84964 AUTHOR ACCEPTED MANUSCRIPT FINAL PUBLICATION INFORMATION Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia The definitive version of the text was subsequently published in Review of Economics and Statistics, 95(5), 2013-12 Published by MIT Press THE FINAL PUBLISHED VERSION OF THIS ARTICLE IS AVAILABLE ON THE PUBLISHER’S PLATFORM This Author Accepted Manuscript is copyrighted by the World Bank and published by MIT Press. It is posted here by agreement between them. Changes resulting from the publishing process—such as editing, corrections, structural formatting, and other quality control mechanisms—may not be reflected in this version of the text. You may download, copy, and distribute this Author Accepted Manuscript for noncommercial purposes. Your license is limited by the following restrictions: (1) You may use this Author Accepted Manuscript for noncommercial purposes only under a CC BY-NC-ND 3.0 Unported license http://creativecommons.org/licenses/by-nc-nd/3.0/. (2) The integrity of the work and identification of the author, copyright owner, and publisher must be preserved in any copy. (3) You must attribute this Author Accepted Manuscript in the following format: This is an Author Accepted Manuscript of an Article by Hallward-Driemeier, Mary; Rijkers, Bob Do Crises Catalyze Creative Destruction? Firm-level Evidence from Indonesia © World Bank, published in the Review of Economics and Statistics95(5) 2013-12 http://creativecommons.org/licenses/by-nc-nd/3.0/ © 2014 The World Bank DO CRISES CATALYZE CREATIVE DESTRUCTION? FIRM-LEVEL EVIDENCE FROM INDONESIA Mary Hallward-Driemeier and Bob Rijkers* Abstract—Using Indonesian manufacturing census data (1991–2001), this tating recovery and growth and the paucity of available paper rejects the hypothesis that the East Asian crisis unequivocally improved the reallocative process. The correlation between productivity firm-level studies on the impact of crises on resource alloca- and employment growth did not strengthen, and the crisis induced the exit tion, empirically discriminating between these competing of relatively productive firms. The attenuation of the relationship between paradigms is important. productivity and survival was stronger in provinces with comparatively lower reductions in minimum wages, but not due to reduced entry, chan- This paper uses plant-level manufacturing census data ging loan conditions, or firms connected to the Suharto regime suffering from Indonesia to examine the impact of the East Asian cri- disproportionately. On the bright side, firms that entered during the crisis sis on the process of creative destruction. Aggregate pro- were relatively more productive, which helped mitigate the reduction in aggregate productivity. ductivity growth decompositions help assess macrolevel trends in the contributions of entry, exit, and reallocation to productivity growth during and after the crisis. These de- I. Introduction compositions are a prelude to firm-level analysis of exit and employment growth patterns, which assesses whether and W HILE crises are periods of intensified adjustment, firm-level evidence on the impact of crises on resource allocation is limited. Moreover, their impact is the- how firm dynamics during the crisis were different from those operating in pre- and postcrisis periods. Thus, we examine both the impact of the crisis on resource allocation oretically uncertain. The idea that crises may accelerate the and the reallocative process itself. Schumpetarian (1939) process of creative destruction by Previous work on allocative efficiency has largely ‘‘cleansing’’ out unproductive arrangements and freeing up, avoided periods of crisis, and work on crises has rarely resources for more productive uses features prominently in examined firm responses. By using firm-level data to exam- macromodels (see Hall, 1995; Caballero & Hammour, ine the impact of a major crisis on resource allocation, this 1994, 1999; Gomes, Greenwood, & Rebelo, 1997). On the paper combines and contributes to several strands of litera- other hand, some recent papers suggest that rather than ture that have evolved fairly separately up until this point. being cleansing, crises scar the economy and undermine The evidence on the impact of crises on resource allocation long-run productivity growth by exacerbating market typically relies on aggregate data. The few previous at- imperfections and destroying productive firms (see Barlevy, tempts to empirically validate the predictions of the cleans- 2002 and Ouyang, 2009). Which view is correct matters for ing paradigm using firm-level data, discussed in detail in policy, since at issue is whether there is a trade-off between section II, have either failed to examine changes in the real- minimizing the short-term impact of crises and maximizing locative process itself or suffer methodological shortcom- long-run growth prospects. If crises are cleansing policies ings. Second, existing plant-level studies of reallocation to dampen short-term impacts, they may obstruct long-run dynamics have demonstrated that business cycles are impor- recovery and be not only costly but also counterproductive. tant determinants of both the pattern and pace of reallocation By contrast, if crises are scarring, policies to minimize (see the surveys by Bartelsman & Doms, 2000; Caves, short-term impacts are consistent with maximizing long-run 1998; and Syverson, 2010) but typically deliberately ex- growth prospects (see Paci, Revenga, & Rijkers, 2012). clude crisis periods. It is therefore not clear to what extent Given the critical role of the reallocative process in facili- the conclusions based on them generalize to crisis times. Third, cross-country studies based on aggregate data suggest Received for publication August 17, 2010. Revision accepted for publi- that plant dynamics are a key determinant of the depth and cation May 24, 2012. * Driemeier and Rijkers: World Bank. duration of crises (Bergoeing, Loayza, & Repetto, 2004; We thank Andrew Waxman for assistance in the preparation of the data Collier & Goderis 2009), but they do not examine how these set and paper and Ana Fernandes, Fitria Fitrani, Leo Iacovone, Kai Kai- dynamics matter. Fourth, microstudies of the impact of ser, Alex Korns, David Newhouse, Rifa Rufiadi, and William Wallace for their assistance in obtaining the data and discussing the policy environ- crises on labor market outcomes mostly rely on household ment in Indonesia. We are also indebted to Mushfiq Mobaraq and Denni and labor market data (see McKenzie 2003, 2004; Fallon & Purbasari for generously sharing their data on political connectedness. Lucas, 2002; Manning, 2000, Beegle, Frankenberg, & Tho- We thank seminar participants at the IZA/World Bank Conference on Economic Development, the World Bank Economists’ Forum, a World mas, 1999), which are ill suited for analyzing the impact of Bank Macroeconomics and Growth seminar, the 2010 NEUDC confer- crises on reallocation dynamics and labor demand. Firm- ence at MIT, Eric Bartelsman, John Giles, Pierella Paci, Luis Serven, and level data are better suited for this purpose and enable us to two anonymous referees for their useful comments. Jagadeesh Sivadasan kindly shared the code used to implement the Ackerberg-Caves-Frazer document heterogeneity in firm vulnerability and adjust- (2006) procedure. The views expressed here are our own and do not ment patterns. necessarily represent the views of the World Bank, its executive board, or Examining the impact of the East Asian crisis on the member countries. A supplemental appendix is available online at http://www.mitpress Indonesian manufacturing sector provides an interesting journals.org/doi/suppl/10.1162/REST_a_00407. case study of how financial crises reverberate through the The Review of Economics and Statistics, December 2013, 95(5): 1788–1810 Ó 2013 World Bank DO CRISES CATALYZE CREATIVE DESTRUCTION? 1789 real economy. Lessons could be relevant for recovery from tive, high-productivity firms drive relatively unproductive the current global crisis, which is also characterized by a firms out of business. Some prominent macromodels predict sharp contraction of demand, reduced access to credit, and that recessions may speed up this process by ‘‘cleansing’’ uncertainty (although with less dramatic depreciations). In out unproductive firms and freeing up resources for more addition, our analysis may be relevant for those concerned productive uses (see, e.g., Caballero & Hammour, 1994, with the impact of transitions to democracy, such as those 1999). While longitudinal manufacturing firm-level studies currently taking place in the Middle East, on private sector provide empirical evidence that the creative destruction dynamics, as we are able to assess whether the regime process facilitates productivity growth, they also raise ques- change that accompanied the crisis differentially affected tions about how this process varies over time. Decomposi- firms connected with the Suharto regime.1 tions of aggregate productivity growth only weakly support The main findings can be summarized as follows: the the hypothesis that allocative efficiency increases during data do not unequivocally support the cleansing paradigm. cyclical downturns (see Griliches & Regev, 1995; Baily The crisis led to a spike in exit, a slowdown in entry, and et al., 1992). Yet such studies typically exclude extreme excessive employment reallocation. Productive firms on economic events, and it is therefore uncertain whether les- average experienced lower employment losses, but the cor- sons based on them apply during times of crisis. relation between productivity and employment growth did In the presence of market imperfections, downturns may not strengthen. Rather than raising the productivity thresh- hamper, rather than facilitate, adjustments and protract the old for survival, the crisis was more indiscriminate in terms recovery process (see, e.g., Loayza, Perry, & Serven, 2005). of the productivity of firms driven out of business. Firms Distortionary labor market regulations and policies govern- more vulnerable to changing credit market conditions were ing firm dynamics appear particularly detrimental to the much more likely to exit during the crisis, but the attenua- efficiency of the reallocative process (see Collier & God- tion of the link between productivity and survival did not eris, 2009; Haltiwanger, Scarpetta, & Schweiger, 2008). increase with vulnerability to changing loan conditions. Barlevy (2002, 2003), moreover, has pointed out that crises The attenuation was also not an artifact of reduced entry can obstruct the process of creative destruction by exacer- rates or driven by political transition. The results are robust bating credit market imperfections, which may hurt effi- to excluding firms with connection to the Suharto regime. cient firms disproportionately as such firms are likely to Moreover, if anything, among the politically connected have higher financing needs, contradicting the creative firms, productivity became a more important determinant of destruction hypothesis. Empirical evidence supporting the firm survival. By contrast, the attenuation was particularly idea that crises exacerbate credit constraints is provided by pronounced in provinces with comparatively high minimum Blalock, Gertler, and Levine (2008) who, using the same wages, suggesting that labor regulations obstructed effi- data set as considered in this paper, show that foreign- ciency-enhancing reallocation. The effects are not all nega- owned firms, which are arguably less vulnerable to liquidity tive, however. The protective power of productivity against constraints, fared much better during the crisis than com- exit was restored postcrisis, and the crisis appears to have parable domestically owned firms.2 weeded out the weakest potential entrants; while entry rates Discriminating between the competing predictions of the were lower during the crisis, those firms that entered were cleansing and scarring paradigms requires firm-level analy- on average much more productive, which helped mitigate sis on the impact of crises on resource allocation, as it the loss in aggregate productivity. requires one to analyze the link between productivity, exit, The remainder of this paper is organized as follows. The and firm growth. Although such analysis is scant, a few stu- next section reviews related literature and describes the con- dies shed light on the debate. Liu and Tybout (1996) com- text. Section III summarizes our hypotheses and explains pare the performance of continuing plants and exiting plants our approach. Section IV describes the data, while section V in Chile and Colombia from 1980 to 1985 but find no evi- presents descriptive statistics on job flows and decomposi- dence for systematic covariance of the efficiency gap tions of aggregate productivity growth. Section VI which between these two groups of firms over the business cycle constitutes the core of the paper, presents firm survival and in either country, even though Chile suffered a recession in employment growth models. A final section concludes. 1982. Exit rates of Chilean firms increased only modestly during the recession. Casacuberta and Gandelsman (2012) examine the impact of the 2002 banking crises in Uruguay II. Related Literature and Context on resource allocation. Even during the crisis, productivity A. Related Literature 2 Similarly Oh et al. (2009) find that the Korean credit guarantee According to Schumpeter (1939), business cycles are dri- scheme for SMEs implemented in response to the Asian crisis stifled the ven by a process of creative destruction by which innova- creative destruction process by enabling relatively inefficient firms to sur- vive and maintain their size. By contrast, Borensztein and Lee (2002) find evidence that in response to the East Asian crisis, Korean banks reallo- 1 We use the data on political connections collected by Mobaraq and cated credit from conglomerate (chaebol) firms to relatively more effi- Purbasari, 2008. cient firms. 1790 THE REVIEW OF ECONOMICS AND STATISTICS was negatively correlated with exit, although the evidence regulation. During the New Order government, minimum suggests the crisis may have attenuated the link between wages had been low and enforcement was fairly lax. In real productivity and exit somewhat. Nishimura, Nakajima, and terms, they collapsed during the crisis, yet they recovered Kiyota (2005) use a cohort analysis comparing productivity very quickly. Moreover, enforcement became more strin- of entrants and survivors to examine the impact of the Japa- gent (World Bank, 2010). In short, Indonesian labor mar- nese recession on productivity growth and find that the kets during the Suharto areas were flexible but became 1996–1997 banking crisis induced the exit of some rela- more rigid postcrisis. In our empirical analysis, we will tively efficient firms among the youngest cohorts. Similarly, explore the impact of such changing labor regulations on Eslava et al. (2010) demonstrate that Colombian firms the reallocative process. dependent on external credit were more likely to exit during Indonesia provides a useful testing ground to examine the 1998–2001 recessions, even if they were highly produc- the impact of crises on resource reallocation. The unex- tive. Using manufacturing data from Russia, Brown and pected nature of the crisis facilitates identification of firm Earle (2002) show that the recession induced by the transi- responses, and Indonesia has a very detailed manufactur- tion from communism to a market-based economy coin- ing-level census, discussed in detail in section IV, which cided with an improvement in the reallocative process. allows us to measure productivity, entry, and exit dynamics In short, theoretical models yield competing predictions while controlling for a rich set of firm characteristics. More- regarding the impact of crises on resource allocation and over, although the extent to which reallocation dynamics in the empirical evidence on the impact of crises on firm developing countries resemble those in developed countries dynamics is limited and ambiguous. is an actively researched issue (see, e.g., Aw, Chung, & Roberts, 2002, and Eslava et al., 2004, for evidence of the B. Context: The Indonesian Crisis importance of creative destruction in Taiwan and Colom- bia, respectively), in their comparative analysis of harmo- The East Asian crisis struck after an extended period of nized firm-level data from seventeen developing and devel- industrialization and economic growth, which was in part dri- oped countries Bartelsman, Haltiwanger, and Scarpetta ven by expansion of labor-intensive exports (Dwor-Frecault, (2004) conclude that reallocation dynamics in Indonesia are Colaco, & Hallward-Driemeier, 1999). After a reduction in very similar to those operating in developed countries and FDI flows in response to the depreciation of the Thai baht, the other Asian economies. Thus, our findings are likely to be rupiah depreciated dramatically, precipitating a sharp in- relevant for other countries. crease in inflation rates. Interest rates were raised to defend the currency, which exacerbated the decline in demand. GDP III. Hypotheses and Approach growth contracted severely in 1997 and fell in absolute terms by over 13% in 1998. Manufacturing was one of the first and A. Hypotheses hardest-hit sectors due to its greater reliance on imported inputs, exposure to changes in foreign demand (particularly The cleansing and scarring paradigms yield competing given the importance of intraregional trade), and greater reli- testable predictions at both the macro- and the microlevels. ance on external financing, often in foreign currency, which If the cleansing view is correct, one would expect crises to became an enormous burden after devaluation. The drop in accelerate the weeding out of unproductive firms, resulting manufacturing output both preceded and exceeded the drop in a stronger association between productivity and survival in aggregate GDP. In fact, the census data suggest that manu- at the microlevel; in other words, unproductive firms would facturing suffered its largest decline in 1997. be disproportionately affected. Furthermore the correlation The crisis also led to the end of the Suharto regime, which between firm productivity and employment growth would exposed firms with ownership connections to Suharto to be expected to strengthen, as less productive firms should greater competitive pressure (Fisman, 2001; Mobarak & contract more in response to shocks. At the macrolevel, one Purbasari, 2008). To the extent that such firms had been able would expect to see a corresponding increase in the contri- to generate high profits (and consequently record high pro- bution of exit and, possibly, entry to aggregate productivity ductivity) during the Suharto era by virtue of such connec- growth, as well as stronger correlations between productiv- tions, the removal of Suharto might attenuate the relation- ity and changes in market share. ship between observed productivity and firm survival, even By contrast, if crises are scarring, one would anticipate though the likely concomitant reduction of cronyism asso- the efficiency of resource allocation to deteriorate and the ciated with connectedness is efficiency enhancing rather link between productivity, exit, and employment growth to than scarring. By virtue of having detailed information on attenuate, undermining aggregate allocative efficiency. To political connectedness with the Suharto regime at the firm- the extent that these scarring effects arise because of in- level compiled by Mobaraq and Purbasari (2008), we are creased credit market imperfections, one might expect firms able to explore this issue. more reliant on finance to be more severely affected by the The fall of Suharto also sparked prolabor pressures and crisis and the attenuation of the link between productivity precipitated the introduction of more stringent labor market and survival to be especially strong for firms more vulner- DO CRISES CATALYZE CREATIVE DESTRUCTION? 1791 able to changing loan conditions. Likewise, if they are dri- tive firms gaining market share—as well as higher contribu- ven by labor market regulation, one would expect these tions from exit and proportionate entry and increases in the effects to be particularly strong for firms located in pro- cross term—firms that are experiencing larger productivity vinces with more stringent labor regulations. Finally, if losses suffering simultaneous reductions in market share. attenuation is driven by regime change, one would expect Since the process of creative destruction may take time, we the attenuation effect to be strongest for firms with connec- present decompositions using both one- and three-year time tions to the Suharto regime. horizons. The longer time horizon offers the additional advantage of shielding against the impact of measurement error (since the signal-to-noise ratio is higher over longer B. Macrolevel Analysis: Decomposing Productivity Growth time horizons). This is important because aggregate produc- To assess whether crises catalyze or retard efficiency tivity growth decompositions are very sensitive to measure- enhancing reallocation and to analyze how industry dynamics ment error. Suppose, for example, that output is measured during crises differ from pre- and postcrisis dynamics, the with error. This will result in a spuriously high cross term evolution of aggregate productivity is decomposed using an and an underestimation of the within, and, to a lesser extent, extended version of the Foster-Haltiwanger-Krizan decompo- between terms. Thus, our decompositions will have to be sition (1998) proposed by Brown and Earle (2008): interpreted cautiously. P P P DP ¼ htÀk Dpit þ Dht ðpitÀk À PtÀk Þþ Dhit Dpit i2C i2C i2C C. Microlevel Analysis P Within P between P cross þ hit ðPt À PtÀk Þ þ hit ðpit À Pt Þ þ hitÀ1 ðpitÀk À PtÀk Þ; i2N i2N i2X Firm survival: Basic test. To examine whether crises proportionate entry disproportionate entry exit catalyze creative destruction and to examine how the deter- minants of firm-survival varied over time, a discrete-time where Pt represents average productivity at time t, D proportional hazards survival model is used (Cox, 1972). denotes changes between period Æ–k, and period t, pit Period-specific hazard rates, lit(t), are modeled as a func- represents the productivity of firm i at time t, and yit is the tion of firm productivity Pit and other covariates xit, which market share of firm i at time t. C denotes the set of incum- we interact with dummies for the crisis and the recovery, to bent firms surviving from period t – 1 to period t, N denotes assess how, during the crisis and subsequent recovery peri- the set of entrants, and X is the set of firms that exited. The ods, the relationship between exit and covariates differed first term in this decomposition represents the ‘‘within’’ from the precrisis process. Our estimable equation is effect, the contribution of within-establishment productivity growth of surviving firms, weighted by initial market share. kit ðtÞ ¼ log k0 ðtÞ þ b0xi xit þ b0CÂxi Crisis  xit The second term reflects the ‘‘between effect,’’ the contribu- þb0RecÂxi Recovery  xit þ b0p Pit þ bCP 0 Crusis  Pit tion of market share reallocation to productivity growth. þb0RecP Recovery  Pit þ b0C Crisis þ b0Rec Recovery þ vit The third term represents the ‘‘cross-effect,’’ the covariance between the within and the between effect. The fourth terms where Crisis is a dummy variable for 1997 and 1998 and captures the ‘‘proportionate entry effect,’’ the change in Recovery a dummy for the period 1999 to 2001.4 This test- average sector productivity weighted by entrants’ market ing strategy is very general as all parameters of the hazard share. The fifth term measures the ‘‘disproportionate’’ entry function are allowed to vary over time. The proportional effect, defined as the difference between entrants’ produc- hazard specification is convenient since it enables us to test tivity and average sector productivity in year t. As Brown whether firms with certain characteristics were dispropor- and Earle (2008), explained the latter term provides a better tionately more or less likely to exit in certain periods. measure of the relative contribution of entrants than the Under the null hypothesis of no short-run differential entry term in the original FHK decomposition which cov- effect of crises on creative destruction, ebCP ¼ 1. If crises aries with aggregate productivity growth because it is the catalyze creative destruction, ebCP > 1, while ebCP < 1 if sum of the disproportionate and proportionate entry terms.3 they hamper it. At the risk of belaboring the point, if ebCP ¼ The final term presents the contribution of exit. 1, this does not mean that crises are not weeding out produc- If crises catalyze creative destruction, one would expect tive firms; whether this happens also depends on bP. The the relative contribution of within-firm adjustment to aggre- interaction term tells us whether productive firms were over- gate productivity growth to be proportionately smaller than represented among the exiters relative to other periods. during less turbulent times. Rather, one would expect to see an increase in the ‘between’ term—relatively more produc- Accounting for attenuation: Finance, labor market re- gulations, reduced entry, and regime change. Salient 3 This implies that the entry effect can be positive (or negative) even if explanations for the attenuation between productivity and the average productivity of entrants is identical to that of incumbents in each year. In years where average productivity growth increases, the con- 4 tribution of net entry will be exaggerated, and in years where it decreases, Alternatively we used year dummies to allow greater flexibility in cap- the relative contributions of entrants will be underestimated. turing changes over time. The results are very consistent. 1792 THE REVIEW OF ECONOMICS AND STATISTICS survival are credit market imperfections and labor market Finally, we assess to what extent the attenuation is driven frictions impeding efficient adjustment, reduced entry, and by firms that had been benefiting from ownership connec- regime change. To test these explanations, a difference-in- tions with Suharto losing their privileged status. If this is difference approach is used. Although we do not observe the explanation for the attenuation effect, the attenuation which firms are credit constrained and which ones are not, should be especially strong for firms with such connections. we compare the precrisis, crisis, and postcrisis performance of firms that are likely to differ in their exposure to chan- Employment growth. To examine which firms grow ging credit market conditions by exploiting information on fastest and assess whether employment growth became differences in dependence on external finance following more strongly associated with productivity during the crisis, Rajan and Zingales (1998) and asset tangibility following or whether, as is the case with survival, the link between Braun (2003). These measures capture different aspects of employment and productivity was attenuated, we estimate firms’ financing needs. Indicators of external financing the following employment growth model: dependence predominantly relate to firms’ long-run finan- DLitþ1 ¼ c0P Pit þ cCP 0 Crisis  Pit þ cRecP Recovery  Pit cing needs, whereas measures of tangibility are likely to correlate with access to credit since assets that are more tan- þcx xit þ cCx Crisis  xit þ cRx Recovery  xit gible offer investors more protection against default from þcC Crisis þ cRec Recovery þ ui þ vit ; borrowers (as they offer more collateral).5 where DLitþ1 is firm growth from period t to t þ 1 and mi is We include indicators of financial characteristics, Fit, a firm-fixed effect. Under the null hypothesis that the crisis interacted with period dummies, and, moreover, interac- did not improve the allocative efficiency of employment re- tions of these measures with our productivity measure: allocation among continuing firms gCRISIS  P ¼ 0, whereas kit ðtÞ ¼ log k0 ðtÞ þ bxi 0 xi þ b0CÂxi Crisis  xi gCRISIS  P > 0 (gCRISIS  P < 0) under the alternative þb0RecÂxi Recovery  xi þ bP 0 Pit þ bCP 0 Crisis  Pit hypothesis that the crisis enhanced (diminished) the impor- tance of productivity as a determinant of firm growth. þb0RecP Recovery  Pit þ bF 0 Fit þ b0CF Crisis  Fit Serial correlation in the error term, in conjunction with þb0RecF Recovery  Fit þ bFP 0 Fi  Pit þ bCFP 0 Crisis the presence of lagged size as an explanatory variable, ÂFi  Pit þ b0RecFP Recovery  Fi  Pit would render OLS estimates of the employment growth þb0C Crisis þ b0Rec Recovery þ vit : equation biased. To address this concern, we also use a fixed-effects estimator. The fixed-effects transformation is If changing credit conditions are driving the attenuation biased due to the correlation between the transformed error effect, one would expect that firms more exposed to such changes to be more likely to exit. bCF > 0, and the attenua- and the transformed explanatory variables (Nickell, 1981), but as the OLS and fixed-effects estimators are biased in tion would be especially pronounced for firms more vulner- opposite directions, they provide a confidence interval able to such changing conditions, bCFP > 0. The protective impact of productivity should become stronger once expo- within which the true parameters lie (Bond 2002).6 sure to changing credit market conditions is accounted for. Analogous regressions are run using real minimum IV. Data wages, MWjt as a proxy for labor market regulation. Mini- The Indonesian Manufacturing Census (1991–2001) col- mum wages are a suitable proxy for labor regulation lected by the Indonesian Statistical Agency, BPS (Badan because they are politically salient, because they increased Pusat Statistik), provides the empirical basis for our analy- substantially in the aftermath of the crisis, and because they sis. It contains information on all Indonesian manufacturing varied both over time and by province, which facilitates establishments with more than twenty employees and spans identification of their impact. The null hypothesis is that the pre- and postcrisis periods, as well as the crisis itself. It minimum wages do not affect reallocation dynamics (bMW ¼ has very detailed information on employment, inputs and bCMW ¼ bRecMW ¼ bMWP ¼ bCMWP ¼ bRecMWP. outputs, industrial classification, exporting, ownership, in- In addition, we examine the impact of reduced entry. vestment behavior, and capital stock, which we measure as Having fewer entrants could result in an attenuation of the the replacement value of machinery and equipment at the link between productivity and exit in aggregate since end of the calendar year. Employment is measured as the entrants tend to be both less productive and more likely to exit. We examine this possibility by including dummies for 6 While the difference and systems GMM estimators developed by Are- whether a firm was an entrant and allowing for a differential llano and Bond (1991) are in principle capable of yielding unbiased esti- relationship between productivity and survival for entrants. mates, these estimators are not well suited for our data. The difference GMM estimator is likely to result in poorly behaved estimates when vari- ables are highly persistent, as is the case with our data (for surviving 5 Our results are also robust to using alternative measures of access to firms, the correlation between lnLt and lnLt-1 is 0.98—in both crisis and finance such as liquidity needs (Raddatz, 2006), which capture firms’ noncrisis years), while the systems GMM estimators rely on a mean sta- short-term financing needs, as well as measures of reliance on loans to tionarity assumption that is palpably undesirable in the context of a crisis finance investment. Results are omitted to conserve space but available (see Roodman, 2006, for a discussion). We therefore eschew this ap- from the authors on request. proach. DO CRISES CATALYZE CREATIVE DESTRUCTION? 1793 average number of workers per day. We augmented the data addition to conventional endogeneity concerns, our TFP with industry-level measures of financial dependence, asset estimates may be biased because our capital measure, which tangibility, employment turnover, and the natural rate of is partially imputed, is not perfectly synchronized with out- establishment entry obtained from secondary sources put and employment measures, which creates potential bias (Braun, 2003; Micco & Pages, 2004) and information on in TFP estimates. The magnitude of this bias is correlated provincial-level minimum wages (World Bank, 2010). with the size of price movements and is likely to peak during In addition, we complemented the data with two measures crisis times, when prices were most volatile. Although of political connectedness constructed by Mobaraq and Pur- value-added per worker is only a partial productivity mea- basari (2008). The first builds on an insight by Fisman (2001) sure, it does not suffer this drawback. Moreover, it is avail- and identifies firms traded on the Jakarta Stock Exchange able for a larger number of observations. (JSX) whose stock returns responded negatively to news Nonetheless, it is important to recognize that measuring reports about Suharto’s health. Mobaraq and Purbasari productivity in volatile times is tricky. Measurement error (2008) identify the major shareholders on the boards of these might induce a spurious attenuation in the relationship firms and all conglomerates run by these entrepreneurs, as between productivity and firm survival and employment well as firms owned by these conglomerates, and classify growth and could thus bias our regressions against finding those as connected. This measure, however, may identify evidence for cleansing. Productivity growth decompositions only those firms for which connections mattered or spur- are even more vulnerable to measurement error as they rely iously include firms for which an adverse stock market valua- on the accurate measurement of both productivity and mar- tion spuriously coincided with news reports about Suharto’s ket share of all firms. To ensure our results are not an artifact health. The second proxy, an indicator of whether a firm has a of measurement error, we have removed all anomalous relative of Suharto on its board, overcomes these limitations. observations from our data set (see the online appendix for a (For more information, see Mobaraq & Purbasari, 2008.) detailed discussion on how anomalous observations were The survey design affects the definitions of key explana- identified). In addition, we conduct a large number of tory variables. Entry is defined as entry into the survey; it is robustness checks, presented in section VIB, including using when establishments cross the twenty-employee threshold, a range of alternative productivity proxies, focusing exclu- not necessarily when they began operations. Conversely, sively on long-run survival using precrisis productivity exit is defined as exit from the survey; we cannot distin- (to avoid having to rely on measures of productivity guish whether firms go out of business or continue operat- obtained during the crisis) and controlling for sector-specific ing with fewer than twenty employees.7 Information on the shocks (to check our results are not driven by inappropriate capital stock was not collected in 1996. We use data from deflators). (See the online appendixes for more information 1991 to 1995 to predict the capital stock based on output, on the construction of our data and key explanatory vari- investment, material inputs, labor usage, ownership charac- ables.) teristics, whether the firm exports, province, and lagged capital.8 We also confirmed the robustness of our results by V. A Bird’s Eye-View of Reallocation: Job Flows and omitting 1996 from the regressions. Aggregate Productivity Dynamics Our preferred proxy for productivity is value-added per worker.9 We also examine the robustness of our results A. Job Flows, Entry, and Exit using TFP computed by means of the Solow and Ackerberg- Caves-Frazer (2006) procedures.10 It should be noted that in Table 1 and Figures 1 and 2 present aggregate entry, exit, and employment growth statistics. Average exit over the entire period is 8.8%, while average entry is 11.1%. Firm 7 As establishments are not required to report their closure, exit is exit spiked during the crisis—in 1997, 10.8% of firms inferred from establishments ceasing to file reports to BPS. We do not count as exits temporary lapses in reporting. Temporary exits account for exited, while in 1998, 11.2% of all firms exited—and 0.6% of all the data. dropped precipitously during the recovery in 1999 and 8 For firms that enter in 1996, we use data from 1997 to 2000 to back- 2000, to peak again in 2001 (see appendix A4). Employ- cast their capital stock using the same set of explanatory variables (but using leads rather than lags where appropriate). ment growth followed a similar trend but did not spike in 9 In our analysis, we use the log of value-added per worker. Over the 2001; before the crisis, manufacturing employment grew entire sample period, on average 10.1% of all firms reported negative quite rapidly. The crises induced substantial job losses; on value-added. In 1997, 10.2% of all firms reported negative value-added, while in 1998, 12.2% of all firms reported negative value-added. These average firms shrank employment by 1.4% in 1997 and firms are excluded from analyses that use the log of value-added per 3.7% in 1998. Employment growth recovered in 1999 and worker as a proxy for productivity. 2000 but dropped in 2001. 10 Of the two, the Solow method is our preferred TFP estimator in this context since it does not require lagged information on factor inputs and The high job losses during the crisis were driven by both thus can be computed for a greater number of observations. Moreover, the a slowdown in job creation and a spike in job destruction, ACF estimator assumes that productivity evolves according to a Markov predominantly accounted for by employment adjustment by process, thereby implicitly assuming stationarity, which may not be appropriate in the context of a crisis. By contrast, the Solow procedure incumbents (see figure 2). The share of job flows accounted allows factor shares to vary over time. for by firm entry and exit is likely to be underestimated, 1794 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 1.—ENTRY, EXIT, AND EMPLOYMENT GROWTH Employment Growth (Survivors) % Negative No Employment % Positive Entry Exit Mean SD Employment Change Change Employment Change 1992 14.54% 9.74% 2.12% 32.02% 30.03% 12.66% 41.49% 1993 10.51% 9.02% 4.27% 28.78% 28.34% 13.31% 46.98% 1994 11.70% 7.41% 2.13% 26.94% 29.75% 18.52% 39.37% 1995 17.96% 6.19% 0.80% 24.70% 26.14% 20.35% 34.69% 1996 14.41% 8.97% 0.61% 23.49% 28.44% 22.73% 33.38% 1997 8.07% 10.78% À1.41% 24.60% 36.29% 22.40% 32.39% 1998 7.93% 11.21% À3.74% 26.90% 44.05% 16.48% 30.55% 1999 7.61% 5.06% 0.60% 25.30% 30.17% 28.34% 32.14% 2000 5.65% 5.20% 0.76% 24.02% 26.11% 37.72% 29.54% 2001 9.36% 13.66% À0.81% 25.37% 33.28% 29.20% 26.46% Entry and exit are defined as entry into the survey and exit from the survey (see the appendix). Entry and exit statistics are based on the raw data; all other statistics are based on the sample that excludes outliers. FIGURE 1.—FIRM ENTRY AND EXIT FIGURE 2.—AGGREGATE JOB FLOWS however, since we observe only firms with at least twenty employees. The amount of gross reallocation far exceeded the amount required to achieve net employment adjustment, biased because of measurement error. The between term leading to enormous excess churning. The increase in was generally negative, but became less so during the crisis excess churning attests to the importance of heterogeneity (and was only positive, just, in 1998); the contribution of across firms: even during the crisis, almost a third of all reallocation of market share from less productive to more firms reported expanding employment. Nevertheless, a productive firms was very modest (though measurement striking feature of the data is how persistent employment is; error may bias the between term downward). The increase over the entire period considered, on average 21% of firms in the disproportionate entry term is also indicative of did not change their labor input each year. The figure also cleansing. Although there were fewer entrants, they were on shows a longer trend toward lower job creation in both average more productive than incumbents, and this helped aggregate and in net creation rates. mitigate the overall loss in average productivity. Yet the more negative contribution of exit during the crisis, which B. Decomposing Aggregate Productivity Growth was especially pronounced in 1997, suggests that relatively productive firms were more likely to exit, which is indica- Figure 3 presents a decomposition of the annual growth tive of scarring. in average value-added per worker. The decomposition is a Figure 4 displays the same decomposition using a three- weighted average of industry-specific decompositions con- year time window to minimize measurement error and ducted at the two-digit industry level with weights propor- avoid underestimation of the contributions of entry and exit. tional to each industry’s contribution to total output. The Lengthening the window to three years does smooth out the crisis is associated with a pronounced increase in the contri- series and increases the contributions of entry and exit, thus butions of the cross, between, and disproportionate entry underscoring that the long-run contribution of turnover to terms, as well as a decrease in the contributions of propor- productivity growth is likely to exceed its initial contribu- tionate entry, exit, and within-firm productivity growth. tion (see Liu & Tybout, 1996). However, using a longer The decomposition is only partially consistent with the window does not substantially alter the qualitative pattern cleansing paradigm. On the one hand, the improvement in of results. the contribution of the cross term, in conjunction with the Figure 5 presents a similar graph using TFP as our proxy decreased contribution of the within term, suggests that the for productivity. The graph resembles figure 4, although the firms that experienced the largest declines in productivity relative magnitude of the decrease in the contribution of the also suffered the largest reductions in market share, within term is larger, whereas the improvements in the although the magnitude of the cross term may be upward- between, cross, and proportionate entry terms appear smaller. DO CRISES CATALYZE CREATIVE DESTRUCTION? 1795 FIGURE 3.—FHK DECOMPOSITION OF THE GROWTH OF REAL VALUE-ADDED PER WORKER, OUTPUT MARKET SHARE Decompositions are conducted at the two-digit industry level and subsequently aggregated with weights proportional to each industry’s contribution to total output. FIGURE 4.—FHK DECOMPOSITION USING THREE-YEAR WINDOW: REAL VALUE-ADDED PER WORKER Decompositions are conducted at the two-digit industry level and subsequently aggregated with weights proportional to each industry’s contribution to total output. FIGURE 5.—FHK DECOMPOSITION OF THE GROWTH OF TFP (SOLOW METHOD), OUTPUT MARKET SHARE Decompositions are conducted at the two-digit industry level and subsequently aggregated with weights proportional to each industry’s contribution to total output. Overall, the aggregate productivity decompositions only period. Firms that exit are on average less productive, smal- partially support the cleansing hypothesis. The facts that ler, younger, smaller, less capital intensive, employ propor- relatively productive firms appear to have suffered some- tionately more unskilled workers, are less likely to be gov- what less, that the correlation between changes in produc- ernment owned, are more likely to be foreign owned, and tivity and changes in market share strengthened, and that are less likely to export than firms that survive. entrants were relatively more productive than incumbents Comparing across columns enables one to examine how are consistent with the cleansing hypothesis. However, the average productivity differences between surviving firms contribution of exit to aggregate productivity growth also and exiting firms evolved. The productivity gap between became negative, contradicting the cleansing hypothesis. exiting and surviving firms narrowed during the crisis; Bear in mind, however, that the results obtained using these before the crisis, the difference in the average log value- decompositions have to be interpreted with caution as they added of surviving firms compared to exiting firms was are vulnerable to measurement error. about .370. It narrowed to .205 during the crisis yet in- creased to .487 during the recovery. These productivity gaps between exiting and continuing firms are significantly VI. Firm-Level Analysis different from each other at the 1% level. The shrinking of the gap between productivity and exit accounts for an A. Firm Survival aggregate average loss of value-added per worker of about 4% over the course of the crisis (since the exit rates in both Descriptive statistics. Table 2 presents summary statis- 1997 and 1998 were approximately 12% and the gap nar- tics for survivors and exiting firms, disaggregated by time rowed by approximately 17%). 1796 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 2.—DESCRIPTIVE STATISTICS (MEANS), BY PERIOD Descriptive Statistics by Period (1 of 2) Period Precrisis Crisis Recovery ln (V/L) All 7.757 7.968 8.019 Surviving(1) 7.787 7.991 7.996 Exiting(2) 7.417 7.786 7.509 Difference(1 À2) 0.370*** 0.205*** 0.487*** ln (V/L)(demeaned by sector) All À0.133 0.070 0.112 Surviving(1) À0.110 0.093 0.079 Exiting(2) À0.395 À0.119 À0.299 Difference(1 À 2) 0.284*** 0.212*** 0.378*** ln (V/L)(demeaned by sector-year) All 0.000 0.000 0.000 Surviving(1) 0.022 0.023 0.031 Exiting(2) À0.256 À0.184 À0.351 Difference(1À2) 0.278*** 0.206*** 0.383*** TFP—Solow All 2.277 2.483 2.400 Surviving(1) 2.281 2.475 2.348 Exiting(2) 2.230 2.557 2.357 Difference(1 À 2) 0.051*** À0.082*** À0.009 TFP—Solow (demeaned by sector) All À0.075 0.118 0.039 Surviving(1) À0.070 0.117 À0.009 Exiting(2) À0.137 0.129 À0.036 Difference(1À2) 0.067*** À0.012 0.026* TFP—Solow (demeaned by sector-year) All 0.000 À0.000 0.000 Surviving(1) 0.004 À0.002 0.002 Exiting(2) À0.057 0.015 À0.038 Difference(1À2) 0.062*** À0.016 0.040** Ln (Y/L) All 8.981 9.089 9.081 Surviving(1) 9.013 9.120 9.073 Exiting(2) 8.623 8.839 8.502 Difference(1À2) 0.390*** 0.281*** 0.572*** TFP-ACF All 6.726 6.767 6.735 Surviving(1) 6.728 6.768 6.737 Exiting(2) 6.701 6.756 6.711 Difference(1À2) 0.027** 0.012 0.026 Firmage All 11.409 11.682 13.349 Surviving(1) 11.554 11.981 13.165 Exiting(2) 9.817 9.261 12.443 Difference(1À2) 1.737*** 2.720*** 0.722*** lnL All 4.216 4.149 4.178 Surviving(1) 4.254 4.211 4.216 Exiting(2) 3.801 3.655 3.644 Difference(1À2) 0.453*** 0.557*** 0.571*** Unskilled Ratio All 0.856 0.861 0.860 Surviving(1) 0.855 0.860 0.861 Exiting(2) 0.871 0.875 0.889 Difference(1À2) À0.016*** À0.016*** À0.028*** Foreign owned All 0.050 0.057 0.072 Surviving(1) 0.053 0.060 0.074 Exiting(2) 0.025 0.032 0.038 Difference(1À2) 0.028*** 0.029*** 0.036*** Government owned All 0.031 0.025 0.021 Surviving(1) 0.032 0.026 0.020 Exiting(2) 0.019 0.014 0.012 Difference(1À2) 0.013*** 0.012*** 0.008*** Exporter All 0.164 0.157 0.135 Surviving(1) 0.166 0.164 0.129 Exiting(2) 0.144 0.103 0.094 Difference(1À2) 0.022*** 0.061*** 0.035*** Ln (K/L) All 6.819 6.712 6.619 Surviving(1) 6.837 6.750 6.650 Exiting(2) 6.598 6.383 5.928 Difference(1À2) 0.239*** 0.367*** 0.721*** Financial dependence (RZ) All 0.163 0.169 0.227 Surviving(1) 0.164 0.169 0.231 Exiting(2) 0.146 0.174 0.193 Difference(1À2) 0.019*** À0.005 0.038*** Liquidity needs All 0.048 0.048 0.048 Surviving(1) 0.048 0.048 0.049 Exiting(2) 0.046 0.049 0.045 Difference(1À2) 0.002*** À0.001** 0.003*** DO CRISES CATALYZE CREATIVE DESTRUCTION? 1797 TABLE 2.—(CONTINUED) Descriptive Statistics by Period (1 of 2) Period Precrisis Crisis Recovery Tangibility All 0.314 0.313 0.316 Surviving(1) 0.315 0.315 0.316 Exiting(2) 0.304 0.298 0.316 Difference(1À2) 0.011*** 0.016*** 0.000 Minimum wages All 4.152 4.594 4.344 Surviving(1) 4.152 4.593 4.301 Exiting(2) 4.153 4.604 4.325 Difference(1À2) À0.002 À0.012*** À0.024*** Significance of the difference between surviving and exiting firms: *90% confidence interval, **95% confidence interval, ***99% confidence interval for two-tailed tÀtest. Differences in bold indicate 90% confi- dence level in rejecting that the difference between surviving and existing firms in the indicated period is not different from the difference between surviving and exiting firms precrisis. Those in italic bold indicate the 95% confidence level. ‘‘Exiting’’ implies the firm is not in the data in year t þ 1’ ‘‘surviving’’ implies the firm is present in year t þ 1. For example, firms that disappear from the data in 1997 are marked as exiting in 1996. Accordingly, ‘‘crisis’’ refers to the years 1996 and 1997, and the ‘‘precrisis’’ and ‘‘recovery’’ periods to the years before and after that; respectively. Demeaning value-added per worker by sector or by sec- survival in developing countries and this data set (see Ber- tor-year yields a similar but slightly less dramatic pattern. nard & Sjoholm 2003, Frazer, 2005, So ¨ derbom, Teal, & Thus, the aggregate attenuation effect is due to a combina- Harding, 2006); size, age, and productivity all increase the tion of more productive sectors being more severely probability of survival. Interestingly, foreign-owned firms affected by the crisis, as well as a decrease in the productiv- are less likely to exit, while exporters are more likely to ity differential between continuing and surviving firms exit, ceteris paribus. within sectors. TFP proxies exhibit a similar pattern. However, these effects are not stable over time. Starting Table 2 furthermore shows that young and small firms with the result of focal interest, the conditional correlation were especially vulnerable during the crisis, whereas expor- between productivity and crisis seems to be attenuated by ters were less likely to exit (compared to other years), per- the crisis; the crisis-productivity interaction term is always haps because of favorable exchange rate movements. During positive and significant at the 1% level, regardless of which the crisis, survivors were not, on average, more likely to productivity proxy we use. This is not consistent with operate in sectors highly dependent on external finance, cleansing; while more productive firms remain less likely to whereas pre- and postcrisis they were, suggesting that the exit, the protective impact of productivity is significantly crisis hit sectors more dependent on external finance rela- weaker than it was precrisis. On the bright side, the attenua- tively harder. Similarly, firms in sectors with higher liquidity tion effect did not last; postcrisis, the conditional correla- needs and lower levels of assets tangibility appear to have tion between productivity and survival is not significantly been particularly exposed. In short, firms’ financial charac- different from what it had been before the crisis: the protec- teristics were correlated with vulnerability to the crisis. tive impact of productivity is restored postcrisis. Young and small firms were especially vulnerable to the Modeling survival: Baseline results. Table 3 present crisis. By contrast, exporting firms did relatively well, our baseline firm-survival model, which models firm exit as perhaps because they benefited from increased international a function of the log of the age of the firm, firm size and its competitiveness due to the depreciation of the rupiah. square to allow for nonlinearity in the size-survival relation- Although exporting was associated with a higher propensity ship, the proportion of blue-collar workers in the total to exit during other periods, exporters were not ceteris pari- workforce (the ‘‘unskilled ratio’’), foreign and government bus more likely to exit during the crisis. During the recov- ownership, whether the firm exports, and productivity. All ery, some of these effects were reversed: firm age was even of these variables are interacted with crisis and recovery less strongly correlated with exit than it had been before the period dummies to assess which firms were more vulner- crisis, while exports were once again more likely to exit able to the crisis. In addition, industry, year, and province than nonexporters ceteris paribus. dummies are included to eliminate time, industry, and loca- tion effects. The first column uses value-added per worker Robustness. Tables 4, 5, and 6 present alternative spe- as our proxy for productivity and is our preferred specifica- cifications and robustness checks. To conserve space, we tion. The second column uses TFP estimated by means of report the coefficients on only our preferred productivity the Solow method, while the third column uses TFP esti- proxies, but the regressions include all explanatory vari- mated by means of the ACF procedure. Note that using ables that are included in table 3 unless indicated other- TFP leads to a substantial reduction in sample size, espe- wise. cially when we use the ACF estimates (inter alia because First, to alleviate concerns that the weakened association TFP can only be computed for firms for which we have between productivity and exit is an artifact of the difficul- information on their current and lagged capital stock). ties of measuring productivity during turbulent times table The results are consistent with the descriptive statistics 4A presents models that use deeper lags of value-added per presented in the previous section and other studies of firm worker and TFP as our productivity proxy. Since productiv- 1798 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 3.—FIRM SURVIVAL Logistic Survival Model: Baseline Model Value-Added TFP -Solow TFP-ACF Coeffficient/SE Coefficient/SE Coefficient/SE ln(V/L) 0.834*** (0.011) Crisis  ln(V/L) 1.187*** (0.025) Recovery  ln(V/L) 0.990 (0.020) TFP (Solow) 0.755*** (0.033) Crisis  TFP (Solow) 1.556*** (0.090) Recovery  TFP (Solow) 1.111* (0.069) TFP (Solow) 0.724*** (0.034) Crisis  TFP (Solow) 1.144** (0.071) Recovery  TFP (Solow) 1.075 (0.059) lnfirmage 0.829*** 0.832*** 0.813*** (0.013) (0.015) (0.017) Crisis  lnfirmage 0.845*** 0.869*** 0.975 (0.021) (0.031) (0.048) Recovery  lnfirmage 1.154*** 1.069 1.131** (0.032) (0.044) (0.055) lnL 0.243*** 0.273*** 0.237*** (0.024) (0.027) (0.026) Crisis  lnL 0.565*** 0.588*** 0.810 (0.091) (0.103) (0.164) Recovery  lnL 0.530*** 0.301*** 0.458*** (0.084) (0.064) (0.082) ln L2 1.111*** 1.095*** 1.111*** (0.011) (0.010) (0.013) Crisis  ln L2 1.046*** 1.046*** 1.004 (0.018) (0.018) (0.023) Recovery  ln L2 1.051*** 1.101*** 1.055*** (0.017) (0.023) (0.020) Unskilled Ratio 0.764** 1.113 0.847 (0.081) (0.161) (0.124) Crisis  Unskilled Ratio 1.468** 0.987 1.396 (0.245) (0.221) (0.410) Recovery  Unskilled Ratio 1.880*** 1.799** 3.577*** (0.324) (0.485) (1.240) Foreign Owned 0.826** 0.703*** 0.821* (0.076) (0.074) (0.090) Crisis  Foreign Owned 1.217 1.365* 1.464 (0.172) (0.227) (0.375) Recovery  Foreign Owned 1.419** 1.299 1.266 (0.194) (0.254) (0.293) Government Owned 0.955 0.936 0.804 (0.105) (0.112) (0.136) Crisis  Government Owned 1.206 1.424 1.726 (0.225) (0.307) (0.617) Recovery  Government Owned 1.482** 1.269 1.519 (0.292) (0.477) (0.609) Exporter 1.314*** 1.279*** 1.285*** (0.059) (0.063) (0.071) Crisis  Exporter 0.690*** 0.677*** 0.611*** (0.052) (0.060) (0.089) Recovery  Exporter 0.834** 1.308*** 1.372*** (0.065) (0.129) (0.144) Province dummies Yes Yes Yes Period dummies Yes Yes Yes Industry dummies Yes Yes Yes N 153,115 95,966 73,196 Pseudo-R2 0.075 0.072 0.058 ***p < 0.01, **p < 0.05, and *p < 0.1 Standard errors for specifications that use TFP as a proxy for productivity (columns 2 and 3) are bootstrapped using 100 replications. DO CRISES CATALYZE CREATIVE DESTRUCTION? 1799 TABLE 4.—ROBUSTNESS CHECKS I: LAGGED PRODUCTIVITY Logistic Survival Model: Robustness Checks: Alternative Productivity Measures Odds Ratios: Relative Probability of Exit Value-Added TFP-Solow Coefficient/SE Coefficient /SE Coefficient /SE Coefficient /SE A Lagged Productivity Lag Length One Year Two Years One Year Two Years ln(V/L) tÀ1 0.866*** (0.014) Crisis  ln(V/L) tÀ1 1.157*** (0.028) Recovery  ln(V/L) tÀ1 0.956** (0.021) ln(V/L) tÀ2 0.907*** (0.017) Crisis  ln(V/L) tÀ2 1.120*** (0.031) Recovery  ln(V/L) tÀ2 0.979 (0.024) TFP (Solow) tÀ1 0.740*** Crisis  TFP (Solow) tÀ1 (0.033) 1.470*** Recovery  TFP (Solow) tÀ1 (0.091) 1.114* TFP (Solow) tÀ2 (0.072) 0.767*** Crisis  TFP (Solow) tÀ2 (0.049) 1.492*** Recovery  TFP (Solow) tÀ2 (0.114) 1.380*** Controls Yes Yes Yes Yes N 134,318 110,583 70,595 58,733 Pseudo-R2 0.080 0.076 0.081 0.078 B. Coarse Productivity Ranking (by sector-year) Model Empty Full Empty Full Low Productivity (1st tercile) 1.724*** 1.330*** 1.375*** 1.341*** (0.059) (0.049) (0.060) (0.058) Medium Productivity (2nd tercile) 1.172*** 0.970 0.920* 0.905** (0.042) (0.036) (0.041) (0.041) Crisis  Low Productivity 1.365*** 6.390*** 1.091* 5.604*** (0.049) (2.604) (0.050) (2.787) Crisis  Medium Productivity 1.424*** 6.887*** 1.508*** 7.508*** (0.057) (2.814) (0.086) (3.807) Crisis  High Productivity 1.478*** 7.600*** 1.542*** 7.925*** (0.061) (3.105) (0.081) (3.946) Recovery  Low Productivity 1.100*** 4.149*** 0.683*** 12.051*** (0.037) (1.704) (0.036) (7.050) Recovery  Medium Productivity 0.922** 3.557*** 0.815*** 14.301*** (0.037) (1.465) (0.056) (8.490) Recovery  High Productivity 0.858*** 3.792*** 0.730*** 13.380*** (0.037) (1.561) (0.049) (7.971) Controls No Yes No1 Yes N 153,115 153,115 95,966 95,966 Pseudo-R2 0.024 0.074 0.020 0.072 ***p < 0.01, **p < 0.05, and *p < 0.1. Standard errors for specifications that use TFP as a proxy for productivity (columns 3 and 4) are bootstrapped using 100 replications. Controls include province, period, and industry dummies, as well as the following variables and their interactions with crisis and recovery dummies: lnL, lnL2 Unskilled Ratio, Foreign Owned, and Government Owned. 1 The ‘‘empty’’ model presented in part B includes province dummies. ‘‘Low Productivity,’’ ‘‘High Productivity,’’ and ‘‘Medium Productivity’’ denote firms in the bottom, middle, and top productivity terciles (this ranking obviously varies with the pro- ductivity proxy used). ity is very strongly correlated over time (the autocorrelation the pattern of results is robust to using these lagged produc- coefficients on log value-added per worker and the Solow tivity measures. We interpret this as strong evidence that residual are .85 and .79, respectively), lagged productivity the attenuation effect is genuine and not driven by measure- is a good proxy for current productivity. Moreover, because ment error. Incidentally, since lagged productivity measures it is measured during noncrisis times, it is arguably less vul- are not available for entrants, this table also demonstrates nerable to measurement error. While the differences in the that the attenuation effect is not solely driven by differential protective power of productivity become somewhat smaller, survival dynamics for entrants. 1800 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 5.—ROBUSTNESS CHECKS II: ALTERNATIVE SPECIFICATIONS, SAMPLE RESTRICTIONS, AND EXIT THRESHOLDS Logistic Survival Model: Additional Robustness Checks Odds Ratios: Relative Probability of Exit VA TFP (Solow) VA TFP (Solow) Coefficient/SE Coefficient /SE Coefficient /SE Coefficient /SE A. Alternative Specifications Controlling for ln(K/L) Including Sector Year Dummies ln(V/L) 0.823*** 0.837*** (0.014) (0.012) Crisis  ln(V/L) 1.212*** 1.125*** (0.035) (0.027) Recovery  ln(V/L) 1.000 0.986 (0.030) (0.022) TFP (Solow) 0.671*** 0.747*** (0.027) (0.050) Crisis  TFP (Solow) 1.642*** 1.508*** (0.099) (0.152) Recovery  TFP (Solow) 0.967 1.146 (0.068) (0.135) Ln(K/L) 1.020* 0.929*** (0.011) (0.012) Crisis  Ln(K/L) 0.953** 1.041* (0.019) (0.023) Recovery  Ln(K/L) 0.903*** 0.900*** (0.018) (0.021) Controls Yes Yes Yes Yes N 111,331 95,966 153,115 95,966 Pseudo-R2 0.078 0.074 0.082 0.078 B. Alternative Sample Restrictions Excluding Firms with Outliers Raw Data ln(V/L) 0.830*** 0.836*** (0.014) (0.010) Crisis  ln(V/L) 1.229*** 1.170*** (0.032) (0.022) Recovery  ln(V/L) 0.989 1.026 (0.026) (0.018) TFP (Solow) 0.731*** (0.036) 0.943*** Crisis  TFP (Solow) 1.629*** (0.016) (0.115) 1.148*** Recovery  TFP (Solow) 1.149 (0.031) (0.114) 1.047 Controls Yes Yes Yes Yes N 67,250 37,871 180,660 108,752 Pseudo R2 0.068 0.083 0.069 0.069 C. Different Exit Thresholds Thirty-People Threshold Fifty-People Threshold ln(V/L) 0.827*** 0.827*** (0.010) (0.010) Crisis  ln(V/L) 1.160*** 1.150*** (0.021) (0.021) Recovery  ln(V/L) 1.008 0.984 (0.017) (0.017) TFP (Solow) 0.828*** 0.943*** (0.029) (0.016) Crisis  TFP (Solow) 1.375*** 1.148*** (0.074) (0.031) Recovery  TFP (Solow) 1.107* 1.047 (0.058) (0.035) Controls Yes Yes Yes Yes N 153,115 95,966 153,115 95,966 Pseudo R2 0.073 0.076 0.056 0.055 ***p < 0.01, **p < 0.05, and *p < 0.1. Standard errors in specifications that use TFP as a proxy for productivity (columns 2 and 4) and/or control for capital per worker (panel A) are bootstrapped using 100 repli- cations. Controls include province, period, and industry dummies (except when sector-year dummies are included in which case industry and period dummies were dropped), as well as the following variables and their interactions with crisis and recovery dummies: lnL, lnL2 Unskilled Ratio, Foreign Owned, and Government Owned. DO CRISES CATALYZE CREATIVE DESTRUCTION? 1801 TABLE 6.—LONG-RUN SURVIVAL Logistic Survival Model: Odds Ratios: Relative Probability of Exit VA TFP (Solow) TFP (Solow) Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE Survival Period ( from – to) Precrisis 1993–1996 Crisis 1996–1999 Precrisis 1993–1996 Crisis 1996–1999 Three-year window ln(V/L) 0.791*** 0.939*** (0.019) (0.019) TFP (Solow) 0.757*** 1.009 (0.052) (0.066) lnfirmage 0.818*** 0.696*** 0.832*** 0.693*** (0.021) (0.015) (0.027) (0.020) lnL 0.261*** 0.114*** 0.271*** 0.123*** (0.041) (0.015) (0.045) (0.017) ln L2 1.096*** 1.177*** 1.090*** 1.166*** (0.018) (0.016) (0.018) (0.016) Unskilled ratio 0.650** 0.740** 0.841 0.806 (0.115) (0.103) (0.172) (0.134) Foreign owned 0.800* 0.950 0.685** 0.925 (0.108) (0.106) (0.101) (0.123) Government owned 1.110 1.249 0.984 1.369** (0.175) (0.191) (0.189) (0.208) Exporter 1.666*** 0.905 1.496*** 0.912 (0.117) (0.060) (0.117) (0.066) Province dummies Yes Yes Yes Yes Period dummies Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes N 14,109 18,025 10,161 11,655 Pseudo R2 0.078 0.109 0.063 0.118 Survival Period (from – to) Precrisis 1991–1996 Crisis 1996–2001 Precrisis 1991–1996 Crisis 1996–2001 Five ¼ year window ln(V/L) 0.811*** 0.885*** (0.017) (0.016) TFP (Solow) 0.723*** 1.055 (0.040) (0.068) lnfirmage 0.917*** 0.767*** 0.909*** 0.760*** (0.021) (0.015) (0.026) (0.019) lnL 0.356*** 0.137*** 0.387*** 0.151*** (0.050) (0.016) (0.056) (0.020) ln L2 1.072*** 1.156*** 1.060*** 1.143*** (0.015) (0.013) (0.015) (0.015) Unskilled ratio 0.702** 0.951 1.029 1.027 (0.115) (0.120) (0.161) (0.164) Foreign owned 0.725** 1.055 0.629*** 0.962 (0.095) (0.100) (0.092) (0.107) Government owned 0.903 1.246* 1.057 1.253 (0.123) (0.165) (0.131) (0.194) Exporter 1.339*** 1.047 1.256*** 1.046 (0.089) (0.060) (0.095) (0.071) Province dummies Yes Yes Yes Yes Period dummies Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes N 12,489 18,025 9,278 11,655 Pseudo R2 0.076 0.106 0.063 0.105 ***p < 0.01, **p < 0.05, and *p < 0.1. Standard errors in specifications that use TFP as a proxy for productivity (columns 3 and 4) are bootstrapped using 100 replications. Second, we discretize our productivity measure into ter- that controls for all other covariates, ‘‘full’’ models. The ciles, defined by sector and year. The resulting ranking is results demonstrate that while all firms are much more likely arguably less sensitive to measurement error. Moreover, it is to exit during the crisis, firms in higher-productivity terciles ordinal, which helps shield against the impact of mismea- suffered the largest increases in exit propensity. Thus, the surement that is common across all firms in a sector in a crisis decreased the survival prospects of both productive given year, such as using inappropriate deflators. Table 4B and unproductive firms yet hit productive firm disproportio- presents specifications that control only for productivity, nately hard. By contrast, the cleansing hypothesis predicts province and industry dummies, ‘‘empty’’ models, and ones that such firms should have suffered relatively less. 1802 THE REVIEW OF ECONOMICS AND STATISTICS Third, we use alternative specifications, which are pre- B. Accounting for Attenuation sented in table 5A. Controlling for capital intensity does not alter the pattern of results as shown in columns 1 and 2 in Given the robust evidence for the attenuation effect, we table 5A.11 Columns 3 and 4 furthermore demonstrate that now examine the most prominent possible explanation for the results are robust to inclusion of sector-year dummies, this effect: credit market failures,12 labor regulations, re- which control for sector-specific shocks and also shield duced entry, and political regime change. against the impact of using inappropriate deflators. Note that the attenuation effect weakens. Thus, it seems that it is Finance and firm survival. Table 7 presents regressions partially, but not exclusively, the result of more productive that examine the link between firm survival and a firm’s sectors being more severely affected by the crisis. financial characteristics. The relationship between firm sur- Fourth, table 5B shows that the pattern of results remains vival and external financing dependence, proxied by the when we include anomalous observations (see appendix B) share of assets that is financed with external funds (fol- that we have excluded from our estimation sample. The lowing Rajan & Zingales, 1998), is examined in the top half results are also robust to dropping all plants for which at of the table. The bottom half focuses on asset hardness using least one of the observed values is anomalous (we typically Braun (2003)’s indicator of asset tangibility. The usefulness exclude such anomalous observations but do not drop other of the U.S.-based external financing dependence and asset plant-year observations from plants characterized by such tangibility measures relies critically on the assumption that outliers). the U.S. rankings can be extrapolated to Indonesia, which in Fifth, by setting arbitrary size thresholds for exit and turn depends on the assumption that there are certain techno- examining how our results change, we assess whether it is logical factors that are industry specific.13 The other expla- likely that our results are due to sample selection bias natory variables are the same as those presented in table 3, because we are observing only firms with twenty or more although the first six columns exclude industry dummies but employees. Table 5C aggravates such sample selection bias control for a sector’s financial dependence, competitiveness, by excluding firms with, respectively, fewer than thirty and contestability, proxied by the Herfindahl index, the (columns 1 and 2) and fifty (columns 3 and 4) employees ‘‘natural’’ rate of entry and of employment turnover. and defining a firm as having exited when it either disap- The results provide ample evidence that changing credit pears from the data altogether or fails to report having more conditions were an important driver of firm exit during the than, respectively, 29 or 49 employees in any of the subse- crisis; firms operating in industries more dependent on quent years. The results are robust to using different exit external finance and with lower asset tangibility were ceteris thresholds and, if anything, using higher thresholds for exit paribus significantly more likely to exit during the crisis, reduces the attenuation effect; to the extent there is sample both in absolute terms and relative to other periods, as is evi- selection bias because we are only observing firms with denced between the significant interactions between the cri- more than 20 employees, it appears to be making it harder sis dummy and these financial characteristics. These results to reject cleansing. hold using both productivity proxies. However, controlling Finally, to address the concern that exit is a protracted for these financial characteristics does not eliminate the process and that by focusing on short-term effects, we are attenuation effect. missing the action, table 6 presents estimates using longer In addition, the protective impact of productivity in time horizons using our baseline model but without crisis industries more dependent on external finance rose signifi- and recovery interaction terms. The top panel presents mod- cantly (columns 2 and 4). By contrast, the crisis interactions els of the likelihood of being in business in 1996 for firms between productivity and asset hardness, as well as liquidity operating in 1993 and juxtaposes those with models of the needs, are not statistically significant (columns 2 and 4). likelihood of being in business in 1999 conditional on oper- Thus, it appears that firms in sectors that are more sensitive ating in 1996. The bottom panel presents models of surviv- to changing credit conditions were hit harder, but this does ing from 1991 until 1996 and from 1996 to 2001. The pat- not account for the observed attenuation of the conditional tern of results does not change when longer time horizons correlation between productivity and survival. are considered; productivity offers less protection during the In summary, firms more exposed to fluctuations in credit crisis than it did precrisis, ceteris paribus. Incidentally, this market conditions were hit harder by the crisis, yet these exercise also offers another check against the influence mea- effects cannot fully account for the attenuation of the link surement error; since our productivity measures are based between productivity and exit during the crisis. If anything, on the precrisis periods, they are not vulnerable to potential 12 mismeasurement of productivity during crisis times. They may also explain the attenuation between firm growth and pro- ductivity and explain reduced entry (see Aghion, Fally, & Scarpetta, 2007, for evidence on the impact of credit constraints on firm entry and 11 Our preferred specifications do not control for capital because this growth). 13 would introduce sample selection bias as the response rate for capital is The impact of liquidity needs proxied by the inventories to sales ratio far lower for measures of the capital stock than for other variables. It is (following Raddatz, 2006) does not rely on this assumption as this mea- also missing in the 1996 data, requiring it to be estimated for that year sure was constructed using the SI data. Again, while not shown, the (see the appendix). results with this measure give the same results. DO CRISES CATALYZE CREATIVE DESTRUCTION? 1803 TABLE 7.—FINANCE AND FIRM SURVIVAL Logistic Survival Model: Additional Robustness Checks Odds Ratios: Relative Probability of Exit VA TFP (Solow) Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE A. Financial Dependence (RZ) Baseline Extended Baseline Extended RZ_fin_dependence 0.988 0.313*** 1.184** 0.914 (0.070) (0.108) (0.092) (0.349) Crisis  RZ_fin_dependence 1.329** 6.727*** 1.167 1.049 (0.148) (3.818) (0.156) (0.696) Recovery  RZ_fin_dependence 0.745*** 1.198 0.446*** 0.169** (0.077) (0.567) (0.067) (0.133) ln(V/L) 0.836*** 0.819*** (0.011) (0.012) Crisis  ln(V/L) 1.160*** 1.194*** (0.026) (0.029) Recovery  ln(V/L) 0.991 0.999 (0.021) (0.023) RZ  ln(V/L) 1.168*** (0.053) Crisis  RZln(V/L) 0.805*** (0.059) Recovery  RZln(V/L) 0.939 (0.059) TFP (Solow) 0.780*** 0.759*** (0.029) (0.040) Crisis  TFP (Solow) 1.592*** 1.578*** (0.098) (0.138) Recovery  TFP (Solow) 1.173** 1.052 (0.081) (0.112) RZ  TFP (Solow) 1.126 (0.196) Crisis  RZ  TFP (Solow) 1.028 (0.270) Recovery  RZ  TFP (Solow) 1.549 (0.531) Controls Yes Yes Yes Yes N 153,115 153,115 95,966 95,966 Pseudo-R2 0.074 0.074 0.071 0.071 B. Tangibility Baseline Extended Baseline Extended Tangibility 0.962 7.892** 1.297 3.230 (0.203) (8.020) (0.341) (2.844) Crisis  Tangibility R 0.341*** 0.298 0.329*** 0.022** (0.115) (0.511) (0.141) (0.034) Recovery  Tangibility 2.651*** 0.018** 1.082 0.405 (0.851) (0.029) (0.484) (0.766) ln(V/L) 0.836*** 0.911** (0.011) (0.039) Crisis  ln(V/L) 1.161*** 1.150** (0.025) (0.080) Recovery  ln(V/L) 0.981 0.799*** (0.020) (0.054) Tangibility  ln(V/L) 0.759** (0.099) Crisis  Tangibility  ln(V/L) 1.026 (0.222) Recovery  Tangibility  ln(V/L) 1.934*** (0.398) TFP (Solow) 0.786*** 0.892 (0.030) (0.106) Crisis  TFP (Solow) 1.581*** 1.116 (0.100) (0.222) Recovery  TFP (Solow) 1.170** 1.021 (0.082) (0.266) Tangibility  TFP (Solow) 0.674 (0.233) 1804 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 7.—(CONTINUED) Logistic Survival Model: Additional Robustness Checks Odds Ratios: Relative Probability of Exit VA TFP (Solow) Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE B. Tangibility Baseline Extended Baseline Extended Crisis  Tangibility  TFP (Solow) 2.923* (1.736) Recovery  Tangibility  TFP (Solow) 1.527 (1.139) Controls Yes Yes Yes Yes N 153,115 153,115 95,966 95,966 Pseudo-R2 0.074 0.074 0.070 0.070 ***p < 0.01, **p < 0.05, and *p < 0.1. Standard errors in specifications that use TFP as a proxy for productivity (columns 3 and 4) are bootstrapped using 100 replications. Controls include province, period, and industry dummies, as well as the following variables and their interactions with crisis and recovery dummies: lnL, lnL2 Unskilled Ratio, Foreign Owned, and Government Owned. TABLE 8.—MINIMUM WAGES AND FIRM SURVIVAL Logistic Survival Model: Labor Regulations and Firm Survival Odds Ratios: Relative Probability of Exit VA TFP (Solow) TFP (Solow) Levels Interacted Levels Interacted Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE ln(V/L) 0.820*** 0.847*** (0.011) (0.014) Crisis  ln(V/L) 1.189*** 1.049 (0.026) (0.036) Recovery  ln(V/L) 1.040* 1.005 (0.021) (0.023) MW  ln(V/L) 1.194*** (0.062) MW  Crisis  ln(V/L) 1.176*** (0.064) MW  Recovery  ln(V/L) 0.819*** (0.049) TFP (Solow) 0.753*** 0.746*** (0.031) (0.034) Crisis  TFP (Solow) 1.555*** 1.417*** (0.094) (0.144) Recovery  TFP (Solow) 1.111 1.137* (0.074) (0.081) MW  TFP (Solow) 0.950 (0.137) MW  Crisis  TFP (Solow) 1.489** (0.296) MW  Recovery  TFP (Solow) 0.492** (0.154) Minimum Wage (log) 0.821 0.215*** 0.725* 0.794 (0.111) (0.088) (0.124) (0.280) Crisis  Minimum Wage (log) 0.958 0.795 1.247 1.060 (0.144) (0.134) (0.262) (0.252) Recovery  Minimum Wage (log) 0.168*** 0.189*** 0.117*** 0.130*** (0.028) (0.033) (0.035) (0.041) Controls Yes Yes Yes Yes N 153,115 153,115 95,966 95,966 Pseudo R2 0.076 0.076 0.074 0.074 ***p < 0.01, **p < 0.05, and *p < 0.1. Standard errors in specifications that use TFP as a proxy for productivity (columns 3 and 4) are bootstrapped using 100 replications. Controls include province, period, and industry dummies, as well as the following variables and their interactions with crisis and recovery dummies: lnL, lnL2 Unskilled Ratio, Foreign Owned, and Government Owned. MW denotes demeaned real mini- mum wages. to the extent that the crisis had a differential impact on Labor regulations. Table 8 presents survival models firms with different financial characteristics, it induced a that control for provincial-level real minimum wages and cleansing effect among firms more exposed to the changing allow their impact to vary over time (the baseline models credit conditions. presented in columns 1 and 3) as well as to interact with DO CRISES CATALYZE CREATIVE DESTRUCTION? 1805 TABLE 9.—ENTRY AND ATTENUATION Logistic Survival Model: Entrants and Firm Survival Odds Ratios: Relative Probability of Exit Value-Added TFP (Solow) Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE 0.836*** 0.815*** ln(V/L) (0.011) (0.012) 1.182*** 1.204*** Crisis  ln(V/L) (0.025) (0.028) 0.988 1.007 Recovery  ln(V/L) (0.020) (0.021) 1.110*** Entrant  ln(V/L) (0.031) 0.930 Crisis  Entrant  ln(V/L) (0.046) 0.977 Recovery  Entrant  ln(V/L) (0.061) 0.757*** 0.730*** TFP (Solow) (0.031) (0.032) 1.554*** 1.605*** Crisis  TFP (Solow) (0.093) (0.107) 1.109 1.130* Recovery  TFP (Solow) (0.075) (0.079) 1.141* Entrant  TFP (Solow) (0.082) 0.887 Crisis  Entrant  TFP (Solow) (0.129) 1.028 Recovery  Entrant  TFP (Solow) (0.250) 1.141* Entrant 1.302*** 0.599** 1.499*** 1.118 (0.052) (0.129) (0.079) (0.181) Crisis  Entrant 0.577*** 0.985 0.582*** 0.758 (0.041) (0.378) (0.052) (0.285) Recovery  Entrant 0.591*** 0.694 0.525*** 0.482 (0.053) (0.331) (0.070) (0.299) Controls Yes Yes Yes Yes N 153,115 153,115 95,966 95,966 Pseudo-R2 0.075 0.076 0.073 0.073 ***p < 0.01, **p < 0.05, and *p < 0.1. Standard errors in columns 3 and 4 are bootstrapped using 100 replications. Controls include province, period, and industry dummies, as well as the following variables and their interactions with crisis and recovery dummies: lnL, lnL2 Unskilled Ratio, Foreign Owned, and Government Owned. productivity (the extended specifications presented in col- pronounced during the crisis; the crisis interaction term umns 2 and 4). Since our models include province dum- between minimum wages and productivity is strongly sta- mies, the impact of minimum wages is essentially identified tistically significant using both log value-added per worker off within-province temporal variation in real minimum and the Solow residual, indicating that relatively productive wage levels. firms confronted with relatively low reductions in real mini- The baseline models suggest that firm survival was not mum wages were proportionately more likely to exit. Note on average strongly correlated with minimum wage levels also that including this interaction reduces the coefficient before and during the crisis, though postcrisis firms facing on the crisis-productivity interaction term. In other words, higher minimum wages were less likely to exit. Controlling labor regulations appear to provide a partial explanation for for minimum wages does not reduce the attenuation effect. the attenuation effect. Yet the extended specifications suggest that the baseline specifications hide significant heterogeneity; firms in pro- Reduced entry. Table 9 presents regressions that vinces with higher minimum wages are on average less include a dummy for whether a firm is an entrant, interacted likely to exit, yet the odds ratio associated with the interac- with period dummies and productivity. The results suggest tion between value-added per worker and demeaned pro- that while entrants’ survival is typically less strongly corre- vince-level minimum wage levels is significantly larger lated with productivity, this result was not different during than 1, indicating that highly productive firms are more the crisis. Allowing for differential productivity-survival likely to exit when minimum wages are relatively high. dynamics for entrants does not reduce the attenuation effect. Labor regulation thus appears to interfere with market Thus, the attenuation effect documented in this paper is not selection. Importantly, this distortionary effect is especially simply an artifact of reduced entry. 1806 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 10.—CONNECTIONS AND CREATIVE DESTRUCTION Logistic Survival Model: Impact of Political Connections on Survival, 1997–2001 Odds Ratios: Relative Probability of Exit VA TFP (Solow) Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE A. Connected—Based on JSX Response Baseline Extended Baseline Extended ln(V/L)’96 0.798*** 0.802*** (0.028) (0.028) TFP (Solow) 1.110 1.147 (0.186) (0.190) Connected, JSX regressions 1.513 92.554 1.374 596.378*** (0.586) (255.060) (0.643) (1,406.310) Connected, JSX regressions  0.636 ln(V/L)’96 (0.197) Connected, JSX regressions  0.078** TFP (Solow) (0.082) Controls Yes Yes Yes Yes N 11,877 11,877 6,957 6,957 Pseudo-R2 0.101 0.101 0.078 0.080 B. Suharto Family Member on the Board Baseline Extended Baseline Extended ln(V/L)’96 0.799*** 0.801*** (0.028) (0.028) TFP (Solow) 1.113 1.125 (0.107) (0.108) Connected, Suharto 1.244 51.671 1.622 117.730* (0.669) (209.327) (0.594) (338.755) Connected, Suharto  0.658 ln(V/L)’96 (0.305) Connected, Suharto  0.158 TFP (Solow) (0.204) Controls Yes Yes Yes Yes N 11,877 11,877 6,957 6,957 Pseudo-R2 0.101 0.101 0.078 0.078 C. Suharto Family Member on the Board and/or JSX Connection (A&B) Baseline Extended Baseline Extended ln(V/L)’96 0.798*** 0.804*** (0.028) (0.028) TFP (Solow) 1.113 1.160 (0.107) (0.160) Connected (JSX/Suharto) 1.390 227.982** 1.622 1,871.488* (0.482) (576.428) (0.594) (7,284.427) Connected, (JSX/Suharto)  0.565** ln(V/L)’96 (0.164) Connected (JSX/Suharto)  0.046 TFP (Solow) (0.093) Controls Yes Yes Yes Yes N 11,877 11,877 6,957 6,957 Pseudo-R2 Yes Yes 0.078 0.081 ***p < 0.01, **p < 0.05, and *p < 0.1, standard errors in columns 3 and 4 block bootstrapped using 100 replications. Controls include province dummies, as well as lnL, lnL2 Unskilled Ratio, Foreign Owned, Government Owned, and Exporter. Connectedness and creative destruction. Another pos- ship information from 1995 to 1997. Thus, to test the role sible explanation of the attenuation effect is that the fall of of political connections, we look at survival patterns from Suharto may have hurt firms affiliated with the Suharto 1997 to 2001, controlling too for factors in our baseline spe- regime disproportionately (see Fisman, 2001, and Mobaraq cifications presented in table 3.14 Also note that relatively & Purbasari, 2008). If these firms were highly productive, few firms are identified as being politically connected to their exit may result in an attenuation of the link between productivity and survival. Table 10 examines this possibi- 14 Since the proxy for connectedness is in part based on stock market lity. The data on political connectedness are available for reactions over the period 1995 to 1997, conditioning on connectedness in firms in 1997, constructed using JSX and board member- prior periods might thus induce survivor bias. DO CRISES CATALYZE CREATIVE DESTRUCTION? 1807 TABLE 11.—EMPLOYMENT GROWTH Ln (V/L)empvafe TFP (Solow) OLS FE OLS FE Coefficient/SE Coefficient/SE Coefficient/SE Coefficient/SE ln(V/L) 0.023*** 0.009*** (0.001) (0.002) Crisis  ln(V/L) À0.005** 0.002 (0.002) (0.002) Recovery  ln(V/L) À0.001 0.004** (0.002) (0.002) TFP (Solow) 0.004 0.025*** (0.003) (0.005) Crisis  TFP (Solow) À0.004 À0.018*** (0.005) (0.006) Recovery  TFP (Solow) À0.004 À0.021*** (0.004) (0.006) Lnfirmage À0.018*** 0.032*** À0.022*** 0.028*** (0.001) (0.006) (0.001) (0.008) Crisis  lnfirmage À0.000 À0.009*** 0.002 À0.012** (0.002) (0.003) (0.003) (0.005) Recovery  lnfirmage 0.004 À0.011** 0.008*** À0.017*** (0.002) (0.005) (0.003) (0.007) lnL À0.069*** À0.570*** À0.028*** À0.490*** (0.009) (0.022) (0.008) (0.029) Crisis  lnL À0.118*** À0.133*** À0.124*** À0.128*** (0.014) (0.013) (0.014) (0.015) Recovery  lnL À0.023* À0.082*** À0.038*** À0.085*** (0.012) (0.013) (0.014) (0.017) lnL2 0.004*** 0.014*** 0.001 0.006* (0.001) (0.002) (0.001) (0.003) Crisis  ln L2 0.010*** 0.011*** 0.010*** 0.010*** (0.001) (0.001) (0.001) (0.002) Recovery  L2 0.002* 0.007*** 0.003** 0.008*** (0.001) (0.001) (0.001) (0.002) Unskilled Ratio 0.013 0.029** À0.016 0.025 (0.010) (0.014) (0.011) (0.018) Crisis  Unskilled Ratio 0.009 0.013 0.023 0.004 (0.018) (0.018) (0.023) (0.023) Recovery  Unskilled Ratio À0.022 0.013 À0.008 0.014 (0.014) (0.017) (0.016) (0.021) Foreign Owned 0.034*** À0.013 0.049*** À0.021 (0.006) (0.013) (0.006) (0.013) Crisis  Foreign Owned À0.005 0.031*** À0.003 0.040*** (0.011) (0.012) (0.012) (0.013) Recovery  Foreign Owned À0.023*** 0.023** À0.031*** 0.021* (0.009) (0.011) (0.011) (0.012) Government Owned À0.011 0.000 À0.011 À0.005 (0.009) (0.013) (0.008) (0.016) Crisis  Government Owned 0.032* 0.010 0.042*** 0.015 (0.017) (0.018) (0.013) (0.022) Recovery  Government Owned 0.001 À0.025 0.008 À0.028 (0.018) (0.021) (0.024) (0.024) Exporter 0.015*** À0.002 0.012*** À0.003 (0.004) (0.005) (0.004) (0.005) Crisis  Exporter 0.030*** 0.018** 0.033*** 0.018** (0.007) (0.007) (0.007) (0.008) Recovery  Exporter À0.007 0.000 0.004 0.002 (0.006) (0.008) (0.006) (0.008) Province dummies Yes Yes Yes Yes Period dummies Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes N 138,997 138,997 88,530 88,530 R2 0.029 0.219 0.025 0.213 Adjusted R2 0.029 0.219 0.024 À0.048 P < 0.01, **p < 0.05, and *p < 0.00. Standard errors of specifications that use TFP as a proxy for productivity (columns 3 and 4) are bootstrapped using 100 replications. Suharto; 210 firms that are connected based on the stock sion. Panel A uses the connectedness measure that is based market response to adverse news about Suharto’s Health, on the response of the Jakarta Stock Exchange to news and 97 firms have a Suharto family member on their boards, about Suharto’s health and the business networks of those which may make it hard to estimate the effects with preci- adversely affected by this news, while Panel B uses as a 1808 THE REVIEW OF ECONOMICS AND STATISTICS proxy for connections whether a firm had a Suharto family VII. Conclusion member on its board. Panel C combines these two defini- tions. The baseline regressions presented in columns 1 and While crises are recognized to be periods of intensified 3 merely control for connectedness, while the extended spe- adjustment and aggregate studies suggest that firm dynamics cifications presented in columns 2 and 4 include interac- are a key determinant of the depth and duration of crises, tions between connectedness and productivity. firm-level evidence on their impact on the efficiency of The baseline regressions demonstrate that firms with resource allocation is scant. Perhaps because of the paucity political connections to Suharto were not, on average, more of the empirical evidence, there is an active debate as to likely to exit over the period 1997 to 2001 ceteris paribus. whether crises have a silver lining by improving resource However, the extended models suggest that among firms allocation. On the one hand, a host of macroeconomic mod- with connections to the Suharto regime, the least productive els is predicated on the idea that the additional competitive ones were most at risk of exiting as the correlation between pressure induced by the crisis will hurt inefficient producers productivity and survival strengthened, while being con- disproportionately. They predict that the crisis will nected became associated with an increased risk of exit. ‘‘cleanse’’ out unproductive firms and reallocate resources Note, however, that the estimated coefficients in these toward more efficient producers. On the other hand, a series regressions are large and imprecisely estimated, reflecting of recent papers point out that in the presence of market the fact that few firms were identified as being politically imperfection, these conclusions may be overturned and that connected. Thus, the attenuation effect does not appear to crises may scar the economy by driving productive firms out be driven by regime change. of business. Using Indonesian manufacturing census data from 1991 to 2001 to examine the impact of the East Asian crisis on C. Employment Growth resource allocation, this paper rejects the hypothesis that the Firm survival and employment growth are determined by crisis unequivocally improved the reallocative process. the same data-generating process, notably the one determin- Decompositions of aggregate productivity growth reveal ing firm size, and we therefore use the same explanatory that firms that suffered the largest productivity losses also variables as in the firm survival models. The results are pre- suffered the largest reductions in market share. In addition, sented in table 11. Columns 1 and 3 present OLS specifica- market share reallocation between firms contributed more tions, while columns 2 and 4 present fixed-effects models, positively to average productivity than during noncrisis which effectively estimate deviations from growth trends. times, although the magnitude of this contribution was mod- Columns 1 and 2 use value-added per worker as the proxy est and positive in absolute terms only in 1998. However, for productivity, whereas columns 3 and 4 use TFP, mea- the correlation between productivity and employment sured by the Solow residual. growth did not strengthen, which is concerning given the It is very difficult to predict employment growth as evi- excessive amount of job reallocation taking place during the denced by the consistently low R2s, even though we are crisis. More worrying, the link between productivity and including a rich set of firm characteristics and dummy vari- exit of existing firms was significantly attenuated during the ables. The results are generally consistent with the literature crisis, suggesting that the crisis was less discriminating in on firm growth: younger firms, smaller firms, and more pro- terms of the productivity of firms driven out of business. ductive firms grow faster (see Bigsten and Gebreeyesus, Fortunately, postcrisis, the link between firm survival and 2007). The crisis appears to have attenuated the link be- productivity was restored, suggesting that the crisis did not tween employment growth and productivity somewhat permanently scar the Schumpetarian process of creative (column 1), but this finding is not very robust; it does not destruction. In addition, although there were fewer entrants, obtain when we estimate the growth model by OLS and use the contributions of entrants rose; the firms that entered were TFP as a proxy for productivity or control for fixed-effects typically more productive than incumbents, and this helped and use value-added per worker as our productivity proxy. mitigate the loss in productivity. In other words, the crisis The finding that more productive firms that survived were appears to have weeded out the weakest potential entrants, not less likely to shed labor, ceteris paribus, suggests that which helped mitigate the loss in aggregate productivity. employment reallocation among surviving firms was not Labor market imperfections more so than credit market especially efficiency enhancing during the crisis. Interest- imperfections help account for the attenuation effect. Firms ingly, the relationship between productivity and employ- in sectors more dependent on external finance and with lower ment growth appears to continue to be attenuated postcrisis. levels of asset tangibility were indeed disproportionately Other results accord with intuition: exporters fared sig- more likely to exit during the crisis. However, controlling nificantly better during the crisis, whereas large firms shed for these financial characteristics did not reduce the condi- more labor, in part because such firms were less likely to go tional attenuation effect, suggesting that changing credit out of business (our regressions are conditional on firm sur- market conditions do not account for the attenuated link vival). Government firms also appear to have shed less between productivity and exit. By contrast, labor regulations labor. provide a possible explanation for the attenuation of the link DO CRISES CATALYZE CREATIVE DESTRUCTION? 1809 between productivity and firm survival; during the crisis, Blalock, Garrick, Paul Gertler, and David I. Levine, ‘‘Financial Con- straints on Investment in an Emerging Market Crisis,’’ Journal of productivity was a relatively less important determinant of Monetary Economics 55:3 (2008), 568–591. survival in provinces with high minimum wages, suggesting Bond, Stephen R., ‘‘Dynamic Panel Data Models: A Guide to Micro Data that labor regulations possibly distorted the adjustment pro- Methods and Practice,’’ Portuguese Economic Journal 1: 2 (2002), 141–162. cess. Finally, the attenuation is not a statistical artifact due to Borensztein, Eduardo, and Jong-Wha Lee, ‘‘Financial Crisis and Credit a reduction in entry and is not driven by firms that had been Crunch in Korea: Evidence from Firm-Level Data,’’ Journal of able to achieve high profits by virtue of their connections Monetary Economics. 49:4 (2002), 853–875. with Suharto being differentially affected. If anything, the Braun, Matias, ‘‘Financial Contractibility and Asset Hardness,’’ UCLA mimeograph (2003). crisis induced a cleansing effect among these firms. Brown, J. David, and John Earle, ‘‘Gross Job Flows in Russian Industry This work has focused on the impact of the East Asian before and after Reforms: Has Destruction Become More Crea- crisis on Indonesia. Because it is the fourth most populous tive?,’’ Journal of Comparative Economics 30 (2002), 96–133. ——— ‘‘Understanding the Contributions of Reallocation to Productivity country and widely portrayed as a ‘‘tiger cub’’ with a bright Growth: Lessons from a Comparative Firm-Level Analysis,’’ IZA future, understanding the crisis dynamics in Indonesia is of working paper 3683 (2008). inherent interest. Focusing on a developing country may be Caballero Ricardo, and Mohamad Hammour, ‘‘The Cleansing Effect of Creative Destruction,’’ American Economic Review 84 (1994), testing the cleansing hypothesis in a context where it is 1350–1368. more likely to be rejected. However, with crises more pre- ——— ‘‘Improper Churn: Social Costs and Macroeconomic Conse- valent in developing countries, this can be a particularly quences,’’ MIT mimeograph (1999). Casacuberta, Carlos, and Nestor Gandelman, ‘‘Protection, Openness, and relevant case to examine. The shock that hit Indonesia was Factor Adjustment: Evidence from the Manufacturing Sector in also large, although the recent financial crisis shows that far Uruguay,’’ Economic Development and Cultural Change 60 more developed countries have not been immune to signifi- (2012), 597–629. Caves, Ricard E., ‘‘Industrial Organization and New Findings on the cant financial shocks. Assessing the extent to which our Turnover and Mobility of Firms,’’ Journal of Economic Literature findings generalize to other crisis episodes is a promising 36:4 (1998),1947–1982. area for further research. Collier, Paul, and Benedikt Goderis, ‘‘Structural Policies for Shock-Prone Developing Countries,’’ Oxford Economic Papers 61:4 (2009), 703–726. Cox, David R., ‘‘Regression Models and Life Tables,’’ Journal of the Royal Statistic Society 34 (1972), 187–202. REFERENCES Dwor-Frecault, Dominique, Francis Colaco, and Mary Hallward-Driemeier, Asian Corporate Recovery (Washington DC: World Bank, 1999). Ackerberg, Daniel, Kevin Caves, and Garth Frazer, ‘‘Structural Identifica- Eslava, Marcela, Arturo Galindo, Marc Hofstetter, and Alejandro tion of Production Functions,’’ UCLA mimeograph, (2006). Izquierdo, ‘‘Scarring Recessions and Credit Constraints: Evidence Aghion, Philippe, Thibaut Fally, and Stefano Scarpetta, ‘‘Credit Con- from Colombian Firm Dynamics,’’ Universidad de Los Andes/ straints as Barriers to the Entry and Post-Entry Growth of Firms,’’ CEDE working paper 007711 (2010). Economic Policy 22 (2007), 731–779. Arellano, Manuel, and Stephen Bond, ‘‘Some Tests of Specification for Eslava, Marcela, John Haltiwanger, Adriana Kugler, and Maurice Kugler. Panel Data: Monte Carlo Evidence and an Application to Employ- ‘‘The Effects of Structural Reforms on Productivity and Profitabil- ment Equations,’’ Review of Economic Studies 58:2 (1991), 277– ity Enhancing Reallocation: Evidence from Colombia,’’ Journal of 297. Development Economics, 75:2 (2004), 333–371. Aw, Bee Yan, Sukkyun Chung, and Mark J. Roberts, ‘‘Output, and Fail- Fallon, Peter R., and Robert E. B. Lucas, ‘‘The Impact of Financial Crises ure: A Comparison of Taiwanese and Korean Manufacturers,’’ on Labor Markets, Household Incomes, and Poverty: A Review of Economic Journal 113 (2002), 485–510. Evidence,’’ World Bank Research Observer 17:1 (2002), 21–45. Baily, Martin, Charles Hulten, David Campbell, Timothy Bresnahan, and Fisman, Raymond, ‘‘Estimating the Value of Political Connections,’’ Richard E. Caves, ‘‘The Distribution of Productivity in Manufac- American Economic Review 91 (2001), 1095–1102. turing Plants,’’ Brookings Papers: Microeconomics (1992), 187– Foster, Lucia, John Haltiwanger, and Cornell John Krizan, ‘‘Aggregate 267. Productivity Growth: Lessons from Microeconomic Evidence,’’ Barlevy, Gadi, ‘‘The Sullying Effect of Recessions,’’ Review of Economic NBER working paper 6803 (1998). Studies 69 (2002), 65–96. Frazer, Garth, ‘‘Which Firms Die? A Look at Manufacturing Firm Exit in ——— ‘‘Credit Market Frictions and the Allocation of Resources over Ghana,’’ Economic Development and Cultural Change 53 (2005), the Business Cycle,’’ Journal of Monetary Economics 50 (2003), 585–617. 1795–1818. Gomes, Joao, Jeremy Greenwood, and Sergio Rebelo, ‘‘Equilibrium Bartelsman, Eric, and Mark Doms ‘‘Understanding Productivity: Lessons Unemployment.,’’ University of Rochester, mimeograph (1997). from Longitudinal Microdata,’’ Journal of Economic Literature 38 Griliches, Zvi, and Haim Regev, ‘‘Firm Productivity in Israeli Industry: (2000), 569–595. 1979–1988,’’ Journal of Econometrics. 65:1 (1995), 175–203. Bartelsman, Eric, John Haltiwanger, and Stefano Scarpetta. ‘‘Microeco- Hall, Robert E., Henry Farber, and John Haltiwanger, ‘‘Lost Jobs,’’ nomic Evidence of Creative Destruction in Industrial and Develop- Brookings Papers on Economic Activity 1 (1995), 221–273. ing Countries,’’ World Bank Policy Research working paper 3464 Haltiwanger, John, Stefano Scarpetta, and Helena Shweiger, ‘‘Assessing (2004). Job Flows across Countries: The Role of Industry, Firm Size and Beegle, Kathleen, Elizabeth Frankenberg, and Duncan Thomas, Regulations,’’ NBER working paper 8766 (2008). ‘‘Measuring Change in Indonesia,’’ RAND working paper 99-07 Liu, Lili, and James Tybout, ‘‘Productivity Growth in Chile and Colom- (1999). bia: The Role of Entry, Exit, and Learning,’’ In M. Roberts and J. Bergoeing, Raphael, Norman Loayza, and Andrea Repetto, ‘‘Slow Recov- Tybout (eds.), Industrial Evolution in Developing Countries eries,’’ Journal of Development Economics 75 (2004), 473–506. (Oxford: Oxford University Press, 1996). Bernard, Andrew, and Fredrik Sjoholm, ‘‘Foreign Owners and Plant Sur- Loayza, Norman, Guillermo Perry, and Luis Serven, ‘‘Schumpeter in the vival,’’ NBER working papers 10039 (2003). Tropics: Regulation and Macro Performance,’’ World Bank Bigsten, Arne, and Mulu Gebreeyesus, ‘‘The Small, the Young, and the mimeograph (2005). Productive: Determinants of Manufacturing Firm Growth in Ethio- Manning, Chris, ‘‘The Impact of Financial Crises on Labor Markets, pia,’’ Economic Development and Cultural Change 55:4 (2007), Household Incomes, and Poverty: A Review of Evidence,’’ Bulle- 813–840. tin of Indonesian Economic Studies 36 (2000), 105–136. 1810 THE REVIEW OF ECONOMICS AND STATISTICS McKenzie, David J., ‘‘How Do Households Cope with Aggregate Shocks? Ouyang, Min, ‘‘The Scarring Effect of Recessions,’’ Journal of Monetary Evidence from the Mexican Peso Crisis,’’ World Development 31 Economics 56:2 (2009), 184–199. (2003), 1179–1199. Paci, Pierella, Ana Revenga, and Bob Rijkers, ‘‘Coping with Crises: Poli- ——— ‘‘Aggregate Shocks and Urban Labor Market Responses: Evi- cies to Protect Employment and Earnings,’’ World Bank Research dence from Argentina’s Financial Crisis,’’ Economic Development Observer 27 (2012), 106–141. and Cultural Change 52 (2004), 719–758. Raddatz, Claudio, ‘‘Liquidity Needs and Vulnerability to Financial Micco, Alejandro, and Carmen Pages, ‘‘Employment Protection and Underdevelopment,’’ Journal of Financial Economics 80 (2006), Gross Job Flows: A Differences-in-Differences Approach,’’ RES 677–722. working papers 4365 (2004). Rajan, Raghuram G., and Luigi Zingales, ‘‘Financial Dependence and Mobarak, A. Mushfiq, and Denni Purbasari, ‘‘Protection for Sale to Firms: Growth,’’ American Economic Review 88 (1998), 559–586. Evidence from Indonesia,’’ Yale University, mimeograph (2008). Roodman, David, ‘‘How to Do xtabond2: An Introduction to ‘Difference’ Nickell, Stephen, ‘‘Biases in Dynamic Models with Fixed Effects,’’ Econ- and ‘System’ GMM in Stata,’’ Center for Global Development ometrica 49 (1981), 1417–1426. working paper 103 (2006). Nishimura, Kiyohiko G., Takanobu Nakajima, and Kozo Kiyota, ‘‘Does Schumpeter, Joseph, Business Cycles (New York: McGraw-Hill, 1939). the Natural Selection Mechanism Still Work in Severe Reces- ¨ derbom, Mans, Francis Teal, and Alan Harding, ‘‘The Determinants of So sions? Examination of the Japanese Economy in the 1990s,’’ Jour- Survival among African Manufacturing Firms,’’ Economic Devel- nal of Economic Behavior and Organization 58:1 (2005), 53–78. opment and Cultural Change 54:3 (2006), 533–556. Oh, Inha, Jeong-Dong Lee, Almas Heshmati, and Gyoung Choi, ‘‘Eva- Syverson, Chad, ‘‘What Determines Productivity?,’’ NBER working luation of Credit: Guarantee Policy Using Propensity Score Match- paper 15712 (2010). ing,’’ Small Business Economics 33 (2009), 335–351. World Bank. Indonesia Jobs Report (Jakarta: World Bank, 2010).