement crisis generates significant changes in prices. The probability of these changes to be positive or negative is approximately similar but associated to the type of product or service observed. Prices for food and rents tend to increase whereas prices for labor intensive products and services tend to decline. As compared to the basic theory illustrated at the outset of the paper, these results show that prices do not unequivocally increase. The main surprise and in contrast to popular views, in some cases prices decline. These findings remain working hypotheses based on very preliminary data and future research will need to be much more comprehensive in coverage of different products and countries. It will also be essential to expand research on price elasticities, including cross elasticities, and relate this research with the research on household well-being. 4.2.2 Employment and Wages Employment. The database on employment contains 334 observations where employment can be a rate, a status or a probability. In all these cases a positive value indicates that employment among host communities has improved as a result of forced displacement. Results in Table 7 show that between 12 and 14% of results are positive and significant, between 65 and 67% are non-significant and between 19 and 24% are negative and significant. Therefore, the majority of results is non-significant and this reaches 76- 81% of all results if we add positive and significant results. However, negative results outweigh positive results. To see whether we could trace results to some specific population characteristics, we constructed dummy variables for positive and negative results and run multivariate logit regressions using these dummies as dependent variables. Predictors are dummies representing population groups including males and females, formal and informal, young and old, and low-skilled and high-skilled workers. Table 8 shows the results.21 Only two results are consistent across weighted and unweighted regressions. One is that males show a negative association with negative results indicating that this group is not well represented among negative results as compared to its representation among positive or non-significant results.22 However, since females also show negative signs, we cannot conclude that gender helps to explain positive or negative results. The second is that older workers are associated with positive coefficients indicating that this group is highly represented among positive and significant results. The variable young is dropped because there are no observations among positive coefficients. Unlike gender, this indicates that age is a critical factor in explaining positive results with younger workers clearly disadvantaged vis-à-vis older workers. As a robustness test, we run the same regressions proposed above in a bivariate mode also adding the time lag between shock and evaluation, the average share of refugees, the countries’ classification in HICs, MICs and LICs, and the 14 crises considered. Results are shown in Table 9 where, for simplicity, we only report 21 These results should be treated with caution as the number of results for population sub-groups is small. 22 Note that zero values in the dependent variable represent positive and non-significant values. 24 t-statistics with values significant at 90% level or above highlighted in bold. As before, we only comment results that are robust to the weighted and unweighted specifications. Males, young and informal workers have a positive association with negative values whereas old, formal and low-skilled workers have a higher association with positive values (except for the weighted t-stat for old which has a negative value). As compared to Table 8, results for males in Table 9 are not consistent with the sign changing between the two tables. This is explained by the existing collinearity between results on males and females (p-corr. coefficient of -0.12 significant at the 1% level).23 This confirms that results on gender are not robust. However, results on young and old and formal and informal workers are consistent in showing younger and informal workers being disadvantaged as compared to older and formal workers. It is also relevant to observe that low-skilled workers have a negative sign against negative results and a positive sign against positive results. This is a rather strong and counterintuitive indication that low-skilled workers do not perform poorly. Results on high skilled workers are instead non-conclusive. Also important to note is the fact that many crises seem associated with either positive or negative results indicating the importance of crises specificities (and the weight of individual crises on the overall results) in understanding these findings. Overall, the likelihood of a forced displacement crisis to have a negative impact on employment for the host population is below 25%. Negative and significant results are associated with young and informal workers. Studies on gender remain scarce and results are inconclusive. Employment is the most researched of the four outcomes considered by this paper but results on population sub-groups remain scarce and future research will need to address this gap. Basic theory could not predict these outcomes whereas popular believes expect the employment effects to be mostly negative, a belief clearly discredited by these results. Even if forced displacement contributes to decrease overall employment only in a minority of cases, the number of workers affected may be large and concentrated among few groups. From a policy perspective, it is essential to better understand the distributional effects on employment, a question that this review could not address because of lack of data. Wages. The database contains 244 observations on wages where wages can be expressed in different forms in terms of time unit, they can be gross or net, or can be defined as earnings. In all these cases, results with positive signs indicate an improvement in wages. Table 7 provides the number of results divided into positive, non-significant and negative. Between 13 and 22% of results are positive and significant, between 55 and 59% are non-significant and between 22 and 27% are negative and significant. Results of the multivariate logit regressions on wages (Table 8) show only one result that is robust to weighted and unweighted estimates. This is females showing a positive association with positive coefficients. Essentially, females would seem better represented among positive coefficients than among negative and non-significant results taken together. However, results on the bivariate regressions (Table 9) show opposite results with females showing a positive association with negative values. Table 9 also shows males having a negative association with positive values. Therefore, gender results in the multivariate setting are spurious due to the collinearity already explained whereas results from the bivariate regressions are not conclusive on gender differences with neither males of females dominating the other gender. The only clear results that emerges from Table 9 are results on the time lag between initial shock and outcome and results on the difference between HICs on the one side and MICs on the other side. The time lag between the shock and the period evaluated seems to be an important explanatory factor. The longer the time gap, the lower is the association with negative values and the higher is the association with positive values. This would indicate that negative results tend to vanish in the long-term. Also, HICs have a negative and consistent sign against negative values and a positive and consistent sign against positive values and vice-versa for MICs. Therefore, HICs clearly outperform MICs. It is also clear from Table 9 that selected 23 There are no other significant correlations among all regressors in Table 7. 25 crises are associated with negative or positive values indicating that results can be very crisis specific (see Annex 2 for a discussion of the results by crisis). The overall evidence for wages shows that only about one in four results are negative and negative results seem to turn into positive in the long-term. Wages for hosts in HICs perform well as compared to wages for hosts in MICs. This may be evidence of skills’ complementarities in HICs and skills substitutability in MICs. As for employment, research on population sub-groups remains largely insufficient to derive firm conclusions on the role of individual characteristics. As for employment, results on wages could not be predicted by basic theory and are in clear contrast with popular beliefs, 5. Conclusion The paper reviewed 49 empirical studies that focused on estimating the impact of forced displacement on host communities. This literature covers 17 different displacement situations in high, medium and low- income countries covering the impact on the labor and consumer markets. A total of 762 results have been used for the meta-analysis. To our knowledge, this is the first comprehensive review of this literature. The empirical modelling analysis highlighted the main traits of this literature. By definition, all studies operate ex-post, after the displacement crisis has taken place. The unexpected nature of the crisis and the randomness of the allocation of displaced persons are two elements used to defend the natural experiment assumption. However, all papers address the central question of endogeneity. The instrumental variable approach is the dominant method to address endogeneity issues and instruments tend to focus on either distance from the shock or previous location of migrants. Double difference and linear elasticity models are the dominant choice of estimation models with matching and placebo counterfactuals often supporting these choices. Cross-section econometrics is the predominant approach (mostly dictated by the type of data available), few papers use time-series models whereas panel data models are the exception. Most papers are set in a partial equilibrium framework, but a few papers use Computable General Equilibrium (CGE) models. The meta-analysis of empirical results shows that negative and significant outcomes are less than 1 in 5 overall. Between 45 and 52% of results show that household well-being for the hosts increases as a result of forced displacement. An additional 34-42% of results are found to be non-significant and only in between 6 and 20% of results are found to be negative and significant. The review of studies on employment and wages shows that, when taken together, between 12 and 20% of results are positive and significant, around 63% are non-significant and only between 22 and 25% are negative and significant. When we zoomed in on negative results on employment and wages, we found that these related mostly to young and informal workers in MICs whereas results on gender and skills remained inconclusive, possibly because our limited number of observations for these population characteristics. Results on prices show asymmetric behavior across types of products. As compared to the basic predictions made in the theory section results largely reflect predictions for young and informal workers. No predictions were made for the net effect on household well-being, employment and wages and the results provided above represent the first evidence derived from a global analysis. The review also identified some recurrent findings once we consider small and large crises, poor and rich host countries, low and high skilled persons and short and long-term effects. Studies on larger crises in richer countries with more developed labor markets provided evidence that low-skilled informal workers are the main losers during a displacement crisis. In line with the literature on economic migration, studies that compared short and long-term effects found that negative effects tend to disappear in the long-run. Studies on the overall well-being of hosts in the aftermath of a displacement crisis in poor countries show rather consistent positive results also explained by the inflow of international aid and its relative importance in a poor country. These are preliminary but consistent findings that emerged from the review. 26 Some selected evidence on policies can also be derived from the review by crisis provided in Annex 2, even if few studies analyze the mechanisms through which forced displacement impacts labor and consumer markets. Prices increase because supply might be non-elastic, at least in the short-term, in poor and isolated areas and for non-tradables, like housing. To increase the price elasticity of supply for food items in poor and isolated areas, investments by the government, donors and humanitarians can help connect these places to markets. The improved road network seems to have a positive impact on household welfare even after the forced migrants return (Maystadt and Verwimp 2018). An improved business and investment climate will also speed up the reaction of the private sector to an increase in demand. An increase in the issuance of construction permits, notably for social housing, can help buffer effects on the housing market, at least in the medium term. If construction permits for high-income housing crowd out construction permits for social housing instead, the negative incomes on lower income hosts are reinforced (Depetris-Chauvin and Santos 2018). There is some evidence that negative impacts on employment of hosts might be stronger in countries with more rigid labor markets (Angrist and Kugeler 2003). Restrictions on the right to work usually mean that refugees of all skills are limited to compete with low-skilled workers in the informal sector, potentially increasing negative impacts on already vulnerable groups (as the studies on Turkey and Jordan show). Allowing refugees to work will disperse the impacts across different sectors and skill levels. As two papers on Turkey exemplify, enterprises created by refugees themselves can contribute to these efforts, if policies and regulations allow them to (Akgündüz, van den Berg, and Hassink 2018; Altindag, Bakis, and Rozo, 2018). The papers we reviewed that looked at tasks complexities and the question of substitution versus complementarities between refugees and natives found occupational upgrading among natives as a result of the refugee inflow (Akgündüz, van den Berg, and Hassink 2018, Akgündüz and Torun 2018, and Foged and Peri 2015). Policies can reinforce these complementarities between forced migrants and native workers and increase the productivity of native workers by providing incentives to upgrade their skills. As a number of studies showed, internal migration helps dissolve some of the impacts on the labor market and could be incentivized by policy makers. Capital flows can help re-equalize capital/labor ratios within the country, if allowed to do so. In general, policies are needed to counterbalance the distributional impacts of a forced displacement inflow on the labor and consumer markets. Despite recent research efforts and the findings described, research in this area remains in its infancy. Studies have focused on selected markets, first round, short and medium-term effects, selected methodologies and selected displacement crises. Very little work is available on second-round and long- term and dynamic effects, on the production side of the economy and on the impact of forced displacement on primary services such as water, electricity, education or health. Results on household well-being, which should be the most important outcome to study, remain few. Panel data, which are the most promising type of data for this type of analysis, covered only a few segments of a few crises. We could not find evaluations that used regression discontinuity designs even if forced displacement crises can potentially lend themselves to this type of instrument. Some crises, such as the Rohingya or the Venezuelan crises, have not been covered by the literature because they may be too recent but other major displacement crises, such as the repeated crises in the Democratic Republic of Congo (DRC), the Central African Republic, Afghanistan, Pakistan or Iraq, have been largely ignored by the economics profession. These are all areas that will require greater research efforts to complement the existing literature. Results are also derived from a multitude of models and case-studies and their comparability remain a challenge. We could not provide, for example, reliable summary figures on the size of the measured effects forcibly limiting our analysis to the significance level of the econometric estimates. More research is also needed to help us understand the channels through which the influx of forced migrants determines impacts on outcomes and whether policies have had any role in this process, notably policies regarding the access to the labor market or the mobility of forced migrants as well as the general business and investment climate. None of the studies covered by this review explicitly measured the effects of policy 27 changes on host communities. Policies affect outcomes and the different policies administered cross- country represent a confounding factor when results are pulled together and compared. Equally important is to have a much better understanding on the process of local integration of displaced persons among host communities in the medium and long-term to better understand when displaced persons can stop to be considered as displaced and are finally counted as integral part of the population. The level of their economic and social integration will also influence their impacts on the host community and change it over time. While our findings show that negative impacts on host communities tend to decline in the long-term, the studies reviewed in this paper did not really delve into this issue. Only few papers studied changes in migration of locals into and out of the area affected by the forced migration inflow as one important adaptation mechanism and even fewer looked at skills-upgrading among hosts, changes in production technologies of firms, and new investments. Finally, there can be noticeable differences between the measured impacts on host communities and perceptions of these impacts. The empirical evidence on the impact of forced displacement on host communities that we discussed in this paper is clearly at odds with the public discourse. To our knowledge, only one study looked at these differences (Kreibaum, 2015) and found this difference to be sizable. Subjective well-being can be a powerful driver for change and understanding its relation with objective well-being is key from a policy perspective. New data collection and research efforts should take this aspect into account. 28 References Akgündüz, Yusuf Emre, and Huzeyfe Torun. 2018. “Two and a half million Syrian refugees, skill mix and capital intensity.” GLO Discussion Paper, No. 186. Akgündüz, Yusuf Emre, Marcel van den Berg, and Wolter Hassink. 2015. “The Impact of Refugee Crises on Host Labor Markets: The Case of the Syrian Refugee Crisis in Turkey.” IZA Institute of Labor Economics Discussion Paper No. 8841. Akgündüz, Yusuf Emre, Marcel van den Berg, and Wolter Hassink. 2018. “The Impact of the Syrian Refugee Crisis on Firm Entry and Performance in Turkey.” World Bank Economic Review 32: 19-40. Al Hawarin, Ibrahim, Ragui Assaad, and Ahmed Elsayed. 2018. Migration Shocks and Housing: Evidence from the Syrian Refugee Crisis in Jordan.” Economic Research Forum Working Paper No. 1213. Alix-Garcia, Jennifer, and Anne Bartlett. 2012. “Occupations under fire: the labor market in a complex emergency.” Oxford Economic Papers 67(3): 687–714. Alix-Garcia, Jennifer, and David Saah. 2010. “The Effect of Refugee Inflows on Host Communities: Evidence from Tanzania.” World Bank Economic Review 24 (1): 148–70. Alix-Garcia, Jennifer, Anne Bartlett and David Saah. 2012. "Displaced Populations, Humanitarian Assistance and Hosts: A Framework for Analyzing Impacts on Semi-urban Households." World Development 40 (2): 373–86. Alix-Garcia, Jennifer, Sarah Walker, Anne Bartlett, Harun Onder, and Apurva Sanghi. 2018. “Do refugee camps help or hurt hosts? The case of Kakuma, Kenya.” Journal of Development Economics 130 (2018) 66–83. Altındag, Onur, Ozan Bakisy, and Sandra Rozo. 2018. Blessing or Burden? The Impact of Refugees on Businesses and the Informal Economy.” https://www.sandravrozo.com/uploads/2/9/3/0/29306259/abr_june2018.pdf Altonji, Joseph G., and David Card. 1991. “The Effects of Immigration on the Labor Market Outcomes of Less Skilled Natives.” In, Immigration, Trade, and the Labor Market, edited by John M. Abowd and Richard B. Freeman, 201-34. Chicago: University of Chicago Press. Amuedo-Dorantes, Catalina, Cynthia Bansak, and Susan Pozo. 2018. Refugee Admissions and Public Safety: Are Refugee Settlement Areas More Prone to Crime? IZA Institute of Labor Economics Discussion Paper No. 11612. Anastasopoulos, Jason, George J. Borjas, Gavin G. Cook, and Michael Lachanski. 2018. Job Vacancies and Immigration: Evidence from Pre- and Post-Mariel Miami. NBER Working Paper Series 24580. Angrist, Joshua D. and Adriana D. Kugler. 2003. “Protective or Counter-Productive? Labor Market Institutions and the Effect of Immigration on EU Natives .” The Economic Journal 113(June), F302-F331. Angrist, Joshua D., and Alan B. Krueger. 1999. “Empirical Strategies in Labor Economics.” In Handbook of Labor Economics, Vol 3A, edited by Orley Ashenfelter and David Card, 1278-1357. Asali, Muhammad. 2013. “The Effect of Immigration on Unskilled Native Workers: Evidence from a Natural Experiment.” Southern Economic Journal 80(2): 345-65. Assaad, Ragui, Thomas Ginn, and Mohamed Saleh. 2018. “Impact of Syrian Refugees in Jordan on Education Outcomes for Jordanian Youth.” Economic Research Forum Working Paper No. 1214. 29 Aydemir, Abdurrahman, and Murat Kirdar. 2017. “Quasi-Experimental Impact Estimates of Immigrant Labor Supply Shocks: The Role of Treatment and Comparison Group Matching and Relative Skill Composition.” European Economic Review 98: 282-315. Baez, Javier E. 2011. "Civil Wars beyond their Borders: The Human Capital and Health Consequences of Hosting Refugees." Journal of Development Economics 96(2): 391-408. Balkan, Binnur, and Semih Tumen. 2016. “Immigration and prices: quasi-experimental evidence from Syrian refugees in Turkey.” Journal of Population Economics 29:657–86. Balkan, Binnur, Elif Ozcan Tok, Huzeyfe Torun, and Semih Tumen. 2018. “Immigration, Housing Rents, and Residential Segregation: Evidence form Syrian Refugees in Turkey.” IZA Institute of Labor Economics Discussion Paper Series No. 11611. Bodvarsson, Örn B., Hendrik F Van Den Berg, and Joshua J. Lewer. 2008. “Measuring Immigration's effect on Labor Demand: A Reexamination of the Mariel Boatlift.” Labor Economics 15(4): 560–74. Borjas, George J. 2017. “The Wage Impact of the Marielitos: A Reappraisal.” Industrial and Labor Relations Review 70(5): 1077–110. Borjas, George J., and Joan Monras. 2017. “The Labor Market Consequences of Refugee Supply Shocks.” Economic Policy (July), 361–413. Braun, Sebastian Till, Anica Kramer, and Michael Kvasnicka. 2017. “Local Labor Markets and the Persistence of Population Shocks.” IZA Discussion Paper Series No. 11077. Braun, Sebastian, and Henning Weber. 2016. “How do regional labor markets adjust to immigration? A dynamic analysis for post-war Germany.” Deutsche Bundesbank Discussion Paper No 05/2016. Braun, Sebastian, and Michael Kvasnicka. 2014. “Immigration and structural change: Evidence from post- war Germany.” Journal of International Economics 93: 253–69. Braun, Sebastian, and Toman Omar Mahmoud. 2014. “The employment effects of immigrati on: Evidence from the mass arrival of German expellees in post-war Germany.” The Journal of Economic History 74 (1), 69-108. Calderón-Mejía, Valentina, and Ana María Ibáñez. 2016. “Labour market effects of migration -related supply shocks: evidence from internal refugees in Colombia.” Journal of Economic Geography 16: 695– 713. Card, David. 1990. "The Impact of the Mariel Boatlift on the Miami Labor Market." Industrial and Labor Relations Review 43 (2): 245-57. Carrington, William J., and Pedro J. F. de Lima. 1996. "The Impact of 1970s Repatriates from Africa on the Portuguese Labor Market." Industrial and Labor Relations Review 49(2): 330–47. Ceriani, Lidia and Paolo Verme. 2018. “Risk preferences and the decision to flee conflict.” The World Bank Policy Research Working Paper Series No 8376. Ceritoglu, Evren, H. Burcu Gurcihan Yunculer, Huzeyfe Torun, and Semih Tumen. 2017. “The Impact of Syrian refugees on natives' labor market outcomes in Turkey: Evidence from a quasi experimental design.” IZA Journal of Labor Policy 6(5): 1-28. Clemens, Michael A., and Jennifer Hunt. 2016. “The Labor Market Effects of Refugee Waves: Reconciling Conflicting Results.” NBER Working Paper Series 23433. 30 Cohen-Goldner, Sarit and Daniele Paserman. 2011. “The dynamic impact of immigration on natives' labor market outcomes: Evidence from Israel.” European Economic Review 55 (8): 1027-45. del Carpio, Ximena, and Mathis Wagner. 2015. “The Impact of Syrian Refugees on the Turkish Labor Market.” Policy Research Working Paper WPS 7402. Depetris-Chauvin, Emilio and Rafael J. Santos. 2017. “The Impacts of Internal Displacement Inflows on Host Communities in Colombia.” KNOMAD Working Paper 27. Depetris-Chauvin, Emilio and Rafael J. Santos. 2018. “Unexpected guests: The impact of internal displacement inflows on rental prices in Colombian host cities.” Journal of Development Economics 134, 289-309. Depetris-Chauvin, Emilio, and Rafael J. Santos. 2018. Followed by Violence: Forced Immigration and Homicides. Instituto de Economia Pontificia Universidad Catolica de Chile Documento de Trabajo 513. Dustmann, Christian, Uta Schönberg, and Jan Stuhler. 2016. “The Impact of Immigration: Why Do Studies Reach Such Different Results?” Journal of Economic Perspectives 30(4): 31-56. Dustmann, Christian, Uta Schönberg, and Jan Stuhler. 2017. “Labor Supply Shocks, Native Wages, and the Adjustment of Local Employment.” The Quarterly Journal of Economics (2017): 435–83. Eckstein, Zvi, and Yoram Weiss. 2004. On the Wage Growth of Immigrants: Israel, 1990-2000.” Journal of the European Economic Association 2(4): 665-95. El-Mallakh, Nelly and Jackline Wahba. 2018. Syrian Refugees and the Migration Dynamics of Jordanians: Moving in or moving out? Economic Research Forum Working Paper No. 1191. Esen, Oguz, and Ayla Ogus Binatlı. 2017. “The Impact of Syrian Refugees on the Turkish Economy: Regional Labour Market Effects.” Social Sciences 6(4): 1-12. Fakih, Ali, and May Ibrahim. 2015. “The Impact of the Syrian refugees on the Labor Market in Neighboring countries: Empirical Evidence from Jordan.” Defence and Peace Economics: 1-23. Fallah, Belal, Caroline Krafft, and Jackline Wahba. 2018. “The Impact of Refugees on Employment and Wages in Jordan.” Economic Research Forum Working Paper No. 1189. Foged, Mette and Giovanni Peri. 2015. “Immigrants' Effect on Native Workers: New Analysis on Longitudinal Data.” American Economic Journal: Applied Economics 8(2): 1–34. Fransen, Sonja , Carlos Vargas-Silva, and Melissa Siegel. 2018. ”The impact of refugee experiences on education: evidence from Burundi.” IZA Journal of Development and Migration 8(6). Friedberg, Rachel M. 2001. “The impact of mass migration on the Israeli labor market.” Quarterly Journal of Economics (November): 1373-1408. Ghosh, Sucharita, and Ali Enamib. 2015. “Do refugee-immigrants affect international trade? Evidence from the world’s largest refugee case.” Journal of Policy Modeling 37: 291–307. Gimenez-Nadal, Jose Ignacio, José Alberto Molina, and Edgar Silva-Quintero. 2018. “On the Relationship between Violent Conflict and Wages in Colombia.” The Journal of Development Studies. Glitz, Albrecht. 2012. “The Labor Market Impact of Immigration: A Quasi-Experiment Exploiting Immigrant Location Rules in Germany.” Journal of Labor Economics 30(1): 175-213. Hercowitz, Zvi, and Eran Yashiv. 2002. “A Macroeconomic Experiment in Mass Immigration.” IZA Institute of Labor Economics Discussion Paper No. 475. 31 Hornung, Erik. 2014. “Immigration and the Diffusion of Technology: The Huguenot Diaspora in Prussia.” American Economic Review 104(1): 84–122. Hunt, Jennifer. 1992. “The impact of the 1962 repatriates from Algeria on the French labor market.” Industrial and Labor Relations Review 45(3): 556-72. Jaeger, David A., Joakim Ruist, and Jan Stuhler. 2018. “Shift-Share Instruments and the Impact of Immigration.” IZA Institute of Labor Economics Discussion Paper No. 11307. Kondylis, Florence, and Valerie Mueller. 2014. “Economic consequences of conflict and environmental displacement.” In International Handbook on Migration and Economic Development, edited by Robert E.B. Lucas, 388-424. Kreibaum, Merle. 2016. “Their Suffering, Our Burden? How Congolese Refugees Affect the Uga ndan Population.” World Development 78: 262–87. Lach, Saul. 2007. “Immigration and Prices.” Journal of Political Economy 115(4): 548-87. Lewis, Ethan. 2004. “How Did the Miami Labor Market Absorb the Mariel Immigrants?” Working Papers 04-3 Federal Reserve Bank of Philadelphia. Mabiso, Athur, Jean-François Maystadt, Joachim Vandercasteelen, and Kalle Hirvonen. 2014. “Refugees, food security, and resilience in host communities: Transitioning from humanitarian assistance to development in protracted refugee situations.” International Food Policy Research Institute 2020 Conference paper 2. Makela, Erik. 2017. “The effect of mass influx on labor markets: Portuguese 1974 evidence revisited.” European Economic Review 98: 240-63. Malaeb, Bilal, and Jackline Wahba. 2018. “Impact of refugees on immigrants’ labor market outcomes.” Economic Research Forum Working Paper Series 1194. Malaeb, Bilal, and Jackline Wahba. 2018. “Impact of refugees on immigrants’ labor market outcomes.” Economic Research Forum Working Paper Series 1194. Mansour, Hani. 2010. “The Effects of Labor Supply Shocks on Labor Market Outcomes: Evidence from the Israeli–Palestinian Conflict.” Labour Economics 17(6): 930–39. Martin, Susan F., Douglas A. Howard, Lahra Smith, Nili Sarit Yossinger, Lara Kinne, and Mark Giordano. 2017. “Environmental Impact of Refugee Camps: Lessons from Ethiopia and Djibouti.” Unpublished. Masterson, Daniel, and Vasil I. Yasenov. 2018. “Does Halting Refugee Resettlement Reduce Crime? Evidence from the United States Refugee Ban.” Immigration Policy Lab Working Paper No. 18-03. Mayda, Anna Maria, Chris Parsons, and Pierre-Louis Vézina. 2017. “Refugee resettlement: The impact on foreign direct investment.” U.S. Department of State Office of the Chief Economist Working Paper 2017 - 06. Mayda, Anna Maria, Chris Parsons, and Walter Steingress. 2017. “Refugee Resettlement: The impact on international trade flows.” U.S. Department of State Office of the Chief Economist (OCE) Working Paper 2017-03. Mayda, Anna Maria, Chris Parsons, Giovanni Peri, and Mathis Wagner. 2017. “The Labor Market Impact of Refugees: Evidence from the U.S. Resettlement Program.” U.S. Department of State Office of the Chief Economist Working Paper 2017-04. 32 Maystadt, Jean-François, and Gilles Duranton. 2018. “The Development Push of Refugees. Evidence from Tanzania.” Journal of Economic Geography: 1–36. Maystadt, Jean-François, and Philip Verwimp. 2014. “Winners and losers among a refugee-hosting population.” Economic Development and Cultural Change 62(4), 769-809. Morales, Juan S. 2018. “The Impact of Internal Displacement on Destination Communities: Evidence from the Colombian Conflict.” Journal of Development Economics 131: 132–50. Moser, Petra, Alessandra Voena, and Fabian Waldinger. 2014. “German Jewish Emigres and US Invention.” American Economic Review 104(10): 3222–3255. Murard, Elie, and Seyhun Orcan Sakalli. 2018. “Mass Refugee Inflow and Long-Run Prosperity: Lessons from the Greek Population Resettlement.” IZA Institute of Labor Economics Discussion Paper Series No. 11613. Ogude, Helidah. 2018. The Impact of Refugee Presence on Host Populations in Tanzania. A Desk Review. Washington, DC: World Bank, UNHCR. Özden, Çağlar, and Mathis Wagner. 2018. Moving for Prosperity: Global Migration and Labor Markets. Policy Research Report. Washington, DC: World Bank. Parsons, Christopher, and Pierre-Louis Vezina. 2018. “Migrant Networks and Trade: The Vietnamese Boat People as a Natural Experiment.” The Economic Journal 1-25. Paserman, M. Daniele. 2013. “Do high-skill immigrants raise productivity? Evidence from Israeli manufacturing firms, 1990-1999.” IZA Journal of Migration 2(6): 1-31. Peri, Giovanni, and Vasil Yasenov. 2017. “The Labor Market Effects of a Refugee Wave: Synthetic Control Method meets the Mariel Boatlift.” IZA Institute of Labor Economics Discussion Paper No. 10605 Peters, Michael. 2017. Refugees and Local Agglomeration - Evidence from Germany’s Post-War Population Expulsions. Unpublished. https://pdfs.semanticscholar.org/92c8/7655808506311c7fbed760f8c917e556e9b6.pdf Rozo, Sandra, and Micaela Sviastchiy. 2018. “Are Refugees a Burden? Impacts of Refugee Inflows on Host’s Consumption Expenditures.” Working Paper Unpublished. https://www.sandravrozo.com/uploads/2/9/3/0/29306259/rsw_july2018.pdf Ruiz, Isabel and Carlos Vargas-Silva. 2013. “The Economics of Forced Migration.” The Journal of Development Studies 49(6): 772-84. Ruiz, Isabel and Carlos Vargas-Silva. 2015. “The Labor Market Impacts of Forced Migration.” American Economic Review: Papers & Proceedings 105(5): 581–86. Ruiz, Isabel and Carlos Vargas-Silva. 2017. “The impact of hosting refugees on the intra-household allocation of tasks. A gender perspective .” United Nations University World Institute for Development Economics Research Working Paper 2017/66. Ruiz, Isabel, and Carlos Vargas-Silva. 2016. The labour market consequences of hosting refugees.” Journal of Economic Geography 16: 667–94. Saiz, Albert. 2003. “Room in the Kitchen for the Melting Pot: Immigration and Rental Prices.” The Review of Economics and Statistics 85 (3): 502-21 Sarvimäki, Matti. 2011. “Agglomeration in the Periphery.” Helsinki Center of Economic Research Discussion Paper No. 332. 33 Schumann, Abel. 2014. Persistence of Population Shocks: Evidence from the Occupation of West Germany after World War II. American Economic Journal: Applied Economics 6(3): 1989-205. Semrad, Alexandra. 2015. “Immigration and educational spillovers: evidence from Sudeten German expellees in post-war Bavaria.” Munich Discussion Paper No. 2015-7. Taylor, J. Edward, Mateusz J. Filipski, Mohamad Alloush, Anubhab Gupta, Ruben Irvin Rojas Valdes, and Ernesto Gonzalez-Estrada. 2016. “Economic Impact of refugees.” PNAS 113(27): 7449-53. Tumen, Semih. 2018. “The Impact of Low-Skill Refugees on Youth Education.” IZA Institute of Labor Economics Discussion Paper No. 11869. Tumen, Semih. 2016. “The Economic Impact of Syrian Refugees on Host Countries: Quasi -Experimental Evidence from Turkey.” American Economic Review: Papers & Proceedings 106(5): 456–60. Verme, Paolo. 2016. “The Economics of Forced Displacement: An Introduction.” Region et Development 44: 141-63. Verwimp, Philip, and Jean-Francois Maystadt. 2015. Forced Displacement and Refugees in Sub-Saharan Africa. An Economic Inquiry. World Bank Policy Research Working Paper 7517. Wyrwich, Michael. 2018. “Migration restrictions and long-term regional development: evidence from large-scale expulsions of Germans after World War II.” Jena Economic Research papers 2018-002. 34 Figure 1 – Shocks in the Consumer and Labor Market Shock to the consumer market Shock to the labor market p w S S D D c l Legenda: D=Demand; S=Supply; p=prices; w=wages; c=consumption; l=labor. 35 Table 1 – Summary of Results Variable Effect Consumer prices Up Wages (skilled) Up Wages (unskilled) Down Net wage effect ? Employment (skilled) Up Employment (unskilled) Down Net employment effect ? Household well-being (owners) Up Household well-being (non- Down owners) Net household well-being ? 36 Table 2 – Comparative Review of Models’ Specification No. Paper Estimator Unit Fixed Eff. Dep.Var. Forced Displ. Var. Instrumetal Var. 1 Akgündüz et al. (2015) OLS i, t t, r EM, PR FD presence; FD; isf(FD) none 2 Akgündüz and Torun (2018) OLS, 2SLS i, r, t r, t EM FD/pop Sum(((Syrians_t-1/pop)*FD)/d)) 3 Alhawarin et al. (2018) OLS i, t i, r, t WB (FD/pop)*TM none 4 Alix-Garcia and Bartlett (2015) D i n.a. WB Simple diff with matching none 5 Alix-Garcia and Saah (2010) OLS i, t mkt, y/m PR, WB 1/d_mk*(FD/pop)*100 none 6 Alix-Garcia et al. (2012) OLS w, m, t t PR FD none 7 Alix-Garcia et al. (2018) DMSP-OLS v, r, t r, t PR, WG, WB Sum_d{(ihs(FD)*d)} none 8 Angrist and Kugler (2003) OLS g, r, t g, r, t EM ln(FD/g_t) d 9 Balkan and Tumen (2016) OLS item, r, m, t i tem, r, m, t PR (FD/pop)*TM none 10 Balkan et al. (2018) OLS h, r, t r, t PR, WB (FD/pop)*TM none 11 Braun and Kvasnicka (2014) OLS r none EM, WB FD/pop none 12 Bodvarsson et al. (2008) 3SLS, CGE r none WG FD/pop FD_t-1 13 Borjas (2017) OLS r, t r, t WG TR*TM none 14 Borjas and Monras (2017) OLS r, s r, s EM, WG FD/LF (FD/WAP)_t-1 15 Braun and Mahmoud (2014) OLS j, r j EM FD/LF Sum_r(FD*occ)/(Sum_r(FD*occ)+natives*occ) 16 Calderon-Mejia and Ibanez (2016) OLS i, r, t r*t, t WG FD/WAP Sum_r(Casualties/d) 17 Card (1990) D r n.a. EM, WG Simple diff with matching none 18 Carrington and de Lima (1996) OLS t or r none EM, WG FD/pop none 19 Ceritoglu et al. (2017) OLS i, r, t none EM, WG TR*TM none 20 Clemens and Hunt (2017) OLS r, s r, s WG his(FD) stock of prior migrants 21 Cohen-Goldner and Paserman (2011) OLS i, j, t j, t, j*t EM, WG FD/j FD/E_t-1 22 Del Carpio and Wagner (2015) OLS i, r, t r, t EM FD/WAP + d Sum_r(FD*(FD_t-1/pop)/d 23 Depetris-Chauvin and Santos (2017) OLS r, t r, t PR, WB FD flow_t-1 sum_r(FD outflow/d) 24 Depetris-Chauvin and Santos (2018) OLS r, t r, t PR FD flow_t-1 sum_r(FD outflow/d) 25 Esen and Binatli (2017) OLS i, t none EM FD; FD/pop none 26 Fakih and Ibrahim (2016) VAR t none EM n.a. none 27 Fallah et al. (2018) OLS i, r, t none EM, WG FD/pop (FD/pop)/d 28 Foged and Peri (2015) OLS i, j, r, t t*j; t*r, i*u EM, WG FD/E sum_r(FD/WAP) 29 Friedberg (2001) OLS i, j, t j EM, WG FD/natives FD/E_t-1 30 Glitz (2012) OLS s, r, t s*t, r*t EM, WG Delta(s/LF) (FD/s*FD/WAP*Delta(FD))/LF_s 31 Hercowitz and Yashiv (2002) SUR, CGE t none EM, PR Delta(FD)/pop none 32 Hunt (1992) OLS r, s r WG FD/LF Temperature and FD_t-1 33 Kreibaum (2015) LPB hh, v, t t, r WB Diff_(t-t-1)(FD/1000pop) FD/d 34 Lach (2007) OLS item, store, item, r, t store, r, t PR FD/natives FD_t-1/natives 35 Makela (2017) OLS g, t none WG TR*TM none 36 Mansour (2010) OLS i, t t, s, j, r WG FD dummy migrants 37 Mayda et al. (2017) OLS r, t r, t EM, WG f(FD) FD_t-1 38 Maystadt and Duranton (2018) OLS h, v, t h, t, t*strata PR, WB ln(Sum_c(FD/d)) none 39 Maystadt and Verwimp (2014) OLS h, v t WB ln(1+FD/d) none 40 Morales (2017) OLS i, r, t r, t WGFD/pop*100; Sum(expulsions*migrations) (100/pop)Sum_r(FD*(FD/pop_r_t-1)) 41 Murard and Sakalli (2018) OLS r r WG, WB FD/pop_t-1 none 42 Peri and Yasenov (2017) OLS g, t none WG TR*TM none 43 Rozo et al. (2018) OLS i, r, t r, t EM, WG, WB FD/(FD_r*d) (FD_t-1/pop_t-1)*FD 44 Ruiz and Vargas-Silva (2015) OLS i, t i, t EM ln(1/d) none 45 Ruiz and Vargas-Silva (2016) OLS i, t i, t EM ln(1/d) none 46 Ruiz and Vargas-Silva (2017) OLS i, h, t I, h, t EM ln(Sum(FD/d)) none 47 Saiz (2003) OLS r, t none PR T none 48 Taylor et al. (2016) CGE n.a. n.a. WB n.a. n.a. 49 Tumen (2016) OLS i, r, t none EM, PR, WG TR*TM none Legenda: D=Dependent variable; FD=Forcibly Displaced population; EM=Employment or Employment Rate; WG=wages; PR=Prices; WB=Well-being (income, consumption or expenditure); LF=Labor Force; FE=Fixed Effects; OLS=Ordinary Least Square; LPB=Linear probability Model; DD=Differences in Difference estimator; TR=Treatment dummy; TM=pre-post treatment dummy; TR*TM=Generally refers to DD estimators; i=individuals; h=households; p=prices; t=time or year; r=region or location; d=distance from shock (camp, country of origin); v=village or community; c=camp; hp=host population; w=week; m=month; y=year; mk=market; nl=night luminosity; ihs=inverse hyperbolic sign; ae=adult equivalent; s=skills or education level; g=population group; s=sector; j=sector, occupation; industry or labor market segment. 2SLS=Two-Stage Least Square ; 3SLS=Stage Least Square; DMSP= Defence Meteorological Satellite Program; CGE=Computerised General Equilibrium Model ; SUR=Seemingly Unrelated Regression ; pop=population. 37 Figure 2 – Number of papers per year of publication 10 Syrian crisis num 5 0 1990 2000 2010 2020 year 38 Table 3 – Number of Results by Paper and Outcomes Paper Journal Outcomes Crisis EM PR WG WB 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 T 1 Akgunduz et al. (2015) IZA Discussion Papers 20 12 32 32 2 Akgündüz and Torun (2018) GLO Discussion Paper 14 14 14 3 Alhawarin et al. (2018) ERF Working Paper 8 8 8 4 Alix-Garcia and Bartlett (2015) Oxford Economic Papers 2 2 2 5 Alix-Garcia and Saah (2009) World Bank Economic Review 4 4 8 8 6 Alix-Garcia et al. (2012) World Development 4 4 4 7 Alix-Garcia et al. (2018) Journal of Development Economics 2 2 6 10 10 8 Angrist and Kugler (2003) The Economic Journal 12 12 12 9 Balkan and Tumen (2016) Journal of Population Economics 14 14 14 10 Balkan et al. (2018) IZA Discussion Papers 14 2 16 16 11 Barun and Kvasnicka (2014) Journal of International Economics 2 2 4 4 12 Bodvarsson et al. (2008) Labour Economics 12 12 12 13 Borjas (2017) Industrial and Labor Relations Review 8 8 8 14 Borjas and Monras (2017) Economic Policy 10 8 12 2 2 2 18 15 Braun and Mahmoud (2014) The Journal of Economic History 18 18 18 16 Calderon-Mejia and Ibanez (2016) Journal of Economic Geography 44 44 44 17 Card (1990) Industrial and Labor Relations Review 4 4 8 8 18 Carrington and de Lima (1996) Industrial and Labor Relations Review 4 4 8 8 19 Ceritoglu et al. (2017) IZA Journal of Labor Policy 20 20 40 40 20 Clemens and Hunt (2017) NBER Working Papers 6 2 4 6 21 Cohen-Goldner and Paserman (2011) European Economic Review 24 24 48 48 22 Del Carpio and Wagner (2015) World Bank Policy Research Working Papers 26 26 26 23 Depetris-Chauvin and Santos (2017) KNOMAD Working Paper 12 2 14 14 24 Depetris-Chauvin and Santos (2018) Journal of Development Economics 12 12 12 25 Esen and Binatli (2017) Social Sciences 8 8 8 26 Fakih and Ibrahim (2016) Defence and Peace Economics 2 2 2 27 Fallah et al. (2018) Economic Research Forum Working Papers 16 4 20 20 28 Foged and Peri (2015) AEJ: Applied Economics 32 12 44 44 29 Friedberg (2001) Quarterly Journal of Economics 2 12 14 14 30 Glitz (2012) Journal of Labor Economics 6 6 12 12 31 Hercowitz and Yashiv (2002) IZA Discussion Papers 4 4 8 8 32 Hunt (1992) Industrial and Labor Relations Review 2 2 2 33 Kreibaum (2015) World Development 2 2 2 34 Lach (2007) Journal of Political Economy 4 4 4 35 Makela (2017) European Economic Review 8 8 8 36 Mansour (2010) Labour Economics 4 4 4 37 Mayda et al. (2017) US Department of State Chief Economist WP 6 6 12 12 38 Maystadt and Duranton (2018) Journal of Economic Geography 12 16 28 28 39 Maystadt and Verwimp (2014) Economic Development and Cultural Change 2 2 2 40 Morales (2017) Journal of Development Economics 20 20 20 41 Murard and Sakalli (2018) IZA Discussion Papers 2 2 4 4 42 Peri and Yasenov (2017) IZA Discussion Papers 18 18 18 43 Rozo et al. (2018) Mimeo 28 16 12 56 56 44 Ruiz and Vargas-Silva (2015) American Economic Review: Papers and Proceedings 6 6 6 45 Ruiz and Vargas-Silva (2016) Journal of Economic Geography 10 10 10 46 Ruiz and Vargas-Silva (2017) WIDER Working Paper 56 56 56 47 Saiz (2003) The Review of Economics and Statistics 10 10 10 48 Taylor et al. (2016) Proceedings of the National Academy of Sciences (US) 4 4 4 49 Tumen (2016) American Economic Review: Papers and Proceedings 4 16 2 22 22 Total 334 120 244 64 110 6 70 4 12 4 22 80 14 90 6 4 44 10 12 16 258 762 Legenda: EM=Employment or Employment Rate; WG=wages; PR=Prices; WB=Well-being (income, consumption or expenditure). For the list of crises 1-17 see table 5. _ 39 Table 4 – Papers, Results and Gini by Outcome Papers Results Res/Paper Gini Employment 24 334 13.9 0.35 Prices 13 120 9.2 0.16 Wages 23 244 10.6 0.37 Well-being 13 64 4.9 0.35 Total/Average 21.0 762 36.2 0.32 The Gini index is calculated using number of papers (as “individuals”) and number of results (as “income”). A lower Gini indicates more spread of results across papers. _ 40 Table 5 – Crises and Countries by Income Group (in percentage of all results) Caseload HICs LICs MICs Total FD pop (%) Est.Time* Burundian and Rwandan refugees in 1 Tanzania 0.0 14.4 0.0 14.4 53.3 13.0 2 Congolese refugees in Rwanda and Uganda 0.0 0.8 0.0 0.8 17.4 16.7 3 Cuban refugees in Miami 9.2 0.0 0.0 9.2 8.1 8.2 Ethnic Germans from EE and FSU to 4 Germany 1.6 0.0 0.0 1.6 3.5 14.0 5 Ethnic Greeks from Turkey to Greece 0.0 0.0 0.5 0.5 20.0 58.0 6 Escapees from Algeria to France 0.5 0.0 0.0 0.5 3.3 6.0 Expellees from East Europe to West 7 Germany 2.9 0.0 0.0 2.9 16.5 6.5 8 FSU escapees to Israel 10.5 0.0 0.0 10.5 9.4 6.3 9 FY refugees to Europe 1.8 0.0 0.0 1.8 0.3 8.3 10 IDPs in Colombia 0.0 0.0 11.8 11.8 10.4 5.0 11 IDPs in Sudan (Darfur) 0.0 0.8 0.0 0.8 30.0 3.7 12 Palestinians from West Bank to Israel 0.0 0.0 0.5 0.5 50.0 4.0 13 Refugees in Denmark 5.8 0.0 0.0 5.8 4.7 14.0 14 Refugees in Kenya (Turkana) 0.0 1.3 0.0 1.3 10.0 20.0 15 Refugees in the USA 1.6 0.0 0.0 1.6 0.1 30.0 Returnees from Angola and Mozambique to 16 Portugal 0.0 0.0 2.1 2.1 5.1 13.0 17 Syrian refugees in Jordan and Turkey 0.0 0.0 33.9 33.9 6.3 3.3 Total 33.9 17.3 48.8 100.0 14.6 13.5 Legenda. FD (%) indicates the number of forcibly displaced persons (refugees or IDPs) as a percentage of the host population in a given geographical area affected by refugees or IDPs. (*) Est. Time shows the average time gap between the beginning of the influx and the year for which the impacts are measured in each study. 41 Table 6 – Papers, Results and Impact Factor by Journal Journal Papers Results Imp.Fact. AEJ: Applied Economics 1 44 3.61 American Economic Review: Papers and Proceedings 2 28 0.01 Defence and Peace Economics 1 2 0.07 ERF Working Paper 1 8 0.05 Economic Development and Cultural Change 1 2 0.73 Economic Policy 1 18 2.25 Economic Research Forum Working Papers 1 20 0.05 European Economic Review 2 56 1.24 GLO Discussion Paper 1 14 0.01 IZA Discussion Papers 5 78 0.66 IZA Journal of Labor Policy 1 40 0.35 Industrial and Labor Relations Review 4 26 0.48 Journal of Development Economics 3 42 1.90 Journal of Economic Geography 3 82 0.45 Journal of International Economics 1 4 2.85 Journal of Labor Economics 1 12 3.01 Journal of Political Economy 1 4 6.64 Journal of Population Economics 1 14 0.02 KNOMAD Working Paper 1 14 0.01 Labour Economics 2 16 1.00 Mimeo 1 56 0.00 NBER Working Papers 1 6 1.35 Oxford Economic Papers 1 2 0.58 Proceedings of the National Academy of Sciences 1 4 0.01 Quarterly Journal of Economics 1 14 8.40 Social Sciences 1 8 0.00 The Economic Journal 1 12 2.27 The Journal of Economic History 1 18 0.27 The Review of Economics and Statistics 1 10 2.38 US Department of State Chief Economist WP 1 12 0.01 WIDER Working Paper 1 56 0.08 World Bank Economic Review 1 8 0.57 World Bank Policy Research Working Paper 1 26 0.01 World Development 2 6 0.29 Total 1.96 762 0.98 42 Table 7 – Results on Sign and Significance Unweighted Weighted by impact factor Freq. Percent Cum. Freq. Percent Cum. Well-being Positive 29 45.3 45.3 17.2 52.2 52.2 Nonsignificant 22 34.4 79.7 13.9 42.2 94.4 Negative 13 20.3 100.0 1.9 5.7 100.0 Total 64 100.0 32.9 100.0 Prices Positive 45 37.5 37.5 49.0 46.2 46.2 Nonsignificant 28 23.3 60.8 19.2 18.1 64.2 Negative 47 39.2 100.0 38.0 35.8 100.0 Total 120 100.0 106.1 100.0 Employment Positive 39 11.7 11.7 39.1 14.1 14.1 Nonsignificant 216 64.7 76.4 186.7 67.3 81.4 Negative 79 23.7 100.0 51.7 18.7 100.0 Total 334 100.0 277.5 100.0 Wages Positive 32 13.1 13.1 74.5 22.5 22.5 Nonsignificant 144 59.0 72.1 181.8 55.0 77.5 Negative 68 27.9 100.0 74.4 22.5 100.0 Total 244 100.0 330.7 100.0 Employment and Wages Positive 71 12.3 12.3 113.5 19.6 19.6 Nonsignificant 360 62.3 74.6 368.5 63.8 83.4 Negative 147 25.4 100.0 126.2 21.8 105.2 Total 578 100.0 608.2 105.2 43 Figure 3 – Share of Results by Outcome Well-being Well-being (weighted) Prices Prices (weighted) Employment Employment (weighted) Wages Wages (weighted) 0 20 40 60 80 100 120 Positive (increase) Nonsignificant (no correlation) Negative (decrease) 44 Table 8 – Multivariate Logit Regressions for Negative and Positive Results Employment Wages Negative Coeff. Positive Coeff. Negative Coeff. Positive Coeff. Unweig. Weigh. Unweig. Weigh. Unweig. Weigh. Unweig. Weigh. males -0.948 -1.024 -0.404 -0.161 -0.086 -0.173 -42.756 -61.361 (2.13)* (3.64)** -0.58 -0.55 -1.03 -1.53 -0.24 -0.45 females -0.715 -1.462 -0.305 -0.225 -0.139 -0.252 489.929 471.309 -1.17 (4.63)** -0.29 -0.78 -1.65 (2.34)* (2.71)* (4.12)** young 0.322 1.908 -0.54 (7.17)** old -0.477 0.847 2.51 2.589 -42.128 -60.748 -0.62 (2.21)* (3.23)** (3.61)** -0.2 -0.92 formal -0.847 0.42 -0.161 -42.782 -61.402 -0.73 -0.79 -0.57 -0.2 -0.25 informal -0.58 0.256 -1 -0.45 lowskilled -0.401 -0.776 -0.08 0.26 -0.091 -0.107 -41.595 -60.215 -0.46 -0.72 -0.13 -1.55 -0.95 -0.89 -0.26 -0.87 highskilled -0.858 -0.205 -0.196 -0.116 -0.144 -0.152 -41.443 -60.063 -1.02 -0.04 -0.19 -0.57 -0.69 -0.76 -0.26 -0.87 _cons 1.137 0.583 0.355 0.276 0.3 0.421 42.798 61.418 (4.09)** (3.38)** -0.99 -1.92 (6.18)** (7.80)** -0.56 -1.41 R2 0.1 0.63 0.29 0.4 0.08 0.16 0.26 0.26 N 79 51 39 39 68 74 32 74 45 Table 9 – Bivariate Logit Regressions for Negative and Positive Results Employment Wages t-stat > 1.3 in bold neg pos neg pos Unw Wei Unw Wei Unw Wei Unw Wei eg. gh. eg. gh. eg. gh. eg. gh. Time of evaluation since crisis -1.38 0.22 0.20 2.62 -3.22 -3.06 2.94 4.03 Refugee share -0.21 -0.54 0.78 1.29 1.79 0.14 -0.85 -2.85 Males 2.75 2.60 -0.29 -1.23 0.94 0.93 -1.52 -2.57 Females 0.85 3.07 -0.88 -0.95 2.59 4.90 -1.13 -1.93 Young 2.56 4.73 . . . . . . Old 0.42 1.33 1.40 -1.45 . . . . Formal -1.42 . 4.68 2.10 . . 0.65 -0.11 Informal 3.36 2.17 . . . . . . Lowskilled -0.83 -1.84 2.62 6.81 -0.84 -1.24 -0.87 -0.50 Highskilled -0.62 -0.82 -0.39 2.07 -1.91 -2.82 0.73 0.69 HIC 0.43 0.06 0.00 0.39 -1.39 -4.43 3.08 3.71 MIC -0.15 -0.79 -0.54 -0.62 1.55 4.73 -3.30 -3.79 LIC -0.32 1.11 0.66 0.26 . . 1.34 1.21 Burundian and Rwandan refugees in Tanzania -0.32 1.11 0.66 0.26 . . . . Congolese refugees in Rwanda and Uganda . . . . . . . . Cuban refugees in Miami . . . . -0.68 1.61 0.68 -0.41 Ethnic Germans from EE and FSU to Germany 2.58 5.20 . . 1.18 2.71 . . Ethnic Greeks from Turkey to Greece . . . . . . . . Expellees from Algeria to France . . . . 0.68 0.61 . . Expellees from East Europe to West Germany 3.97 1.21 1.17 3.57 . . . . FSU escapees to Israel -2.09 -2.56 -0.30 -1.42 0.11 -3.80 1.24 -0.06 FY refugees to Europe 2.77 5.28 . . . . . . IDPs in Colombia . . . . 4.45 4.32 . . IDPs in Sudan (Darfur) . . . . . . . . Palestinians from West Bank to Israel . . . . 0.96 0.14 . . Refugees in Denmark -2.20 -4.11 1.84 3.06 . . 4.39 6.55 Refugees in Kenya (Turkana) . . . . . . 1.34 1.21 Refugees in the USA . . . . . . . . Returnees from Angola and Mozambique to Portugal . . . . 1.69 3.00 0.37 -1.14 Syrian refugees in Jordan and Turkey 0.09 -0.62 -0.39 -0.47 . . -1.23 -0.76 46 Annex 1 – Review of Empirical Models 1.1 Well-being and prices The papers covering the consumer market and household well-being are predominantly related to low and middle-income countries with few exceptions. They also either focus on consumer prices or on household well-being measured in terms of household expenditure or income per capita or some form of wealth indicator. In this section, we describe the main prototypes of these models by focusing on selected papers. Alix-Garcia and Saah (2010) model the impact of the refugee presence on prices as: 38 84 (, ) = + 1 + 2 + 3 + , Γ + ∑ + ∑ ψ + , =1 =1 Where B and R are the share of Burundian and Rwandan refugees on the population of the area, D is the inverse of the distance to the closest refugee camp from market i, p is price, F is food aid, X are weather controls, M are market level fixed effects, and ψ are year/months fixed effects. More precisely, the shock is measured as the share of refugees over the total population of refugees weighted by the inverse of the distance of the refugees from the closest markets as follows: 1 = ( ) ∗ 100 ⅆ + where t is time and m is market. The model estimates therefore price elasticities to changes in refugees stocks controlling for distance. Also noteworthy is the introduction of a control for food aid given the role of international aid in a low-income country such as Tanzania. Alix-Garcia, Bartlett, and Saah (2012) model the impact of IDPs in South Sudan on prices as: ,, = 0 + 1 ,, + 2 ,, + 3 , + 4 yr2006 + 5 2007 + 6 + 7 ℎ + Where p is the natural log of prices; w, m and y stand for week, month and year respectively, a is the total amount of aid, s is the natural log of the aid related to the good analyzed, r is the number of IDPs and “hungry” is a dummy variable for the hungry season. The equation is estimated with an OLS estimation where the error term has a flexible correlation structure that includes up to three lags correlations. The key identifying assumption is that there is no simultaneity between prices and IDPs or aid and the shock measure is simply the number of IDPs counted in any particular month. As for Alix-Garcia and Saah (2009), this is a price elasticity model controlling for international aid and also for seasonality. Unlike the previous paper, this is a weekly time model with no cross-section variation. Alix-Garcia et al. (2018) study the impact of the Kakuma refugee camp in Kenya on the economic well- being of surrounding areas. They use an equation where the dependent variable is the inverse hyperbolic sine of the DMSP-OLS luminosity index at the village level: ℎ = + ∑ ∗ ( ) + ⅆ + + + Where v is village, l is location, t is time, refugees is the inverse hyperbolic sine of the refugee population in camp in year t and road is the inverse hyperbolic sine of the distance of each village to the nearest road interacted with a year fixed effect. ⅆ are the village and time fixed effects. Standard errors are clustered by broad geographic regions. As a shock variable, the authors use parametric and semi-parametric measurements of the distance between the Kakuma refugee camp in Kenya and neighboring villages which they interact with the refugee population in each year (the argument of the sum sign in the equation above). The parametric specifications of f(.) are the inverse hyperbolic sine of the distance to Kakuma or the inverse 47 distances where d=1, whereas the semi-parametric specification considers six bands of distances (d=1, 2,…6) where f(.) is a series of dummy variables representing the bands. This is a cross-section village model estimated over time and the only paper that uses night luminosity as a proxy of economic well-being, which is suitable to study situations where the displaced people are all in camps and not dispersed among hosts. In this case, international aid is administered within the camp and its effect is captured in the overall luminosity effect. Maystadt and Verwimp (2014) look at Burundian and Rwandan refugees in Tanzania during the 1993-1994 crisis with a consumption model as follows: (ℎ, ) = 0 + 1 (1 + (ℎ), ) + 2 ℎ,1991 ∗ (1 + (ℎ), ) + 3 ℎ, + 4 (ℎ), + 5 + ℎ, where V is household consumption per adult equivalent, h is household, v is village, t is time and RI is the refugee shock described as the population of refugees in camps divided by the distance of the host village from camps. Activity is the main initial occupation of each household, Z and Q are household and village specific characteristics and is a time dummy. This is therefore a household consumption model used to measure household well-being. It is suitable for situations where refugees are located in camps and hosts are located in separate villages. Maystadt and Duranton (2018) look at the same crisis using a time-space variation equation to model consumption as follows: ℎ, ( ) = 0 + 1 (ℎ,), + ℎ + + ∗ + ℎ, (ℎ,), Where ℎ, is household nominal consumption in year t; (ℎ,), is the price level in village v in year t; (ℎ,), is a refugee index, ℎ , and are household, time and strata*time fixed effects. The authors use robust standard errors clustered at the initial village level to account for correlation within villages. As for Maystadt and Verwimp (2014), the refugee index is constructed using the population of refugees in the camps and the distance between refugee camps and villages nearby. More precisely, they take the log of the sum of the refugee population across refugee camps divided by the distance of these camps from neighboring villages: , = (∑13 1 ), , where pop=population, c=camp, v=village, t=time. Therefore, this is a household real consumption model used as a measure of household well-being and, similarly to the previous paper, what is measured is the elasticity of household consumption in local villages to changes in the refugee camp index. Kreibaum (2015) studies refugees in Uganda and estimates household consumption with a longitudinal linear probability model trying to capture short and long-term effects of the refugee shock as follows ;, = 0 + 1 _, + 2 ℎ, + 3 , + 4 , + + + , where Y is the log of consumption, i, c and t are household, communities and years respectively, d is district, refugee_level is the number of refugees per thousand local inhabitants (a measure interpreted as the long- term impact of refugees) and shock is the difference of this measure between two successive surveys (a measure interpreted as the short-term refugee impact). and are the time and district fixed effects. This is again a consumption model and one of a few papers that tries to distinguish between short-term and long- term shocks using the same equation. 48 Akgündüz, van den Berg, and Hassink (2015) use a spatial-time equation to model all the outcomes they consider including inflation as follows = + + + + Where is inflation for food, housing or hospitality sectors, is the refugee shock, i are provinces or regions; t is time, T and R are time and region fixed effects respectively. The shock or treatment effect is either a binary variable describing the presence of refugees, the number of refugees or the inverse sine function of the number of refugees (to normalize the variable with respect to wealth where many observations may take zero value). Unlike other price models that consider prices of individual products, this paper uses inflation indexes constructed on groups of products representing sectors. Balkan and Tumen (2016) model the impact of immigration on prices with a DD approach as ,,, = + (, ∗ , ) + + + + + ,,, Where i, r, y and are consumption items, regions, years and months respectively, p is price, f(.) are fixed effects, and P is the pre-post immigration period. The shock T is the share of immigrants over the local population and the parameter gives the average impact of immigration on prices in the treatment region in logs. Balkan et al. (2018) use the same difference-in-difference equation and adapt it to assess the impacts of Syrian refugees on housing rents in Turkey: (,, ) = + ( ∗ ) + + + ,, + ,, Where r and y are regions and years of observation, i indexes households, f(.) are fixed effects, and X is a vector of dwelling characteristis (including size, number of rooms, and existence of kitchen, indoor toilet, bath or shower, piped water and hot water system). P is a dummy variable which is 1 in the post- immigration period, T is 1 for the treatment region Southeastern Anatolia (and in a second specification also includes the Mediterranean region), and 0 for all other regions in Turkey, except the Mediterranean and Southeastern Anatolia region. The parameter gives the average impact of the refugee influx on housing rents in the treatment region in the post-immigration period in logs. Saiz (2003) studies the Mariel boatlift case in Miami and looks at rental prices estimating the following DD model: = + + + + where R is the rent for unit i at year t, is a unit fixed effect, and D are dummies for time and location taking values of 1 for the post-treatment period and Miami. In this case, we have a simple DD approach where the number or share of displaced persons do not have a role. In contrast, Depetris-Chauvin and Santos (2018) exploit variation in the intensity of quarterly displacement inflows over time t and between 13 main cities c in Colombia to estimate the impact on rental prices with the following equation: ln(, ) = + ln(,−1) ) + ′ , + ⅆ + ⅆ + , where d are city and quarter fixed effects. , is a vector of controls, which includes city-level linear trends, and interactions between remoteness and time dummies. For robustness checks, they also included additional (potentially endogenous) determinants of rental prices. The error term , is clustered at the city- year level. They weight all the regressions by population and lag IDP inflows by one quarter. The point ̂ can be interpreted as a standard elasticity. Inflows measure the total number of IDPs arriving in estimate city c in quarter t-1. To address endogeneity concerns, they use the sum of IDP outflows from all 49 municipalities M (except the receiving host city c) during quarter t weighed by the inverse of the road distance between the host city c and each municipality of origin m. This describes the receptivity measure used as instrument: −1 , = ∑ , ,′ \{} 1.2 Employment and wages The first papers that explored the question of the impact of displaced people on host populations focused on labor markets effects in high-income countries using a simple Difference (D) or Differences-in- Difference (DD) approach. Card (1990) looked at the impact of the 1980 Mariel boatlift operation carrying Cubans to Miami on the local residents. The paper compares hourly wages, employment to population and unemployment rates before and after the boatlift covering the period 1979-1985. The author disaggregates by different population groups including Whites, Blacks, Hispanics and Cubans and benchmarks these indicators with those of other comparable US cities for the same population groups. The model used is not formally outlined but the paper estimates outcomes by year and for each ethnic group between treated (Miami) and non-treated (comparison cities) and provides simple differences (D) in tabular form. Angrist and Krueger (1999) use the Card (1990) work to formalize a DD approach to such studies as follows: = ′ 0 + + + + where Y is the individual (i) employment (unemployment) outcome, X is a vector of individual economic characteristics, M is an interaction term of the post-treatment period (after 1980) and the treated group (Miami) and and are time and cities fixed effects. The coeffcient of interest or the DD estimator is and the estimation method is generally a linear OLS model. Borjas (2017) reasseses the wage effect of the Mariel boatlift case also using a DD approach described as ̅ = + + ( ∗ − ) + ̅ is the mean age adjusted log wage of male high-school drop-outs, and are city and year fixed where effects and captures the DD effect intercating location (Miami) and time (post-Mariel) variables. Different constructed non-treated locations (placebos) are used as counterfactual to Miami. Amongst others the author also uses the Synthetic Control Method (SCM, see also Peri and Yasenov 2017) to construct the synthetic city. The model is a DD model similar to Angrist and Krueger (1999) but is not an individual model. It is a a spatial model based on city-time cells. This is a very significant departure from Card (1990) and Angrist and Krueger (1999) and one of the reasons why results of this paper cannot be compared to the previous two papers (see more on this in the empirical results section). Clemens and Hunt (2017) looking at the Mariel boatlift and the FSU immigration to Israel argue that the wage effects estimated by Borjas (2017) are spurious (Kronmal, 1993)24 and that “Because the city-year averages are pre-adjusted by city and year, the resulting regressions run by Borjas test not for a difference- in difference of the average wage level, as Borjas incorrectly states, but instead for a difference-in- difference of the relative wage of workers with less than high school (compared to the average worker at any other education level).” (p.13). The authors then propose a correction of the Borjas model described as ∆ = + + (ℎ1 ) − ′ (ℎ1 ) + 24Among various other results, Kronmal (1993) shows how estimations that consider ratios with the same denominator on the two sides of a linear equation or a ratio as dependent variable and the denominator of this ratio as independent variable can be spurious. 50 Where r and s stand for region and skills level, asinh is the inverse hyperbolic sine and the endogenous refugee supply shock (ℎ1 ) is instrumented by the predetermined stock of prior migrants (ℎ1 ). Worth noting is the fact that the spurious correlation discussed by Kronmal (1993) applies to many of the models reviewed in this paper as the dependent and independent variables are often ratios with population size on the denominator of both the dependent and independent variables. Peri and Yasenov (2017) also reassessed the Mariel boatlift case using SCM methods (Abadie and Gardeazabal 2003) where wages for the treated group in Miami are compared with those of a synthetic control group constructed out 43 non-treated cities. Similalry to matching methods, the synthetic panel is constructed using weights that minimize the difference between wage predictors for treated and control cities. Results are then simply illustrated graphically comparing average wages for treated and non treated groups to see whether the pre-post longitudinal series show any discrepancies between the two groups. The authors do not estimate confidence intervals and standard errors25 and, as a complementary analysis, propose instead an estimator described as: = + ∑ + ∑ + ∑ ( ∗ ) −79 −79 −79 + ∑ ( ∗ ) + −79 where is the average log weekly wage of high school dropouts in group i at time t, Miami is a dummy for the treated group in Miami versus the Synthetic group, and is a set of 3-year dummies representing three years’ periods (p) before and after the shock. The shock year is in-built in the constant and the coefficients of interest are therefore (80−82 in particular as this is the first 3-year period after the 1979 shock). This model improves on the previous ones in that the SCM ensures a better matching between treated and non-treated groups whereas the differentiations by three-years time periods allows to capture short, medium and long-term effects. We remain, however, in the realm of D and DD models with the D estimations illustrated graphically or in tabular form and the DD estimations provided econometrically with linear modelling. Hunt (1992) studies the effect of the French repatriates from Algeria after Algerian independence in 1962 on unemployment and wages of the local residents as follows: − ⅆ1968, = − 1968, (1968, , ⅆ1968, , 1968, ⅆ 1968, ) where the repatriate variable is expressed as a proportion of the labor force in 1968 (post-treatment), education is the proportion of the population not in education with a secondary school degree or higher, age is the proportion of young people (15-24 y.o.) in the labor force, and the department structure refers to the employment shares across the main economic sectors. The empirical equation also includes regional dummies. The wage equation is the same with the exception of the dependent variable which is defined as the natural logarithm of wages in 1967 (the year before the one considered for the independent variables). The model is estimated for 1962 and 1968 and the first difference between coefficients of the two years is the measured effect. Therefore, here we have linear estimations of D followed by manual estimations of the DD effect. 25This is an anomaly of this paper given that stochastic dominance theory provides the theory and empirical tools that make these estimations possible (see for example Araar and Verme, 2016). 51 Friedberg (2001) studies immigration to Israel from the FSU and models wages of local residents on the national level in a cross-section framework as follows: + = + + Where W is the average native log wage in occupation j, X is a vector of occupation-sepecific factors that could affect the level of wages and r is the ratio of immigrants to native workers. In this case is the shock, is the coefficient of interest and the cross-occupation equation is the novelty. However, it is not a D or DD model. To address possible endogenenity issues, the same equation is also specified in dynamic terms where the change in wages over time is regressed on the inflow of immigrants over time (all the elements of the equation above are defined in terms of changes over time rather than stocks). In addition, the author uses an instrumental variable approach where the instrument is immigrants’ previous occupation in Russia and a wage equation based on individual data specified as follows: = + + ∑ + + =1 where w is the log earnings of individual i in occupation j, X is a vector of control variables, are year dummies, occ are a set of occupation dummies and r is the ratio of immigrant to native workers. Cohen-Goldner and Paserman (2011) also study FSU migrants into Israel with the same specification for hourly wages and employment described as = 0 + 1 + 2 + 3 + + + + where y is log of hourly wages or employment, i, j and t are individuals, labor markets (cells) and quarters respectively, IMM is the ratio of immigrants in segment j, Z and X are vectors of macroeconomic and individual characteristics , and are the segment, quarter and segment-quarter combined fixed effects and is the error term. Standard errors are clustered at the segment-quarter level. The equation is also extended to take into account immigrants with different levels of tenure in Israel and to better capture long-term effects. Differently from previous discrete DD models that use pre- and post, treated and non- treated groups, this model use the immigrants’ intensity by location as shock and can measure therefore the elasticity of wages and employment to the ratio of immigrants. Hence, the question is not whether immigration has an impact but how much of an impact has each level of immigration. As before, estimations models are linear with fixed effects. Borjas and Monras (2017) propose to use a standard model and empirical approach to study the employment, unemployment and wage impact of displaced people on host communities in the context of four different crises: The Mariel boatlift of Cubans to Miami, the Jews immigration to Israel from the Former Soviet Union after the desegregation of the Union in 1991, the immigration of former Yugoslavia citizens to Europe during the Balkan wars of the 1990s and the exodus to France of French and Algerian people from Algeria after Algerian independence in 1962. The empirical model is: 1 ∆ = + − ( ) − + 0 Where w is wage, r and s are regions (labor markets) and skill-type (educational level) respectively, (.) are the fixed effects for these two dimensions, L is the number of workers before (0) and after (1) the shock, is the wage elasticity and m is the share of immigrants in L. Therefore, this is also a linear model that aims at measuring elasticities of wages to immigration levels with fixed effects. To account for possible endogeneity, the authors also proposed a reduced form equation of th type ∗ ∆ = + − (1 + ) + . 52 The reduced form equation derives from the first-order Taylor’s expansion of the log change in the size of − the native worksforce ( 1 0 ) which is transformed, in turn, into a labor supply expression for natives 1 1 + where is a parameter that measure the labor supply response. The reduced equation allows 1 to desegregate the wage elasticity and the labor supply parameter that would otherwise be confounded into one coeffcient. This is the only paper we found that addresses specifically this issue. Angrist and Kugler (2003) consider employment of natives and immigrants in high income EU countries hosting displaced people from former Yugoslavia during the 1990s and measure the short-run impact on natives’ employment as ( ) = + + + ( ) + Where the dependent variable y is the log of the employment to population ratio for natives, i, j and t are demographic groups, country and year, and the shock is the log of the immigrant share s over the demographic group at year t. The estimation is a cross-country OLS model which includes fixed effects for demographic group i, country j and year t. The paper also uses a second equation where s is instrumented using distances of receiving countries from former Yugoslavia and a third equation where immigration is interacted with countries’ institutions. As for previous models, the authors here use an OLS li near model with fixed effects complemented by an IV model to study the elasticity of natives’ employment to immigration intensity. Foged and Peri (2015) studied the inflow of refugees from conflict areas to Denmark between 1991 and 2008 using individual fixed effects regressions and a DD model based on municipality data. The FE model is described as ′ = + + , + , + , + where y is one of three outcomes for natives (NAT) including the complexity of the task performed, hourly wages and the fraction of a year worked (a measure of labor supply), i, j, m and t represent individuals, establishments, municipalities and time respectively, x is a vector of time-varying individual characteristics, is the refugee-country immigrant share of employment in municipality m at time t, , and , are industry by year and region by year effects and , are fixed effects for individuals and units (u) combined. By varying u one can measure the effect of S on outcomes y for different units of analysis. The paper estimates three equations where u is establishments, municipalities or nothing. The OLS estimation is also complemented by a 2SLS estimation where is instrumented using the refugee dispersal policy adopted by Denmark during the period considered. The instrument is the following ̂ = ( ∑ ̂ ) /1988 ∈ Where ̂ is the imputed working-age population of immigrants from refugee country c in municipality m at time t and 1988 is the total working-age population in municipality m in 1988. The DD estimator is described as follows: −1 14 ′ = + ∑ ( = ) + ∑ ( = ) + , + , + , + , =−3 =1 + + 53 where M is the treatment equal to 1 if individual i is in the upper quartile of the difference in predicted refugee flow and 0 if is in the lower quartile, D are year dummies and the rest are industry, region, education and occupation time specific fixed effects and municipalities fixed effects. The pre-treatment period is 1991-1994, a period that did not see a major inflow of refugees. Again, this model falls into the linear elasticity models group where the main innovation is represented by the structure of the instrument in the IV model that complements the simple OLS model. Mayda et al. (2017) exploit the variation in the number of newly resettled refugees in the U.S. across commuting zones i and over time t to analyzes the impacts on wages and employment with the following linear equation: = ( ) + + + ∗ + + Where t = 1990, 2000, 2010, and are commuting zone and year fixed effects, ∗ captures commuting zone time trends, and is a vector of additional time-varying control variables (such as initial commuting zone populations and the growth of local employment and wages predicted by industrial composition). ( ) is a function of the presence of refugees in a given commuting zone i and decade t. As the authors have only data on new refugee arrivals, they use changes in the stock of refugees due to these new arrivals as a proxy for overall changes in the refugee stock, and estimate the specification in first differences as follows: = + + + + where the treatment dummy I takes the value of 1 for those commuting zones and decades in which the change in refugee stock (standardized by the initial population of the commuting zone) was larger than the mean by 0.1%. To address endogeneity threats to identification, as refugees are likely to settle in commuting zones with better wage and employment prospects, the authors instrument refugee arrivals with the number of initial refugees with no U.S. ties. To control for non-random allocation of refugees on the part of the placement agency, these authors use matching to select a sample of control commuter zones in the pre- treatment period, an approach similar to Dustmann, Schoenberg, and Stuhler (2017). Makela (2017) used the same SCM method used by Peri and Yasenov (2017) and Borjas (2017) to study the impact of returnees from Angola and Mozambique to Portugal in 1974. In this case, donors for the synthetic group are countries rather than cities and the analysis in based on camparing Portugal with comparable countries that did not experience similar levels of immigration during the period considered. The outcome variables considered are average annual labor productivity, average annual wage per worker and the unemployment rate. The same author also combines the SCM approach with a Difference approach (essentially using the synthetic control group for an econometric difference estimation during the post- shock period) working with Portuguese regions rather than countries and focusing on the agriculture and construction sectors. Moreover, a fixed effects and and a generalized synthetic control method are used as robustness tests. Calderon-Mejia and Ibanez (2016) study the impact on IDPs in Colombia on hourly wages of the host populations using household data and an OLS and IV approach. The wage equation is defined as follows: = + + + + where w is the hourly wage, i, c and t denote individuals, cities and time, and are the city-time and time fixed effects, X are individual characteristics and S is the supply shock defined as 54 ∑ = ( ) 12−65 where the numerator is the cumulative sum of IDPs (M) entering city c starting from the year 1999 to year t and the denominator is the working-age population in city c at time t. The authors then instrument S using = ∑ 0 which is the cumulative number of massacres in city c at time t weighted by the inverse of the distance from the site of the massacre to city c. Here again, the model falls into the OLS+IV approach designed to estimate elasticities of wages to immigration intensity with the original contribution being the particular instrument designed for the IV equation. Looking also at the impact of IDPs in Colombia on wages, Morales (2017) uses a labor force survey, census data and registry data to study short and long-term effects as follows: ℎ − : = + ⅆ−1 + + + + + + − : = + ⅆ + + + + where y is the log of wages, i, m, and i are individuals, municipalities and time respectively, are individual controls, is the log of total population or other municipality controls, and are time and municipality fixed effects, are municipality time trends, are department fixed effects and d is the inflow of IDPs defined as 100 ⅆ = where is the total number of IDPs arriving in municipality m at time t. The same variable without the t subscript is used for the long-run effects equation. The author also estimates an IV equation with a migration network or enclave type of instrument defined as 100 ⅆ = (∑ 1993 ) where is the total number of expulsions from municipality j at time i, 1993 is the share of migrants from municipality j who lived in municipality m in 1993. The instrument relates to migration decisions taken prior to 1993, which is precedent to the period considered in the study. As for the previous study, here we are again in the domain of OLS+IV estimations of wages-displaced elasticities with the major methodological contribution being the construction of the instrument. A number of recent studies estimate the impact of Syrian refugees on neighboring countries’ labor markets. As for other crises, we have a mix of DD and elasticities models. Ceritoglu et al. (2017) and Tumen (2016) study the influx of Syrian refugees in Turkey after the start of the Syrian civil war in 2011 and propose the following DD model: ,, = + ∗ ( ∗ ) + ′ ,, + , + + + ,, 55 where y is the labor market outcome of interest (formal and informal employment to population ratios, unemployment to population ratios, and labor force participation rate) R and T are the dummy variables for treatment and pre-post treatment periods respectively, i, j and t are individuals, regions and years, X is a vector of individual level charactersitics and Z is a region and time specific proxy for economic activity. Note that, by dropping R and T and including region and year fixed effects, the equation can also be estimated as ,, = + ∗ ( ∗ ) + ′ ,, + , + + + ,, Del Carpio and Wagner (2015) study the impact of Syrian refugees in Turkey using the following equation: = + ( ) + ( ) + + + where Y can be total employment in the working age population or various employment disaggregations including formal and informal, regular and irregular and full and part-time employment, i, t and r are individuals, subregions and year respectively. ( ) is a function of the distance from the Syrian border which serves as a control to compare subregions that have equal chance of receiving Syrian refugees based on distance from the border. R is the shock defined as the number of Syrian refugees normalized by the working age population for each subregion in year t, and and are the subregion and year fixed effects. 1 In a second model, R is also instrumented as = ∑ , where is the total number of registered Syrians in Turkey and is the fraction of the Syrian population in each Syrian governorate before the shock and is the travel distance from each Syrian governorate capital s to the most popoulous cities in the Turkish subregions r. These models fall in the OLS+IV tradition to estimate elasticities of various outcomes to the displacement crisis. Del Carpio and Wagner (2015) also study wages using a decomposition approach of mean wages into the part that is explained by changes in employment composition of Syrians and non-Syrians and the part that is explained by other factors. This is the only paper of this review that follows this approach. Esen and Ogus Binatli (2017) study the impact on employment and unemployment of Syrian refugees in Turkey using a fixed and a random effects models as follow: = = ∑ , + ∑ + ; = = + ∑ , + where Y is formal or informal employment or unemployment, X are the explanatory variables, D are dummies for regions and i and t represent regions and time. In the random effects model, represents the composite error term for regions, time and random effects. The two equations are then compared with a Hausman test. The shock is represented by two variables as part of X. One is the total number of Syrian refugees per year and the second is a categorical variable based on classes of density of refugees values. Therefore, these authors follow the linear estimation approach but instead of using the OLS+IV approach they opt to compare FE and RE models. This is also the only paper using a RE model. Akgündüz, van den Berg, and Hassink (2015) model the employment rate in Turkey with a DD equation: = + + + + Where is the employment rate, i=provinces or regions, t=time, T and R are time and region fixed effects. As for the price model of the same authors, the shock or treatment effect I is either a binary variable describing the presence of refugees, the number of refugees or the inverse sine function of the number of refugees. Therefore, the authors estimate both a DD estimator with the indicator variable and marginal changes with the continuous refugee variable. This is the only paper we found that follows this approach. 56 Fallah et al. (2018) study the impact of the Syrian refugees in Jordan on host communities covering wages, employment and unemployment. The model is a linear DD model and is the same for all these outcomes: = 0 + + + + ∗ + where Y is wage, unemployment or employment, i, t and l are individuals, time and locality respectively, S is the share of refugees in localities, X is a set of control variables, and is the DD coefficient of interest. In this case, the DD estimator includes a continuous variable rather than a dummy indicating the presence of refugees resulting as a sort of hybrid between the other DD estimators illustrated and the linear OLS+IV elasticities approaches although the variables are not in log form. The authors also instrument for the locality share of refugees based on the distance from the main refugee camp (Zaatari) and use a discrete- time hazard model to study duration of school to work transitions. This is another example of an instrument constructed with distances whereas the time hazard model is an innovation of this paper. Ruiz and Vargas-Silva (2015, 2016 study the employment impact of the Burundian and Rwandan refugees on Tanzania host communities. They estimate a linear probability model as = 0 + 1 + 2 + 3 + 4 () + 5 + where Y is employment or occupation status, i are individuals, t is time, and are individual and area fixed effects, X is a set of individual, household and regional controls and D is a measure of intensity of forced migration. This is the log of the inverse of the distance between host communities and the border with the countries of origin of refugees. Ruiz and Vargas-Silva (2015) use the same components to measure the shock as Ruiz and Vargas-Silva (2016) but as an additional measure add a more elaborate index measuring the distance between the host community and the refugee camps over time. The index is described as 13 1 ℎ = (∑ [( ) ( 13 )( )]) ⅆ, ∑1 14 =1 where d is the distance from the camp and pop is the population in camps where the total refugee population is the sum of the refugee population in the 13 camps considered. In this model, the refugee population and the distance from the refugee camps play an equally important role. This index is evidently suitable to study host populations who live around refugee camps rather than situations where the displaced people live among host communities. Ruiz and Vargas-Silva (2017) study the same caseload but focus on differential impacts on engaging in household chores, farming, and employment outside the household across gender and skill level. To proxy the refugee shock S for each household j, they use the distance (D) of each host community to each refugee camp r weighted by the peak population P of each camp without adding a time dimension. 13 = (∑ ) , =1 They also add distance to the border of Burundi, Rwanda and Uganda log(1/distance) as controls , , ⅆ to the following model: ,, = 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 ( ∗ ) + 9 ( ∗ ∗ ) + + Where the dependent variable is either a dummy with the value of 1 if individual i from household j engaged in a given task during the previous week or the number of hours the individual dedicated to the task. They include the household fixed effect and the month as a control to capture seasonal effects and 57 as a series of individual and household controls. The time dummy takes the value 1 for 2004, and the gender dummy if the individual is a woman. Note here the interest in employment at the extensive and intensive margin. Braun and Mahmoud (2014) studied the Germans that fled or were expelled from Eastern Europe after World War 2 using OLS and IV models. The OLS estimation is described as = + + + ⅆ + where y is the share of employed native males among all native males in occupation j and state i, m is the share of male expellees in the total male labor force in state-occupation cell ij, x is a vector of state specific control variables and d is a set of occupation dummies. The IV model instruments m exploiting regional variations in pre-war distribution of occupations and the distance of the expellees’ origin from West Germany. The instrument is described as follows 1939 ∑(̂ ∗ ) ̂ = 1939 1939 1939 ( ∗ ) + ∑(̂ ∗ ) where ̂ is the estimated number of expellees from a sending region s who have settled in state i in West 1939 1939 Germany, is the pre-war population size in a state and is the occupational structure before the war. Alix-Garcia et al. (2018) look at the impact of the Kakuma camp in Kenya on neighboring communities and model the probability of having a wage earner in the family and the wage earned during the previous 12 months with an OLS model. Alix-Garcia and Bartlett (2015) examine the impact of IDPs in the Darfur region of Sudan using a matching method comparing individuals in the treated town (those affected by internal displacement) with similar individuals in a non-treated town. They look at changes in employment between 2000 and 2010, which captures the period before and after the 2003-2009 conflict, therefore using a DD type of identification strategy. A variety of labor market outcomes are considered including the probability of becoming employed, unemployed, high and low skilled, and manual laborer and, vice-versa, the probability of abandoning manual labor or medium and high-skilled jobs. This is the only paper reviewed that uses a standard matching method on individuals. Finally, a few authors used time-series econometric models to assess the impact of forced displacement on the local labor market. Carrington and de Lima (1996), for example, looked at the returnees from Angola and Mozambique to Portugal using several time-series models in an effort to establish the relation between the immigration rate and the unemployment rate, the employment to population rate and wages. Fakih and Ibrahim (2016) studied the impact of Syrian refugees on the Jordanian labor market using a Vector Autoregressive Model (VAR) attempting to capture Granger causality. A VAR system is made of a set of time series stationary variables expressed as a linear function of their lags as follows: = 0 + 1 −1 + 2 −2 + ⋯ − + using unemployment, employment and labor force status as outcomes of interest. Bodvarsson, Van den Berg and Lewer (2008) develop a general equilibrium model for wages where the ceteris paribus effect of an immigration shock on native wages is the sum of a “consumer demand effect” and an “input substitution effect”. They test the model for retail wages in Miami after the Mariel Boatlift, using a simultaneous-equations regression model in which the independent variable (i.e. Cuban immigrant density in each city) affects the dependent variable (i.e. weighted average native retail wages ) through these two channels (i.e. the weighted average immigrant wage and the retail sales per capita ). It consists of the aggregate equation: = 0 + 1 ( ) + 2 () + 3 () + 58 and the two channel equations = 0 + 1 ( ) + 2 () + = 0 + 1 ( ) + 2 () + Each equation has an additional vector of controls (i.e. , ⅆ ). They apply 3SLS to estimate the equations. 59 Annex 2 – Empirical Results by Crisis Although many of the models reviewed share similarities, none of the results arising from these models are entirely comparable, either because of differences in the structure of the models or because of data, estimation techniques, instruments used or crisis treated. This makes the comparative review of empirical results a daunting task. We opted to proceed in two steps. In this annex, we provide an overview of results organized by crisis on the ground that this is the most defining feature that could explain differences across results. In particular, the level of economic development of host countries, the absolute and relative scale of the crises and the timing of the inflow are very important factors in determining outcomes. We therefore group results by crisis and also by level of economic development of host countries. In the text, we instead group all results from all papers and provide a meta-analysis. 2.1 High Income countries Expellees from Eastern Europe to West Germany after World War II26 An estimated 12m Germans fled or were expelled from Eastern Europe between the last stages of World War 2 and 1950. This mass migration increased the population of West Germany from 39m in 1939 to 48m in 1950 and was seen by hosts as a major threath to their well-being during the difficult post-war reconstruction period. Expellees were close substitutes to workers from West Germany across the entire skill distribution, being German native speakers and having very similar education levels. The impact of this group of displaced people on their hosts has been recently studied in a few papers. Braun and Mahmoud (2014) focus on male employment and find that a 10 percentage point increase in the share of German expellees is associated with a reduction in the native employment rate by 2.6 percentage points with an OLS estimation and by 4 percentage points with an IV estimation. These effects are mainly driven by labor market segments (defined by occupation and states) that experienced very high inflows of expellees and they are found to subside in later periods with the percentage change in the employment rate declining to 1.7 by 1953 from 3.8 percentage points in 1950. Based on their findings, the authors conclude that regions and occupations that experience relatively small inflows of refugees or migrants should show only small or no employment effects and that these effects should be expected to be short lived. The shortage of physical capital in West Germany after the war also limited the absorption capacity of the labor market. Braun and Weber (2016) measure longer-term dynamic effects on employment and labor income until 1970 using a dynamic structural search and matching model that accounts for movements between regional labor markets. They show that it took regional labor markets at least a decade to adjust but that the expected discounted lifetime income of native workers declined by only 1.38 percent. Besides unemployment benefits, this was due to internal migration, which played an important role in diffusing the impacts over time. The inflow also seems to have contributed to sectoral change away from agriculture and thus to an increase in output per worker (between-sector effect), but to a decline in the output per worker within sectors (Braun and Kvasnicka 2014). Other papers also look at the longer-term impact of this inflow, exploiting the fact that the expellees were unevenly distributed across Germany based on available housing and that they were not allowed to settle in the French occupation zone. They find that the inflow had a positive impact on population growth and agglomeration that can still be measured several decades later (Wyrwich 2018; Schumann 2014), leading to increases in manufacturing employment and income per capita (Peters 2017). Braun, Kramer and Kvasnicka (2017) do, however, find that this effect is only persistent within large local labor markets. 26This case is covered among high-income countries as West Germany was the receiving country but the income per capita in West Germany after the war may well have been considered as low or medium. We opted to keep it in this group as there was no optimal choice. 60 Ethnic Germans from Eastern Europe and Former Soviet Union to Germany after 1987 Ethnic Germans arriving in Germany with the lifting of emigration restrictions in central and eastern Europe and the Former Soviet Union (FSU) after 1987 did not face legal barriers on the labor market, as they were granted German citizenship upon arrival. Glitz (2012) exploits an exogeneous placement policy by the government which did not take labor market needs or skill composition into account. Looking at annual effects in the period between 1996 and 2001, the author finds a positive and significant short-run displacement effect but no conclusive effect on relative wages between OLS and IV results. The lack of short-term wage effects in the IV results might be explained by Germany’s strong union coverage at the time. Repatriates and escapees from Algeria to France In the aftermath of Algerian independence in 1962, up to 900,000 people fled to France including 750,000 French nationals living in Algeria, about 100,000 naturalized Jews and several thousands pro-French Algerians who fought against the independentists. The French and Algerian caseloads were different in size but also in education level with the Algerians having the quasi totality of adults with less than primary education and the French having only about a quarter with this level of education. Both groups mostly settled in the Southern municipalities of France along the Mediterrean coast. A number of measures facilitated the integration of French repatriates into the labor market, including special benefits for up to a year to support the job search, a lump sum towards housing, and priority for certain jobs. Nevertheless, they had higher unemployment rates than natives in 1968 (Hunt 1992). Using the 1962 and 1968 French censuses, Hunt (1992) studied this case focusing on French nationals and found a positive and significant small effect on the average locals’ unemployment rate and a negative and weakly significant effect on salaries at the national level, but does not differentiate between skill groups. Borjas and Monras included this caseload in their 2017 study distinguishing between French repatriates and Algerian emigrants. They find that the Algerian low skilled emigrants had a positive and significant effect on the unemployment rate and a negative and significant effect on the employment rate of male locals whereas the repatriation of French nationals had no effect on either employment or unemployment in the OLS regressions but a weak statistically significant positive effect on unemploymenet in their IV estimates. They also find that, for unemployment, these results are mainly driven by low skilled workers (less than primary education) whereas the impact on the employment rate affects all skills groups. They do not find any beneficial complementarities of skilled French natives with the low-skilled Algerians. These two studies overlap only in relation to French repatriates and only unemployment and the results can be considered similar, even if the effect estimated is conceptually slightly different, as Clemens and Hunt (2017) note. Clemens and Hunt (2017) confirm the results of Borjas and Monras for Algerian nationals. We are not aware of any paper that contributes to the interpretation of these results. One of the reasons for the weak impact might have been the economic boom in France during the time of the influx and strong labor demand, as reflected by other important labor immigration trends happening around the same time. Hunt (1992), for example, shows that even if internal migration to areas with many repatriates might have decreased, this was offset by increased international migration to these same areas. Cuban refugees in Miami From May 1980 to June 1981 an estimated 120-126,000 Cubans arrived in Miami on boats as Castro suddenly allowed Cubans to leave the country from the port of Mariel on April 20, 1980. Approximately half of this population settled in Miami contributing to increase the labor force by about 7 percent (Card, 1990). Based on the Cuban Adjustment Act of 1966 and The Immigration and Nationality Act Amendments of 1976, Cubans were given refugee status, they were allowed to work and could also be granted permanent residency after one year. The high number of previous Cuban and other Hispanic immigrants meant that local networks were available to help with job seeking and that language was not an obstacle to work. 61 Card (1990) was the first to study the labor market impact of this caseload of refugees and concluded that there was no significant impact on the employment to population ratio, the unemployment rate or hourly wages for any population group including Whites, Blacks, Cubans or other Hispanics, and low skilled. Looking closer at these results one can observe differences for wages, employment and unemployment between Miami and comparison cities for selected years but these effects are not consistent in sign during the post-shock period with no clear trend. 27 Angrist and Krueger (1999) expand these results using a Difference-in Differences approach whereas Peri and Yasenov (2017) use a synthetic control group to improve on the matching of the control group. Both papers find no effects on any of the labor market outcomes considered. Using the same four comparison cities and time frame (1979-1985) as Card (1990), Bodvarsson, Van den Berg, and Lewer (2008) find that the net effect of Cuban immigration on native wages in the retail industry is positive and significant on average for whites, and positive but non-significant for blacks and Hispanics. Borjas and Monras (2017) analyse the same crisis and, similary to previous studies find no significant effect on employment or unemployment. However, both Borjas (2017) and Borjas and Monras (2017) find a negative and significant impact on relative wages for a certain subroup of low skilled individuals defined as non-Hispanic male high-school dropouts aged 25-59 who have worked and received wages, in sharp contrast with high school graduates that show a positive and significant gain. Anastasopoulos et al. (2018) complement this work by showing that the number of help-wanted ads published in local newspapers, which are most strongly correlated with local labor market conditions for high-school dropouts, decreased in Miami until the end of 1982 relative to different alternative control cities. The contrast between Borjas (2017) study and other studies in relation to low skilled local residents has been the object of debate. A closer look at these papers and more recent work shows that this is due to the different definitions of subgroups considered: different authors look at different subgroups of the low- skilled population in terms of the age range, sex (only men or men and women), and race (Blacks, non- Hispanic or non-Cuban). Card first makes and analysis subdividing Whites, Blacks, Cubans and other Hispanics and then focuses on Blacks low and high skilled whereas Borjas (2017) and Borjas and Monras (2017) consider all male non-Hispanic natives together and focus on desegregating skills elevels and Peri and Yasenov focus on non-Cuban men and women with no high chool degree between 19 and 65 years old. Card also divides all non-Cubans in Miami by predicted wage quartiles but finds no evidence of a decline in the wage of workers in the lowest quartile compared to workers in the upper quartile Peri and Yasenov (2017) showed that nearly all other sub-groups of individuals with less than high school education perform better than the sample selected by Borjas. They also argue that Borjas’ results are not robust because the sample size for this subgroup is very small and the measurement error is sizable. Borjas (2017), however, argues that including other groups in the sample ignores the changing composition of the workforce which occurred at the same time, as new Hispanic immigrants arrived after 1980 and an increasing number of women entered the workforce in 1980s. Clemens and Hunt (2017) criticize Borjas population selection and the separation of those who never finished high school from those with only high school degree because the sample shifted to include substantially more black male workers with relatively low wages. They show that this shift fully accounts for the decline in wages as found by by Borjas. Considering the different categorization used, Card (1990), Angrist and Krueger (1999), Peri and Yanesov (2017), Borjas (2017) and Borjas and Monras (2017) results are in fact rather similar. All authors agree that the Mariel boatlift to Miami in 1980 had no effect on employment and unemployment of natives overall or for different sugroups of natives. They also agree that there are no clear effects on wages overall and they also do not find negative impacts on low-skilled workers overall. The apparent discrepancy occurs only in relation to wages of a specific subset of low-skilled workers, for which Borjas (2017) and Borjas and Monras (2017) find a negative and significant effect on wages. Therefore, the negative effect of the refugee 27 This is visible if one calculates the differences between Miami and comparison cities in Tables 3 and 4 (not shown in the paper). 62 crisis, if any, is limited to a subset of low skilled workers, a result that Peri and Yanesov (2017) and Clemens and Hunt (2017) also disputed. Interstingly, all these papers ignore the question of household well-being. Whether the decrease in wages for a subset of low skilled workers ultimately results in a decrease or increase in average household well- being depends on production and productivity. Therefore, these results do not provide any evidence on whether living standards of the local residents of Miami have improved or not in the aftermath of the Mariel boatlift. This has been an important gap in this specific debate. The only paper that looks at outcomes beyond the labor market is Saiz (2003), who finds that rental prices increased in the short and medium run (1979-1983) by between 7 and 11 percent more in Miami than in comparison municipalities. These findings were limited to lower-quality housing and show that demand for this type of housing remained strong over the period. The literature also offers some explanations for the lack of impact of the Mariel boatlift on the labor outcomes of the host community. The structure of the industry in Miami (notably a relatively high share of textile and apparel industry) offered jobs for low skilled non-English speaking labor (Card 1990). Using a confidential micro data version of the Annual Surveys of Manufacturers, Lewis (2004) finds little evidence that these industries expanded their production of labor intensive goods relative to other productions (like a Heckscher-Ohlin open economy model would have predicted). However, he also shows that these industries opted for labor intensive technologies in the aftermath of the crisis and adopted computers more slowly than industries in comparison cities expanding employment at the extensive rather than intensive margin. The new arrival of Cubans also helped to compensate for a net decline in internal migration to Miami (Card 1990) whereas the overall increase in population increased local consumer demand, which in turn increased labor demand. Bodvarsson, Van den Berg, and Lewer (2008) find that the positive impact of Cuban immigration on retail sales per capita (demand effect) outweighed the negative impact of the immigrant wage on native wages (substitution effect on the labor market). Former Soviet Union (FSU) escapees to Israel In the aftermath of the fall of the Berlin wall in 1989, the lifting of emigration restrictions in the Soviet Union and the desengregation of the Soviet Union in 1991, many Jews from the Former Soviet Republics migrated to Israel. It is estimated that between 1989 and 1995, 610,100 immigrants arrived in Israel from the Former Soviet Union increasing the size of the population by 13.6%, with nearly half of this population having tertiary education and significant work experience (Borjas and Monras, 2017). Alhtough these immigrants were not classified as refugees after 1989, the sudden nature of this phenomenon with the unexpected desegregation of the Soviet Union in 1991, the substantial push and pull factors at play and the scale of the migration flow made this crisis a relevant case-study for several of the scholars working on forced displacement. Friedberg (2001) was the first to assess the impact of this migration flow on the labor market of local residents. The paper finds a negative impact on wages and a non significant impact on employment levels. However, when using an Instrumental Variable approach (IV) and subdividing the workforce into occupational groups, the author finds no evidence that the influx of FSU citizens has adversely affected the wage growth of local workers. When a distinction is made between high skilled and low skilled workers, it is found that high skilled workers’ wages gained whereas the effect on low skilled workers and overall employment was non significant. Cohen-Goldner and Paserman (2011) also study FSU migrants to Israel and find a significant negative effect on wages in the short-term but not in the long-term. A 10 percent increase in the share of immigrants lowers native’s wages in the short-run by 1-3 percent, an effect that disappears after 4-7 years. The short- term effect is explained by the impact on low skilled blue collars whereas there are no short or long-term effects on high skilled white-collar workers. These effects are also found to be similar for men and women. The differences in results between Friedberg (2001) and Cohen-Goldner and Paserman (2011) can be 63 explained by the time frame studied: Friedberg looks at results after five years, and Cohen-Goldner and Paserman show that after four years the effect is close to zero. By contrast, Borjas and Monras (2017) focus on earnings and find a negative and significant effect overall explained by the the very high skilled (university completed education) whereas the low skilled (less than primary education) are found to benefit from the influx of immigrants. Friedberg (2001) and Borjas and Monras (2017) in particular have clearly opposite results in relation to high and low skilled workers although the two papers are not entirely comparable.28 A study that attempted to understand these differences (Clemens and Hunt, 2017) showed that the difference between the two papers can be explained by the difference in the specification of the instrument used. Using a placebo approach, these authors show that Borjas and Monras IV results could be reproduced with a placebo instrument whereas Friedberg results could not, providing some evidence in favour of the latter paper. However, both Borjas and Monras (2017) and Cohen-Goldner and Paserman (2011) criticize Friedberg’s instrument on the ground that the occupation status in the FSU might only be weakly correlated with the actual occupation in Israel due to occupational downgrading (see also Eckstein and Weiss 2004 on this point). Overall, the evidence presented in this section remains inconclusive. The three papers reviewed are not entirely comparable in terms of population groups and time-frame and they all reach different conclusions on high and low skilled workers. There is also no agreement on the optimal instrument to use. There are no discrepancies among these papers on the impact of the migration flow on native employment but, using a general equilibrium model, Hercowitz and Yashiv (2002) find a negative impact on native employment questioning whether partial equilibrium models are suitable for studying the overall impact on employment. Results on prices are also inconclusive. Using data at the national level until 1999, Hercowitz and Yashiv (2002) find a negative impact on the relative price of imports lagged by 4-5 quarters, but results for other quarters are not significant. Exploiting city level variation in inflows and using a simple difference- approach with data from 1990, Lach (2007) finds a negative impact on prices in the short-run, which does not seem to be caused by the increased number of consumers (size effect) but by the high price elasticity and low search costs of this new group of consumers (composition effect). Palestinians from West Bank to Israel Two papers looked at the impact of the second Intifada of 2000 and the sudden inability of Palestinian workers to commute to their jobs in Israel. Asali (2013) studies the impacts of the sudden drop in labor supply in Israel. He finds no effect on the employment and wages of unskilled Israeli Jewish workers, positive effects on Israeli Arab workers with less than primary school mostly in the short-run, and negative effects on Israeli Arab workers with middle or high school.29 Mansour (2010) studies the impacts of the increased supply of Palestinian workers on the labor market in the West Bank. He finds that an increase in the supply of low- and high-skilled workers both decrease the wages and increase unemployment of low- skilled workers but has no significant effects on high-skilled workers. Refugees from Former Yugoslavia to the EU After the fall of the Berlin wall in 1989 and the collapse of the Soviet Union in 1991, Yugoslavia split into the five costituent republics in 1991 and 1992 leading to a series of conflicts lasting a decade and generating 28 Both papers use an OLS and IV approach but using different observational units and equations (see emprirical modelling part). Results are similar in the two papers for the OLS approach but Friedberg shows that this approach is biased (by immigrants entering occupations with low wages and low wage growth) whereas Borjas and Monras (2017) trust both OLS and IV approaches and further subvide the unit of observation by education groups. Therefore, equation specifications, unit of observation and instruments are different. Yet, both authors consider occupations and different skills levels and they reach opposite conclusions about high and low skilled workers. 29 This is an interesting counterfactual to the increase in labor supply studied by the other papers in this review. As the results are not directly comparable, they are, however, not included in the dataset used for the meta-analysis. 64 outflows of refugees who mostly settled in selected European countries. It is uncertain how many refugees the Balkan wars generated but using census data from seven European countries Borjas and Monras (2017) find about 259,000 people who were born in former Yugoslavia and moved to Europe during the decade. All considered, this is not a massive inflow of people but certain municipalities had a sizable increase in labor supply. The refugees had similar levels of education compared to the native population, but were faced with a lack of language skills.30 They were offered different types of residence status by the different host countries but most of them were allowed to work. Angrist and Kugler (2003) studied this caseload and, measuring aggregate differences across 18 European countries, find a negative and significant effect on employment with both an OLS and an IV approach. An increase in the share of immigrants by 10 percent reduces native employment rates by 0.2-0.7 of a percentage point with men and younger workers being the most affected groups.31 These effects tend to be accentuated in localities with more rigid labor and product markets, weak institutions and stagnant labor markets. Borjas and Monras (2017) also cover refugees from Former Yugoslavia in seven of the 18 European countries studied by Angrist and Kugler (2003) and, differentating by education level and within- country regions, find a significant positive effect on the unemployment rate of the locals with an OLS estimation and no significant effect with an IV estimation and a non significant effect on the employment rate. The results of these two papers are not comparable because of the models used, categorizations of variables and sample covered and they also provide different results on employment. Clemens and Hunt (2017), carrying out a Kronmal correction on the instrument used by Borjas and Monras (2017), find a a statistically insignificant effect on unemployment. Refugees in Denmark Foged and Peri (2015) studied the inflow of refugees from conflict areas to Denmark between 1991 and 2008 accounting for up to 4.7 percent of the labor force in 2008. The paper exploits labor market administrative data following individuals continuously over time in a panel setting and a refugee dispersal policy that allocated refugees across the country between 1986 and 1998. Using OLS and 2SLS models, the paper finds that immigration increases the complexity and wages of jobs for the low-skilled natives. An increase in refugee-country immigrants by 1 percentage point increases the complexity of native jobs between 1.3 and 3.1 percent and wages by 1-1.8 percent. The authors interpret these results as immigrants pushing natives to more complex and better paid occupations either within the same establishments or by migrating to other establishments. They also find that total labor supply of natives either increases or is stable. These effects are not very large when compared to the overall changes of these parameters over the period considered but they show complementarities rather than competition between immigrants and natives. Similar effects are observed for high-skilled natives. In this case, the effects are smaller in terms of occupational complexity but larger for wages. These results are also supported by the DD cohort- municipality model proposed by the same authors. Those cohorts living in municipalities with higher immigration experience a larger shift towards more complex occupations and better wages with these effects persisting in the short and long-run. Young and low-tenure, low-skilled natives are also shown to respond to immigration with stronger transitions towards higher occupational complexity and better wages with no negative effect on employment supply. Refugees in the United States Similar to Foged and Peri (2015), Mayda et al. (2017) look at the long-term labor-market impacts of refugee inflows and exploit a refugee placement policy in the country of destination. They study the impact of resettled refugees on natives’ wages and employment in commuter zones in the U.S. over three decades, between 1980 and 2010. They find very small, mostly insignificant effects on wages and employment on 30In the countries analyzed by Borjas and Monras (2017) they were disproportionately middle-skilled (secondary education). 31 Note that, in a country where 5 percent of the labor force is foreign, a 0.5 reduction of a percentage point implies 83 native workers losing their jobs for every 100 immigrant workers finding a job. Therefore, at the higher end of this estimations, the rate of substitution between native and foreign workers is around one to one, a very large effect on a per capita basis. 65 average for both low skilled and high skilled native workers. The point estimates do not vary much between the full control sample and the matched control samples. The reduced form and 2SLS regressions are also robust to broader or narrower definitions of the treatment shock (i.e. commuter zones with refugee inflows larger than 0.05% or 0.2% of the population) and the omission of some controls. They note that the average refugee inflow was not large, and that the different skill-set of resettled refugees compared to natives might have led to a high complementarity between the two groups. 2.2 Middle-income countries Returnees from Angola and Mozambique to Portugal After the 1974 military coup, the newly installed government of Portugal granted independence to Angola and Mozambique generating a flow of “retornados” (returnees), people of Portuguese or European descent who felt unsecure in the former colonies and decided to return to Portugal. Prior to 1974, the population of Portugal was decreasing due to emigration whereas it grew by 5 percent per year in 1974 and 1975. Estimates of the total number of returnees during the period vary between 0.5 and 1m people but many returnees moved on to other countries. According to the census, there were about 0.5m returnees in 1981, predominantly working age males formerly engaged in service activities, and relatively well educated as compared to the local population. Together with returning soldiers, they increased the Portuguese labor force by over 15 percent in three years.32 Most of them were native Portuguese speakers. They arrived with few resources but received benefits, including cash subsidies, which made up about 11 percent of total government spending at the time (Makala 2017). Compared to France, who received a comparable returnee flow from Algeria, Portugal was less developed at the time, and the economy was in a downturn. Two published studies looked at the impact of these returnees on the local labor market. Carrington and de Lima (1996) used time series econometrics to observe the evolution of the Portuguese labor market before and after 1974. They find a sharp deterioration of labor market indicators after 1974. However, the immigration rate had no impact on the unemployment rate, the employment to population rate or wages except for a small one year lagged effect on the unemployment rate and real wages. Interestingly, these results hold whether the authors control for Portuguese macroeconomic indicators or the Spanish labor market, which had similar characteristics to the Portuguese labor market in the 1970s. Using a separate longitudinal model and focusing on the construction sector, the same authors find a large and significant effect on earnings in this sector. However, the authors’ challenge their own results considering the unobserved heterogeneity that affects the latter model leaving conclusions somewhat open to interpretation. They also argue that the persistence of the effects raises question if the returnees were the cause. Makela (2017) uses a SCM approach based on comparator countries (see empirical models section) to estimate the impact of the returnees on average annual productivity, wages and the unemployment rate. The author finds a significant negative effect of immigration on all three outcomes. The estimate impact on productivity is around 26 percent in the five years after the shock, the one on wages is from 8 up to 55 percent from 1977 to 1985 whereas the unemployment rate rises by about 2.3 percentage points between 1975 and 1980 but declines by a similar amount between 1980 and 1985. The author also combines the SCM approach with a Difference approach working with Portuguese regions rather than countries and focusing on the agriculture and construction sectors. In addition, a fixed effects and and a generalized synthetic control method are used as robustness tests. All these estimations are consistent in finding a significant negative effect on agricultural and construction wages. On average, a one percentage increase in the returnees population share leads to a decrease in wages of 4.13-9.53 percent in these sectors. 32It should also be noted that the emigration of Portuguese guest workers ended in 1973 with the oil crisis. This added to the labor supply shock caused by the returnees from the former colonies. 66 In summary, the two available studies for the 1974 returnees to Portugal are rather consistent in finding a significant negative effect on wages in the years following 1974 for the agriculture and construction sectors whereas the evidence on employment and unemployment is non conclusive. IDPs in Colombia Colombia has a long history of internal violence that claimed hundreds of thousands of lives since the late 1950s. Such violence has been mainly linked to the emergence of powerful revolutionary groups including the Revolutionary Armed Forces of Colombia (FARC) and the National Liberation Army (ELN) and to paramilitary groups that initially emerged to contrast these revolutionary groups. In the 1980s, internal violence intensified due to the expansionary ambitions of the revolutionary groups that led to a civil war against the state and the increasing violence perpetrated by military and paramilitary groups. As a consequence of this violence, many civilians who had been caught in the fighting were forced to flee. The conflict affected mostly the North-East of Colombia and almost five million people have been estimated to have fled this area since the early 1980s. These internally displaced persons were mostly from rural areas and settled mostly in urban areas and had a level of education comparable with low-skilled workers in urban areas. Calderon-Mejia and Ibanez (2016) use household survey data and an IV approach to assess the impact of IDPs on the hourly wages of host communities. They find that a 10 percent increase in the share of IDPs reduces hourly wages by 0.88% with this effect being larger for women as compared to men. The effect is smaller (0.63%) but still negative and significant for manual male workers and management and professional female workers (0.64%) whereas is non-significant for female manual labor and male management/professional labor. The most affected workers are independent/self-employed workers with females (2.28%) suffering more than males (1.31%), particularly those with high school education or less (2.0%). Morales (2017) use survey, census and registry data covering the 1993-2005 period to study the same caseload of IDPs in Colombia. Short-run effects indicate a negative impact on wages. A one percent increase in population due to IDPs results in a 1.4 % reduction in local wages with this effect being larger for women, particularly low skilled women (2.2%). The effect is also negative and significant for high skilled men but smaller (0.7%). In this case, OLS and IV estimations concord with IV estimations showing larger effects. The long-run effects show instead a positive effect with OLS estimations and a negative effect for low-skilled women and no effects for other groups with IV estimations. There is therefore some evidence that the negative effects tend to disappear in the long-run but not for all groups. This study is also one of the few studies that considers the potential impact on outmigration. It finds that an increase of 1% in population due to IDPs generates an outmigration of 0.2-0.3 people per 100 residents. The two studies reviewed on labor market impacts in Colombia are therefore very consistent in finding a negative effect on wages with the effect being larger for women and low skilled workers, notably in the informal sector. This effect also seems to be attenuated in the long-run, possibly due to outmigration and other labor market adjustments. Using administrative panel data on quarterly IDP flows and rental prices by income level between 1999- 2014 for 13 cities in Colombia (which received 66% of all IDPs), Depetris-Chauvin’s and Santos’ (2018) OLS estimates show a significant positive impact of IDP inflows on average rental prices for low- and middle-income housing, which may last up to 10 quarters. Results for high-income housing are non- significant. Their IV approach does not show statistically significant impacts on rental prices on average; but rental prices for low-income housing increase while they decrease for high-income housing. This is one of the few papers that explores potential channels through which the IDP inflow might impact the variable of interest. They provide evidence that the heterogeneous impact on rental prices might be due to an increase in supply of high-income housing, as licenses for new non-social interest housing increase while those for social interest housing decrease, and wages in the construction sector decrease, reducing construction costs. 67 At the same time, they find evidence of a large housing deficit in the low-income areas of the host cities. The second channel might be an increase in crime associated with the inflows of IDPs, measured by the homicide rate in the host cities, which has a negative impact on high-income rental prices. In a separate paper from Depetris-Chauvin and Santos (2017), the authors’ OLS and IV estimates indicate that the inflow of IDPs decreases food prices, although the authors express concern about reverse causality, as lower food prices might increase violence and internal displacement. They also find some evidence of a decrease in per capita consumption expenditures. This might be the result of increased rental prices and negative impacts on labor market outcomes. Syrian refugees in Turkey and Jordan Following a stream of mass protests in central Syria and the subsequent crackdown on the part of the Syrian authorities in the spring of 2011, Syria slid into a complex civil war that is still raging at the time of writing in 2018. Within three years from the beginning of the conflict over five million Syrians had left the country most of which settled as refugees in neighboring countries including Jordan, Lebanon, Turkey and Iraq. The bulk of the exodus occurred between 2011 and 2013 resulting in approximately 0.25 m refugees in Iraq, 0.67m in Jordan, 0.95 m in Lebanon and 3.6 m in Turkey.33 It is also estimated than an additional 6 m people have been displaced within Syria. Overall, more than 10 m people – about half of the pre-conflict Syrian population - has been displaced during the conflict thus far. The impact of refugees on host communities has been studied mainly for Turkey in a string of studies that focused on labor market outcomes. Akgündüz et al. (2015) study the influx of Syrian refugees in South- East Turkey and find a non-significant effect on employment whether employment is broken down by region, province or skills’ level. Del Carpio and Wagner (2015) using the Turkish Labor Force Survey (LFS) and an IV approach find a negative and significant impact on local employment in the informal sector but a positive and significant impact on the formal sector. Similarly, Ceritoglu et al. (2017) and Tumen (2016) find Syrian refugees to have a positive and significant effect on formal employment explained by the performance of older workers and a negative effect on informal employment explained by a negative performance of younger workers. Using a regional panel data set for the period 2004-2016, Esen and Binatli (2017) is the only study that finds an increase in unemployment and a decrease not only in informal but also in formal employment as a result of the refugee influx. The results of Akgündüz and Torun (2018) suggest that the refugee inflow led to occupational upgrading of natives, as their task complexity increased, particularly for medium-skilled natives. Three studies on Turkey also looked at the impact on prices. Akgündüz et al. (2015) find a positive and significant effect on food and housing prices and a non-significant effect on hospitality prices. Balkan and Tumen (2016) find instead that prices have declined as a result of the refugee influx due to an increase in cheap labor supply particularly in the informal sector. Using a simple DD approach, Balkan et al. (2018) find that housing rents increased in the range of an additional 3.5-5.5 percent in refugee receiving regions in the short run (2012-2013) compared to control regions, as housing supply is inelastic in the short run. The effect is negative and statistically insignificant for below-median rents, but positive and statistically significant for high-rent housing. They interpret this as a sign of residential segregation, with natives moving out of lower-priced neighborhoods where refugees settled. The only other country affected by the Syrian crisis where studies are available is Jordan. Fallah et al. (2018) look at Syrian refugees in Jordan and find that locals have not experienced negative labor market outcomes if one considers labor market participation, employment, employment by type or wages. They find no difference between the labor market outcomes of locals living in areas with a high share of refugees and those who don’t. Fakih and Ibrahim (2016) look at Syrian refugees in three governorates of Jordan and use a longitudinal vector autoregressive model (VAR) to assess the impact of refugees flows on local employment. Similarly to Fallah et al. (2018) they find no correlation between refugee flows and local 33 Based on UNHCR data as of January 2019 (https://data2.unhcr.org/en/situations/syria). 68 trends in employment, unemployment or labor force participation. On the other hand, results by Malaeb and Wahba (2018) show that previous immigrants to Jordan were more likely to work informally, work less hours and had lower wages after the influx of Syrian refugees. In the Jordan’s segregated labor market, Syrian refugees seem to be closer substitutes to immigrants than to natives. El-Mallakh and Wahba (2018), however, show that the probability of Jordanians migrating out of the regions with higher numbers of Syrian refugees increased. While confirming no significant effect on labor market outcomes of salaried Jordanians, Rozo and Sviastchi (2018) find negative effects of refugee exposure on self-employment. Two papers looked at the impact of Syrian refugees on prices in Jordan. Alhawarin et al. (2018), finds no evidence of impacts on predicted rental prices on average. Depending on the dataset used, the impact on rental prices in regions closer to the Syrian border was insignificant or positive, and in regions distant from the border insignificant or negative. They do, however, find a negative and significant impact on a housing quality index. Rozo and Sviatschi (2018) confirm these results, finding larger expenditures on housing for individuals living closer to refugee camps, at the expense of other types of expenditures, and higher rental and property income in these areas. These latter authors do not find any evidence of impacts on overall consumption expenditures. Overall, the evidence on the Syrian crisis points to no visible overall effects on the labor market in neighboring countries although there is evidence of competition between former immigrants and informal workers on the one side and the new wave of Syrian immigrants on the other side. Most studies show a positive impact on formal employment of natives in Turkey, also providing evidence for professional upgrading. The absorption of Syrian refugees in the Turkish labor market seems facilitated by an increase in the number of new Syrian-owned businesses (Altindag, Bakis and Rozo 2018; Akgündüz, van den Berg, and Hassink 2018) with firms substituting capital with Syrian workers (Akgündüz and Torun 2018). Results on prices in Turkey remain unclear with different studies showing positive, negative or non-significant results. Only in Jordan there is some evidence for increases in rental prices in areas closer to refugee camps. Ethnic Greeks from Turkey to Greece After the Greco-Turkish war of 1919–1922, 1.2 million Greek Orthodox were forcibly resettled from Turkey to Greece, increasing the Greek population by more than 20% within a few months. Murard and Sakalli (2018) look at the impact of this resettlement on local municipalities almost 100 years after the event. They find that localities with a greater share of refugees in 1923 have today higher earnings, higher levels of household wealth, greater educational attainment and larger financial and manufacturing sectors. These results are similar to the positive long-term effects due to agglomeration economies found for German expellees (Braun and Kvasnicka 2014; Schumann 2014; Wyrwich 2018) and for the forced population relocation in rural areas within Finland after World War II (Sarvimäki 2011). 2.3 Low income countries Burundian and Rwandan refugees in Tanzania Following the assassination of the Burundian president in 1993 and the Rwandan genocide in 1994, a large number of Burundian and Rwandan refugees settled in the Kagera region of neighboring Tanzania. By 1995, this region of 1.5 m inhabitants hosted about 0.7 m refugees from these two countries alone. This is the largest crisis in terms of incidence of refugees over the host population and also the crisis that has the largest number of empirical studies in Sub-Saharan Africa thanks to the considerable number of household surveys that have been conducted in the affected region over the years. Alix-Garcia and Saah (2009) was one of the first studies that looked at the impact of the Burundian and Rwandan refugees in Tanzania using a mix of USAID, WFP and DHS data sets on food aid, prices and household assets. They find positive effects on prices of non-aid foods and smaller but positive effects on aid-related food items. This is one of the few studies that looked at both aid and refugee effects separately finding that the aid effect is considerably smaller than the refugee effect. The authors also find a positive 69 and significant household wealth effect (measured in terms of household assets) for rural households living close to refugee camps and negative wealth effects for households in urban areas. Thanks to the World Bank Kagera Health and Development panel Survey (KHDS) a number of studies were able to assess the impact of these refugees on host communities focusing mainly on household well- being and labor markets outcomes. Maystadt and Duranton (2018) find that all types of local workers gained from the refugee presence, although the positive effects are weaker for agricultural workers and the self- employed in non-agricultural activities. They find a positive and significant effect on household consumption in 9 of the 16 models’ specifications they propose with the rest of the results being all positive and non-significant. Maystadt and Verwimp (2014) use six different specifications of a similar model and find four of these specifications with a positive and significant effect on consumption with the remaining two specifications being positive and non-significant. They also show a differentiated impact between agricultural laborers and self-employed with the former suffering from high prices and competition on jobs and the latter benefitting from higher prices and cheap labor. Overall, doubling the refugee presence is found to increase per adult equivalent consumption of host households by 6-8 percent. The effect is also positive but lower (2-3 percent) for agricultural workers and self-employed in non-agricultural activities. Using the same data, Ruiz and Vargas-Silva (2016) find that the forced migration shock led to an increase in the likelihood of Tanzanians working outside the household as caretakers and a lower likelihood of working outside the household as employees. This is particularly true for agricultural employees suggesting a certain substitution effect between locals and refugees. Ruiz and Vargas-Silva (2015) complement these results by finding that the general impact on employees is negative and significant but the impact on professionals and government employees is positive and significant. Looking closer at the relative differential impacts by gender and skill-level, Ruiz and Vargas-Silva (2017) find that, on average, the refugee influx led to women being less likely to engage in employment outside the household and to work less hours outside the household relative to men. The results seem to be driven by those of 30 years of age or younger. Women who were literate and had basic math skills, however, were more likely to engage in outside employment. The authors suggest that the channel for this impact is that the refugee influx increased the availability of cheap domestic workers. Overall, findings for the Kagera region of Tanzania show that the employment effects tend to be positive for formal high skilled workers and negative for informal unskilled workers. Unlike other studies on high- and middle-income countries these studies also looked at household well-being and generally find a positive impact, particularly for self-employed and residents close to camps. Refugees in Kenya (Turkana) The Kakuma refugee camp in Kenya was initially established in 1991 to host refugee children fleeing Somalia. Its population grew steadily over the years due to conflict in neighboring countries and by 2016 the camp hosted more than 180,000 refugees, one of the largest refugee camps that ever existed. One study (Alix-Garcia et al., 2018) focused on the impact of the Kakuma refugee camp on neighboring communities using night lights as a measure of economic activity and distance from the camp as identification strategy. The study finds that the Kakuma refugee camp increases economic activity of neighboring villages. A linear DD estimation finds that a one percent increase in the distance to the Kakuma refugee camp (∼ 1.2 km) at the mean level of refugee inflows (∼ 69,000 refugees) results in a 1.8 percent reduction in the nighttime lights index. For the sample of villages with a population of 5,000 or more in 1989, a one percent increase in the distance to Kakuma at the mean level of refugee inflows is associated with a 2.3 percent reduction in the nighttime lights index. Using a Tobit specification, the authors find that a 10% increase in distance from Kakuma (12 km) at the average population of refugees corresponds to a 3.3 to 4.2 percentage point reduction in the probability of observing any nighttime lights. In essence, the closer a village is to the refugee camp, the higher is the nighttime luminosity. The authors are also able to transform these estimates in the impact on household consumption. If a 10% increase in 70 refugee population is associated with a 3.7% increase in the luminosity index within 10 km of the camp, the equivalent effect on consumption is approximately (0.015 × 0.037) *100 = 5.5%. Therefore, the paper provides rather strong evidence that vicinity to the camp increases economic activity and household consumption for local residents. These findings are also consistent with those outlined above for the Kagera region of Tanzania. Congolese refugees in Uganda and Rwanda The Democratic Republic of Congo (DRC, ex-Zaire) has been the center of the first (1996-1997) and second Congo (1998-2003) wars that saw several countries and several armed groups involved and the largest death toll for a single war since World War II. Up to 5.4 m people may have perished as a consequence of these wars, although the actual number of casualties remains a disputed issue. In addition to generating death and mass displacement within Congo, hundreds of thousands of Congolese fled to neighboring countries including Uganda and Rwanda. Two studies looked at the impact of these refugees on host communities in these two countries. Kreibaum (2015) looked at Uganda and finds that, overall, the refugee presence increases monthly consumption of the host population. Increasing the number of refugees per 1,000 inhabitants by 10 increases consumption on average by 3 percent, which is about equivalent to one day’s income for the local population. Interestingly, this does not match the perception of the local population which feels that the presence of refugees decreases well-being. Taylor et al. (2016) look at Congolese refugees in Rwanda and find a positive effect of aid on the local economy. This is the only paper in our knowledge that uses a general equilibrium model supported by microeconomic survey data to assess the impact of cash and food aid assistance to refugees in camps on the local economy identified as host communities living in a 10 km radius from the camps. They find that each adult refugee receiving cash assistance increases the annual real income of host households by 205 to 253 USD, which is more than the value of cash assistance provided to refugees of 120—126 USD. They also find positive impacts on local trade and food aid, although the latter impact is smaller than the impact of cash assistance. Therefore, both studies find a positive impact of refugees on household well-being of local communities, which is in line with the studies on Burundian and Rwandan refugees in Tanzania and the work on the Kakuma refugee camp in Kenya. IDPs in Sudan The wars ravaging across Sub-Saharan Africa have led to the internal displacement of millions of people, particularly in countries such as Nigeria, Sudan, South-Sudan, DRC and Somalia due to the prolonged nature of conflict in these countries. These populations are very difficult to study, or even count, because access to IDPs is limited due to war and restrictions are imposed by governments who often share responsibility for displacement. Population surveys are very scarce and published studies are rare. One exception is represented by two studies on internal displacement in the Sudanese region of Darfur, particularly the city of Nyala, a city that counted almost 3 m residents in 2010. This city became the epicenter of internal displacement in the Darfur region with up to 700,000 IDPs located in camps on the outskirts of the city or scattered around the city. Alix-Garcia, Bartlett and Saah (2011) studied the impact of IDPs on local prices and find a significant association between the growth in IDPs and the rise in food prices. The impact varies across products, from an IDPs/prices elasticity of 0.4 for fava beans to 2.9 for wheat. This relation is found to be significant for key products like sorghum, wheat, fava beans and oil but not for other products such as millet or sugar. Alix-Garcia and Bartlett (2015) return to study the same case focusing this time on occupations. They find that local residents living in Nyala had a higher likelihood of being employed in skilled sectors and a lower likelihood of becoming unemployed relative to a control group drawn from a comparable city not affected by IDPs (el Obeid). Such effect is particularly visible for male older workers but also present for females. The same paper also uses house improvements as a proxy of wealth and finds that people living in Nyala 71 made significantly lower house improvements than people living in el Obeid during the IDPs crisis. The study also finds that this is explained by households dominated by low skilled workers. 72