WPS4490 Policy ReseaRch WoRking PaPeR 4490 Risk Sharing Opportunities and Macroeconomic Factors in Latin American and Caribbean Countries: A Consumption Insurance Assessment Luigi Ventura The World Bank Latin America and the Caribbean Region Office of the Chief Economist January 2008 Policy ReseaRch WoRking PaPeR 4490 Abstract This paper evaluates the degree of consumption insurance is relatively more important than aggregate risk, and that enjoyed by Latin American and Caribbean countries, some countries in the region appear to enjoy a certain with respect to various reference areas, by estimating amount of international risk diversification. a parameter expressing the sensitivity of a country's The paper also identifies some macroeconomic factors consumption growth to a measure of idiosyncratic shocks that may be responsible for a higher or lower degree of to income. The paper surveys common econometric risk pooling (such as international openness, financial implementations of "consumption insurance tests." The depth, and credit availability). The findings show that the author proposes some econometric procedures in order financial development of an economy is a crucial factor to detect the actual presence of international risk sharing, in determining the amount of risk sharing opportunities, as well as to assess the relative impact of idiosyncratic as well as public expenditure. The preliminary results versus aggregate shocks. The evidence suggests that Latin also suggest that trade openness and shocks to terms of American and Caribbean economies have been hit by trade play an important role in determining the degree of non-diversifiable income shocks, that idiosyncratic risk insurability of such risks. This paper--a product of the Office of the Chief Economist in the Latin America and the Caribbean Region--is part of a larger effort in the department to access the issue of Country Insurance, and to analyze ways to reduce systemic vulnerabilities in LAC countries. Policy Research Working Papers are also posted on the Web at http://econ.worldbank. org. The author may be contacted at luigi.ventura@uniroma1.it. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Risk Sharing Opportunities and Macroeconomic Factors in Latin American and Caribbean Countries: A Consumption Insurance Assessment. Luigi Ventura1 1Dipartimento di Scienze Economiche, University of Rome "La Sapienza". The author would like participants in the Pre-conference for the Regional Study on "Country Insurance: Reducing Systematic Vulnerabilities in LAC countries", at the World Bank, february 2007, and in particular Ayhan Kose and Eduardo Levy Yeyati for very useful comments on a previous version of the work. 1. Introduction The aim of this paper is that of assessing the risk sharing opportunities enjoyed by Latin American and Caribbean (LAC) countries, as reflected in co-movements of rates of growth of (aggregate) domestic consumption with rates of growth of national income and rates of growth of aggregate consumption (averaged across a macro-area of interest). The underlying theoretical idea is that when a country has full access to risk sharing opportunities within a given macro area, consumption growth rates should not co-vary with aggregate income growth rates of the same country, but rather with aggregate consumption growth in the reference area (equivalently, consumption growth rates should be perfectly correlated across individual countries). Therefore, only aggregate risk should matter, whereas individual (idiosyncratic) risk should be diversified away. If the possibility of insuring against idiosyncratic risks is something desirable in all economies, this is particularly true in developing countries, where lack of insurance may lead to more inequality and to lower levels of investments, income, consumption, and savings (Townsend (1994), Greenwood and Jovanovic (1990), Banerjee and Newman (1993)). Increasing inequality, in turn, may constitute a relevant obstacle to the development of growth processes in such countries (Townsend (1994)). Of course, there are many reasons why idiosyncratic (at an individual country level) risk may not be wiped off, including incomplete financial or real markets, limited participation in real or financial markets, absence of intra or inter generational transfers, limited saving opportunities, etc. In particular, in the last few years growing attention has been paid, in this stream of literature, to the effects of financial integration on the risk sharing opportunities enjoyed by individual economies, both in the case of developed and developing countries. Imbs (2006) shows that financial integration increases both correlations in GDP fluctuations and in consumption fluctuations, but that the effect is larger on the former. This helps explain the so called "quantity puzzle", i.e. what appears to be a persistent gap between the two groups of correlations (whereas theory would rather predict an inverse gap, with consumption correlations being larger than output correlations). Kose et al. (2006), on the other hand, show that emerging market economies have not benefited, in terms of improved insurance opportunities, from what seems to be a substantial increase in international financial flows in the last two decades, suggesting a sort of "threshold" effect for the gains from financial integration. Over the last fifteen years or so, economists have resorted to various tests of consumption insurance, run both at a microeconomic level, and at a macroeconomic level, to evaluate the degree of risk sharing within a country or within a broader area. 2 The microeconomic analyses have been conducted by analyzing consumption at an individual, more often household, level, by using survey data. Macroeconomic analyses, on the other hand, have used macro-data on consumption and income, sometimes organized as a panel (of countries or regions within a country) data-set. The main goal of this paper is therefore that of evaluating the degree of consumption insurance enjoyed by LAC countries, with respect to various reference areas (LAC countries themselves, OECD countries, etc...). This will be done essentially by estimating a parameter expressing the sensitivity of a country's consumption growth to a measure of idiosyncratic shocks to income. Common econometric implementations of this test, used in recent contributions, will also be surveyed and compared to the original methodology followed in this work. A subsequent, though still preliminary, step in the analysis will be to try and identify some macroeconomic factors which may be responsible for a lower degree of risk pooling (international openness, some measures of financial depths, credit availability, etc...). Finally, an attempt will be done to estimate the potential welfare gains of non insured countries from risk-pooling with other countries in the area, or more developed countries in different areas, following the approach by van Wincoop (1994, 1999). 2. Empirical implications of risk sharing Since the pioniering work by Cochrane (1991) and Mace (1991), the hypothesis of market completeness (or of complete insurability of idiosyncratic risks) has been implemented by estimating an equation of the form: logcit = 0 + 1ISit + 2log(cat)+it (1) where the left hand side is the logarithmic difference of national per capita (non durable) consumption, log(cat ) is the logarithmic difference of per capita aggregate (over a reference set of countries) consumption, and ISi is any variable capturing an idiosyncratic shock to country i. Most often, as will also be done in the sequel, the variable ISi will be replaced by a variable expressing idiosyncratic shocks to log differenced income, log(yit ) , leading to the specification: id logcit = 0 + 1log yit + 2log(cat)+it id (2) 3 The test would then consist in checking that the coefficient 1 is equal to zero, while the coefficient 2 is equal to one. If that was not the case, we might conclude that countries cannot fully offset, by using a suitable insurance mechanism, idiosyncratic shocks to their endowments, which are consequently transmitted to consumption. Remarkably enough, some researchers (see, for a few examples, Grimard (1997), and Jalan and Ravallion (1999)) have gone even further (than simply rejecting the hypothesis of full insurance if 1 turns to be statistically significant) by considering the relative size of 1 as a kind of measure of the degree of insurance. An alternative procedure to estimating (2) has also been followed, taking stock of the contributions by Crucini (1999), and Crucini and Hess (2000), Lewis (1997), and Kim et al. (2005), which account for the fact that an individual country may only be allowed to pool a fraction of its aggregate income (and consequently share the corresponding risk). It would therefore be of some interest to assess the magnitude of this factor, which has been done by running a regression of the type: logcit = + logcat +(1-)log yit +it id (3) where all the variables keep the same meaning as in (2). Although it is surely tempting to evaluate the magnitude of this coefficient, we will stick in the empirical exercise to the estimation of equation (2) or variations thereof, as we will show in the next section that the approach implicit in (3) is not devoid of serious econometric drawbacks (essentially consisting in potentially imposing false constraints on data). There is, however, an additional question we need to address, before we move on to a more appropriate empirical implementation of these "consumption insurance" tests. Equation (1), as it stands, implies that we expect positive and negative realization of the idiosyncratic shocks variable to have the same impact, i.e. no systematic impact at all, on consumption growth. This is so because with complete markets a country will have the opportunity to average out consumption across states of the world, and consumption growth will always be equal to aggregate consumption growth, regardless of the nature of the shock (in other words, consumption growth would be constant across all states of the world). A fortiori, coefficient 1 should not turn out to be significant, even if we regressed equation (1) on separate sub samples, for example (but not necessarily) consisting only of positive, or negative, realizations of the shock variable2. However, this would only be so if all the variables involved were stationary. This would possibly be the case, for example, if we could observe many realizations of the shock variable to the same country in the same situation (for example, exactly at the same moment in time) and used 4 these data to actually estimate equation (1). In practice, available data refer to the same country over different periods of time (as in time series analyses), or to different countries at given points in time, as in cross section analyses, or combinations of the two, as in panel estimations. For example, in the case of time series estimations, although we may reasonably suppose (and test) that rates of income growth (and consumption) be stationary over long periods of time, it might equally be possible that non stationarity be found over shorter time spans. Developing countries, for example, might experience growing rates of growth for a relatively long period. If that were the case, even with perfect insurance we might reasonably expect growth rates in consumption to co-vary more strongly with income growth in case of positive innovations (which partly embody an increase in permanent per capita consumption), and more moderately in the case of negative innovations (which would be largely offset by insurance mechanisms). Consequently, on average we might detect lack of insurance (i.e., a statistically significant coefficient1) even when insurance is in fact complete, particularly when positive shocks prevailed over negative ones. Likewise, lack of stationarity might be induced by taste shifts, entailing similar consequences. When insurance is severely restricted, on the other hand, the opposite (i.e., detecting more insurance than it is in fact the case) might also realize, as consumption growth would strongly co-vary with negative income innovations, and would be less sensitive to positive ones. That is why it might be sensible to split the shock variable in two variables, reflecting negative and positive income innovations, and focus on the estimate of coefficient 1 , attached to the - former. To sum up, by restricting the coefficients of the two sub-variables to be equal (i.e., if we did not split the shock variable), we might fail to detect full insurance (when present), or make the opposite mistake, of detecting more insurance than it is in fact the case. On the other hand, by splitting the shock variable in any sensible way, nothing would change in the standard, full insurance and stationary case, corresponding to equation (1): both coefficients should turn out to be statistically non significant. Put another way, the procedure followed in the rest of the paper will be without loss of generality, with respect to the maintained theory. Hence, we will carry out our consumption insurance test by explicitly distinguishing between "negative" and "positive" realization of the shock variable, where the terms "negative" and "positive" will be better clarified in the next section, and the equation to be estimated will take the form: logcit = 0 + 1ISit + 2ISit + 3log(cat)+it . + - (4) 2In section 3 a methodology to distinguish between negative and positive realizations of the shock variable is proposed. 5 3. Consumption insurance tests: more on methodology Let us come back, for a while, to equation (2), which constitutes the basis for our test of consumption insurance. We already noticed that the dependent variable is the log difference of country i's aggregate consumption growth, and the two regressors are, respectively, the innovation to the log difference of national GDP, and the log difference of a given set of countries' aggregate per capita consumption. Indeed, it is important to use innovations to the log difference of national GDP, instead of national GDP itself, if one wants to evaluate the impact of idiosyncratic risks onto consumption. To see why, suppose we did not decompose the variable log(GDPit ) in its two components; then, if risk pooling were indeed at work, and if the aggregate component of log(GDPit ) did not exert any or a small effect on idiosyncratic consumption growth, the estimated coefficient of this aggregate GDP growth variable would most likely be downward biased, as the weight of the small (or zero) effect of the aggregate component might easily overcome the weight of the idiosyncratic one. In view of the importance of this decomposition, the idiosyncratic component of GDP growth will be computed as the (estimated) residual in the following equation: log(GDPit) = log(GDPat)+it , (5) where the variable on the right is the logarithmic difference of total GDP of the same set of countries as in equation (4). The (approximate) rate of growth of a given country's GDP will therefore be decomposed in two orthogonal components: in fact, log(GDPit) = ^log(GDPat)+ eit, ^log(GDPat) eit. (6) This procedure has been adopted accounting for the fact that in equation (2) we cannot possibly include the aggregate component of GDP growth, due to potentially severe multicollinearity problems with aggregate per capita consumption growth, and that the latter should be orthogonal to the former, to avoid a potentially serious omitted variable bias on 1. On the other hand, it is quite likely, always in view of the collinearity hinted at above, that any effect of the aggregate component of national income growth on idiosyncratic consumption shocks will be captured by the aggregate consumption term. Moreover, filtering log(GDPit ) as in (6) might turn out to be useful to remove 6 endogeneity problems affecting innovations in income as an explanatory variable (one might argue, in fact, that aggregate income growth of the pool of countries is not likely to be correlated with innovations affecting any single country's income growth). The more standard practice (see, for a few examples, the contributions by Obstfeld and Rogoff (2004), Sorenson and Yosha (1998), and the more recent ones by Sorenson et al. (2006) and Artis and Hoffman (2006)), consisting in decomposing income innovations in the two components (idiosyncratic vs. aggregate) by simply subtracting the average (across countries in a wide economic area) rate of GDP growth from individual countries' ones leads to the following specification: logcit = 0 + 1(log(GDPit )-log(GDPat))+ 2log(cat)+it (7) which does not guarantee, per se, this orthogonality property. On the other hand, subtracting log(GDPat) from log(GDPit) may strongly (and positively) bias the estimation of 2, possibly inducing into the false belief that risk pooling is indeed occurring, and to a large extent. This is so for a simple reason, which we may sum up as follows: if the subtracted term, log(GDPat ) , keeps some explanatory power, once the effect of log(cat ) has been accounted for, and if 1 is nonzero, we end up in a case of omitted variable. As the term log(GDPat ) is usually correlated with log(cat) , the coefficient of the latter will incorporate part or all of the effect of the omitted variable onto the dependent variable, which might be responsible for the bias. In section (4.2) it will be shown that these two procedures may indeed yield very different results. Importantly, the formulation of the consumption insurance test adopted in this work is also different from a more common methodology (cfr. Kose et al. (2006) and references cited therein), which is based upon estimating the equation: logcit -log(cat) = 0 + 1(log(GDPit)-log(GDPat))+it (8) both because of the way the idiosyncratic component of GDP growth is computed, as described above, and because in (8) the coefficient on aggregate consumption growth, log(cat) , is constrained to be equal to 1. Unfortunately, as can easily be reckoned from actual data, this can by no means be taken for granted, and furthermore this assumes that the researcher has correctly identified the set of countries sharing the same pool of risks. Needless to say, the consequences on the estimated parameters of introducing a non valid constraint may also be very serious. Importantly, the filtering procedure based on (6) can also be used to decompose logcit in an 7 idiosyncratic and an aggregate component, just like for income growth. This would provide us with a test equation similar to (8), which does not suffer from the potential pitfalls hinted at above, and which does yield, in a certain number of remarkable cases, results which are quite far from the ones we obtain by using (8). A more thorough discussion of this is outside the scope of this paper, and is contained in a more methodological, companion paper (Pierucci and Ventura, 2007). Finally, an alternative version of the consumption insurance test, which has been designed to account for the possibility that countries only pool a fraction of their income, has been mentioned in section 2, and is based upon equation (3). Now, it can be easily shown that equation (3) is approximately equal to equation (8), if the dynamics of aggregate consumption is close to the dynamics of aggregate income. In fact, by subtracting log(cat ) from both sides of equation (2), and rearranging, we obtain: logcit -logcat = +(1-)(log yit -logcat)+it , (9) which coincides with (3) if log(cat ) = log(GDPat ) , and is therefore subject to the same potential criticisms. On all these grounds, in the sequel we are implementing a test based on equation (4), partially modified to account separately for negative and positive shocks. The final estimated equation will therefore be of the kind: logcit = 0 + 1 GDPit + id ++ 1 GDPit - id -+ 3log(cat)+it (10) where the variable log(GDPit ) has been split in two separate variables, containing respectively id the "positive" and the "negative" realizations of the original variables, and zero elsewhere. To distinguish between positive and negative realizations of the variable log(GDPit ), we use a concept of output gap, in a very straightforward fashion. Assuming that the trend output (as measured, for example, by real GDP filtered according to the Hodrick and Prescott (1997) method) is the level of output that a country wishes to secure, we define as "negative" components of the variable log(GDPit ) those corresponding to periods of negative output gap (when actual GDP is below its trend level), and "positive" components those corresponding to periods of positive output 8 gap. This, let us notice, does not imply that those components be actually negative or positive, as the opposite might well be the case. This method will allow us to capture unfavorable and favorable shocks even when log(GDPit ) is positive at all times, which is almost always the case for developed countries and for many developing countries, as well. However, it is maybe worth stressing once more, this or alternative specifications of negative and positive shocks will not fail to reveal "standard" consumption insurance, if that is indeed the case (i.e., there is no loss of generality with respect to the full insurance case). Starting from the econometric analysis outlined above, one last step of this research will aim at evaluating two important features related to country insurance. The first is whether we may indeed recognize some international risk sharing. Secondly, a measure of individual countries' vulnerability to idiosyncratic and aggregate risk will be proposed and computed. The first question is quite relevant, as it strictly pertains to the interpretation of the results of a consumption insurance test. We might, in fact, recognize a pitfall in standard consumption insurance test, in that the polar case consisting of a statistically non significant coefficient 1, coupled with a unit coefficient 2 might well signal a situation of perfect consumption insurance, but equally well describe a situation in which idiosyncratic risk is irrelevant, while only aggregate risk is relevant, and this for all countries in a given reference area, with no international risk sharing whatsoever. In other words, it might be the case that all countries in a group are hit by the same shocks3, and respond in a similar way, whereas the impact of idiosyncratic risk is negligible, either because some mechanisms of risk sharing (possibly only domestic) are in place, or simply because of income (as opposed to "consumption") smoothing (see on this, for example, the contribution by Morduch (1995)). When, on the other case, we estimate a significant coefficient for 1, and an equally significant, and possibly large, 2, we know for sure that risk sharing is not complete, but it still remains to be understood whether or not there is some international risk sharing at work. Therefore, we need a methodology to understand whether this is the case, and we propose the following procedure: 1) regress domestic consumption growth on the three following variables: idiosyncratic and aggregate income growth, computed as in (6), and aggregate (over the reference area) per capita consumption growth; 2) test the significance of the last variable, and conclude in favor of international risk sharing if the coefficient is statistically different from zero. 3I am grateful to Eduardo Levy Yeyati for having so much stressed this important point. 9 The rationale of this test is the following: for international risk pooling to be in place, it must be the case that aggregate per capita consumption growth retains some explanatory power over and above what is implicit in the aggregate component of domestic income growth, which should bring about the statistical significance of the corresponding coefficient. A second interesting question (always relevant for the interpretation of the coefficients of (10)) concerns the effective relevance of idiosyncratic vs. aggregate risk in determining consumption dynamics. We cannot possibly assess this relevance by simply looking at the significance and/or magnitude of 1. We might, in fact, obtain a large and significant estimate for 1, but the corresponding variable might possibly explain a very tiny portion of consumption variability, in which case we should conclude that the country in question is little affected by idiosyncratic risk. Hence, the idea is to measure the relevance of idiosyncratic and aggregate shocks (what we will term "vulnerability") by computing the decrease (it might actually be an increase) in adjusted R2 obtained by alternatively removing the idiosyncratic component or the aggregate component of domestic income growth. 4. Consumption insurance in Latin American and Caribbean countries: some evidence from macro data We are now going to actually estimate equation (10) using data from Latin American and Caribbean countries, and carry out some additional analyses. Before that, however, a brief introduction to data sources and actual variables used in the empirical analysis is in order. 4.1 Data Data for the empirical analyses are obtained from the most recent World Bank database "World Development Indicators", and from the International Monetary Fund database "International Financial Statistics", June 2005 edition. Data from the "World Development Indicators" database include: final consumption expenditure, household final consumption expenditure, GDP, the three of which constitute the main variables in the following analysis, as well as additional economic and structural variables, such as foreign direct investments, receipts from international tourism, liquid liabilities (M3), net domestic credit (both as a total and stemming from specific institutions), market capitalization of listed companies, official and real exchange rates, total value of stocks traded, long term debt, internet users, net current transfers from abroad, net foreign assets, population, and others. Most of these additional variables are measured in terms of percentage of GDP. 10 For a few countries, final consumption expenditure had to be computed as a product of a variable expressing final consumption expenditure (including also statistical discrepancies) as a percentage of GDP, and GDP measured in constant 2000 US dollars. Household final consumption expenditure, on the other hand, did not need any additional manipulation. The longest period covered by the data is 1965-2005, though regrettably this is not true for all countries. As a result, the longest estimation horizon has been that of 1971-2005. Due to the need to aggregate consumption and GDP across countries, data on consumption and GDP were measured in constant 2000 US dollars. Consumption insurance tests were performed on a wide set of 93 countries from the following (geo- economic) areas: East Asia and Pacific, Sub-Saharan Africa, Latin American and the Caribbean, and OECD. In particular, the following LAC countries have been covered (due to data availability): Argentina, Belize, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, El Salvador, Trinidad and Tobago, Uruguay and Venezuela. The analysis on welfare gains from international risk diversification has been performed on a wider group of LAC countries, including also Barbados, Dominica, Dominican Republic, El Salvador, and Grenada. To perform the welfare analysis, data on money market rates and treasury bills rates, and on CPI growth rates were obtained from the "International Financial Statistics" database, for the period 1994­2003, and cover all the countries in the larger group, except for Costa Rica, Ecuador, Honduras, Nicaragua, and Peru, for which only data on the official discount rate were available. Bahamas was also excluded from this analysis, as consumption data were not available for recent time periods. 4.2. A preliminary look at data and some empirical results To get a first grasp of the phenomenon under scrutiny, we may give a look at Graphs 1 and 2, containing scatter diagrams where the x-variable is represented by normalized per capita income (where the mean and standard deviation has been computed over the whole sample of countries) and the y-variable is represented by the values of coefficients 1 in equation (10). - 11 Countries are grouped according to geo-economic classification in graph 1 (the three geographic areas mentioned above, and OECD), and an economic criterion (high income, upper middle income, lower middle income, low income, and emerging economies4). Insert Graphs 1 and 2 here Many interesting insights may be derived from a quick inspection of these two graphs (for clarity, only LAC labels have been added to graph 1). On the one hand, it seems pretty clear that there is a direct relationship linking income to insurance of idiosyncratic shocks. In fact, countries with higher per capita incomes display a lower coefficient 1 , as the interpolating line suggests. On the other hand, however, this relationship does not seem - to hold, if not very roughly, at low levels of per capita income (graph 2), or even for emerging economies. High income countries, on the other hand, seem to be fairly well characterized by this regularity. Graph 1 shows the relative positions of countries in some geographic and economic areas, and we may immediately notice that Sub-Saharan countries feature the lowest levels of income and the widest dispersion in 1 coefficients: at a visual level, no signs of positive or - negative relationships between insurance and income seem to exist. Per capita incomes are more differentiated in the case of LAC countries, as well as positions in terms of risk pooling; also in this case, the relationship between risk pooling and income is not very definite, and a good share of countries in the group feature fairly large values of 1 . Quite different is the case of countries in - the South East and Pacific group, and especially in the OECD group. In both cases we may observe a positive relationship between income and insurance, and in a good percentage of countries the value of 1 is around or below the 0.5 level. - Graphs 1 and 2 are plotted by using total consumption changes as the dependent variable. The picture would not change substantially if we used household final consumption dynamics as the dependent variable (the scatter diagram would be slightly more dispersed, though, and slightly shifted above). Let us now explore, in more detail, the case of LAC countries. Tables 1 and 2 contain, respectively, the coefficients of correlation (computed on recursive samples, 1971-2003) between individual LAC countries' per capita GDP growth and average GDP growth (over the LAC region), and the coefficients of correlation between individual LAC countries' per 4Our set of countries include the following emerging economies: Argentina, Brazil, Chile, China, Cote d'Ivoire, Dominican Republic, Ecuador, Hungary, Korea, Mexico, Malaysia, Panama, Peru, Philippines, El Salvador, South Africa, Thailand, Uruguay, Venezuela. 12 capita consumption growth and average consumption growth. Table 1 is based over total consumption, whereas table 2 is based over household final consumption. A rapid inspection of tables 1 and 2 shows that those correlations are not very strong, as they seldom raise above 0.6, and remain fairly stable over the period 1992 ­ 2003. When they do display some dynamics, these are most often decreasing, which points at an increasing role played by idiosyncratic risks in LAC economies. Moreover, the correlations of consumption growth rates are normally lower than those of GDP growth, which is an indication of the existence of the so called "quantity puzzle", that was mentioned in the introduction. Importantly, the gap between the two sets of correlations does not appear to significantly decrease over the period 1992 ­ 2003. To understand the dynamics of consumption insurance in LAC countries we ran several times the econometric tests illustrated in section 2 and 3. In table (3) are reported the results of the estimation of equation (10) for 23 Latin American and Caribbean countries. The first section of the table contains results concerning rates of growth of total consumption expenditures, i.e. including general government final consumption expenditures, while the second section displays results relative to household final consumption expenditures. Each section covers the case in which pooling of risk is envisaged within the LAC region, or within the OECD area. Insert table (3) here As anticipated by looking at graph 1, a cursory inspection of table (3) reveals that in no case may we observe perfect risk sharing in LAC countries, regardless of whether we use total or household consumption growth as dependent variable; importantly, in quite a few cases (Bahamas, Belize, Haiti, Jamaica, Nicaragua, Panama, Suriname, Trinidad and Tobago) no risk pooling at all seems to be in place, as aggregate consumption at the LAC or OECD level is not significant as an explanatory variable5. Looking at the significance of the latter, we may also recognize the cases in which some risk pooling is carried out in the LAC area, or rather in the OECD area. While the vast majority of LAC countries seem to pool risk within the LAC region, as can be reckoned from the fact that the corresponding variable for OECD is statistically significant in far fewer cases, for some countries the contrary seems to hold; this is the case, for instance, of the Dominican Republic and of Venezuela, but only if household final consumption is taken into account. Moreover, the value of the estimated coefficient of idiosyncratic innovations in income displays a fairly large amount of variability across countries. It ranges from 0.541 and 0.602 of, respectively, Nicaragua and Venezuela (0.098 and 0.588 of Bolivia and Paraguay in the case of household final 5Remarkably, we would have not realized this, had we followed an approach based on equation (11). 13 expenditures) to 1.623 and 1.287 of, respectively, Chile and Uruguay (1.799 and 2.134 of Chile and Trinidad and Tobago, when only household final consumption is considered). In the Appendix we have also included table (4), which contains results of the estimation of equation (4), in which the variable representing idiosyncratic shocks is not split into its negative and positive components, and table (5), which contains a comparison of the estimated coefficients of negative and positive idiosyncratic shocks, and can be used to understand the (possibly large) differences between results presented in tables (3) and (4). It is also interesting to give a look at figures (1) to (4), where recursive and rolling estimates6 of 1 - (the coefficient of "negative" idiosyncratic risk) have been plotted for the various countries with respect to the two versions of consumption aggregates7. To obtain a (hardly) sufficient number of observations for each estimation, the 1 's are computed only for the time span 1992­2003. This is - not a long period to look at, albeit one which may provide us with an opportunity to distinguish some interesting trends in risk sharing, and to check whether currency or other types of crises have had an appreciable impact on that. Insert figures (1) to (4) here For example, looking at recursive 1 's computed on household final consumption expenditures' - growth, we may observe countries like Ecuador, Paraguay, Trinidad and Tobago, and Venezuela, showing an upwards trend in the estimated coefficient (signalling a decrease in insurance of idiosyncratic shocks), whereas countries like Guatemala and Peru show a decreasing trend in 1 's. - This also seems to be the case for Costa Rica and Dominican Republic, where we recognize a sudden (and substantial, in the case of the latter) decrease in 1 around 95-96. Quite the opposite is - true for Colombia, showing an abrupt and important increase in 1 in the period 98-99. The effects - of Argentina's and Mexico's crises on insurance performance are also quite evident in the table, as we may observe a rapid increase of 1 in 2001 for the former (and other countries, like Bolivia), - and in 1994-1995 for the latter (and other countries, like Nicaragua). The Brazilian crises in January 1999 had quite a different impact, as it was relatively short (everything went back to normal by the 6 In recursive estimates one more (recent) observation is added at each time, and a new estimation is performed. In rolling estimations a "window" of observations is defined, and new regressions are performed by having this window slide towards the end of the sample (i.e. a new, more recent observation is added, and the oldest is eliminated). In our case, a window of 20 observations has been selected. 7 The implications of the different methods used in computing 1 's are not negligible, as might be argued by - computing correlation coefficients among the different versions of 1 's. - 14 end of the year), and did not turn into a financial crisis; not surprisingly, as we may observe from the graph(s), this crisis is not signaled by the behavior of 1 's. - As was explained in section 3, the methodology followed in this paper to assess the magnitude of consumption insurance significantly differs from the more standard one, and may lead to different results. To see that this is indeed the case, we report in table 6 the correlations between recursive and rolling coefficients computed in this work, and the same coefficients computed according to the methodology implicit in equation (8). Indeed, differences can be substantial. Following the approach outlined at the end of section 3, we also tried to gauge the extent of idiosyncratic and aggregate risk vulnerability of LAC countries. The results are reported in table (7). The table reports (complete) results only for those countries for which it was clear that a decomposition of income growth in an idiosyncratic and an aggregate component could be obtained. When this was the case (as for Chile, Haiti, Jamaica, Nicaragua, Panama, El Salvador and Trinidad), we assumed that income dynamics was generated solely by idiosyncratic shocks, and the correct reference should be table 3. The results reported in table 7 suggest that most LAC countries are relatively more vulnerable to idiosyncratic than to aggregate shocks (except in the case of Bolivia, Brazil, Colombia and Paraguay, for which the reverse would appear to be true), and that in some cases the former do explain a large share of variability in consumption dynamics (as for Chile, Mexico, Argentina, El Salvador, Jamaica and Uruguay, in order of importance). The case of Chile is quite significant, as it is well known that the dynamics of world copper prices is a major determinant of its economic up and downturns. Not surprisingly, then, Chile is the first country in the list. Per se, these indications could be important for policy purposes, if a policymaker wanted to identify the major sources of risk affecting a country or a group of countries, and in as much as instruments to buffer idiosyncratic variability in income (essentially, various types of insurance mechanisms) may be different from those required to counter aggregate risk (essentially, aggregate savings). Importantly, table 7 also shows that some countries do appear to enjoy the effects of some international risk pooling. This is the case of Argentina, Brazil, Guyana, Mexico and Peru (to which we might possibly add Chile, by looking at table 3). 4.3 On (some) macro-economic determinants of idiosyncratic risk sharing The next step in our empirical analysis is to check the existence of a relationship between the magnitude of 1 's and some macroeconomic variables, such as the degree of international - openness, the extent of domestic credit provided by the banking system, the behavior of terms of trade, the amount of public consumption, the volume of exchanges on stock markets as a measure of 15 financial depths, the volume of foreign direct investments, the term structure of internal indebtedness, etc. Some preliminary results are contained in table (8), where we present panel estimates of a regression of 1 's (in their various forms) on a set of macroeconomic variables, with country - specific coefficients. The relevance of fixed effects has been suitably tested, and has proven robust to a number of changes in the specification of equations. Insert table (8) here The estimated coefficients consistently show that an increase in domestic credit provided by banks to the private sector, as a percentage of GDP (or, somehow equivalently, an increase in the percentage of liquid liabilities over GDP, or of liquid liabilities over GDP), an increase in the total value of stocks traded, always as a percentage of GDP, an increase in the percentage of long term indebtedness over GDP and of public expenditures are associated to an improvement in insurance against idiosyncratic shocks. International openness has an important (and negative) impact on insurability, and this is also reinforced by the positive and significant coefficient on shocks to the differential in terms of trade (especially with respect to positive shocks). This is not an encouraging result, in that it might also be linked to the (ambiguous) significance of increasing financial flows in improving the degree of risk sharing. In fact, it has been noted by several authors (see, for example, Oh et al. (2001)) and Lane and Milesi-Ferretti (2003)) that the extent of trade in goods might work pretty well as a predictor for financial flows. The sign and magnitude of a couple of important control variables, namely real per capita income and exchange rate regime should also be noted. In particular, the sign of real per capita income is ambiguous, as it turns out to be positive for recursive estimations, and negative for rolling estimations. Exchange rate regimes have been specified in a de facto fashion, following the contribution by Levy Yeyati and Sturzenegger (2005). In all specifications, only a floating exchange regime (i.e. the corresponding dummy variable) has proven significant, and beneficial in terms of improving insurance opportunities. The relevance of some other macroeconomic variables was also preliminarily tested, yielding at the very least a few ambiguous results. For instance, measures of currency crises were inserted in the regressions, but did not turn out to influence the results. Another remarkable case is that of inflowing foreign direct investments, or net foreign direct investments, which were mildly significant in some versions of the model, and not significant in most others. In any case, even without directly accounting for financial inflows and outflows of assets or income, the picture we get from our preliminary analysis would point at a decisive role played by the banking system and 16 by a country's economy financial depth in providing insurance against idiosyncratic shocks. Moreover, public expenditure seems to be key in providing insurance against idiosyncratic shocks. 5. A few remarks on potential welfare gains from risk sharing In view of the results reported in section 4, it is tempting to provide an assessment of the potential gains from risk pooling that Latin American and Caribbean countries might obtain. This is not an easy task, however, given the number of variables and processes involved in a country's opening to international risk sharing. This is why many different methods (and different hypotheses) to assess benefits from risk pooling have been put forward in the literature, and so diverse results have been found, as reported in van Wincoop (1999), ranging from a small 0.5% to two digit or even three digit gains (see, for a comprehensive survey, the work by Auffret (2001)). The basic idea underlying one of those methodologies (see, for a few examples thereof, Cole and Obstfeld (1991), Obstfeld (1994b and 1995), van Wincoop (1994), Kose (1995), Lewis (1996)), is quite simple, and can be summarized as follows: if the representative individual in a country is endowed with a concave utility function, more variance in consumption will be associated to less (expected) utility. If aggregate consumption in a country is not perfectly correlated with aggregate consumption of a given reference area, then there is scope for diversification, which will lower the variance of aggregate consumption, thereby increasing expected utility. In these contributions, welfare gains are usually computed as the percentage increase in expected consumption corresponding to an increase in welfare brought about by a reduction in variance of consumption. From a different, and more financial, perspective, gains from risk sharing are measured as the ratio of the utilities an investor can obtain under two alternative situations: a situation of autarky, and one of full international diversification (see Lewis (2000) and Mercereau (2006), for some examples). In this case, however, we have more than simple insurance, in the sense that it is not simply a matter of smoothing consumption across states of nature; here expected consumption may also be affected, by resorting to foreign financial markets. This is why we are adopting the first line of reasoning, to try and provide some approximate measures of welfare gains Latin American and Caribbean countries might obtain, in case they had the opportunity to obtain full insurance. Following van Wincoop (1999), let us consider a representative agent with preferences represented by the following expected utility: 17 H V = E e- t (cit ) T 1- dt 1- 0 with H the terminal period, the discount rate and cit the consumption of tradables by country i. It T is assumed that utility is separable between consumption of tradables and non tradables (which may be then be omitted from the analysis as they do not contribute to risk sharing), and that the endowment process for tradables ( yit ) follows a random-walk with drift of the type: T dyit = yit dt +T yit dzi T T T with zi a standard Brownian motion process (with mean and variance respectively equal to 0 and dt ). In the absence of risk pooling consumption would just be equal to the endowment, with a resulting utility, at time 0 , equal to: U = (yi0)1- 1- e- T H , 1- with = + ( -1)( - 0.5T ). 2 (11) On the other hand, when risk pooling is present, the variance of endowments, denoted by TW will typically be smaller. In this case, it can be shown that the welfare gain, measured as the (permanent) percentage increase in expected consumption yielding an equivalent increase in welfare, is equal to: G = -1- H(r - )1- e-( e-(r- )H (12) r- )H0.5r-dT 2 with dT = TW -T 2 2 2 the change (decrease) in the variance of consumption growth, = -0.5 T the risk adjusted growth rate of consumption, rthe risk-free interest rate, and the 2 other symbols have a clear interpretation. To compute the welfare gains of our LAC countries according to equation (12) a few assumptions have been made. First, and most importantly, we suppose that although LAC countries may not achieve, by means of international risk sharing, the aggregate rate of growth of consumption of the reference area, be it LAC or OECD, they can at least get the same variance of aggregate per capita consumption growth. Therefore, we have replaced TW by the variance of per capita consumption growth in the LAC 2 18 area and in the OECD area, without reference to the covariance of each country's aggregate growth with total aggregate consumption growth. As it can be easily shown, the variance of consumption growth is lower in the OECD area than in any LAC country, whereas in a few cases (Bolivia, El Salvador, Guatemala and Jamaica) we observe that the opposite is true; in such cases, of course, international risk sharing could be potentially harmful, if we limited our considerations to pure risk sharing. Secondly, real "risk-free" interest rates have been computed subtracting the CPI growth rate from money market rates, or from Treasury bill discount rates, where the latter were not available. The real interest rate used in the computation was the average of the rates prevailing in the last three years of the time period used for the estimation (2001­2003). Standard deviation of per capita consumption growth rates and average per capita consumption growth rates have been computed by considering the time span 1994­2003. By applying equation (12) to the countries in our sample, we obtain the results displayed in table (9). Insert table (9) here The top section of the table contains estimated gains for a five year horizon, the bottom section those for a ten years horizon. As can be readily seen, gains are always positive if pooling occurs within the OECD area, and almost always positive if they occur at the LAC level (except for the countries mentioned above). For the five year horizon, ad an intermediate value of relative risk aversion ( = 6 ), (positive) gains range from a lowest value of about 0.16% of Brazil to a very high value of about 10% of Trinidad and Tobago. The relative dispersion of gains is brought about by a dispersion in real interest rates (which are used to discount gains) and in standard deviations of consumption growth rates with respect to the reference area. With respect to this kind of analysis, two main criticisms may be raised, both from a micro and a macro perspective. From a micro perspective we may observe that using macro data conceals a lot of information concerning the dynamics of individual consumption. A large share of idiosyncratic variability in consumption is already averaged out by aggregating across individuals, which suggests that the measures obtained for welfare gains by using macro data actually underestimate the gains in welfare obtained by pooling risks, both at a national and at an international level (the question of whether there is a significant correlation between international and national risk pooling remains to be 19 addressed). Addressing the question of welfare gains from a micro perspective, therefore, awaits the availability of more suitable, micro-data. From a more genuine macro perspective, on the other hand, a different criticism can be raised, and this is related to the fact that variability in income, at future dates, should exert an upward pressure on savings, if the third derivative of utility functions is positive. This is the well known result concerning precautionary savings, which was firstly related to convexity of marginal utility by Leland (1978) and subsequently more thoroughly formalized by Kimball (1990). As some recent research has shown (cfr. Ventura (2007)), however, income variability can increase savings even in the absence of prudence, when risk aversion is present, and savings are channeled to risky assets (which is virtually almost always the case). As Devereux and Smith (1994) pointed out, in a multi-country growth model with an infinitely lived representative agent for each country, and no aggregate risk at the world level, international diversification of country-specific income risk may lead to a decrease in the equilibrium saving rate in each country, and to lower growth rates in the same country. This, in turn, might reduce each country's welfare level. In fact, it is possible to get at least a partial and preliminary confirmation of this hypothesis by our data. In table (10) we have reported the results of the regression of gross savings, as a percentage of GDP, over the time series of 1 's, estimated in the previous section, with reference to both total - final consumption and household final consumption expenditure, and fixed (cross section) effects (all of which turn out to be highly significant - results are not reported in the table). Insert table (10) here As is clear from table (10), the effect of 1 's is positive, which might be taken as evidence that less - insurance (higher 1 's) against idiosyncratic shocks may indeed have the effect of raising the - savings rate in a country. The same result also holds with "rolling" 1 's, at least in the case of household final consumption - expenditures. What this analysis does not take into account, however, is the impact of lack of insurance onto the selection of investment plans, and on the volume of investments, which is obviously a key variable in assessing the impact of risk on growth. On the other hand, we may conjecture that the results reported above may hold only in the (more or less total) absence of insurance, whether in the form of national or international risk sharing. In fact, agents would insure if they were given a chance, and would normally use assets which might well increase the mean value of consumption (growth), 20 over and above providing hedging facilities. The amount of resources devoted to insurance activities, therefore, might also constitute beneficial savings, and be channeled to profitable investments, at home or abroad. Devereux and Smith (1994) also notice, among other authors, that integration may lead to a reallocation of savings to investment projects with higher risks and returns. 6. Concluding remarks and extensions The focus of this paper has been that of exploring and gauging the extent of insurance vis à vis idiosyncratic risks enjoyed by LAC countries. This has been done by running consumption insurance tests on most countries in the LAC region over the horizon 1972-2003, to estimate the crucial parameter of insurance. The analysis has been preliminarily performed over a much larger set of countries, and then has focused on LAC economies. This has shown that LAC economies share the features of less mature and less developed or emerging economies with respect to insurability of idiosyncratic risks. Apparently, there is no clear cut relationship between insurability and per capita real income, unlike in the case of more developed, high income economies. A sort of threshold effect seems to be at work. Some interesting insights could be obtained by the more detailed consumption insurance analysis on Latin American countries: firstly, it has been shown that LAC economies are fairly much hit by non diversifiable income shocks, which get therefore transmitted to consumption, and thus to welfare. An original methodology has been employed to show that idiosyncratic risk is relatively more important than aggregate risk, and that some countries do appear to enjoy a certain amount of international risk diversification (though seemingly not to a large extent). More importantly, our analysis suggests that most LAC countries appear to pool some fraction of their risk within the region, rather than outside (for example, within the OECD set of economies). By replicating our consumption insurance test several times for each country, we have built up a dataset of insurance parameters, which reveals some interesting dynamics, especially in view of recent results by Kose et al. (2006), which seemed to suggest that correlations of growth rates of output and consumption with their corresponding world aggregates have not changed substantially during the period of globalization for emerging market economies. Our dataset has then been enriched with a host of macroeconomic variables for the same countries and for the period 1992- 2003, and has been used to test the potential impact of some macroeconomic variables on insurance opportunities available to LAC countries. Some preliminary results show that the financial development of an economy is crucial in determining the amount of risk sharing opportunities, as well as public expenditure, and that trade openness and shocks to terms of trade also play an 21 important role in determining the degree of insurability of such risks. On the other hand, currency crises dummies did not appear to be particularly relevant (actually, they never turned out significant, at least in our regressions, to explain the value of insurance coefficients). The last part of the analysis has been devoted to an assessment of welfare gains from risk pooling, based on the work by van Wincoop (1999), showing that potentially large benefits could be obtained by pooling idiosyncratic risk both within and outside the region. 7. References Artis, M.J. and M. Hoffmann, 2004, Financial Globalization, International Business Cycles and Consumption Risk Sharing, CEPR Discussion Paper n. 4697. Auffret, P., 2001, An Alternative Unifying Measure of Welfare Gains from Risk-Sharing, Policy Research Working Paper n. 2676, The World Bank. Banerjee, Abhijit V & Newman, Andrew F, 1993, Occupational Choice and the Process of Development, Journal of Political Economy, 101,2. Cochrane, J. H., 1991, A simple test of full consumption insurance, Journal of Political Economy, 99, 5. Crucini, M., 1999, On international and national dimensions of risk sharing, Review of Economics and Statistics, 81,1. Crucini, M. and G.D. Hess, 2000, International and Intranational Risk Sharing, in Intranational Macroeconomics, edited by G.D. Hess and E.v. Wincoop, Cambridge University Press, Cambridge. Devereux, M.B. and G.W. Smith, 1994, International Risk Sharing and Economic Growth, International Economic Review, 35-3. Greenwood, Jeremy and B. Jovanovic, 1990, Financial Development, Growth, and the Distribution of Income, Journal of Political Economy, 98(5). Grimard, F., 1997, Household Consumption Smoothing through Ethnic Ties, Journal of Development Economics, 53, 2, pp. 391-420. Imbs, J., 2006, The real effects of financial integration, Journal of International Economics, 68, 296-324. Jalan, J. and M. Ravallion, 1999, Are the Poor Less Well Insured? Evidence on Vulnerability to Income Risk in Rural China, Journal of Development Economics, 58, 1, pp. 61-81. Kim, S., Kim, S.H. and Y. Wang, 2005, Regional versus global risk sharing in East Asia, Asian Economic Papers, 3,3. 22 Kose, M.A., Prasad, E.S. and M.E. Terrones, 2006b, How does Financial Globalization Affect Risk-Sharing? Patterns and Channels, paper presented at the 7th Jacques Polak Annual Research Conference. Kose, M.A., Prasad, E.S., Rogoff, K. and S.-J. Wei, 2006a, Financial Globalization: a Reappraisal, IMF Working Paper n. 06/189. Lane, P., and G.M. Milesi Ferretti, 2001, The External Wealth of Nations: Measures of Foreign Assets and Liabilities for Industrial and Developing Nations, Journal of International Economics, 55, 2, pp 263-294. Levy Yeyati, E. and F. Sturzenegger, 2005, Classifying Exchange Rates Regimes: Deeds vs. Words, European Economic Review, 49, 6, pp. 1603-1635. Lewis, K., 1997, Are Countries with Official International Restrictions "Liquidity Constrained"? European Economic Review, 41, pp. 1070-1109. Lewis, K., 2000, Why do Stocks and Consumption Imply such Different Gains from International Risk Sharing, Journal of International Economics, 52, pp.1-35. Mace, B., 1991, Full insurance in the presence of aggregate uncertainty, Journal of Political Economy, 99, 5. Mercereau, B., 2006, Financial Integration in Asia: Estimating the Risk-Sharing Gains for Australia and Other Nations, IMF Working Paper n. 06/267. Morduch, J., 1995, Income smoothing and consumption smoothing, Journal of Economic Perspectives, 9, 3, pp. 103-114. Obstfeld, M. and K. Rogoff, 2004, Foundations of International Macroeconomics, Cambridge, MIT Press. Pierucci, E. and L. Ventura, 2007, A geography of international risk sharing: methodological issues and an empirical exercise, mimeo. Sorensen, B.E. and O. Yosha, 1998, International Risk Sharing and European Monetary Unification, Journal of International Economics, 45, pp. 211-238. Sorensen, B.E., Wu, Y.-T., and Y. Zu, 2006, Home Bias and International Risk Sharing: Twin Puzzles Separated at Birth, unpublished manuscript, University of Houston, Texas. Townsend, R.M., 1994, Risk and insurance in village India, Econometrica, 62,3. van Wincoop, E., 1994, Welfare gains from international risk sharing, Journal of monetary Economics, 34, 2. van Wincoop, E., 1999, How big are potential welfare gains from international risk sharing? Journal of International Economics, 47, 1. Ventura, L., 2007, A note on the relevance of prudence in precautionary saving, Economics Bulletin, 4, 23. 23 Tables, figures and graphs. 24 25 Table 3. CONSUMPTION INSURANCE COEFFICIENTS Consumption risk-sharing is tested by model (13). Each test is applied twice, using two different measures of change in consumption: growth rates in total final consumption, and growth rates in household final consumption expenditure. Two different aggregates are considered, where total consumption is averaged across LAC countries and OECD countries. P-values are reported in brackets. Countries TOTAL FINAL CONSUMPTION HOUSE Macroarea: Latin America and Caribbean Macroarea: OECD Countries Macroarea: Latin America Caribbean logGDPid - log (ca) logGDPid - log (ca) logGDPid - log (Hca Argentina 1.058 [0.00] 1.283 [0.00] 1.070 [0.00] 0.088 [0.62] 1.112 [0.00] 1.0 Bahamas 0.949 [0.01] -0.105 [0.86] 0.821 [0.05] 1.246 [0.15] - - - Belize 1.137 [0.03] 0.175 [0.84] 1.548 [0.00] 0.206 [0.80] 1.349 [0.13] 1.8 Bolivia 0.461 [0.01] 0.545 [0.00] 0.579 [0.00] 0.031 [0.84] 0.098 [0.66] 0.3 Brazil 0.783 [0.00] 1.271 [0.00] 0.522 [0.01] 0.661 [0.00] 1.120 [0.01] 1.7 Chile 1.623 [0.00] 0.728 [0.06] 1.919 [0.00] 0.918 [0.02] 1.799 [0.00] 1.0 Colombia 0.734 [0.00] 0.537 [0.00] 0.803 [0.00] 0.675 [0.01] 0.954 [0.00] 0.3 Costa Rica 1.173 [0.00] 0.748 [0.02] 1.081 [0.00] 0.319 [0.03] 0.989 [0.00] 0.6 Dom. Republic 0.659 [0.15] 0.910 [0.03] 0.502 [0.16] 0.786 [0.10] 0.362 [0.54 0.5 Ecuador 0.947 [0.02] 0.503 [0.14] 0.833 [0.00] 0.118 [0.71] 0.840 [0.00 0.5 Guatemala 0.928 [0.00] 0.552 [0.00] 1.016 [0.00] 0.593 [0.00] 0.601 [0.00] 0.4 Guyana 1.052 [0.08] 1.510 [0.05] 0.892 [0.05] 0.602 [0.46] 1.110 [0.18] 1.0 Haiti 1.210 [0.01] 0.289 [0.60] 1.214 [0.00] 0.064 [0.89] - - - Jamaica 0.630 [0.05] -0.297 [0.41] 0.725 [0.02] -0.188 [0.62] - - - Mexico 0.982 [0.00] 0.865 [0.00] 0.959 [0.00] 0.644 [0.00] 073 [0.00] 0.7 Nicaragua 0.541 [0.00] 0.084 [0.88] 0.553 [0.00] -0.663 [0.25] - - - Panama 0.602 [0.13] 0.008 [0.99] 0.633 [0.06] 0.564 [0.39] 0.574 [0.29] 0.7 Paraguay 0.793 [0.08] 1.526 [0.00] 1.028 [0.00] 0.401 [0.37] 0.588 [0.21] 1.6 Peru 0.482 [0.02] 0.691 [0.04] 0.387 [0.03] -0.450 [0.25] 0.962 [0.00] 0.4 Suriname 1.428 [0.00] -0.058 [0.95] 1.470 [0.00] 0.435 [0.72] - - - Trinidad 1.629 [0.00] 0.308 [0.62] 1.503 [0.00] 0.363 [0.58] 2.134 [0.00] -0.2 Uruguay 1.287 [0.00] 1.301 [0.00] 1.247 [0.00] 0.428 [0.06] 1.815 [0.00] 1.1 Venezuela 0.602 [0.03] 0.640 [0.15] 0.809 [0.00] 0.841 [0.12] 0.673 [0.03] 0.2 Table 6. Correlations between "standard" and new insurance coefficients, recursive and rolling. ARG BOL BRA CHL COL CRI DOM ECU GTM GUY HND HTI JAM MEX NIC PER PRY Recursive 0.92 0.15 0.17 0.57 0.78 0.75 0.61 0.98 0.32 0.58 0.42 1.00 -0.20 0.98 -0.76 -0.83 0.17 Rolling 0.83 -0.37 0.38 0.99 0.26 0.83 0.51 0.77 0.65 0.59 0.68 1.00 0.17 0.90 0.99 0.79 -0.16 26 Table 7. Risk Pooling and Risk Relevance This table contains an assessment of the presence of (some) international risk sharing and of the relevance of idiosyncratic and aggregate risk (RIR and RAR, respectively) for some LAC countries, following the methodology outlined in section 3. The latter have been computed both with reference to total consumption (C) and household final consumption (H). Country log ca t-value RIRC RARC RIRH RARH Argentina 0.83 2.67 0.60 0.31 0.60 0.29 Belize -0.65 -0.77 0.31 -0.02 0.32 -0.02 Bolivia -0.06 -0.21 0.25 0.34 0.12 0.27 Brazil 2.05 6.94 0.29 0.44 0.09 0.21 Chile - - 0.76 - 0.69 - Colombia -0.05 -0.27 0.28 0.69 0.27 0.24 Costa Rica 0.03 0.05 0.35 0.18 0.35 0.13 Dom. Republic 0.70 1.27 0.16 0.14 0.13 0.10 Ecuador -0.33 -0.46 0.24 0.02 0.24 0.32 Guatemala -0.31 -1.05 0.39 0.39 0.59 0.37 Guyana 3.41 2.12 0.11 0.08 0.05 0.04 Honduras 0.50 0.84 0.23 0.16 0.28 0.15 Haiti - - 0.40 - 0.40 - Jamaica - - 0.49 - 0.49 - Mexico 0.47 2.26 0.67 0.36 0.64 0.33 Nicaragua - - 0.28 - 0.09 - Panama - - 0.18 - 0.12 - Peru 1.47 1.62 0.40 0.02 0.77 0.12 Paraguay 0.50 0.54 0.11 0.28 0.08 0.29 Salvador - - 0.52 - 0.63 - Trinidad - - 0.26 - 0.28 - Uruguay 0.10 0.22 0.46 0.38 0.45 0.36 Venezuela 0.62 1.41 0.26 -0.01 0.19 -0.02 27 Graph1. Insurance and Income. LAC OECD EA SSA 1.8 CHL BLZ TTO 1.3 HTI CRI URY GUY ARG MEX HND ECU GTM enti DOM PRY SLVBRA 0.8VEN PAN effic COL Coencrau JAM BOL NIC PER 0.3 Ins -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 -0.2 -0.7 Normalized per capita income 28 Graph2. Insurance and Income HIC HIC non OECD LIC LMC UMC Emerg. 1.5 tenicifefoC 1 cenarsuIn 0.5 0 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 Normalized per capita income 29 Table 8. Explaining insurance coefficients In this table are reported fixed effects, panel estimations of the impact of domestic credit by banks (DCB), per capita public consumption (PUB CONS) as a percentage of per capita GDP, total value of stocks traded (STOCKS TR.), long term indebtedness (LTD), as a percentage of GDP, the degree of international openness (OPENNESS), positive shocks to the differential in terms of trade (DLTOTP), negative shocks to differential in terms of trade (DLTOTN), and de facto floating exchange rate regime (FLOAT). Whenever possible, data are expressed in logarithms. Fixed effects (whose significance is never rejected) are not reported. 1 - REC 1 - ROL h1 - REC h1 - ROL GDPPC 0.23 -0.43 0.34 -0.66 [0.03] [0.01] [0.02] [0.00] DCB -0.10 -0.18 -0.16 -0.11 [0.00] [0.00] [0.00] [0.09] PUB. CONS. -0.16 -0.10 -0.26 [0.00] [0.01] [0.00] STOCKS TR. -0.03 -0.03 -0.07 -0.09 [0.00] [0.07] [0.00] [0.00] LTD -0.08 -0.083 -0.143 [0.02] [0.00] [0.00] LTD(-1) -0.14 [0.05] OPENNESS 0.18 0.33 0.28 [0.00] [0.00] [0.03] DLTOTP 0.31 0.78 0.43 1.11 [0.02] [0.00] [0.03] [0.00] DLTOTN -0.39 -0.73 -0.71 [0.02] [0.02] [0.04] FLOAT -0.03 -0.08 [0.10] [0.00] Obs. 157 148 151 146 Adj. R2 0.96 0.86 0.97 0.98 30 Table 9. Computing welfare gains from risk pooling. Relative Risk Av. = 3 = 6 = 9 Pooling with: LAC OECD LAC OECD LAC OECD Argentina 1.31% 1.45% 2.61% 2.89% 3.89% 4.31% Barbados 0.78% 0.94% 1.56% 1.87% 2.33% 2.80% Belize 0.48% 0.64% 0.96% 1.27% 1.43% 1.90% Bolivia -0.12% 0.04% -0.23% 0.08% -0.35% 0.11% Brazil 0.08% 0.23% 0.16% 0.45% 0.24% 0.67% Chile 0.15% 0.31% 0.31% 0.63% 0.46% 0.94% Colombia 0.11% 0.27% 0.22% 0.54% 0.34% 0.81% Dominica 0.78% 0.94% 1.56% 1.87% 2.33% 2.79% Dominican Republic 0.55% 0.71% 1.11% 1.42% 1.66% 2.13% El Salvador -0.08% 0.08% -0.15% 0.16% -0.23% 0.24% Grenada 3.77% 3.92% 7.43% 7.74% 10.99% 11.44% Guatemala -0.13% 0.02% -0.27% 0.05% -0.40% 0.07% Guyana 1.51% 1.67% 3.00% 3.32% 4.48% 4.96% Haiti 3.36% 3.51% 6.63% 6.93% 9.82% 10.27% Jamaica -0.03% 0.12% -0.05% 0.24% -0.08% 0.35% Mexico 0.54% 0.69% 1.07% 1.38% 1.60% 2.06% Panama 0.84% 1.00% 1.67% 1.99% 2.50% 2.98% Paraguay 0.70% 0.85% 1.39% 1.70% 2.08% 2.54% Trinidad and Tobago 4.93% 5.09% 9.69% 10.00% 14.28% 14.73% Uruguay 1.64% 1.79% 3.26% 3.55% 4.86% 5.29% Venezuela 0.92% 1.08% 1.84% 2.15% 2.75% 3.21% a) 5 years horizon Relative Risk Av. = 3 = 6 = 9 Pooling with: LAC OECD LAC OECD LAC OECD Argentina 2.35% 2.60% 4.64% 5.14% 6.87% 7.62% Barbados 1.58% 1.90% 3.14% 3.78% 4.67% 5.63% Belize 0.95% 1.25% 1.88% 2.50% 2.81% 3.73% Bolivia -0.22% 0.07% -0.45% 0.15% -0.67% 0.22% Brazil 0.15% 0.41% 0.29% 0.83% 0.43% 1.24% Chile 0.31% 0.63% 0.62% 1.27% 0.93% 1.90% Colombia 0.23% 0.54% 0.45% 1.08% 0.67% 1.62% Dominica 1.53% 1.83% 3.04% 3.64% 4.52% 5.42% Dominican Republic 1.11% 1.42% 2.21% 2.83% 3.29% 4.23% El Salvador -0.15% 0.16% -0.31% 0.32% -0.46% 0.49% Grenada 7.29% 7.58% 14.16% 14.74% 20.62% 21.46% Guatemala -0.27% 0.05% -0.54% 0.09% -0.80% 0.14% Guyana 3.07% 3.40% 6.07% 6.72% 9.01% 9.97% Haiti 6.48% 6.77% 12.62% 13.19% 18.43% 19.27% Jamaica -0.05% 0.21% -0.10% 0.43% -0.15% 0.64% 31 Mexico 1.04% 1.34% 2.07% 2.67% 3.10% 3.98% Panama 1.70% 2.02% 3.37% 4.01% 5.02% 5.97% Paraguay 1.36% 1.66% 2.70% 3.30% 4.02% 4.91% Trinidad and Tobago 9.68% 9.98% 18.67% 19.25% 26.97% 27.82% Uruguay 3.01% 3.28% 5.93% 6.46% 8.77% 9.55% Venezuela 1.81% 2.12% 3.60% 4.21% 5.36% 6.26% b) 10 years horizon 32 Table 10. Gross savings rate and uninsurable idiosyncratic risk In this table are reported the results of a fixed effects, panel estimation, of gross savings rate over risk sharing coefficients (under alternative definitions). Fixed effects are all significant, but not reported in the table. Gross savings rate (% GDP) 1 - 0.610 REC [0.01] 1 - 0.200 ROL [0.12] h1 - 0.188 REC [0.07] h1 - 0.138 ROL [0.07] Obs. 245 252 230 226 R2 0.56 0.53 0.62 0.61 33 Figure 1. RECURSIVE (TOTAL) CONSUMPTION INSURANCE COEFFICIENTS Argentina Bolivia Brazil Chile 1.08 .464 .79 1.68 1.07 .460 .78 1.67 .456 .77 1.66 1.06 .452 .76 1.65 1.05 .448 .75 1.64 1.04 .444 .74 1.63 1.03 .440 .73 1.62 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Colombia CostaRica DominicanRepublic Ecuador .80 1.28 1.0 1.0 0.8 .76 1.26 0.9 0.6 .72 1.24 0.8 0.4 .68 1.22 0.2 0.7 0.0 .64 1.20 -0.2 .60 1.18 0.6 -0.4 .56 1.16 0.5 -0.6 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 ElSalvador Guatemala Guyana Haiti .772 .955 1.28 1.3 .768 .950 1.24 1.2 .764 .945 1.20 1.1 .760 .940 .756 1.16 1.0 .935 .752 1.12 0.9 .930 .748 1.08 0.8 .744 .925 .740 .920 1.04 0.7 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 34 Figure 1. (continues) RECURSIVE (TOTAL) CONSUMPTION INSURANCE COEFFICIENTS Honduras Jamaica Mexico Nicaragua 1.08 .64 1.02 .575 1.04 .62 1.00 .570 1.00 .60 0.98 .565 0.96 .58 0.96 .560 0.92 .56 0.94 .555 0.88 .54 0.92 .550 0.84 .52 0.90 .545 0.80 .50 0.88 .540 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Paraguay Peru Suriname Trinidad and Tobago .88 .49 1.47 1.7 .86 1.6 .48 1.46 1.5 .84 .47 1.45 1.4 .82 .46 1.44 1.3 .80 1.2 .78 .45 1.43 1.1 .76 .44 1.42 1.0 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Uruguay Venezuela 1.32 .70 .65 1.28 .60 1.24 .55 1.20 .50 .45 1.16 .40 1.12 .35 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 35 Figure 2. RECURSIVE (HOUSEHOLD) CONSUMPTION INSURANCE COEFFICIENTS Argentina Bolivia Brazil Chile 1.12 .100 1.14 1.86 1.10 1.12 .096 1.85 1.10 1.08 1.84 .092 1.08 1.06 1.83 .088 1.06 1.04 1.82 1.04 .084 1.02 1.81 1.02 1.00 .080 1.00 1.80 0.98 .076 0.98 1.79 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Colombia Costa Rica Dominican Republic Ecuador 1.00 1.05 1.0 .9 0.95 1.04 0.9 .8 1.03 0.8 0.90 0.7 .7 1.02 0.85 0.6 1.01 .6 0.80 0.5 1.00 0.4 .5 0.75 0.99 0.3 0.70 .4 0.98 0.2 0.65 0.97 0.1 .3 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 El Salvador Guatemala Guyana 1.316 .64 1.22 1.20 1.312 .63 1.18 1.308 1.16 1.304 .62 1.14 1.12 1.300 .61 1.10 1.296 1.08 1.292 .60 1.06 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 36 Figure 2. (continues) RECURSIVE (HOUSEHOLD) CONSUMPTION INSURANCE COEFFICIENTS Honduras Mexico Nicaragua Paraguay .81 1.08 .65 .60 .80 1.04 .79 .64 .59 1.00 .78 .63 .58 .77 0.96 .76 0.92 .62 .57 .75 0.88 .74 .61 .56 .73 0.84 .72 0.80 .60 .55 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Peru Suriname Trinidad and Tobago Uruguay 0.995 1.47 2.2 1.94 0.990 1.92 1.46 2.0 0.985 1.90 0.980 1.45 1.8 1.88 0.975 1.44 1.6 1.86 0.970 1.84 1.43 1.4 0.965 1.82 0.960 1.42 1.2 1.80 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Venezuela .70 .65 .60 .55 .50 .45 .40 92 93 94 95 96 97 98 99 00 01 02 03 37 Figure 3. ROLLING (TOTAL) CONSUMPTION INSURANCE COEFFICIENTS Argentina Bolivia Brazil Chile 1.08 .464 .78 1.8 .77 1.04 .460 1.6 .76 .456 1.4 1.00 .75 .452 1.2 0.96 .74 .448 1.0 .73 0.92 .444 .72 0.8 0.88 .440 .71 0.6 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Colombia Costa Rica DominicanRepublic Ecuador .72 1.3 1.0 1.0 0.8 1.2 0.9 .68 0.6 1.1 0.8 0.4 .64 0.2 1.0 0.7 0.0 .60 -0.2 0.9 0.6 -0.4 .56 0.8 0.5 -0.6 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 ElSalvador Guatemala Guyana Haiti 1.6 1.03 1.30 1.3 1.5 1.02 1.25 1.2 1.4 1.01 1.3 1.00 1.20 1.1 1.2 0.99 1.15 1.0 1.1 0.98 1.0 0.97 1.10 0.9 0.9 0.96 1.05 0.8 0.8 0.95 0.7 0.94 1.00 0.7 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 38 Figure 3. (continues) ROLLING (TOTAL) CONSUMPTION INSURANCE COEFFICIENTS Honduras Jamaica Mexico Nicaragua 1.1 .9 1.04 .6 1.0 .4 .8 1.00 0.9 .2 0.8 .7 0.96 .0 0.7 -.2 0.6 .6 0.92 0.5 -.4 0.4 .5 0.88 -.6 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Paraguay Peru Suriname Trinidad and Tobago 1.4 .48 1.70 1.7 1.3 1.6 1.65 1.2 .44 1.5 1.60 1.1 1.4 1.0 .40 1.55 1.3 0.9 1.2 1.50 0.8 .36 1.1 1.45 0.7 1.0 0.6 .32 1.40 0.9 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Uruguay Venezuela 1.35 .8 .7 1.30 .6 1.25 .5 1.20 .4 1.15 .3 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 39 Figure 4. ROLLING (HOUSEHOLD) CONSUMPTION INSURANCE Argentina Bolivia Brazil Chile 1.04 .110 1.10 2.0 1.00 .105 1.8 1.05 0.96 .100 1.6 0.92 .095 1.00 1.4 0.88 .090 1.2 0.84 .085 0.95 0.80 .080 1.0 0.76 .075 0.90 0.8 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Colombia CostaRica Dominican Republic Ecuador 1.0 1.08 1.0 .9 0.9 0.9 1.04 .8 0.8 0.8 1.00 0.7 .7 0.6 0.7 0.96 .6 0.5 0.6 0.92 0.4 .5 0.3 0.5 0.88 .4 0.2 0.4 0.84 0.1 .3 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 El Salvador Guatemala Guyana 2.4 .67 1.6 2.2 .66 1.5 .65 2.0 1.4 .64 1.8 1.3 .63 1.6 1.2 .62 1.4 .61 1.1 1.2 .60 1.0 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 40 Figure 4. (continues) ROLLING (HOUSEHOLD) CONSUMPTION INSURANCE COEFFICIENTS Honduras Mexico Nicaragua Paraguay .9 1.15 0.8 1.2 .8 1.10 1.1 0.4 1.0 .7 1.05 0.0 0.9 .6 1.00 -0.4 0.8 .5 0.95 0.7 -0.8 .4 0.90 0.6 .3 0.85 -1.2 0.5 .2 0.80 -1.6 0.4 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Peru Suriname Trinidad and Tobago Uruguay 1.00 1.47 2.2 2.10 0.99 1.46 2.0 2.05 0.98 1.8 2.00 1.45 0.97 1.6 1.95 1.44 0.96 1.4 1.90 0.95 1.43 1.2 1.85 0.94 1.42 1.0 1.80 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 92 93 94 95 96 97 98 99 00 01 02 03 Venezuela .9 .8 .7 .6 .5 .4 92 93 94 95 96 97 98 99 00 01 02 03 41 Appendix Table 3 Table 4. CONSUMPTION INSURANCE : EMPIRICAL RESULTS Consumption risk-sharing is tested by model (7). Each test is applied twice, using two different measures of change in consumption: growth rates in total final consumption, and growth rates in household final consumption expenditure. Two different aggregates are considered, where total consumption is averaged across LAC countries and OECD countries. P-values are reported in brackets. Countries TOTAL FINAL CONSUMPTION HOUSEHOLD FINAL CONSUMPTION Macroarea: Latin America and Caribbean Macroarea: OECD Countries Macroarea: Latin American and Macroarea: OECD Countries Caribbean logGDPid log (ca) logGDPid log (ca) logGDPid log (Hca) logGDPid log (Hca) Argentina 0.995 [0.00] 1.292 [0.00] 1.037 [0.00] 0.038 [0.80] 1.034 [0.00] 1.084 [0.00] 1.101 [0.00] -0.010 [0.95] Bahamas 1.209 [0.00] 0.083 [0.88] 1.193 [0.00] 1.392 [0.13] - - - - - - - - Belize 0.930 [0.01] -0.042 [0.96] 1.288 [0.00] 0.063 [0.94] 1.04 [0.09] 1.489 [0.26] 1.701 [0.01] 0.616 [0.63] Bolivia 0.618 [0.00] 0.532 [0.00] 0.654 [0.00] 0.040 [0.79] 0.445 [0.01] 0.315 [0.05] 0.509 [0.00] 0.065 [0.69] Brazil 0.990 [0.00] 1.336 [0.00] 0.894 [0.00] 0.732 [0.00] 1.162 [0.00] 1.787 [0.00] 0.693 [0.01] 0.684 [0.09] Chile 1.467 [0.00] 0.648 [0.10] 1.437 [0.00] 0.930 [0.04] 1.582 [0.00] 0.914 [0.03] 1.644 [0.00] 1.012 [0.03] Colombia 0.693 [0.00] 0.531 [0.00] 0.718 [0.00] 0.639 [0.00] 0.895 [0.00] 0.319 [0.00] 1.045 [0.00] 0.526 [0.00] Costa Rica 1.023 [0.00] 0.768 [0.01] 1.141 [0.00] 0.768 [0.02] 0.976 [0.00] 0.664 [0.02] 1.066 [0.00] 0.675 [0.02] Dom. Republic 0.726 [0.02] 0.905 [0.03] 0.622 [0.03] 0.894 [0.04 0.612 [0.10 0.500 [0.30] 0.574 [0.09] 0.996 [0.04 Ecuador 0.964 [0.00] 0.498 [0.13] 0.842 [0.00] 0.123 [0.68] 0.642 [0.00 0.591 [0.00] 0.749 [0.00] 0.138 [0.42 Guatemala 0.880 [0.00] 0.553 [0.00] 0.921 [0.00] 0.566 [0.00] 0.800 [0.00] 0.472 [0.00] 0.841 [0.00] 0.491 [0.00] Guyana 1.185 [0.00] 1.427 [0.05] 1.012 [0.00] 0.707 [0.36] 1.150 [0.04] 1.017 [0.25] 0.963 [0.04] 0.899 [0.84] Haiti 1.077 [0.00] -0.104 [0.82] 1.069 [0.00] 0.019 [0.96] - - - - - - - - Jamaica 0.717 [0.00] -0.235 [0.47] 1.024 [0.00] -0.062 [0.86] - - - - - - - - Mexico 0.963 [0.00] 0.861 [0.00] 1.000 [0.00] 0.650 [0.00] 1.052 [0.00] 0.728 [0.00] 1.072 [0.00] 0.653 [0.00] Nicaragua 0.577 [0.00] 0.117 [0.83] 0.568 [0.00] -0.643 [0.23] - - - - - - - - Panama 0.654 [0.04] -0.021 [0.97] 0.609 [0.02] 0.548 [0.38] 0.776 [0.07] 0.635 [0.45] 0.893 [0.02] 0.931 [0.27] Paraguay 0.743 [0.02] 1.515 [0.00] 1.042 [0.00] 0.401 [0.36] 0.532 [0.10] 1.603 [0.00] 1.069 [0.00] 0.370 [0.41] Peru 0.921 [0.00] 0.591 [0.13] 0.777 [0.00] -0.197 [0.63] 0.939 [0.00] 0.478 [0.00] 0.836 [0.00] -0.090 [0.50] Suriname 1.361 [0.00] -0.086 [0.92] 1.362 [0.00] 0.307 [0.78] 1.258 [0.01] 0.243 [0.68] 1.219 [0.02] 0.275 [0.73] Trinidad 1.129 [0.00] 0.152 [0.80] 1.010 [0.00] 0.402 [0.54] 1.525 [0.00] -0.270 [0.67] 1.381 [0.00] -0.014 [0.98] Uruguay 1.324 [0.00] 1.285 [0.00] 1.190 [0.00] 0.398 [0.07] 1.641 [0.00] 1.161 [0.00] 1.301 [0.00] 0.380 [0.10] Venezuela 0.603 [0.00] 0.640 [0.12] 0.730 [0.00] 0.734 [0.13] 0.571 [0.02] 0.311 [0.49] 0.700 [0.00] 0.497 [0.00] 42 Table 5.CONSUMPTION INSURANCE COEFFICIENTS: POSITIVE AND NEGATIVE SHOCKS Countries TOTAL FINAL CONSUMPTION HOUSEHOLD FINAL CONSUMPTION Macroarea: Latin America and Caribbean Macroarea: OECD Countries Macroarea: Latin American and Macroarea: OECD Countries Caribbean logGDPid logGDPid logGDPi + - logGDPid logGDPid logGDPi + - logGDPid logGDPid logGDPid + - logGDPid logGDPi logGDPi + Argentina 0.867 1.058 0.995 [0.00] 0.991 1.070 1.037 [0.00] 0.883 1.112 1.034 [0.00] 1.050 1.137 1.101 [0.00] Bahamas 1.382 0.949 1.209 [0.00] 1.455 0.821 1.193 [0.00] - - - - - - - - Belize 0.648 1.137 0.930 [0.01] 1.010 1.548 1.288 [0.00] 0.648 1.349 1.04 [0.09] 2.088 1.339 1.701 [0.01] Bolivia 0.754 0.461 0.618 [0.00] 0.734 0.579 0.654 [0.00] 0.749 0.098 0.445 [0.01] 0.740 0.299 0.509 [0.00] Brazil 1.174 0.783 0.990 [0.00] 1.163 0.522 0.894 [0.00] 1.197 1.120 1.162 [0.00] 0.899 0.409 0.693 [0.01] Chile 1.086 1.623 1.467 [0.00] 0.876 1.919 1.437 [0.00] 0.998 1.799 1.582 [0.00] 1.010 2.182 1.644 [0.00] Colombia 0.657 0.734 0.693 [0.00] 0.603 0.803 0.718 [0.00] 0.842 0.954 0.895 [0.00] 1.034 1.054 1.045 [0.00] Costa Rica 0.875 1.173 1.023 [0.00] 1.220 1.081 1.141 [0.00] 0.962 0.989 0.976 [0.00] 1.339 0.862 1.066 [0.00] Dom. Republic 0.764 0.659 0.726 [0.02] 0.850 0.502 0.622 [0.03 0.756 0.362 0.612 [0.10 0.939 0.379 0.574 [0.09 Ecuador 0.988 0.947 0.964 [0.00] 0.864 0.833 0.842 [0.00 0.365 0.840 0.642 [0.00 0.848 0.710 0.749 [0.00 Guatemala 0.826 0.928 0.880 [0.00] 0.744 1.016 0.921 [0.00] 1.025 0.601 0.800 [0.00] 0.974 0.771 0.841 [0.00] Guyana 1.315 1.052 1.185 [0.00] 1.241 0.892 1.012 [0.00] 1.191 1.110 1.150 [0.04] 1.154 0.862 0.963 [0.04] Haiti 0.154 1.210 1.077 [0.00] 0.467 1.214 1.069 [0.00] - - - - - - - - Jamaica 0.909 0.630 0.717 [0.00] 1.212 0.725 1.024 [0.00] - - - - - - - - Mexico 0.930 0.982 0.963 [0.00] 1.047 0.959 1.000 [0.00] 1.016 073 1.052 [0.00] 1.143 1.010 1.072 [0.00] Nicaragua 0.735 0.541 0.577 [0.00] 0.611 0.553 0.568 [0.00] - - - - - - - - Panama 0.735 0.602 0.654 [0.04] 0.564 0.633 0.609 [0.02] 1.120 0.574 0.776 [0.07] 1.074 0.798 0.893 [0.02] Paraguay 0.700 0.793 0.743 [0.02] 1.055 1.028 1.042 [0.00] 0.479 0.588 0.532 [0.10] 1.107 1.031 1.069 [0.00] Peru 1.615 0.482 0.921 [0.00] 1.444 0.387 0.777 [0.00] 0.904 0.962 0.939 [0.00] 0.860 0.823 0.836 [0.00] Suriname 1.261 1.428 1.361 [0.00] 1.199 1.470 1.362 [0.00] - - 1.258 [0.01] - - 1.219 [0.02] Trinidad 0.788 1.629 1.129 [0.00] 0.742 1.503 1.010 [0.00] 1.121 2.134 1.525 [0.00] 1.044 2.004 1.381 [0.00] Uruguay 1.368 1.287 1.324 [0.00] 1.101 1.247 1.190 [0.00] 1.405 1.815 1.641 [0.00] 1.175 1.382 1.301 [0.00] Venezuela 0.603 0.602 0.603 [0.00] 0.558 0.809 0.730 [0.00] 0.345 0.673 0.571 [0.02] 0.345 0.818 0.700 [0.00] 43