r -~~~~~~~~~~~~~~~~- PO'LICY RESEARCH \XYJ0RKING PAPER 2 8 rinanciai Crises, rinanciai DJepeniuence, and Industry Growth Luc Laeven Daniela Kiingebiei The World Bank Financial Sector Strategy and Policy Department OU I June 2002 1 PULilC RXESEARCH W iORKiNG P APER 28055 Abstract Laeven, Klingebiel, and Kroszner investigate the link financial systems. They hypothesize that the deepening of ke-LW,een financial crises anA inAustry growth. Thley the financial systemL allows sectors dependent on external analyze data from 19 industrial and developing countries finance to obtain relatively more external funding in that have experienced financiat crises during tne past 30 normal perioas, so a crisis in sucn counrries wouia nave a years to investigate how financial crises affect sectors disproportionately negative effect on externally dependent on external sources of finance. Specifically, dependent sectors. In contrast, since externally the authors examine whether the impact of a financial dependent firms tend to obtain relatively less external crisis on externally dependent sectors varies with the financing in shallower financial systems (and hence have depth of the financial system. They find that sectors relatively lower growth rates in such countries during highly dependent on external finance tend to experience normal times), a crisis in such countries has less of a a grpntpr cnntrnction of valiuie addelped tduring n rrisis in dispnronrtinnntelv neantive effect on the grnwth of deeper financial systems than in countries with shallower externally dependent sectors. This paper-a product of the Financial Sector Strategy and Policy Department-is part of a larger effort in the department to study the link between financial development and economic growth. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Rose Vo, room MC9-624, telephone 202-473-3722, fax 202-522-2031, email address hvo 1 @worldbank.org. Policy Research Working Papers are also posted on the Web at http:/ /econ.-orldbank.org. The author, may be contacted at llaevenunworldbannL orga dlingebiel@aworldbank.org, or randy.kroszner@gsb.uchicago.edu. June 2002. (26 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about Th Polic - - _- - . . _ L Z'____ -'LI - ---Z -- --_ -- - - rL.-L II_r .1. _ 1.1 -L I aevelopment issues. An uobjective of ie series is w gei the irndngs ouI quic,kly, even if thc pt,esent"tiUns are les .han ful;y pvoi'ed. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this pape ar en:.-!y hoseof be-u,+-rs They, do no, n..eessan.l- rpent the viwf thelW..d Bank,,4 i- ts 17-,cutiv Drct, or-t the countries they represent. l Produced by the Research Advisory Staff Financial Crises, Financial Dependence, and Industry Growth Luc Laeven, Daniela Klingebiel and Randy Kroszner** Preliminary drart Kevwords: financial criseq- financing constraints- financial development; industrv growth ** The authors are at the World Bank, the World Bank, and the University of Chicago, the Council of Econornic Advisors and the National Bureau of Economic Research, respectively. Luc Laeven is corresponding author. E-mail: llaeveni)worldbank.org. Tel. 202 458 2939. We thank Stijn Claessens and Inessa Love for comments and Ying Lin for research assistance. Tihe opinions expressed do not necessarily reiiect euause of the uriu 'Wor: vi thc Co-unti: oi Economic Advisors. Comments welcome. L. Introduction While it is widely accepted that financial crises have adverse consequences for the economy as a whole, relatively little empirical work investigates the mechanisms by which financial crises generate problems in the real sector. In this paper, we analyze data from developed and developing countries that have experienced financial crises during the last 30 years to investigate the impact of cri1scs onI irLU1sUIal roLWU L i I oL&ULUIes. Understar dIingU UUZi i111pact is C.ucial for ruILIaUinLg11 policies to mitigate the costs of a crisis in the financial sector to the economy as a whole and contributes to the literature on the mechanisms linking financial shocks and real economic activity. Much theoretical work has been done on how financial intermediaries and financial markets facilitate investment by firms and, hence, promote economic growth (see Levine (1997) and Rajan and Zingales (1998) for comprehensive overviews). Financial intermediaries and financial markets are grvaially UIVLou=L Lt reUuc ieiLiv al aZatu aiu adverGVe sIVle6LIVU jJIVLUeUI LIW.L U c Lar mak raising extemal funds difficult and expensive for firms. Well-functioning and well-developed financial intermediaries and markets thus should disproportionately benefit firms that are most dependent on extemal funds to finance their growth. Conversely, crises in the financial sector should have a disproportionately negative impact on firms that rely heavily on external sources of finance. Specifically, we investigate whether the impact of a financial crisis on sectors dependent on extemal sources of finanLcing varies witiL u1v lvevl W. UoVf UPelII.etL o UIV, MiBArLidI syLVItLI. Lv evaiuatL ui. empirical relevance of this theoretical mechanism, our empirical work focuses on the differential impact of financial crises on sectoral growth. To preview our results, we find that in well developed and deep financial systems, sectors highly dependent on extemal finance tend to experience a greater contraction of value added during a I crisis than do externally dependent sectors in countries with shallower financial systems. As has been shown in previous work (Rajan and Zingales, 1998), the depth of the financial system appears to relax credit constraints to permit externally dependent sectors to grow faster during normal periods. To explain our results, we hypothesize that the depth of the financial system allows sectors dependent on external finance to obtain relatively more external funding in normal periods; so a ;Lsis wJUl hIave a UdFi VVIVtioateLY U&aLl V 1IeILVct onU VAL.Lally UIUepUeUVL LJlUs. In V.ULUdbL, since externally dependent firms tend to obtain relatively less external financing in shallower financial systems (hence, we observe relatively lower growth rates in externally dependent sectors in such countries during normal times), a crisis in such countries has less of a disproportionately negative effect on the growth of these sectors. These results provide evidence supporting the existence of a "credit channel" through which firms dependent on external finance are harmed LLiOJ-oL oi.iL JLmaely dLLU L1 peIrO1AJiod oJf fLJL-La.iG UisUCrs. In the next section, we provide a more detailed motivation for the approach we are taking and relate our work to the existing literature. Section m explains our econometric approach. Section IV then describes the data and, in particular, how we measure financial dependence and how we define financial crises. Section V contains the results. Section VI presents a number of caveats and qualifications with respect to our analysis. Section VII concludes. H. Motivations and relation to previous work There exists a large body of empirical literature on the link between finance and growth. Levine and Zervos (1998) study whether stock markets and banks promote economic growth. They find that measures of stock market liquidity and private sector credit have a strong independent 2 effect on growth. Controlling for potential biases, Beck, Levine, and Loayza (2000) argue that there exists a clear empirical relation between the level of financial intermediary development and economic growth. They find that financial intermediaries exert a large, positive impact on total factor productivity growth, which feeds through to overall economic growth. Jayarante and Strahan (1996) provide evidence that financial markets can directly affect econorric g Io,%?2b atIdylAthie rea.o f bar. vorgih re+;;cion ;r."^ 'he TT-;teA S0+t.- T.M- find that the rates of real, per capita growth in income and output increase signiticantly following intrastate branch reform. Improvements in the quality of bank lending, not increased volume of bank lending, appear to be responsible for faster growth. Rajan and Zingales (1998) examine whether financial development facilitates economic growth by reducing the costs of external finance to firms. They find that industrial sectors that are relat.ively morn.e in need of e W i.Ance devel.op `ropor.tion oly 4 fat ;r.i con.es "a mth..ore- developed financial markets. As we discuss in more detail in the next section, they also overcome some of the identification problems embedded in standard cross-country growth regressions by using an interaction between a country characteristic (financial development of a particular country) and an industry characteristic (external financial dependence of a particular industry) in addition to country indicators and industry indicators. Demirga9-Kunt an.d M ?irl.o1snovZc (1997) show tuhat w,ell-developed finan.cial system.s are associated with extemally financed fimn growth. rhey also find that differences in legal systems affect firms' use of external financing to fund growth: in countries with efficient legal systems, a greater proportion of firms use long-term external financing. There exists also a large empirical literature on the existence of a credit channel. This 3 literature tries to investigate to what extent adverse shocks to a borrower's net worth increase the cost of extemal financing, and through which channels these adverse effects occur. For households, Mishkin (1977) provides evidence of effects of household balance sheet conditions on consumer expenditures in the US during the 1973-75 depression. Kashvap and Stein (2000) study the monetarv-transmission mechanism using quarterly data on every insured US comm.ercial bank for the period 19764993. They find that the im.pact of monetary policy on lending is stronger for banks with less liquid balance sheets, i.e., banks with lower ratios of securities to assets. Moreover, this pattem is largely attributable to the smaller banks. Their results support the existence of a "bank lending channel" of monetary transmission, though they do not allow us to make precise statements about its quantitative importance. Peek and Rosengren (2000) use the Japanese banking crisis as a natural experiment to test WCUhICI a ivan. suply shJLJck CLI Lafe I .ea LA ecor.or... act v .t. B uts was" 1I'J...I"lY CLiI V.J to US credit markets, yet connected through the Japanese bank penetration of US markets, this event allows one to identify an exogenous loan supply shock and ultimately link that shock to construction activity in U.S. commercial real estate markets. They exploit the variation across geographically distinct commercial real estate markets to establish conclusively that loan supply shocks emanating from Janan had real effects on economic activity in the United States. tiAggrea teL 4aIU LiULarciai shvcAs canar.%AL ui1c thevcopvratLe seoLrU by ULL LailJing ceitGAUL, iUr.L1u1g working capital and trade financing, to borrowers with valuable trading and investment opportunities (see Kashyap and Stein (1994) for a review). Real, financial or regulatory shocks can cause a real or perceived shortage of capital for banks. As a result, banks may become unwilling to lend even to viable companies and instead may prefer to invest excess liquidity in safe assets such as government 4 WPS2854 Rich and Powerful? Subjective Michael Lokshin June 2002 C. Cunanan Power and Welfare in Russia Martin Ravallion 32301 bonds. A credit crunch can originate from weak financial institutions or from tightened regulation and supervision. Increased uncertainty about whether and at what price loans will be available can also result in a shortage of loanable funds (Stiglitz and Weiss, 1981). These effects can be particularly severe for bank lending because banks are more likely than other financial intermediaries or markets to lend to firms that suffer from a greater degree of informational asymmetries. JU n audi-11ra A z ll l.-,ce sheet e,e car ltr ai.wplifi the pffooft of shckhnsr nin corporations (see Bemanke and Gertier (i995) tor a review). Agenor and Aizenman (Ii999), Chan- Lau and Chen (1998), Kim and Stone (1999) in a domestic context and Greenwald (1999) in an international context show that generally weaknesses in the financial sector along with tighter regulation and supervision appear to contribute to corporate distress by curbing credit. Ding, Domac and Ferri (1998) and Ghosh and Ghosh (1999) provide some empirical evidence that tighter rules for 11i1llarlcal LllOtIL.^-JIOn zfetd h upply MA larAU1,al L%nds%& insvrlco-;s ha.e n ntw. of corporates were also likely to have been important in reducing the supply of financing. Empirical research on identifying tools for the resolution and management of banking crises that are effective in resolving the crisis while limiting adverse economic spillover to the rest of the economy is sparse and most research in this area is limited to individual cases. Honohan and Klingebiel (2002) use cross-countrv evidence to determine whether specific crisis containment and resoluLiVIn pol1icies syt...atiLcallJLLLy inluenceLL,JAV 141e fiscall costs owf rewsolvir.g _a criisiss. Tr1.-.e .fi.n.d hAct accommodating policies - such as blanket deposit guarantees, open-ended liquidity support, repeated recapitalizations, debtor bailouts, and regulatory forbearance - significantly increase fiscal costs of resolving a crisis. Claessens, Klingebiel and Laeven (2001) review the literature on crises resolution strategies. 5 III. Method We apply the method in Rajan and Zingales (1998) to investigate the link between extemal financial dependence and industrial growth during financial crises. Rajan and Zingales (1998) relate re.al grnwth in va u e ndded nf a sertnr tn an intprqrtinn ti'rm that inrhuiii1 a pnrnoyv f foin2n,,-ial development and an index of external financiai dependence. Tney snow tiat nrms that are reiatively more dependent on extemal finance develop disproportionally faster in countries with more- developed or deeper financial markets, that is, they find a positive relation between the interaction term and real growth in value added. Their index of extemal dependence is constructed at the industry level based on data of US firrms Thevychonoe the finanrcial structure of US industriesaStheir be-ncimirk because the relatively open, sophisticated, and developed US financial markets shouid allow US firms to face the fewest obstacles to achieving their desired financial structure. This approach offers a valid and exogenous way to identify the extent of extemal dependence of an industry anywhere in the world under the assumption that there are technological and economic reasons why some industries depend more on extemal finance than others, and that these differences persist across countries. They also overcome some of thte identifirqtion prnblem.s emh.edded in standard cross-country growth rperessions by usingg an interaction between a counry cnaractenstic knnanciai aeveiopment of a particuiar country) and an industry characteristic (external financial dependence of a particular industry) in addition to country indicators and industry indicators. This approach allows them to isolate the impact of financial development on industry growth after controlling for cross-country and within-country differences, and is therefore less subject to criticism about an omitted variable bias or model 6 specification than traditional approaches. Our main innovation is to apply this approach to industries in countries experiencing financial crises to be able to investigate the real impact of shocks to the financial system in a country over time. First, we estimate the basic model in Rajan and Zingales (1998) for our sample of countries (model 1). RVAGR, = Ci + INDj + A * SHARE, + P2 * FDi *EDj + -,O (1) where R VA GR5, is the real growth in value added of sectorj in country i, C, is a country dummy for country i, ii-vD is an industry dummy for industryj, SHYLIEiJ is the snare ot sectorj in the totai value added of country i, FDi is the development of the financial system of country i, EDj is the external dependence ratio of sector ] according to Rajan and Zingales (1998). The specification thus ;nlVude ie o.tyrdi.us efcs euevLe at.ltvpUisfru. leveloffnrca Ui!.UU~lkA~U AJUI_I O.IIU LULUUOUJ '1 O YY TV UOV UiILV il~LVLidLiaVCJVA JLVO ILU UIV LVI U! 1UI1JAM4 development of a country: total credit to GDP, private sector credit to GDP, and M2 to GDP. The main differences with the Rajan and Zingales (1998) setup is twofold. First, we estimate the model for two sub-periods, namely, before and during a financial crisis. When estimating the model for the crisis period, we use the pre-crisis levels of share in value added and our proxies for financial development to avoid potential endogeneity problems. Second, we estimate the model for crLisis coLU Ucs or.ly, UIaL is, for lviu.uiv ulaL arVe liLsteU :... 'aprii 4aU Illlgvila (20u0.) as IIdvil,g experienced a financial crisis (and for which we have data). Note that by including country indicators into the regressions, we control for country-and industry-specifics. By including country 7 indicators for the crisis period, we are effectively controlling for the general severity of the crisis in each country. We are also interested in the link between the interaction of financial dependence and financial development on the one hand and the difference in real growth in value added between the cr.isie ,pr:nd "nd thp nr,-.ric ipriond non the nthpr,khane Ti in an altrrn,tivp enpAfirnt.nn nfmnroPl (1), we therefore use the difference in real growtn in value added between the crisis period and the pre- crisis period as a dependent variable. ARVAGRU = Ci +NDj + Aj *SHAREij +p82 *FDi *EDj +eŁ (2) where AR VAGR, is the difference in real growth in value added of sectorj in country i between the crisis period and the pre-crisis period. In other words, AR VA GRi = R VA GRU Crais - R VA GRU,pre-crisis X where RVAOR- iS, the re1 Qronwth in v21uenrl2ddedr nfePr-tnr i in nniintrv i Am-rinc thr s4is npri.od and RVAGRy,pre-Crisij iS the real growth in value added of sectorj in country i during the pre-crisis period. To avoid potential endogeneity problems, we use the pre-crisis levels of share in value added and our proxies for financial development. B101ecause +r - ...ar + she i .fion FDi- * EDjL -a the real growth in value added of sectorj in country i, R VAGR0, we also estimate the following model: iVA = Ci + lNL/j + p * + i * HighEDj + ji (3) 8 where HighEDj is a dummy variable for "High External Dependence" that takes value of one if sectorj is among the top-50% of most financially dependent sectors of all sectors considered by Raian and Zingales (1998), and zero otherwise Thiq sefip provides a robustness che-k that controls for UiLeasuremenet error in uie extermai dependence ratio-of each sector estimated by Rajan and Zingales (1998). In other words, we may expect that the most financially dependent sectors show a different growth pattern in well-developed countries on average, but there may not necessarily be a different effect for the most financially dependent sector and the second most financially dependent sector. Similar to model (2), we also estimate the following model: . fTT A fl -lT . f . * (Di X T T T. r Hh T. mn iK VA tif j L. = Ct 17'VLj t plSJHAR, + (4) Li ~ Pi 01 U 2 +. IV. Data The industry data is from Rajan and Zingales (1998). We use their measure of financial depe.n.denre bh sector based on US frn.-level data. Financial or external dendence is colculat as the fraction oI capital expenditures not financed with cash flow from operations. The sectors considered by Rajan and Zingales (1998) are a mix of three-digit and four-digit ISIC (International Standard of Industrial Classification) level industries. Rather than use the mix of four-digit level sectoral breakdowns for some industries and three-digit level sectorail breakdowns for other industries in Rajan and Zingales (1998) is somewhat arbitrary, we use external dependence ratios for sectors on a t.hree-digT4T*' SIC lee only. WA f kotv.ntfo *e=ciSrulateI t 1he w h ed aveage exte.al- dependence figure for the four-digit !SIC sectors on a tnree-digit ISIC level. For the sectors that are 9 already on a three-digit ISIC level in Rajan and Zingales (1998), we simply use their external dependence figures. For the sectors that are not already on a three-digit ISIC level, we apply the same method as in Rajan and Zingales (1998) to financial data on US finns from Compustat to estimate external dependence figures. Table 1 lists the three-digit ISIC level external dependence figures across sectors in the United States during the 1980s. We use these external dependence figures to cor-uc a hi. exe.aAeed.c .IgL.D)d yval ht+le vlu foei etJi ~AL~LU L U%,a 1L"&11 V"%ALI.JLJaL L.'IU ~ I&%5~ L.JUUIILIIJ V &JLUIO.1, LUCL& "An.vo Y"iu"% j.ii 'jJLIV iJ 0~.VLWJLJ wL among the top-50% of most financially dependent sectors, and value of zero otherwise. Similarly, we construct a low external dependence (LowED1) dummy variable that takes value of one if sectorj is among the bottom-50% of most financially dependent sectors, and value of zero otherwise. As measure of firm performance we use real growth in industry value added (annually compounded), the same measure as in Raian and Zingales (1998). The data on value added for each 4_- r, -nn-, Cnn-d-,S . a a1.*-A 4rn- *1-o T-A-. o4'~ Q*n+ .44+n- VAr-4-nlr A nfnk-ean++n~.. iL1lLU~L JL' L Ill , I.GJ L UJU LLLJ L WULU 1 %ALU L %J I JLII L1I'JA1.LL%ULLLL LL LI LULILLOLVUj A VCLU WW% U "JJ L "UJ o'. jFUL LW&J5UI4 U)' the United National Statistical Division. rhe value added data are corrected for inflation using CPI data from the Intemational Financial Statistics ofthe International Monetary Fund. We calculate the real growth in value added figures for sectors on a three-digit ISIC level as well. We also calculate the industry's share in total value added of the country, a variable used by Rajan and Zingales (1998). Our measures of financial deDth (total credit to GDP. Drivate sector credit to GDP. and M2 to GDP)F.u a,-uu,d he evel of Cj-DP p er c apitLa are fro,,,1 heI JLLtLerr.atioral inLancialC Statistics VI she International Monetary Fund. We use the Caprio and Klingebiel (2002) data set to time crisis and pre-crisis periods. Since it is difficult to identify the crisis period precisely, we use (t- 1, t+l) as the crisis period, where t is the first crisis year reported in Caprio and Klingebiel (2002). To ensure that the pre-crisis period is a 10 aisunct penod not aiiectea by the cnsis, we separate the pre-cnsis penoa from me cnsis penoa Dy three years. We define the pre-crisis period to be (t-8, t-4), if t-8 is available, otherwise as many years towards t-8 as possible, where t is the first crisis year reported in Caprio and Klingebiel (2002). This restricts the pre-crisis period to a maximum of 5 years. We only allow for one crisis period in a country, which is the first crisis mentioned in Caprio and Klingebiel (2002). to avoid identification prbe. 1? case o~f recp;r,-, cr.ses.oo "J*^1%Jarnao ;i ea0% t%f* . AA*r %I* flj0.a We started with the Caprio and Kiingebiei (2002) data set of systemic banking crisis countries. This data set includes 113 banlcing crises from 93 countries since the 1970s. Due to data constraints we need to drop a large number of countries. First, we do not have data on sectoral value added for many crisis countries. Second, we exclude countries for which we do not have data for both the pre-crisis and the crisis periods. This excludes, for example, Poland for which we do not have data fo; +'ke p-.4e=cii period. We also A-o o-le for whic we + dop ---+-- hae-e-vau LIaVY% IWL" IVAl U1%1 FL"-"'010 j)I.LLlJ%A. TV %, "low'. '.&Aj) ZVJUU~II YWIUY%.., WV, UVF llIVJ 114z ' V ,i ,LLt.JL V4LUv, added data for at least five sectors. Tlhis excludes Argentina, for which we have only data available for four sectors during the pre-crisis period. The final data set includes 19 crisis countries, including both developing -and developed countries. Table 2 presents a list ofthese countries. For each country, the table also shows the average real growth in value added and the number of sectors during both the Dre-crisis and crisis period. We do not investigate the post-crisis periods, because we do not have suicieIntIL UaLa on pos-crisis yea.-s foI illaly fU the '-Lu s Hill Ukvu baLuplv. The number of sectors varies widely across countries from 10 sectors in Hungary to 28 sectors in Chile, Finland, Israel or Sweden. To ensure consistency in a country across periods, we examine the same sectors in both the pre-crisis and crisis periods. This excludes a number of sectors for several countries for which we could obtain data in only one sub-period. We note that this setup 11 may lead to a potential selection problem because the data in the Industrial Statistical Yearbook is gradually becoming more comprehensive over time. Another potential selection effect would exist if entire sectors disappear during the crisis period. The latter is however not the case in our sample. Since we are interested in the difference in growth between the pre-crisis and crisis period, we need to use a balanced panel. The final data set contains a total of 448 sector-country observations from 19 cr.sis cou,-,ies. The number of firms within the sectors varies widely over time. In particular we see a large increase in the number of firms within certain sectors at certain points in time. This maybe the result of a re-classification or the inclusion of firms that were previously excluded from the statistics on value added. In both cases, changes in value added are not related to firm performance, and such observations need therefore be excluded from the analysis. We have deleted all sectoral observations if LUe rLU.Ver of LIJLS lWULL UI, OlVi.LL sectoLraULU JL-L1 ULWar. LV0U0 VI -/0 ,/0 (ubLL'U UL oLralIVeUj between the pre-crisis and crisis periods. This criterion deletes around 5% of observations across the different sub-periods. We also have deleted observations if the real growth in value added exceeds 100%, which excludes only a small number of cases. Table 3 presents the summary statistics of some variables that indicate changes in real sector and financial sector activity for both the pre-crisis and during crisis periods. When comparing the summary statistics of tue pre-L[I1b1 d cU Lrbis periods, we irnu Uth folluwigU C w lbi4almuLLisics. During crises periods, on average countries experience lower real GDP growth, lower real growth in sectoral value added (both for sectors that are highly dependent on external finance and sectors that are not), lower real growth of M2, and lower growth of (private sector) credit. Financial crises thus 12 are negat-iiy corrieyc du wiui reut and financial sector activity and performance. T nese statistics aiso indirectly provide some reassurance about the appropriateness of the timing of the crisis periods. Table 4 presents the pre-crisis levels of our proxies for financial development for our crisis countries. The financial development proxies indicate relatively low levels of financial development in countries like Bolivia, Chile, Columbia, and Mexico, and relatively high levels of financial developm.enmt in Hulngar, Japan, and Spain. V. Results First, we investigate the role of financial development on the link between external finance and sector growth for both pre-crisis and crisis periods. To this end, we estimate model (1) both for the pre-crisis period and the crisis period. The regression results are presented in panel A of Table 5. Each "pre-cr.sis" and "crisis" pair of columns Qes a dif,feren^.t r..easij of fin an.cial developm.,e.n.t. sAll standard errors are corrected for heteroskedasticity. Consistent with the findings in Rajan and Zingales (1998), we find for the pre-crisis "normal" period that financially dependent sectors grow on average disproportionally faster in countries with well-developed or deeper financial systems. In our sample, however, this result is statistically significant at the 10 percent level for only the total credit to GDP measure of financial development. This difference could partly be attributed to the fact th.at ,ve hav7e fewver obsei-.ations th.an Po2x,i 9mA Zigae (199) since wefo.cus onc.ssw.ti Another reason could be that we use extemal dependence figures on a three-digit level only. Our coefficient estimates for both the value-added share and the interaction term also are somewhat smaller than in Rajan and Zingales (1998). During crisis periods, we find the opposite relationship, namely, that financially dependent 13 sectors grow disproportionably slower in countries with well-developed or deeper financial systems. In none of our specifications, however, is the coefficient on the interaction term between financial depth and external dependence statistically significantly different from zero. In Panel B of Table 5; we investigate whether the crisis relation differs from the pre-crisis relatinn hv iiuing the differe nce in real grnwth in vallie adiied hetwpen the rrkiq npinnd and the nre- crisis period, AR VALGRI, as dependent variabie (modei (2) in tne previous section). As in Panel A we have three alternative specifications with each using a different measure of financial development. The reduction in growth rate from the pre-crisis period to the crisis period is larger for financially dependent firms in countries with well-developed financial systems. The coefficient of the interact.ion ter is statistically significantly different from zero in two ofthe three specification. in other words, financially dependent firms appear to be hit aisproportionally by a financial cnsis it they operate in countries with developed financial systems. The effect is economically significant. A one standard deviation increase in credit-to-GDP, for example, would reduce the difference in real growth in value added between the crisis period and the pre-crisis period by 1.0 percent (and the mean decline in real growth in value added between the crisis period and the pre-crisis period is 6.0 Next, we use a dummy variable that indicates high or low extemral dependence rather than a continuous variable (models (3) and (4) in the previous section). The regression results are presented in Table 6. The coefficient estimates and levels of statistical significance in Table 6 are very similar to those in Table 5. The main difference with the results in Table 5 is that in Panel B the interaction term between HighAED and FD, is now statistically qianificantlv different frnm zero in 211 three specificalouls. LUCne rIesuls in T ables 5 and 6 suggests that in tiMies of crisis tere is a negative 14 relationsimp between the interacton term ot financial development or depth ot the financial system and external dependence and real growth in value added. VI. Data and Measurement Caveats We note a number of caveats and qualifications with respect to our analysis. First, there is the qus^;no fuh. velibiiyo d,~4ata+ dA..g crse.iMny . leve.v aable rec ,,.L +...; a lag t.ad. oe 'J%Aw~.W. lwax' "&W IWI1A"WA"tjP MJL~ %s"LL"r, %ILL% IVJ.UIIY 1LILLUI-IVV%L VCGLIGUII..O I Va'L WI UI aG IrLU OuVF.I1' shocks. Firm performance variables such as growth in value added tend to respond to financial crises with a lag. Perhaps growth in value added is not a good measure of firm performance, in particular during periods of crisis. In addition, financial development indicators such as credit to GDP tend to increase during periods of crises as GDP decreases to a larger extent than credit. Second, determining the precise timing of crises is difficult, both in terms of identifying the ~~ 'ninA fka p"A nfr a nr.co A -,4cvo -,n k ...A -t ni-nul- .-A -nn. nn I-.a .A IJ~'&W1I6 41A4 II'' '1I1 .Ją~11GW V ..'IOIa11(13UULIAUj OLU V IY 13UIIU 111(a7 LU U~ I N CU U 0'JUh1. beE,=n..g r.dfheer. of c.sis A .si ..ay ld p lol a.drla not be r-esoleson Especially in the case of a typical V-shape pattern of recovery from a crisis it is crucial to get the timing right in order not to under- or overestimate the impact on firm performance. We use the data in Caprio and Klingebiel (2002) to define the beginning of a crisis and allow for a certain degree of mistiming by using a three year period around the Caprio and Klingebiel (2002) year as crisis period. Third, measures for the size of the financial system relative to GDP may not be good proxies for financial dev,elopmren.t. Durg periods of credit boomns, ofen. pr-eceding financial crises (see Kaminsky and Reinhart, 1999), for example, (private) credit over GDP may overstate the level of financial development or depth of the financial system. In addition, the political-economy of the policy responses to a financial crisis could affect the availability of credit in crises (see Kroszner (1998) and Klingebiel, Kroszner, Laeven and Van Oijen (2001)). 15 Finally, one may question the appropriateness of the Rajan and Zingales (1998) proxy for financial dependence for our sample. Their approach uses US external dependence as proxy and assumes the same technological reasons that make a particular industry in the US more dependent on external finance than other industries in the US, also make this particular industry more depeder.t on xtarnol finance in all ot4her co,uis -ow,,i,A +'he ,rorldA A 1*nirv this asslaption is plausible, it may not hold for all countries for country-specific reasons. Many developing countries, for example, support certain industries through subsidies. These industries maybe less dependent on external finance than without those subsidies. VII. Conclusions Trn normal crisic r,pinr.Aa tze f;nA th at fvikof 4that awe ,.l1 i ,n,we rn l awt ,rn vvtrnal ow- ...k L - VT * I b-b _ A -Vr a _ V I J *-SD W^_*_- vlAf _-J* L 0__ grow disproportionally faster in countries with deep financial systems consistent with Rajan Zingaies (1998). When we examine crisis periods, however, we find the opposite relation: crises in the financial sector have a disproportionately negative impact on sectors that rely heavily on external sources of finance in countries with deep financial systems. Our results provide evidence on the mechanisms linking the financial and real sectors in a financial crisis. WTe ma4'.aesie th,at a deeper0, fi.ncia~l s..+... allows, sectors~ depe.nde. nt on.,+,....nl Sfi.nn.-n U-C& aQ II J LJUL. U JIQI'.LUILOj OLJ~ LLJL "JLJL'J VY 0 0%IA,L%J1 0 UV~LL A%JJIL LI.L VJLI VAI.%%J.LLUa IL111a1ln.% to obtain relatively more external funding in normal periods, so a crisis would have a disproportionately negative effect on externally dependent firms in deeper financial systems. In contrast, since extemally dependent firms tend to obtain relatively less external financing in a shallower financial systems (hence the relatively lower growth rates in extemally dependent sectors in such countries during normal times)- a crisis in such countries has less of an effect on the arowth 16 of these sectors. in addition, it could also be that deeper financiai systems are more efficient in enforcing hard budget constraints on firms during a financial crisis than are financial institutions in underdeveloped financial systems. 17 References Agenor- Pierre-Richard and Thqhinm AM7enmnan (1 999) "PFinanni2l Rertnr Tneffiienriet- Qnd Coordination Failures: Implications for Crisis Management", Policy Research Working Paper 2185, World Bank. Beck, Thorsten, Ross Levine and Norman Loayza (2000), "Finance and the Sources of Growth", Journal of Financial Economics 58(1-2), 261-300. Elviua.rn-U, BoVe riIU MMnd %JULLLV1 Gertle (199Jr Ul tDI BLk Box: IHu CredUi Chc-e; of 1VIonctc W>hinrtAn T.C. Kroszner, Randall S. (1998), "On the Political Economy of Banking and Financial Regulatory Reform in Emerging Markets:, Research in Financial Services 10, 33-5 1. Levine, Ross (1997), "Financial Development and Economic Growth: Views and Agenda", Journal of Economic Literature 35(2), 688-726. T.yine, Ross nn.d Sarah Zervos (1998), "(trAw Mtrk-ts nnd nnnnmiG (Iro .h", American Economic Review 88(3), 537-58. Mishkin, Frederic S. (1977), "What Depressed the Consumer? The Household Balance Sheet and the 1973-75 Recession", Brookings Papers on Economic Activity 1, 123-64. Peek, Joe and Eric S. Rosengren (2000), "Collateral Damage: Effects of the Japanese Bank Crisis -- F"--I A -. *1- Y T--.- A -- - - - fl (f%^1 \ 'I^ A e On ReIal iAciVILY in uhe ULULned States, American Ecunomic Review 90(;, 3V-4+5. Rajan, Raghuram G., and Luigi Zingales (1998), 'Financial Dependence and Growth", American Economic Review 88(3), 559-96. Stiglitz, Joseph E. and Andrew Weiss (1981), "Credit Rationing in Markets with Imperfect Information", American Economic Review 71(3), 393-410. 20 Table 1 External Dependence Across Industries in the United States During the 1980s This table reports the median level of external financing for ISIC industries during the 1980's on a three-digit ISIC level. We use the classification of the second revision of the ISIC. External dependence is the fraction of capital expenditures not finance with cash flow from operations. Cash flow from operations is defined as in Rajan and Zingales (1998). For the sectors that are already on a three-digit ISIC level in Rajan and Zingales (1998) we simply use their external dependence figures. For the sectors that are on a four-digit ISIC ievel in Rajan and Zingaies (i998) we re-calculate the weighted average external dependence figure for the four-digit ISIC sectors on a three-digit ISIC level using Cornpustat and tne method in Rajan and Zingaies (i998). ISIC code Industrial sector External dependence 314 Tobacco -0.45 361 Pottery -0.15 323 Leather -0.14 324 Footwear -0.08 372 Nonferrous metal 0.01 322 Apparel 0.03 353 Petroleum refineries 0.04 369 Nonmetal products 0.06 371 Iron and steel 0.09 311 Foodproducts 0.14 341 Paper and products 0.17 321 Textile 0.19 342 Printing and publishing 0.20 355 Rubber products 0.23 332 Fumiture 0.24 381 Metal products 0.24 3051 '--,su-; l ce.al 0.25 331 Wood products 0.28 354 Petroleum and coal products 0.33 384 Transportation equipment 0.36 390 Other industries 0.47 362 Glass 0.53 382 Machinery 0.60 352 Other rhemicals 0.75 383 Electric machinery 0.95 385 Professional goods 0.96 356 Plastic products 1.14 21 Table 2 Average Real Growth in Value Added for AU Sectors Across Countries Thg table --ports the rea! mrnuth in apptn al rnl ,a addAeA ,p,.a=e hy, co-nt,n and fnr nbth pre.-.as; andA crsais, per.o as well as the years of each sub-period and the number of sectors included. The during crisis period is(t-1, t+1) where t is the first crsis year rennrted in Caprio and KlingPbiel (2002). The pre..sis peri odis (t-8, tA), if t-R iq qvniln1bl, otherwise as many years towards t-8 as possible. The sample consists of a total number of 19 countries. Pre-crisis During crisis Coun.-, Real go-nth i.n Years Alu-e oas Numberof value added sectors value added sectors Bolivia 0.046 1978-82 21 -0.079 1985-87 21 Chile 0.1!! 1970-72 28 0.038 1975-77 28 Colombia 0.061 1974-78 27 -0.038 1981-83 27 Eg,mpt 0 042 1983-87 24 0.032 1990-92 24 Finland 0.023 1983-87 28 -0.060 1990-92 28 Hungarv 0.054 19R3-87 10 -0=138 1990-92 10 Israel 0.057 1970-73 28 0.201 1976-78 28 Japan 0.054 1983-87 27 0.012 1990-92 27 Kenya 0.049 1977-81 23 0.038 1984-86 23 Malaysia 0.065 1977-81 22 0.041 1984-86 22 Mexico 0.055 1974-78 15 0.023 1981-83 15 New Zealand -0.021 1979-83 26 -0.012 1986-88 26 Norway -0.024 1979-83 27 -0.008 1986-88 27 Panama 0.019 1980-84 21 -0.206 1987-89 21 Spain 0.108 1970-73 25 0.131 1976-78 25 Sweden 0.035 1983-87 28 -0.222 1990-92 28 Turkey 0.071 1986-90 25 0.010 1993-95 25 Uruguay 0.021 1974-77 20 -0.133 1980-82 20 Zimbabwe 0.055 1987-91 23 0.006 1994-96 23 22 Table 3 Summary Statistics Before and During Crisis Both for the pre-crisis and during crisis periods, this table list the summary statistics of the following variables: real growth in GDP, real growth in total credit, real growth in private credit, real growth in M2, real growth in value added of highly dependent (High ED) sectors, and real growth in value added of not-highly dependent (Low ED) sectors. The highly dependent sectors are those sectors that are among the top-50% of most financially dependent sectors on a three- digit ISIC level according to Rajan and Zingales (1998). Similarly, the not-highly dependent sectors are those sectors that are among the bottom-50% of most financially dependent sectors. The total sample includes 19 countries and 448 industry-country observations. Observations Mean Minimum Maximum Std Dev Before Crisis Pvesl~ Co, s -DP 7 !9 0.052 =003 0.1.60 0.040 Real growth in Total credit 19 0.118 -0.055 0.641 0.161 Real growtu in Private Credit i9 0.108 0.005 0.403 0.093 Real growth in M2 19 0.088 -0.040 0.494 0.125 Real growth in value added of High ED sectors 205 0.058 -0.439 0.333 0.098 Real growth in value added of Low ED sectors 243 0.037 -0.686 0.393 0.102 During crisis Real growth in GDP 19 0.005 -0.076 0.150 0.047 Real gTowth in Total credit 19 0_096 -0=150 0.736 0.227 Real growth in Private Credit 19 0.099 -0.237 0.634 0.222 ID -al gr0_'+U vo,Kf,) I 1 AMA fl19A A.A-. A.4 017 Real growth in value added of High ED sectors 205 -0.018 -0.527 0.618 0.173 Real growth in value added of Low ED sectors 243 -0.0i0 -0.626 0.868 0.195 23 Table 4 Financial Depth Indicators This table reports total credit-to-GDP, private credit-to-GDP, and M2-to-GDP at the beginning ofthe pre-crisis period in each countrv. These variables are used as proxies for financial depth. Data are from the International Financial Statistics of IMF. Since the figures are for the first year of the pre-crisis period for each country, they are not directly comparable across countries. The pre-crisis years can be found in Table 2. Countrv Total credit-to-GDP Total private credit-to-GDP M2-to-GDP (pre-crisis) (pre-crisis) (pre-crisis) Bolivia 20.13% 12.96% 19.26% r1.1 - -, a,'n ~ifn/ IC IIAOI ~~_Ilmrl ~~~~~I 1. LvJ70 O..3,L70 Li .-t iO Colombia 24.41% 14.69% 19.76% Egypt 98.79% 26.02% 82.57% Finland 54.35% 55.56% 44.68% Hungary 100.88% 48.81% 47.69% Israel 50.52% 29.94% 51.32% Japan 113.52% 93.23% 93.55% Ke-nyav 28.30% 19.92% 38.31% Malaysia 31.49% 27.74% 45.95% MIexiAco ;7.25% 4.084% 5.47% New Zealand 31.32% 21.57% 29.44% Norway 54.34% 32.14% 52.94% Panama 61.57% 54.36% 34.53% Spain 75.11% 58.03% 54.13% Sweden 73.77% 40.81% 54.26% Turkey 38.32% 18.51% 28.52% TTt.r, ')I 2700Q 10 .1%0 108 AQO Zimbabwe 25.27% 9.41% 25.58% 24 Table 5 Value Added Growth, Financiai Dependence, and Financial Development: Before and During a Financial Crisis, With Continuous Financial Dependence Variable Dependent variable in panel A is real growth in value added of sector. Dependent variable in panel B is the difference in real crrnwth in value added between the crisiq nerind and the nre-crisis nprind Cnuntrien include Bolivia, Chile, Colombia, Egypt, Hungary, Spain, Finland, Hungary, Israel, Japan, Kenya, Mexico, Malaysia, Norway, New Zealand, Panama; Sweden. Turkey; Uruguay Zimbabwe. Cut-off for difference in growth of number of firms within sector between sub-periods is +100% and -50% (doubled or halved). ED is the external dependency figure in Rajan and Zinzales (1998) on a three-dizit ISIC level (see Table 1). Countrv and industrv durmmies are included. but not reported. We use share in value added, total credit to GDP, private credit to GDP, and M2 to GDP for the first year of the pre- crisis period only. A constant was added, but is not reported. Heteroskedasticity-consistent standard errors between brackets. * significant at 10% level; ** significant at 5% level; *** significant at 1% level. Panel A: 'anable ,,~~Pe-crisis C"X r-r" C rec-- C C In Value AIde .*.**1 * . * 2 * ***0.2'98O A**-OJ3 (0.114) (0.211) (0.113) (0.212) (0.115) (0.208) UT%1 II Tr_..1 I...*. T(2%D *0.-1 ) nl ŁY7'7 -IDIJ * TL4al Cr'etUIL to GJLI* U.07 -0.07I 7 (0.043) (0.066) ID * Private C-.Ae+ 't-DTD A0A.05 Air, (0.046) (0.071) ED * M2r to GlD 0.044 -0.IAA (0.057) (0.099) Prob>F ***0°-°° ***0° °° ***0° °° ***0° °° ***0° °° ***0 °°° R-sq1ared 0.207 0.384 0.204 0.382 0.2n4 0.388 Observations 448 448 448 448 448 448 .r ariel rvi: Variable Crisis vs. Pre-crisis Crisis vs. Pre-crisis Crisis vs. Pre-crisis Share in Value Added 0.215 0.142 0.220 (0.242) (0.240) (0.244) ED * Total Credit to GDP **-0.225 (0.1;1) ED * Private Credit to GDP -0.124 ED * M2 to GDP *-0.280 (A IC A\ P.ob> ***0.000 ***0.0AAA ***A0AAA rI uu- V.UIU u.uuu bJ.bJUV R-squared 0.296 0.288 0.295 Obl-serira ,-;onLs AAQ AAQ AAQ 25 Table 6 Value Added Growth, Financial Dependence, and Financial Development: Before and During a Financial Crisis, With Dummy Variable Indicating High Financial Dependence Dependent variable in panel A is real growth in value added of sector. Dependent variable in panel B is the difference in real arnwth in value added hetwreen the erisis pernnd arnd the nre srise ner:nol Crinttries iniuide Rolivia Chile Colombia, Egypt, Hungary, Spain, Finland, Hungary, Israel, Japan, Kenya, Mexico, Malaysia, Norway, New Zealand, Panama- Sweden Turkev- lJruguyv 7Zimbabwe= Cut-off for difference in grnwth of number of firms within sector between sub-periods is +100% and -50% (doubled or halved). High ED indicates above median external dependence. Country and industry dummies are included, but not reported. We use share in value added, total credit to GDP. private credit to GDP, and M2 to GDP for the first year of the pre-crisis period only. A constant was added, but is not reported. Heteroskedasticity-consistent standard errors between brackets. * siznificant at 10% level; ** significant at 5% level: significant at 1% level. Panel A: Variable rre-crsis Crisis re-c s risis rre-criSis Crisis inare in vFalu auuau A - --u.a3a - - --u - al -- 3i --u. (0.115) (0.216) (0.115) (0.217) (0.117) (0.217) TT-.L !Th W~.-l - - . 1 f-T%"l4 I% nCr Aiigu ED i vL41 Credit Lu GDU *0. 06U.U2 -V.VJJ (0.034) (0.059) *,D * P .. v.e C'..wt. 4.o C-TDP 0 .I 0 0Y70 TdTl.J 11T 1VGL A;+ tlUT AJAJAA.VV U. I V (0.037) (0.061) LT..t 1: * +-I) (A hD A ncn A W7M IU1LL St LYSS. LU -av -v..vI v (0.044) (0.077) Prob>F ***0.000 ***0.000 ***0.000 ***0.000 ***0.000 ***0.000 R=squ-aed 0.210 0.384 0.205 0.384 0.205 0.384f Observations 448 448 448 448 448 448 r4Uli. D. Variable Crisis vs. Pre-crisis Crisis vs. Pre-crisis Crisis vs. Pre-crisis Share in Value Added 0.227 0.193 0.216 (0.244) (0.241) (0.245) High ED * Total Credit to GDP **-0.139 (0.068) High ED * Private Credit to GDP **-0.144 (0.068) High ED * M2 to GDP *-0.158 (0.090) Prob>F ***0000 :::0000 :::0000 R-squared 0.294 0.291 0.292 UOservations 4 4 044 26 WPS2837 Reform, Growth, and Poverty David Dollar May 2002 E. Khine in Vietnam 37471 WPS2838 Economic Mobility in Vietnam Paul Glewwe May 2002 E. Khine in the 1990s Phong Nguyen 37471 WPS2839 Marketing Externalities and M. Shahe Emran May 2002 F. Shilpi Market DeveIop.rent Forhad Shlpin 87476 WPS2840 Public Snending and Outcomes: Andrew Sunil Rajkumar May 2002 H. Sladovich Does Governance Matter? Vinaya Swaroop 37698 WPS2841 Contractual Savings in Countries Gregorio Impavido May 2002 P. Braxton with a Small Financial Sector Alberto R. Musalem 32720 Dimitri Vittas WPS2842 Financial Sector inefficiencies Pierre-Richard Agenor May 2002 M. Gosiengfiao and the Debt Laffer Curve Joshua Aizenman 33363 WPS2843 A Practical Guide to Managing David Scott May 2002 L. Yeargin Syvtemic Financial Crises: A Review 81553 of Approaches Taken in Indonesia, the Republic of Korea, and Thailand WPS2844 Money Demand in Venezuela: Mario A. Cuevas May 2002 M. Geller Multiple Cycle Extraction in a 85155 Cointegration Framework WPS2845 The Spatial Division of Labor Marcel Fafchamps May 2002 F. Shilpi KIn-Npa Forhad cki[pi 87A76 WPS28R46 ! lndia's Fconomic t,rowth Leavinn Gaurav Daft May 2002 C. Cunanan the Poor Behind? Martin Ravallion 32301 WPS2847 The Nature and Dynamics of Poverty Hippolyte Fofack May 2002 P. White Determinants in Burkina Faso 81131 in the 1990s WPS2848 Administrative Barriers to Foreign Jacques Morisset May 2002 M. Feghali Investment in Developing Countries Olivier Lumenga Neso 36177 WPS2849 Pooling, Savings, and Prevention: Truman G. Packard May 2002 T. Packard Mitinatino the Risk of Old Arie 75841 Poverty in Chile WPS2850 Determinants of Commercial Bank David A. Grigorian June 2002 S. Torres Performance in Transition: An Vlad Manole 39012 Application of Data Envelopment Analysis WPS2851 Economic Development and the Bernard Hoekman June 2002 P. Flewitt WorldA Tr-ade Ogn~ --- , n- A4- r okha 3 2724 V V U IU I IOCUV %.IYCXI -IOI~ flLIUI - 10 WPS2852 Reninnal Anreements and Trade in Aaditya Mattoo June 2002 P. Flewitt Services: Policy Issues Carsten Fink 32724 WPS2853 Private Interhousehold Transfers in Donald Cox June 2002 E. Khine Vietnam in the Early and Late 1990s 37471