American Finance Association

Transcription

1 American Fance Association What Do We Know about Capital Structure? Some Evidence from International Data Author(s): Raghuram G. Rajan Luigi Zgales Source: The Journal Fance, Vol. 50, No. 5 (Dec., 1995), pp Publhed : Wiley for American Fance Association Stable URL: Accessed: :20 UTC Your use JSTOR archive dicates your acceptance Terms & Conditions Use, available at fo/about/policies/terms.jsp JSTOR a not-for-prit service that helps scholars, researchers, students dcover, use, build upon a wide range content a trusted digital archive. We use formation technology ols crease productivity facilitate new forms scholarship. For more formation about JSTOR, please contact support@jsr.org. American Fance Association Wiley are collaboratg with JSTOR digitize, preserve extend access The Journal Fance. All use subject JSTOR Terms Conditions

2 THE JOURNAL OF FINANCE. VOL. L, NO. 5. DECEMBER 1995 What Do We Know about Capital Structure? Some Evidence from International Data RAGHURAM G. RAJAN LUIGI ZINGALES* ABSTRACT We vestigate determants capital structure choice analyzg fancg decions public firms major dustrialized countries. At an aggregate level, firm leverage fairly similar across G-7 countries. We fd that facrs identified previous studies as correlated cross-section with firm leverage United States, are similarly correlated or countries as well. However, a deeper examation U.S. foreign evidence suggests that oretical underpngs observed correlations are still largely unresolved. THIRTY SEVEN YEARS AND hundreds papers after Modigliani Miller's semal work, what do we really know about corporate capital structure choice? Theory has clearly made some progress on subject. We now underst most important departures from Modigliani Miller assumptions that make capital structure relevant a firm's value. However, very little known about empirical relevance different ories. Empirical work has uneard some stylized facts on capital structure choice, but th evidence largely based on firms United States, it not at all clear how se facts relate different oretical models. Without testg robustness se fdgs outside environment which y were uncovered, it hard determe wher se empirical regularities are merely spurious correlations, let alone wher y support one ory or anor. Th paper attempts start fillg th gap our knowledge. Our primary objective establh wher capital structure or countries related * Both authors are at Graduate School Busess at University Chicago. Rajan thanks Center for Research on Securities Prices while Zgales thanks Graduate School Busess for fundg. Th project was also made possible with a grant from Center for International Busess Research at University Chicago National Science Foundation (grant SBR ). We thank Patricia O' Brien, Douglas Diamond, Eugene Fama, Jennifer Franc, Steve Kaplan, Anil Kashyap, Richard Leftwich, Mern Miller, Mitchell Petersen, Jay Ritter, James Seward, Robert Vhny for helpful dcussions, participants workshops at University Chicago (Fance & Junior Faculty Lunch Group), HEC-University Montreal, Indiana University, University Maryl, National Bureau Economic Research. Summer Institute, Sckholm School Economics 1994 WFA meetgs for comments. We are debted Eduardo Gonzales for excellent research asstance, Andrew Alford for valuable help gettg us acquated with Global Vantage detailed comments on an earlier draft. Th paper corporates part a Center for Research Security Prices workg paper entitled "Notes on International Capital Structure." 1421 All use subject JSTOR Terms Conditions

3 1422 The Journal Fance facrs similar those appearg fluence capital structure U.S. firms. In dog so, we do not restrict ourselves attemptg reproduce regularities found United States or countries, but we try go deeper underst forces behd m. The use ternational data provides a unique opportunity for th analys. To extent that or countries are similar United States, y provide an dependent sample test received wdom. To extent that y have different stitutional structures, y crease our ability dcrimate among alternative ories. The cost usg an ternational sample that some time has be spent analyzg differences between countries, rangg from accountg practices legal stitutional environments. Thus, we start presentg typical balance sheet each G-7 countries ( United States, Japan, Germany, France, Italy, United Kgdom, Canada). Th analys highlights effects different accountg rules, also pots corrections that need be made so that measures leverage are comparable across countries. Unlike previous studies, we fd that extent which firms are levered fairly similar across G-7 countries, with only United Kgdom Germany beg relatively less levered. Th fdg does not seem be an artifact eir our sample or period it covers. Instead, our conclusions are different from those reached most earlier researchers prcipally because we have more detailed measures, more comparable calculations, leverage. We n analyze major stitutional differences across countries ir likely impact on fancg decions. Although G-7 countries are fairly homogeneous ir level economic development ( addition data availability, th anor good reason focus on m), ir stitutions-as exemplified tax bankruptcy code, market for corporate control, hrical role played banks securities markets-are fairly different. Apart from establhg a framework with which underst between-country differences, review stitutions important because y may affect with-country cross-sectional correlation between leverage facrs such as firm pritability firm size. Th may help us identify true economic forces underlyg facrs. Fally, we compute with-country partial correlations between leverage facrs identified as important United States. It remarkable that se facrs are, general, similarly correlated with leverage or countries also. While constency correlations may dicate that re are deed underlyg forces that fluence capital structure choice, re may also be reason doubt our understg what se forces are or how stitutional differences identified above moderate ir fluence. For example, leverage creases with size all countries except Germany. A possible explanation that larger firms are better diversified have a lower probability beg fancial dtress. Lower expected bankruptcy costs enable m take on more leverage. But a number economts (see for All use subject JSTOR Terms Conditions

4 What Do We Know about Capital Structure? 1423 example, White (1993) Kaer (1994)) argue that Germany bankruptcy code not conducive reorganizg firms, firms enterg bankruptcy are usually liquidated. Sce liquidation values are generally lower than gog concern values, bankruptcy potentially more costly Germany. So we might expect a stronger positive correlation between size leverage Germany. Why n do we observe a significant negative correlation? Th suggests that eir our understg economic underpngs facrs (e.g., that size an verse proxy for expected costs bankruptcy), or our understg fluence stitutions (e.g., bankruptcy laws), or both, flawed.' More research clearly called for. The rest paper proceeds as follows. Section I describes data. Section II computes leverage each country after implementg necessary accountg adjustments. Section III overviews major stitutional differences across G-7 countries exames how se relate differences leverage. In Section IV, we undertake a comparative study crosssectional determants capital structure choices attempt rationalize observed regularities. Section V concludes. A. Data I. Data Description Previous studies that attempt compare capital structures different countries have been hampered lack constent accountg market formation outside United States. A recently compiled database ternational corporations, Global Vantage, helps us, at least partially, address th problem. The database contas accountg data monthly sck prices for approximately 8,000 companies from 31 countries sce Global Vantage started collect data only From that year onward it cluded all companies present Morgan Stanley Capital International Index, Fancial Times Actuaries World Index or local market dex.2 Pre-1987 data were backfilled refore suffer from survivorship bias. However, Global Vantage retas firms even if y are dropped from relevant dex, so long as data available. For th reason we concentrate our analys on period, usg pre-1987 data only as a robustness check. We limit our attention largest economies where re are sufficient firms represented make comparons meangful. In particular, we focus on 1 Th not say that re are no potential explanations pattern Germany. For stance, external fancg may be very costly Germany, leadg firms rely largely on ternal sources. Sce large firms typically have fewer vestment opportunities greater cashflows from extg vestments, one might expect m be able fance more through ternal sources. If th argument true, it still leaves puzzle why external fancg so costly Germany. 2 For United States, Japan, Germany, France, Italy, United Kgdom, Canada ( seven countries we consider) local market dex, respectively, S&P 500, Nikkei 500, FAZ Share Index, CAC General Index, MIB Current Index, FT Actuaries 500, TSE 300. All use subject JSTOR Terms Conditions

5 1424 The Journal Fance non-fancial corporations G-7 countries. In 1991, Global Vantage covers more than two thirds companies (representg more than 90 percent market capitalization) countries with a small sck market (France, Germany, Italy). In or major countries Global Vantage covers between one third one half companies traded, representg more than 75 percent market capitalization. We elimate fancial firms such as banks surance companies from sample because ir leverage strongly fluenced explicit (or implicit) vesr surance schemes such as deposit surance. Furrmore, ir debt-like liabilities are not strictly comparable debt sued nonfancial firms. Fally, regulations such as mimum capital requirements may directly affect capital structure. The fal sample covers between percent companies lted every country, represents more than 50 percent market capitalization each country. There are at least two potential sources bias we should worry about. First, sample selection criterion used Global Vantage biases sample wards largest lted companies each country. Given figures on coverage, th suggests that while sample may do well capturg aggregate leverage a country, it probably not representative average firm. Anor selection bias ares from fact that only lted companies are reported. The fraction lted firms differs widely across different countries, so does average size companies lted.3 It important underst why se differences are, but th paper has more modest aims. The figures above suggest that while companies followed Global Vantage are fairly representative lted companies, se, turn, may represent only a small ( varyg) proportion firms a country.4 While lted companies, tip proverbial iceberg, are perhaps greatest terest fancial community, terests academicians are broader. Unfortunately, it hard establh beyond doubt wher tip iceberg representative larger mass hidden below. But extent that common stitutions with a country fluence both tip mass below, formation gared from an analys tips will have broader implications. We will attempt check for possible biases data throughout th paper. But ultimately, ternational data cannot be made 3 Edwards Fcher (1993) estimate that lted companies accounted for 30.5 percent tal sales corporations United Kgdom 1986, while lted Aktiengesellschafts ( closest German equivalent public limited liability firms) accounted for just 10.6 percent Germany. Pagano Roell (1990) fd that market capitalization average company traded Frankfurt or Milan approximately 60 percent larger than that average company traded London. 4 One might be concerned that, given different stitutional environments, only best firms have access public equity market Germany Italy. We do not thk th true. Very few firms went public Germany after World War II refore beg public can be regarded as exogenous from our pot view. Th true Italy also, though a lesser extent. Furrmore, The Economt (7/24/1993) reports that a McKsey study fds unlted firms Italy perform twice "as well" as lted ones. All use subject JSTOR Terms Conditions

6 What Do We Know about Capital Structure? 1425 Table I Dtribution Firms Followed Global Vantage Size All consolidated firms all G-7 countries are pooled y are placed size deciles accordg 1991 market value ir assets U.S. dollars. Country (Percentage Consolidated Firmsa) United United Decile States Japan Germany France Italy Kgdom Canada Smallest Largest Total number firms a Columns may not sum 100 because roundg errors. perfectly homogeneous, reader will have terpret our results with all caveats md. To explore magnitude se biases homogeneity our sample across countries we sort all companies deciles accordg market value ir assets ( U.S. dollars) at end As Table I shows, size dtribution companies with each country fairly homogeneous across countries with exception Japan. Anglo-American countries ( United States, United Kgdom, Canada) have relatively more firms that are smaller than overall median (respectively 59, 57, 56 percent). By contrast, Germany, France Italy sample slightly tilted wards larger companies (only 35, 34, 46 percent are below median). Only Japan has almost entire sample (97 percent) firms larger than overall median. In presentg results, we will attempt correct for differences size. II. International Comparons Leverage Fancg A. Balance Sheets Considerable sight can be obtaed simply comparg average balance sheets firms our sample. In dog so, we note three major sources differences accountg practices. First, not all countries require 5 More precely quasi-market value assets, defed as book value assets mus shareholders' equity plus market value equity plus book value preferred sck. Market value computed at end correspondg fcal year. End-1991 exchange rates are used. All use subject JSTOR Terms Conditions

7 1426 The Journal Fance firms report consolidated balance sheets, although majority firms each country do it ( 1991, countries with least proportion firms reportg consolidated balance-sheets are Germany Japan with approximately 76 percent each).6 Companies with unconsolidated balance sheets report an affiliate's net assets (under "equity" method reportg affiliates) as a long term vestment on ir balance sheets. Hence se firms would (correctly) appear have lower leverage than orwe identical firms who report consolidated balance sheets. Alternatively, an attempt wdow-dress ir balance sheet, y may place debt y take on less vible affiliated companies n borrow it back via terfirm trade credit.7 For ease comparon, th paper focusses on firms reportg consolidated balance sheets, Table II reports average country balance sheets for all firms sample that reported consolidated balance sheets Second, valuation assets (at hrical cost or current value) may differ substantially across countries. For stance, it generally believed (Nobes Parker, 1991, p. 25) that German accountg places greater emphas on "conservatm" less on "true fair" considerations. Asset values German companies may refore be understated relative asset values many or countries. Conversely, Fance Acts made revaluation compulsory for French companies (Nobes Parker, p. 17). There no easy way correct for th, our results on book values must be terpreted with appropriate caution. The third difference relates what cluded what excluded from a balance sheet different countries. For example, lease reportg varies substantially: fancial leases appear on balance sheet United States, Canada, United Kgdom (especially latter half eighties) but not regularly Japan Contental Europe. As extent leasg creases, however, more se countries are forcg companies report m.8 Anor difference that Germany, unlike practice United States, both funded unfunded portion pension liabilities are reported on balance sheet (as are assets held agast pension liabilities). Furrmore, generally accepted German accountg practices allow firms 6 Towards end our study period, implementation European Commsion's Seventh Directive considerably reduced leeway for firms decidg wher consolidate. 7 Of course, extent that a parent borrows from a subsidiary, consolidation may understate extent leverage. Also, we do not claim that consolidated firms mselves do not pose problems for a study such as ours. Multationals may consolidate foreign subsidiaries. We will correctly attribute all leverage parent firm. Th will aumatically dimh differences between countries. The creasg globalization operations large firms should reduce differences between capital structures firms different countries. 8 Rurford (1986) reports that leasg accounted for 17.1 percent gross capital formation U.S. corporate secr 1980, 8.7 percent United Kgdom, 4 percent Japan, 3.2 percent France, 1 percent Germany. Barclay Smith (1995) fd that lease obligations represent 9 percent tal debt for a large sample U.S. firms drawn from COMPUSTAT. Th figure probably represents an upper bound error produced lease undereportg or countries, where leasg less widespread. An error th magnitude not likely affect our results a major way. All use subject JSTOR Terms Conditions

8 What Do We Know about Capital Structure? 1427 set aside greater provions for future potential liability pritable years. Th reserve n used smooth accountg come lean years. Thus, Germany, reported earngs may be less representative true earngs than those United States (see Alford et al., 1993).9 An dication importance se differences that 29 percent liabilities a German company are cluded category "Liability Or" ( no or country does th item represent more than 8 percent). Approximately 50 percent "Liability Or" represented pension liabilities, remag 50 percent consts special reserves for potential liabilities. We will correct for some se differences when we dcuss leverage. Bearg above caveats md, clear differences emerge between countries Table II. Firms Anglo-American economies have proportionately more fixed assets less current assets ir balance sheet: 40 versus approximately 30 percent for Germany, France, Italy, Japan.10 However, composition current assets differs greatly latter group. Japanese companies have a larger amount cash short term vestments (18 percent assets versus at most 11 percent for or countries), th accounts for most difference current assets with respect United States. In Contental Europe, however, higher current assets are due higher venries accounts receivable. We now turn liability side balance sheet. B. Measures Leverage Given observed differences composition liabilities, before undertakg any vestigation leverage it appropriate defe what we mean th term. Clearly, extent leverage- most relevant measure-depends on objective analys. For stance, agency problems associated with debt (Jensen Mecklg (1976), Myers (1977)) largely relate how firm has been fanced past, thus on relative claims on firm value held equity debt. Here, relevant measure probably sck debt relative firm value. Ors (see Aghion Boln (1992)) have focussed on leverage as a means transferrg control when firm economically dtressed, from shareholders (or ir fiduciaries) bondholders (or ir fiduciaries). Here, important question wher firm can meet its fixed payments, consequently, a flow measure like terest coverage ratio more relevant. Rar than explor- 9 Anecdotal evidence on th sue reveals magnitude problem. Nobes Parker (1991, p. 27) report that AEG Telefunken succeeded generatg exactly zero earngs for three years a row. More recently, 1994, Daimler Benz revealed its earngs restated accordg U.S. stards while seekg a ltg on New York Sck Exchange. Daimler Benz had suffered substantial reverses various markets, so one would expect its smood earngs be higher than true (U.S. restated) earngs. But extent difference, DM3 billion (approximately $2 billion), was surprgly large. All th suggests that objective smoothg come may sometimes conflict with objective presentg a conservative picture. 10 The larger proportion fixed assets Canada may simply be because a dproportionately large fraction (28 percent) Canadian companies sample are oil mg. All use subject JSTOR Terms Conditions

9 1428 The Journal Fance Table II Balance Sheets for Non-Fancial Firms G7 Countries-1991 The value each item calculated as a fraction book value tal assets n averaged across all firms reportg consolidated balance sheets country. Only balance sheets non-fancial firms are cluded. United United States Japan Germany France Italy Kgdom Canada ASSETS Cash short-term vestments Account receivable/debrs Invenries Current assets-or Current assets-tal Fixed assets (tangible) Investments advances-equity Investment advances-or Intangible assets Assets-or Assets-tal LIABILITIES Debt current liabilities Accounts payable/credirs Current liabilities-or Current liabilities-tal Deferred taxes Long-term debt Mority terest Reserves-untaxed Liabilities-or Liabilities-tal Shareholders equity Total liabilities shareholders equity Source: Global Vantage Data Base. g all possible ories ir associated measures leverage, we use ones suggested dcussion above as illustrative. The broadest defition sck leverage ratio tal liabilities tal assets. Th can be viewed as a proxy for what left for shareholders case liquidation. However, it does not provide a good dication wher firm at rk default near future. Also, sce tal liabilities also cludes items like accounts payable, which may be used for transactions purposes rar than for fancg, it may overstate amount leverage.1" Similarly, pension liabilities arg from labor market contracts will fluence th ratio. " However countries, or specific classes firms which use trade credit as a means fancg, accounts payables should be cluded measures leverage. All use subject JSTOR Terms Conditions

10 What Do We Know about Capital Structure? 1429 A more appropriate defition fancial leverage provided ratio debt (both short term long term) tal assets. Th measure, however, fails corporate fact that re are some assets that are fset specific nondebt liabilities-. For example, an crease gross amount trade credit reflected a reduction th measure leverage. Given that level accounts payable accounts receivable may jotly be fluenced dustry considerations, it seems appropriate use a measure leverage unaffected gross level trade credit. We could defe leverage as ratio tal debt net assets, where net assets are tal assets less accounts payable or liabilities. Although th measure not fluenced trade credit, it affected facrs that may have nothg do with fancg. For example, assets held agast pension liabilities may decrease th measure leverage. Therefore, effects past fancg decions probably best represented ratio tal debt capital (defed as tal debt plus equity). One measure rk that equity holders will not be able make fixed payments will have give up control coverage ratio, i.e., ratio earngs before terest taxes (EBIT) terest expense. Th measure appropriate if we believe that vestments equal magnitude depreciation are needed keep firm a gog concern. If no such vestments are needed, a better measure firm's ability service debt ratio earngs before terest, taxes, depreciation (EBITDA) terest expense. A common problem for both measures that y assume that shortterm liabilities like accounts payable short-term debt will be rolled over, which need not be true times dtress. Furrmore, as Jensen (1989) argues, an ability make fixed payments at low levels debt may have very different implications for control firm than an ability make those payments at high levels debt. The former more likely lead liquidation while latter may lead reorganization (especially if debt closely held). Anor problem that se measures are very sensitive come fluctuations. With se caveats md, we report Table III, Panel A all above mentioned defitions leverage for different countries. The sck measures are computed both at book value quasi-market values-where book value equity replaced market value equity. At th stage, we do not attempt adjust measures for differences accountg. Accordg first defition (nonequity liabilities tal assets) lted companies Anglo-American economies have considerably lower median leverage 1991 (about 0.56) than companies Contental Europe Japan (0.70). The means as well as aggregate ratio (obtaed summg tal liabilities across companies a country dividg summed assets) corroborate th. Th measure, a sense, fers an upper limit amount leverage different countries. If market values are used, Japan not considerably more levered than Anglo-American countries, while countries Contental Europe still seem have higher leverage. All use subject JSTOR Terms Conditions

11 1430 The Journal Fance Table III Extent Leverage Adjusted Leverage Different Countries Leverage measures are calculated for all non-fancial companies reportg consolidated balance sheets In 'book' column, equity measured at book value. In 'market' column, equity measured at market value. For stance, th alters value assets as follows: (quasi) market value assets obtaed subtractg book value common equity from book assets addg back market value common equity. In Panel A, Nonequity liabilities tal assets sum all liabilities divided value assets. Debt tal assets value short term plus long term debt divided tal assets. Debt net assets book value debt divided net assets where net assets assets mus accounts payables or current liabilities. Debt capital book value debt divided sum book value debt equity. EBIT earngs before terest taxes. EBITDA earngs before terest, taxes, depreciation. Aggregate ratios are obtaed summg numerar across all reportg firms country dividg denomar summed across same firms. In Panel B, Adjusted liabilities are tal liabilities less pension liabilities ( Germany) less cash. In Panels B C, Adjusted debt book value debt less value cash marketable securities. Adjusted assets are tal assets less cash short term securities, less pension liabilities ( Germany), less tangibles. Adjusted book equity book equity plus provions plus deferred taxes less tangibles. In Panel C, Interest coverage ratio come before terest taxes divided terest expense. Interest coverage earngs before terest taxes divided terest. Panel A: Extent Leverage Different Countries Nonequity Liabilities Debt Debt Debt Interest Coverage Total Assets Total Assets Net Assets Capital Ratio (Medians (Means) (Medians (Means) (Medians (Means) (Medians (Means) (Medians Aggregate) Aggregate) Aggregate) Aggregate) Aggregate) Number EBIT/ EBITDA/ Country Firms Book Market Book Market Book Market Book Market Interest Interest United (0.66) 0.44 (0.44) 0.27 (0.31) 0.20 (0.24) 0.34 (0.38) 0.24 (0.27) 0.37 (0.37) 0.28 (0.32) States Japan (0.67) 0.45 (0.45) 0.35 (0.35) 0.22 (0.24) 0.48 (0.35) 0.27 (0.29) 0.53 (0.52) 0.29 (0.31) Germany (0.72) 0.60 (0.56) 0.16 (0.20) 0.12 (0.16) 0.21 (0.25) 0.15 (0.19) 0.38 (0.39) 0.23 (0.28) France (0.69) 0.64 (0.61) 0.25 (0.26) 0.21 (0.23) 0.39 (0.39) 0.32 (0.33) 0.48 (0.46) 0.41 (0.41) Italy (0.67) 0.70 (0.67) 0.27 (0.28) 0.29 (0.28) 0.38 (0.38) 0.38 (0.39) 0.47 (0.46) 0.46 (0.47) United (0.57) 0.40 (0.42) 0.18 (0.21) 0.14 (0.16) 0.26 (0.31) 0.18 (0.21) 0.28 (0.29) 0.19 (0.23) Kgdom Canada (0.61) 0.49 (0.47) 0.32 (0.36) 0.28 (0.27) 0.37 (0.39) 0.32 (0.31) 0.39 (0.39) 0.35 (0.36) All use subject JSTOR Terms Conditions

12 What Do We Know about Capital Structure? 1431 Italy Japan Canada United France United Germany Country States Kgdom Book (0.56) (0.48) (0.69) (0.75) (0.58) (0.69) (0.61) (Medians Total Nonequity Assets Liabilities Market (Aggregate)) (0.50) (0.35) (0.69) (0.65) (0.46) (0.48) (0.45) Panel B: Extent Book Debt (0.36) (0.13) (0.21) (0.22) (0.05) (0.30) (0.33) (Medians Table Total Adjusted (Aggregate)) Assets Market (0.32) (0.09) (0.21) (0.18) (0.04) (0.21) (0.24) Leverage III-Contued Different Book Debt (0.41) (0.18) (0.33) (0.34) (0.07) (0.44) (0.41) (Medians Net Countries Assets Market (Aggregate)) (0.36) (0.11) (0.33) (0.27) (0.05) (0.27) (0.28) Book (0.44) (0.19) (0.41) (0.46) (0.10) (0.49) (0.45) (Medians Debt Capital Market (Aggregate)) (0.40) (0.13) (0.41) (0.36) (0.06) (0.28) (0.31) All use subject JSTOR Terms Conditions

13 1432 The Journal Fance Table III-Contued Panel C: Adjusted Leverage for Different Subsamples Changes Leverage over Debt Book Comparons with Time for Capital for Studies Usg Consolidated Consolidated The effect OECD Data: Comparon with Firms Reportg Debt levels for Firms Similar consolidation Nonequity OECD Data: through Consolidated Size Across on Debt Liabilities Debt Book Debt Book Firms 1986 Countries Book Capital Book Assets Assets 1982 Capital (Median) (Median) (Median) (Aggregate) (Aggregate) (Aggregate) Firms Firms Debt All Change Capital Largest Smallest Firms Level between Interest (Book Size Size Reduced Global Global 1991 Country Coverage Value) Qutilea Qutile l991b Samplec Rurfordd Borioe Vantage' OECD Vantagef United States Japan Germany France Italy United Kgdom Canada a Consolidated firms from all countries are pooled size qutiles are formed based on market value assets U.S. dollars b Includes firms that present unconsolidated balance sheets also. c Drops most debted consolidated firms such that remag sample contas consolidated firms numberg 76 percent all reportg (consolidated unconsolidated) firms country. For example, U.S. 24% most debted firms are dropped. d Rurford bases her computations on OECD stattics for The ratio termed tal debt assets ratio ratio aggregate non-equity liabilities tal assets. e Borio's computations use OECD stattics for The ratio termed gross debt assets ratio aga ratio aggregate non-equity liabilities tal assets. f Aggregates are computed for all reportg firms Global Vantage database. All use subject JSTOR Terms Conditions

14 What Do We Know about Capital Structure? 1433 Movg on debt tal assets, our fdgs change considerably. Now Germany United Kgdom appear have low levels leverage, both as a fraction book value market value assets. Of course, part low leverage for Germany may be because way pension liabilities are treated. We will correct for th shortly. The debt net assets ratio also suggests that corporations Germany United Kgdom have lower leverage than or countries. If leverage defed as debt over capital, North American countries Germany have a similar median leverage around 38 percent, United Kgdom appears substantially less leveraged (28 percent), while France, Italy, Japan have substantially more leverage (respectively 48, 47, 53 percent). However, figures for Japan may be maly due potential undervaluation assets. In fact, leverage appears a more normal 29 percent if it measured at 1991 market value (recall that at end fcal year 1991, Japanese sck market had fallen approximately 50 percent from its 1989 peak, so th fdg not necessarily beg driven "excessive" sck valuations). Market value measures, however, confirm higher level sck leverage France Italy. The aggregate ratios aga suggest that Germany United Kgdom are relatively underlevered. We also compute median aggregate terest coverage ratios. In second last column Table III, Panel A we report ratio come before terest taxes terest expense. In last column, depreciation added numerar coverage defition. Despite potential downward bias German data, terest coverage figures corroborate our earlier fdgs that Germany United Kgdom have lower leverage than or countries our sample. Although country rankgs are somewhat a function measure used, one major fact emerges: neir German nor Japanese companies are very highly levered U.S. stards. Th surprg light previous research which we will dcuss shortly. Germany, particular, comes across as a surprgly low levered country. But firms United Kgdom also have low leverage, while corporate leverage or five countries seems be quite similar. C. Adjustg Leverage for Differences Accountg Before drawg strong conclusions from th exerce, we have check if se fdgs are robust adjustments for differences reportg stards. We lt mimal adjustments that may be desirable. First, consider cash balances. Although we do not know how much cash short-term vestments are really needed run a busess, it terestg explore implications treatg se as excess liquidity, fsettg m an equivalent amount debt, removg both from balance sheet.12 There are or items which assume special importance some countries because 12 These cash balances may also represent compensatg balances required banks. All use subject JSTOR Terms Conditions

15 1434 The Journal Fance accountg differences. For stance, unlike U.S. firms, German companies do not net out pension assets pension liability ir balance sheets. While we do not know level pension assets, a first approximation assume that pensions are fully funded, we subtract pension liabilities from assets Germany. There are three additional adjustments that need be made. First, deferred taxes should really be considered a component shareholders' equity.13 Given that importance th item varies across countries, it seems appropriate analyze impact addg it book value equity. Second, value U.S. assets may be exaggerated with respect those or countries wave acquitions 1980s. Th because premium paid an acquition recorded as goodwill depreciated over 40 years. Evidence th possible drtion can be found dramatic crease tangibles United States over 1980s (2.2 percent assets percent 1991), much smaller number or countries (except for France). To elimate th potential bias we subtract value tangibles from book value equity all countries ( reduce assets accordgly). Fally, we argued that a large fraction German liabilities composed dubious provions for future liabilities, which are really equity. Therefore, for sake comparon, we reclassify se liabilities as shareholders' equity all countries. The net effect all se adjustments Table III, Panel B. The amount leverage every country except Canada drops substantially. In particular median ratio adjusted debt capital Japan drops from Germany from But, if anythg, results seem strengn our claim; with exception United Kgdom Germany (median adjusted debt book capital , respectively), firms or five countries have similar leverage (median adjusted debt book capital between ). In everythg that follows, we use adjusted measures leverage. D. Additional Checks for Robustness There are a number additional checks that can be performed. To check that our results on leverage are not special year chosen, we look at terest coverage ratio debt capitalization ratio 1986, a year that for many countries represented peak economic expansion.14 As Table 13 The extent which deferred taxes are equity-like may vary across countries. For stance, much sample period (i.e., before FAS 109), U.S. firms used deferral method which (loosely speakg) change deferred tax liability was determed from come statement. On or h, U.K. companies used liability method which level deferred taxes determed after an assessment probable future tax payments, change deferred taxes n calculated. Therefore, deferred taxes United Kgdom may be more debt like. Adjustg for th would not change our results qualitatively. 14 It would appear that best measure leverage for Germany debt capital, where capital defed as sum debt, equity, untaxed reserves. Interest coverage may understate amount leverage because come from pension assets held on balance sheet will also count as come, it may overstate leverage because propensity hide come (though latter effect may be small when we average over firms time periods). All use subject JSTOR Terms Conditions

16 What Do We Know about Capital Structure? 1435 III, Panel C dicates, terest coverage aga high for Germany United Kgdom, while it approximately equal for or countries. The debt book capital measure corroborates th.15 Differences leverage can be attributed different size composition G-7 country sample. For th reason, we compare leverage companies belongg smallest 20 percent largest 20 percent dtribution firms sorted dollar market value assets Independent size firm, firms United Kgdom Germany are less levered while all or countries are approximately at same level. Interestgly, Germany only country where larger firms have lower leverage (as can also be seen Table III, Panel A III, Panel B comparg median or mean leverage with aggregate leverage). Some countries have a greater number lted firms which state has a majority ownership. Firms those countries may appear have higher leverage because we do not account explicitly for state guarantees debtholders. We identified state owned companies our sample for France Italy, countries with largest state secr. There are 21 such companies all. Median adjusted debt capital for se firms 0.83 France, which much higher than median for firms private secr, But for Italy, medians are closer ger, at , respectively. Sce state owned firms are such a small proportion our sample, droppg m does not alter our conclusions. In rest analys, y are dropped from sample. Anor source concern that restrictg our attention firms with consolidated balance sheets, we might have a significantly biased measure leverage for some countries. Durg 1980s, firms were under an creasg pressure present consolidated balance sheets, both from local authorities from fancial markets. In 1989 alone, 109 Japanese firms our sample moved consolidated accounts. An estimate impact consolidation can be obtaed lookg at how level leverage changes when a firm starts report consolidated balance sheets. We estimate th difference country country.16 In all countries except United Kgdom, we fd that year when a firm moves consolidate accounts, its debt capital ratio creases relative previous year about five percentage pots. Th difference always stattically significant. In United Kgdom, difference only two percentage pots not stattically significant. Th suggests that absence consolidated accounts for all companies may lead us underestimate amount leverage especially Japan Germany where approximately 24 percent 15 It also possible that across country differences leverage may be simply due differences dustry composition G-7 sck markets. For stance, 28 percent Canadian firms are mg oil companies (as compared with 6 percent overall sample). We refore recompute measures for a homogenous group dustries. Our results are qualitatively unchanged. 16 In Canada United States all firms report on a consolidated bas. Therefore, our estimates are for remag five countries. In France, 4.3 percent firms do not consolidate, Italy, 9.2 percent, United Kgdom, 1.8 percent. All use subject JSTOR Terms Conditions

17 1436 The Journal Fance sample elimated each country consolidation requirement (for all or countries loss less than 10 percent). Interestgly, firms that do not report consolidated balance sheets se two countries have much lower leverage than firms that do. The median adjusted debt capital ratio for Japanese firms not reportg consolidated accounts 0.1 Japan 0 Germany, which much lower than median ratio for firms reportg consolidated accounts. Obviously, when we clude firms that do not present consolidated balance sheets sample, leverage Japan Germany furr reduced, though overall pattern does not change. It, however, important ask why firms that do not consolidate appear have such low leverage while leverage goes up when firms decide consolidate. One explanation we have emphasized so far that not consolidatg, firms may be concealg debt subsidiaries. But firms that do not consolidate are typically much smaller than firms that do; median firm that has not consolidated 1991 has sales 122 billion yen Japan 356 million DM Germany while median firm reportg consolidated balance sheets has sales 225 billion yen Japan 1271 million DM Germany.'7 The small size firms that do not consolidate suggests anor reason why y might have low leverage: most firms may move reportg consolidated balance sheets only when y have rae external fance (usually debt) domestically or from abroad. Th may account for crease leverage when firms consolidate. Firms that fance ternally are likely have low leverage also not fd much need for reportg consolidated balance sheets. Sce firms that do not consolidate are somewhat more pritable than firms that do (median EBITDA assets Japan 0.1 Germany 0.17 compared , respectively, for firms that consolidate), se may deed be firms which do not need external fance, thus have low leverage. These are, course, only conjectures. In absence more evidence on why firms that do not consolidate have low leverage, let us assume as an exerce that, on average, y have 10 percentage pots more unreported leverage than y actually report ( or words, we assume twice average crease when firms change reportg status). The median adjusted debt capital ratio Japan would go up from , that Germany from Th does not overturn our ma results Obviously, reported sales crease when a firm consolidates accounts. But crease (on average, 18 percent Japan 39 percent Germany) seem o small account for difference size. 18 We assume twice average level because decion start reportg on a consolidated bas may not be dependent amount hidden leverage that a company has. If th case, our estimate 5 percent will represent only a lower bound impact consolidation. Anor way control for th possibility followg test: worst case scenario for results that firms that choose not report on a consolidated bas are most highly levered. Then put Germany Japan on an equal footg with or countries we truncate samples for or countries at 76th percentile leverage. Even with th, Germany All use subject JSTOR Terms Conditions

18 What Do We Know about Capital Structure? 1437 E. Comparon with Results Prior Literature We are, course, not first compare capital structures across countries. With a few notable exceptions, many se studies conclude that companies Japan Contental Europe are more highly levered than companies Anglo-American economies.19 For example, Borio (1990) classifies former countries as "high leverage" latter as "low leverage". Rurford (1988) summarizes previous studies present additional evidence from Organization for Economic Cooperation Development (OECD) data suggestg that firms France, Germany, Japan are more highly levered than firms United States United Kgdom. While she acknowledges that adjustments for accountg differences, move market values, could narrow perceived difference leverage, she concludes that it probably would not alter ma fdg. Economts have explaed se perceived aggregate differences as due differences extent nature fancial termediation (see Borio (1990)), differences stitutional structures governg bankruptcy debt renegotiation (see Frankel Montgomery (1991)), differences market for corporate control (see, for example, Bergl (1990)). Given that so much work based on an empirical regularity that we do not fd, it important that we trace why our results differ from received wdom. There are at least four possible sources difference. The first that our measures differ from previous ones used. The second that adjustments we have made correct for differences accountg were not possible with earlier data. The third that our sample large firms differs from samples used earlier work. Fally, it possible that capital structures different countries have changed over time. As we shall argue, first two sources seem account for much difference our fdgs. Capital structures have also changed over time ways that strengn our fdgs, but differences our fdgs from previous literature exted even at begng our sample period. Both Rurford (1988) Borio (1990) use OECD data arrivg at aggregate measures leverage. Unfortunately, OECD figures for Germany do not report sck debt separately. Instead only nonequity liabilities are reported. So only measure leverage that can be reported for all G-7 countries ratio nonequity liabilities assets. Even though th measure termed tal debt tal assets Rurford, gross debt assets Borio, it actually a composite debt, trade credit, pension liabil- appears be less levered than all or countries except United Kgdom (see Table III, Panel C, column 6), or countries are higher. 19 We should pot out that we are not first question received wdom. For example, Kester (1986) fds that after controllg for a number determants, re are no major differences extent which firms are levered Japan United States. Mayer Alexer (1990) fd that large German firms borrow less than large U.K. firms. Yet no previous study that we know has looked at all G-7 countries detail, nor has any study suggested that supposed leverage differences between Anglo American firms firms Japan Contental Europe are very sensitive way leverage defed. All use subject JSTOR Terms Conditions

19 1438 The Journal Fance ities, deferred taxes, provions, or liabilities. As we have already argued, re are obvious problems drawg ferences about leverage from th measure. Interestgly, as can be seen Table III, Panel C, -country aggregate nonequity liabilities assets ratio 1982 for firms our sample corresponds closely that reported Rurford Borio (we choose 1982 because that only year that common between coverage Rurford's study Global Vantage Database). Sce OECD sample has far greater coverage than Global Vantage database (for stance, stattics for France OECD sample are based on over 800 firms) fact that differences are mor suggests that Global Vantage sample fairly representative aggregate corporate secr. More evidence on th can be obtaed calculatg debt book assets ratio usg OECD data. We cannot compute th ratio for Germany, but ratios for all or countries except Japan are close aggregate ratios computed from Global Vantage sample. While difference for Japan can stem from differences sample ( OECD presents aggregate figures based on over 25,000 firms), re are or explanations: cluded defition debt are bills that are dcounted fancial stitutions-arguably, th not borrowg firm should not enter a calculation leverage. Also, OECD data are not reported on a consolidated bas, so borrowg a subsidiary from a parent would crease aggregate leverage that sample.20 Fally, it possible that changes capital structures over 1980s may partly expla why our fdgs are different. We report changes aggregate leverage for consolidated firms reportg throughout period There an crease debt capital ratios for firms Anglo- American economies a decrease for firms or economies (see Table III, Panel C). The crease leverage most pronounced for U.S. firms while decrease leverage most pronounced for German firms. Yet broad pattern firms Germany United Kgdom havg lower leverage than rest exts even It, refore, appears that our conclusions are different from those reached most earlier researchers prcipally because we have more detailed measures, more comparable calculations, leverage. To summarize our fdgs thus far: usg different measures leverage correctg for major differences accountg, we can conclude that: (i) United Kgdom Germany have lowest leverage among G-7 countries; (ii) all or countries have approximately same amount leverage, with some changes rankg based on specific measure. 20 Our sample large firms typical those used prior ter-country comparons such as Kester's (1986) study 344 Japanese firms lted on first section Tokyo Sck Exchange. The tal debt book assets ratio for firms Kester's sample which not very different from 0.4 for firms our sample. The small difference can be attributed fact that we have more firms our sample; se are likely be smaller are thus likely have lower leverage. All use subject JSTOR Terms Conditions

20 What Do We Know about Capital Structure? 1439 Table IV Sources External Fancg ( ) External fancg as a fraction tal fancg ratio net external fancg sum cashflow from operations net external fancg. The figures for various components external fancg are aggregated across all non-fancial companies country across all years, n normalized net external fancg obtaed firms country durg period Net debt fancg sum net short term debt suances long term debt suances less long term debt reduction. Equity suance cludes sue both common preferred sck conversions debt equity. Net equity fancg sum equity suance less equity reduction. The data are obtaed from flow funds statement. The number firms varies year year. External Fancg Composition External Fancg as a Fraction Net Debt Net Equity Total Fancg Issuance Issuance Global OECDa Global OECD Global OECD Country Vantage Data Vantage Data Vantage Data United States Japan Germany France Italy United Kgdom Canada a OECD data from editions OECD publication "Fancial Statements Non-Fancial Enterpres." F. The Flow Fancg at Aggregate Level Up th pot we have largely restricted our analys sck-based measures extg capital structure. We now analyze choice fancg with flow funds data. A reason for extendg our analys that data on capital structure do not dtguh between equity built through retaed earngs equity obtaed through sck fergs. As Myers (1984) pots out, costs associated with se two forms fancg are very different. In Table IV, we report sources fancg for firms four countries for which we have flow funds data. For United States, United Kgdom, Canada, external fancg smaller than ternal fancg, with firms United States rag least from external sources. But firms Japan constently rae more money externally than ternally.21 Is greater dependence on external fancg also true for countries Contental Europe? Unfortunately, th cannot be answered with Global 21 Usg O.E.C.D. data we also checked that th was true for Japanese firms over decade, , which precedes period covered our sample. Therefore, greater reliance Japanese firms on external fancg not driven re Japanese sck market late 1980s. All use subject JSTOR Terms Conditions

21 1440 The Journal Fance Vantage Database. So second column Table IV, we report fraction external fancg computed from OECD data. Firms Germany, France, Italy rae substantially less from external sources than eir firms United Kgdom or Canada. So, aga, re no clear dtction between Anglo-American economies ors. The reason U.S. firms creased ir leverage over 1980s despite usg so little external fance because external fance has consted entirely debt. The figures for United States are extreme, perhaps because tense activity market for corporate control over th period. But as comparon with OECD data suggests, y may underestimate crease corporate leverage over th period. The Global Vantage database cludes formation only for publicly traded companies while OECD data for all corporations. Therefore, Global Vantage will not clude additions debt ( reductions equity) that accompany leveraged buyouts. For stance, RJR Nabco not cluded sample after 1989 buyout (although it re-enters sample 1991 after reverse leveraged buyout). Fally, extremely low leverage for United Kgdom despite substantial levels external fancg a result a conscious emphas on equity suances rar than debt as a source external fancg. All th suggests that levels leverage that we see different countries do not are romly, but are a consequence conscious fancg choices made firms. III. Institutional Differences Leverage In previous section we showed that differences leverage across G-7 countries are not as large as previously thought. Only firms United Kgdom Germany appear be substantially less leveraged than firms or G-7 countries. These fdgs beg question why firms countries such as Japan United States with such diverse stitutions have a similar amount leverage, why firms countries such as United Kgdom United States with similar capital markets fancial stitutions have such different levels debt. Much previous literature has focussed on a classification countries based on size or power bankg secr, hence term "bank-oriented"(japan, Germany, France, Italy) "market-oriented" countries ( United States, United Kgdom, Canada). In th section, we argue that th just one, perhaps not most important, stitutional difference between G-7 countries. The tax code, bankruptcy laws, state development bond markets, patterns ownership also may matter ( course, se stitutional features may be fluenced bankg secr). Our aim th section not settle question wher stitutional differences are responsible for differences aggregate corporate capital structure, but rar rae questions that need be addressed future research. All use subject JSTOR Terms Conditions

22 What Do We Know about Capital Structure? 1441 Table V Tax Treatment Interest, Dividends, Retaed Earngs Different Countries Tax advantage debt with respect retaed earngs (dividends) computed usg formulas derived Miller (1977). The different rows correspond different assumptions on margal personal tax rate vesr. The "average" vesr assumed head a family three, has three times per capita come. United United States Japan Germany France Italy Kgdom Canada Tax free vesr Tax adv. w.r.t. retaed earngsa Tax adv. w.r.t. dividendsb Invesr p tax bracket 1990 Tax adv. w.r.t. retaed earngs Tax adv. w.r.t. retaed earn. no cap. gas taxc Tax adv. w.r.t. dividends Invesr p bracket 1990 cludg local taxes Tax adv. w.r.t. retaed earngs Tax adv. w.r.t. retaed earn. no cap. gas tax Tax adv. w.r.t. dividends Average Invesr cludg local taxes Tax adv. w.r.t. retaed earngs Tax adv. w.r.t. retaed earn. no cap. gas tax Tax adv. w.r.t. dividends Invesr p bracket 1983 cludg local taxes Tax adv. w.r.t. retaed earngs Tax adv. w.r.t. retaed earn. no cap. gas tax Tax adv. w.r.t. dividends Sources: various editions "Dog Busess...," "Individual Taxes: A Worldwide Summary," "Corporate Taxes: A Worldwide Summary" publhed Price Waterhouse. a [1 - (1 - t )(1 - - where tpe)/(l tp)] t, corporate tax rate, tpe capital ga tax rate tp personal tax rate. b [1 - (1 - t")(1 - - tpd)/(l tp)] where tpd tax rate on dividends after dividend tax credit accounted for. C[1 - (1 - t")/(1 -tp)]. A. The Effect Taxes on Aggregate Leverage We first exame effect tax code on aggregate leverage. The extg empirical literature on ternational capital structure differences (see, for example, Mayer (1990)) claims that taxes have no explanary power. However, as we argue below, th conclusion may be unwarranted if personal taxes are also considered addition corporate taxes. Unfortunately, wher taxes have explanary power or not highly sensitive assumptions about margal vesr's tax rate. Table V reports relative tax advantage debt with respect retaed earngs dividends (see Miller (1977)) as 1990, under different assump- All use subject JSTOR Terms Conditions

23 1442 The Journal Fance tions about personal tax rate recipient. The relative tax advantage debt very sensitive se assumptions. For stance, a tax-exempt vesr fds debt more tax advantaged Germany than United States (tax advantage 50 versus 28 percent). However, th conclusion reversed if we consider an vesr who taxed at p margal tax rate each two countries (-6 versus 28 percent). Clearly, se two cases do not exhaust possibilities. Invesrs care about all taxes y pay, not just taxes levied central government. We clude effect corporate personal taxes levied city where primary sck exchange country located (third row Table V). Th changes magnitude computed tax advantage debt. Furrmore, rar than havg vesrs pay maximum statury rate or no taxes at all, we could assume that y have personal tax rate "average" citizen (somewhat arbitrarily, we defe th as somebody beg sole earner a family three thus earng three times per capita come). Th aga changes rankg countries on how tax advantaged debt. In addition differences personal tax rate margal vesr, one must recognize that most G-7 countries experienced major fcal reforms 1980s. The last row Table V reports tax advantage debt as 1983 for an vesr highest tax bracket. In general, tax reforms creased tax advantage debt with respect retaed earngs, but decreased its advantage with respect dividends. In sum, Table V shows importance not only cludg personal taxes computation tax advantage debt, but also cludg "right" personal taxes. A prece computation effective tax rates, takg account come wealth levels population, margal corporate tax rate for firms, would require an entire study like one undertaken Kg Fullern (1984) for 1970s. Our modest objective here explore wher such a study warranted. We do that examg wher changes tax treatment debt equity are at all associated with changes way operatg prits are channeled vesrs. The first three rows Table VI show how a dollar pre-tax come allocated across debt, dividends, retaed earngs each country periods If taxes matter, we should observe a shift allocation pretax dollar wards route that has creased its after tax value most. At same time route that most penalized (or least advantaged) a tax reform should experience a reduction its flow. 22 Total debt payments are obtaed summg terest expenses across companies, tal dividends are similarly obtaed n grossed up a pre-tax rate multiplyg ratio tal pre-tax come tal after tax come (Th overstates true dividends if tax on dtributed prits different from tax on retaed earngs. Unfortunately, we have no way systematically correctg for th). Total pre-tax retaed earngs are obtaed subtractg (pre-tax) dividends debt payments from pre-tax earngs. All se are normalized tal pre-tax earngs. We calculate se numbers over three year periods so as mimize noe from poor economic conditions any sgle year. The fdgs are not qualitatively different if we restrict ourselves years All use subject JSTOR Terms Conditions

24 What Do We Know about Capital Structure? 1443 Table VI The Allocation Pre-Tax Dollar Various Routes Changes Allocation Over Time The aggregate terest expense, dividends, earngs for an economy are computed summg dividual firm values across firms. The share a pre-tax dollar paid debt economy terest paid (all variables are aggregates) divided come before terest taxes. The share a pre-tax dollar paid through dividends dividends paid grossed up a pre-tax value divided come before terest taxes. The dividends are grossed up a pre-tax value multiplyg earngs after terest before taxes dividg earngs after terest after taxes. The share a pre-tax dollar retaed one mus share paid debt mus share paid dividends. United United States Japan Germany France Italy Kgdom Canada Share a pre-tax dollar paid through route Debt Dividends Capital Gas Route most tax advantaged tax reforms between a Debt Debt Retaed Dividends Dividends Dividends Debt earngs Route least tax advantaged tax reforms between a Retaed Dividends Debt Debt Debt Retaed Dividends earngs earngs Change share pre-tax dollar flow between allocated companies consolidated reportg throughout route most tax advantaged Change share pre-tax dollar flow between allocated companies consolidated reportg throughout route least tax advantaged a From Table V under assumption that capital gas tax paid at statury rate. All use subject JSTOR Terms Conditions

25 1444 The Journal Fance If we use p personal tax rates, th fact what we see. The results are reported Table VI. For stance, United States debt route, which most tax advantaged 1986 reform, creases its share pre-tax dollar from $ $ By contrast, United States retaed earngs are least tax advantaged reform, ir share decreases from $0.35 $0.21. The share pre-tax dollar gog route that most tax advantaged tax reform between creases 5 7 countries-with a mean crease share 6 cents (t = 1.73). Conversely, route least advantaged tax reforms between saw a mean decrease share 6 cents (t = -1.4), with decreases 5 7 countries. A t-test for differences means suggests that re a significant effect taxes at 5 percent level. By contrast, if we repeat exerce usg changes tax advantage different routes for an "average" citizen we do not fd that taxes have a significant effect. All we have shown that one cannot easily dms possibility that taxes fluence aggregate corporate leverage a country. In order reach any conclusion on effect taxes, not only it important that researchers clude both personal corporate taxes, but it imperative y obta right effective rate. B. Bankruptcy Law As Harr Raviv (1992) suggest, bankruptcy law should be regarded as an tegral aspect a debt contract. The G-7 countries vary considerably ir bankruptcy procedures, especially extent which liquidation emphasized over renegotiation claims, extent which management has control durg bankruptcy process. Table VII outles salient features each country (see White (1993) Kaer (1994) for details). Bankruptcy law has a number important effects: Strict enforcement credir rights enhances ex ante contractibility. Furrmore, it commits credirs penalizg management ( equity holders) if firm gets fancial dtress, thus givg management strong centives stay clear it. Fally, strict enforcement reduces costly, long drawn out, hagglg between claimholders that ensues when re a possibility that origal contracts may be violated. By contrast, it may be easier keep pritable enterpres as gog concerns, or provide managers right centives postbankruptcy, if credir rights are violated bankruptcy. Countries differ extent which y manage th trade-f enforcg credir rights. For stance, bankruptcy law United States gives management substantial rights cludg ability propose a reorganization plan with 120 days filg ( period usually extended), a stay on attempts any credir collect, right manage firm durg proceedgs. By contrast, Germany's code much more credir friendly. A debr has present a plan dealg with its solvency with fifteen days learng it, secured credirs are not stayed filg, management replaced All use subject JSTOR Terms Conditions

26 What Do We Know about Capital Structure? 1445 a receiver durg proceedgs. Th, Kaer (1994) argues, leads little reorganization that takes place beg conducted privately under supervion banks. Similarly, Franks Torous (1993) compare U.K. bankruptcy code with that United States, conclude that " U.S. code appears have strong centives keep firm as a gog concern even when it worth more liquidation [while] U.K. code, emphasizg rights credirs- some cases givg priority one credir- likely lead o many premature liquidations." The or countries appear fall between extremes United States Germany or United Kgdom extent which y support credir rights. Is it a cocidence that countries where ex ante contract most strictly enforced are also ones where firms have least debt? Also, do firms efficiently mata low leverage because bankruptcy code results o much liquidation viable firms? Or do firms efficiently mata low leverage because managers fear losg ir firm-specific human capital vestment if firm liquidated? The answers await future research. C. Bank Versus Market Based Countries Contrary previous studies (see, for example, Bergl (1990) references it) we do not seem fd any systematic difference between level leverage so-called bank-oriented countries (Japan, Germany, France, Italy) so-called market-oriented countries (United States, United Kgdom, Canada). Th raes questions wher "bank orientation" a meangful dtction wher differences importance bankg secr have any effect on a firm's fancg decions. There no doubt that re are major differences power banks across G-7 countries.23 The two polar cases are probably represented Germany United States, with all or countries fallg between. In Germany, banks are both allowed underwrite corporate securities own equity dustrial companies. In United States, significant limits are placed on both activities (see Kroszner Rajan (1995), James (1994)). However, above classification bank-oriented markeriented countries does not match very well with extent bank powers. In United Kgdom, banks have most powers German banks have, even if y do not use m. By contrast, France Italy only recently moved a universal bankg system. A better measure importance bankg secr fancg firms ratio bank loans made private secr gross domestic product (GDP). We report ratio calculated for middle period covered our sample, 1986, first column Table VIII. Th measure suggests that bankg secr more important bank-oriented econo- 23 Roe (1994) has a detailed dcussion bank powers Germany, Japan, United States. Derme (1990) lts bank powers several developed economies. All use subject JSTOR Terms Conditions

27 1446 The Journal Fance Tale VII Salient Features Bankruptcy Code Different Countries Management Control Rights Secured Country Forms Liquidation Forms Reorganization Bankruptcy Aumatic Stay Credirs United Chapter 7: Can be Chapter 11: Can be Trustee appoted Aumatic stay on Secured credirs get States voluntary (management voluntary (management Chapter 7. Management any attempts highest priority any files) or voluntary files) or voluntary stays control collect debt once settlement. However, ir (credirs file). (credirs file). Chapter 11. filg takes place. attempts collect payment are also stayed unless court or trustee approves. Japan Court Superved Composition (Wagi-ho), Third party appoted All credirs are Secured credirs have Liquidation (Hasan) Corporate Arrangement except composition stayed except highest priority Special Liquidation (Kaha Seiri) corporate arrangement. court superved greater votg rights (Tokubetsu Sean). The Reorganization (Kaha liquidation renegotiation. However, latter less costly a Kosei-ho). The lt composition where can be subject stay broader set firms are order creasg only unsecured dependg on petition eligible file. eligibility. Only debrs credirs are that filed. file. stayed. Germany Liquidation Composition (Vergleich or Receiver appoted Only unsecured Secured credirs can (Konkursordnung) can be Zwangsvergleich) can be manage firm. credirs are recover ir claims even requested credirs or filed for only debr. stayed. after a bankruptcy filg. debr. Management No stay for secured required file as soon as credirs. it learns it solvent. France Liquidation (Liquidation Negotiated Settlement Debr loses control Stay on all Secured credirs may lose Judiciaire) (Reglement Amiable) liquidation. Debr credirs judicial status if court determes where a court appoted remas control arrangement. security necessary conciliar attempts a orwe but submits for contuation settlement with credirs court appoted busess, or if securg Judicial Arrangement admtrar's decions asset sold as part (Redressement Judiciaire). a judicial arrangement. settlement. All use subject JSTOR Terms Conditions

28 What Do We Know about Capital Structure? 1447 Tale VII-Contued Italy Bankruptcy (Fallimen). Preventive Composition Debr removed from Stay on all Secured credirs stayed (Concorda Preven). control over firm. credirs. bankruptcy, though composition allowed only if enough value exts pay secured credirs full 40% unsecured credir claims. Secured credirs follow admtrative claims priority. United Members' voluntary Admtration, Debr removed from Stay on all Secured credir may Kgdom wdg up, Credirs' Admtrative control except credirs prevent admtration voluntary wdg up, Receivership (usually members' voluntary admtration, on order appotg h Compulsory wdg up. ends sale busess), wdg up. unsecured only own receiver. A credir Voluntary liquidation, no with a fixed or floatg Arrangement. stay a voluntary charge can appot an arrangement until admtrative receiver a proposal realize security approved. pay credir. Canada Liquidation proceedgs Firms can file for Firm control Stay on all Secured credirs have much like Chapter 7 aumatic stay under reorganizations while credirs give 10 days notice United States. Companies Credirs trustee appoted for reorganization. debr tent Arrangement Act or liquidations. Trustee may repossess collateral. Bankruptcy be appoted oversee Repossession even close Insolvency Act. management some bankruptcy filg reorganizations at permitted, but stayed dcretion court. after filg. Sources: Kaer (1994), Lo Pucki Triant (1994), Teichner (1991), White (1993). All use subject JSTOR Terms Conditions

29 1448 The Journal Fance Table VIII Size Capital Markets G-7 Countries 1986 Domestic Bank Credit Sck Market Bond Market Private Secr Sck Market Capitalization Bond Market Capitalization as a Fraction Capitalization as a Fraction Capitalization as a Fraction Country GDP (%) ($ billion) GDP (%) ($ billion) GDP (%) United States Japan Germany France Italy United Kgdom Canada Sources: Sck Market GDP data are from Morgan Stanley Capital International Perspective, Domestic Credit from International Fancial Stattics, Bond Market Data are from Salomon Brors International Bond Markets Analys, mies.24 Th dtction also supported or measures importance fancial markets. In Table VIII, we present tal capitalization each country's corporate equity corporate bond market 1986 normalized GDP same year. Bank-oriented countries have very small fancial markets. The exception Japan where much growth markets for corporate securities came 1980s when strong hold banks over corporate fancg was relaxed (see Hoshi, Kashyap, Scharfste (1990a)). In light our evidence, it would appear that difference between bank oriented countries market oriented countries reflected more choice between public (scks bonds) private fancg (bank loans) than amount leverage. Th not surprg even from a oretical pot view. While it might appear that closer monirg control firm management provided banks should make more debt fancg available bank oriented countries, recent work (Diamond (1991), Rajan (1992), Sharpe (1990)) has emphasized costs excessive bank debt. So despite greater availability debt fance from banks, firms bank-oriented countries may not want borrow beyond a pot. An alternative explanation that banks se countries provide both debt equity fance firms so greater availability fancg does not reflect leverage ratio. Which explanations, if any, correct, a question for future research. 24 Loans private secr cludes both consumer loans loans busesses. However, from Jappelli Pagano (1994), we know that consumer credit much more developed United States Canada than or countries. If we adjust for th, banks become even less important sources fance for firms Anglo-American economies. All use subject JSTOR Terms Conditions

30 D. Ownership Control What Do We Know about Capital Structure? 1449 Anor major stitutional difference across G-7 countries level ownership concentration workg market for corporate control (Bergl (1990) Franks Mayer (1994)). The United States, United Kgdom a much lesser extent, Canada have firms with diffused ownership, but also, an active takeover market. Some economts (see Roe (1994), for example) have even suggested that active takeover market substitutes for control over management provided a concentrated ownership. By contrast, Contental Europe Japan, ownership highly concentrated, thanks use ter-company cross-holdgs, pyramidg ownership dual class sck. As a consequence, hostile acquitions are almost unheard. Franks Mayer (1994) report only three attempts at hostile acquition Germany entire post-world War II period. The effect ownership concentration on capital structure far from obvious. On one h, presence large shareholders on board direcrs should reduce extent agency costs between managers shareholders facilitate equity sues. Furrmore, se shareholders may be undiversified, which may crease ir aversion debt. On or h, if some se large shareholders are banks, y might have a vested terest reducg amount outside sourcg ir clients, forcg m borrowg from banks. Therefore, it may not be surprg that we cannot detect a clear relationship between concentrated ownership that characterizes some countries aggregate leverage. A strong pressure from takeover market may force firms crease leverage. Managers may take on debt so as commit payg out future cashflows (or so as commit restructurg firm). Th, turn, may make firm unattractive raiders (see, for example, Zwiebel (1992)).25 In th respect, United States sts out for tensity takeover pressure durg sample period. In fact, United States only country where equity sues are, on net, negative over period (see Table IV). Moreover, as seen Table III, Panel C, leverage creased considerably over th period even though, as dcussed earlier, our dataset leaves out highly leveraged gog private transactions. Is crease debt United States over 1980s a secular shift wards higher leverage, or will it reverse itself as much-needed restructurg effected? Only future research can tell. Fally, it appears that restructurg activity 1980s also had substantial impact on cross-sectional dtribution leverage (see Bernanke, Campbell, Whited (1990)). As Figure 1 shows, more firms United States have extremely high leverage. Thus dtribution leverage United States has a "fatter" right tail than countries without such pressure. The dtribution also fat-tailed United Kgdom Can- 25 The welfare effects th crease, though, are less clear. While Zwiebel (1992), takeovers push managers value maximizg capital structure, Novaes Zgales (1995) y may duce excessive leverage. All use subject JSTOR Terms Conditions

31 1450 The Journal Fance 21 _~ CD Percentile 9 USA ED Jap ii Germ EE France E3 Italy Ei UK E- Canada Figure 1. Dtribution debt--book-capital ratio. The graph plots level debt-book-capital ratio 1991 at different percentiles with-country cross sectional dtribution for each G-7 countries. The x-ax reports percentile dtribution, y-ax level debt--book-capital ratio. Debt--book capital book value adjusted debt divided sum book value adjusted debt adjusted equity. Adjusted debt book value debt less value cash marketable securities. Adjusted book equity book equity plus provions plus deferred taxes less tangibles. ada, two or countries that experience at least some hostile takeover pressure. Furrmore, right tail dtribution for se countries creased over 1980s (not shown). lv. Cross Sectional Evidence As previous section suggests, differences stitutions do seem have some power explag differences aggregate capital structure. However, it also suggests a broader terpretation stitutions than previous literature has focused on; extent which firms are levered an economy does not seem depend solely on share external fancg that banks account for that economy. Or facrs such as bankruptcy code, tax code, market for corporate control may also fluence aggregate capital structure. We have concentrated thus far on identifyg explag betweencountry differences capital structure. We now shift our focus crosssectional differences between firms a country. Previous studies on U.S. firms have establhed that capital structure cross-sectionally correlated with certa facrs. We first verify that se correlations contue hold for U.S. firms our data set. We n go on exame if se stylized relations hold or countries. Fally, we attempt establh why each facr has correlation with leverage that we document. In or words, use ternational data has two purposes: first simply document that correlations hold generally, while second- more important one- try expla cross-country variation se correlations. If, for All use subject JSTOR Terms Conditions

32 What Do We Know about Capital Structure? 1451 stance, a facr does not "work" predicted way anor country (conditional on no measurement or econometric problems), it must be eir because oretical rationale for facr workg United States spurious, or because stitutional differences alter how facr works. A. The Facrs Correlated with Leverage Accordg Harr Raviv (1991), consensus that "leverage creases with fixed assets, nondebt tax shields, vestment opportunities, firm size decreases with volatility, advertg expenditure, probability bankruptcy, pritability uniqueness product." We focus on four above mentioned facrs: tangibility assets ( ratio fixed tal assets), market--book ratio (usually thought as a proxy for vestment opportunities), firm size, pritability. We limit ourselves se for two reasons. First, se facrs have shown up most constently as beg correlated with leverage previous studies (see Bradley, Jarrell, Kirn (1984), Long Malitz (1985), Harr Raviv (1991)). Second, data severely limits our ability develop proxies for or facrs.26 Theories capital structure suggest how some facrs might be correlated with leverage. If a large fraction a firm's assets are tangible, n assets should serve as collateral, dimhg rk lender sufferg agency costs debt (like rk shiftg). Assets should also reta more value liquidation. Therefore, greater proportion tangible assets on balance sheet (fixed assets divided tal assets), more willg should lenders be supply loans, leverage should be higher. Highly levered companies are more likely pass up pritable vestment opportunities (Myers (1977)). Therefore, firms expectg high future growth should use a greater amount equity fance. As suggested Myers (1977), we use ratio market value assets book value assets as a proxy for growth opportunities. The effect size on equilibrium leverage more ambiguous. Larger firms tend be more diversified fail less ten, so size (computed as logarithm net sales) may be an verse proxy for probability bankruptcy. If so, size should have a positive impact on supply debt. However, size may also be a proxy for formation outside vesrs have, which should crease ir preference for equity relative debt. There are aga conflictg oretical predictions on effects pritability on leverage. Myers Majluf (1984) predict a negative relationship, because firms will prefer fance with ternal funds rar than debt. Jensen (1986) predicts a positive one if market for corporate control effective forces firms commit payg out cash leverg up. If it effective, however, managers pritable firms prefer avoid dciplary role debt, which would lead a negative correlation between pritabil- 26 For stance, magnitude nondebt tax shields or than depreciation not available, re are o few observations get a meangful measure earngs volatility, advertg expenditure R&D expenditure are rarely reported separately ( y are ten capitalized). All use subject JSTOR Terms Conditions

33 1452 The Journal Fance ity debt. On supply side, suppliers should be more willg lend firms with current cashflows. We measure pritability as cashflow from operations normalized book value assets. Fally, re possibility that correlations may stem from perceived mpricg. For stance, if firms typically sue sck when ir price high relative book value (see Korajczk, Lucas, McDonald (1991), Jung, Kim, Stulz (1994) for evidence th kd behavior United States; Loughran, Ritter, Rydqvt (1994) for or countries) one might observe a negative correlation between market--book ratio leverage. We will dcuss some se possibilities what follows. B. Facr Correlations United States The basic regression we estimate Leverage[Firm i] = a +,1 Tangible Assetsi Market Book Ratios + 13 Log Salesi + 14 Return on Assetsi + si We use two measures leverage based on adjusted debt capitalization ratio When equity measured at book value, we term measure book leverage we report regression coefficients Table IX, Panel A. The second measure market leverage where market value equity used computg capitalization. Estimated coefficients are Table IX, Panel B. All regressors are four year averages ( ) correspondg variables.27 The coefficients are estimated usg a censored Tobit model.28 It not surprg that all coefficients for U.S. firms have sign found previous work (see Harr Raviv (1991)), are significant at 1 percent level. A one stard deviation crease tangibility, market-book ratio, log sales, pritability change book leverage 23, -37, 23, -11 percent its stard deviation respectively.29 Column (i) Table IX, Panel B shows that all coefficients reta ir expected sign when dependent variable market leverage We average explanary variables reduce noe account for slow adjustments. We lag explanary variables one period reduce problem endogeneity. Summary stattics on variables for each country are available from authors. 28 In some cases adjustment generates a negative value leverage. To elimate outliers we truncate sample at -1. For th reason we compute Tobit regressions. The ordary least squares (OLS) results are very similar. 29 We measure effect changes on latent variable. 3 It possible that some partial correlations may obta because explanary variables are correlated with some firm specific omitted variables. Given parsimony our specification, it important that we test for th estimatg an OLS regression first differences dependent variable (book leverage 1991 less book leverage 1986) agast first differences explanary variables (we subtract average from average variable). All coefficients have same sign as levels regression. Also, we check wher between dustry variations or with dustry variations are largely drivg estimates. For "between" estimation, observations are average All use subject JSTOR Terms Conditions

34 What Do We Know about Capital Structure? 1453 Table IX Facrs Correlated with Debt Book Market Capital The dependent variable book leverage which adjusted debt adjusted debt plus book value adjusted equity Tangibility ratio fixed assets book value tal assets. Market--book ratio book value assets less book value equity plus market value equity all divided book value assets. Logsale logarithm net sales. Pritability EBITDA divided book value assets. All explanary variables are four year averages ( ). Stard errors are parenses. The regression cludes an tercept whose coefficient not reported. The regression estimated usg maximum likelihood a censored Tobit model. The estimated model : Leverage[Firm i] = a Tangibilityi Market-book Ratioi Log Salesi Pritabilityi + Ei. Country United United Variable States Japan Germany France Italy Kgdom Canada Panel A: Book Capital Tangibility 0.50*** 1.41*** 0.42** 0.53** *** 0.26*** (0.04) (0.18) (0.19) (0.26) (0.23) (0.07) (0.10) Market--book *** *** ** *** -0.11*** (0.01) (0.04) (0.07) (0.08) (0.14) (0.03) (0.04) Logsale 0.06*** 0.11*** -0.07*** *** 0.08*** (0.01) (0.02) (0.02) (0.02) (0.03) (0.01) (0.01) Pritability -0.41*** -4.26*** ** (0.1) (0.60) (0.52) (0.72) (0.85) (0.30) (0.22) Number observations Pseudo R Panel B: Market Capital Tangibility 0.33*** 0.58*** 0.28* ** 0.27*** 0.11 (0.03) (0.09) (0.17) (0.19) (0.22) (0.06) (0.07) Market--book -0.08*** -0.07*** *** -0.15** * -0.06** -0.13*** (0.01) (0.02) (.06) (0.06) (0.11) (0.03) (0.03) Logsale 0.03*** 0.07*** -0.06*** *** (0.00) (0.01) (0.02) (0.02) (0.03) (0.01) (0.01) Pritability -0.6*** -2.25*** ** -0.48*** (0.07) (0.32) (0.47) (0.53) (0.77) (0.24) (0.17) Number observations Pseudo R *, **, ***, significant at 10, 5, 1 percent level, respectively. two-digit Stard Industrial Classification dustry. In 'with' estimation, observations are differences from dustry means. From magnitude coefficients (estimates not reported), United States tangibility, market book ratio, size seem be proxy for both dustry firm, idiosyncratic charactertics firm itself. Interestgly, negative relationship between pritability leverage appears be specific withdustry regression. For between dustry regression, coefficient positive. All use subject JSTOR Terms Conditions

35 1454 The Journal Fance C. Cross-Sectional Correlations International Data Rar than exame each country separately, we outle broad patterns across countries, n draw attention exceptions.31 Tangibility always positively correlated with leverage all countries (both for book leverage market leverage regressions). The market--book ratio enters with a negative coefficient all countries, always significant at conventional levels market leverage regressions. Interestgly, market--book seems both have a between dustry component a with dustry component (estimates not reported). So idiosyncratic component market--book ratio for a firm matters as much as dustry market--book. Size positively correlated with leverage except Germany where it negatively correlated. Recall from Table III, Panel C that largest qutile firms Germany had much lower median leverage than smallest qutile, so th correlation not simply driven outliers. The correlation contues hold with dustry regressions suggestg that th not simply an dustry effect. Fally, pritability negatively correlated with leverage all countries except Germany. It economically significant France. Overall, facrs found be correlated with leverage United States appear be similarly correlated or countries as well. These facrs expla, on average, about 19 percent cross-sectional variation or countries ( explanary power ranges from 5 30 percent, with higher explanary power general for market leverage regressions). Th suggests that observed correlations are not completely spurious. However, we know that relationship between ories empirical proxies, at best, weak. Therefore, before concludg that extg ories have significant power explag capital structure, we should exame se correlations more carefully. D. What Behd se Facrs? D. 1. Tangibility The similarity correlations across countries may actually be a cause for concern about our understg oretical underpngs those facrs. For stance, an important facr seems be ratio fixed tal 31 Early attempts explore cross sectional determants capital structure different countries were undertaken Remmers, et al. (1974) Snehill, et al. (1975). Both studies analyze a sample large firms from four selected dustries five countries (United States, Japan, France, Norway, The Nerls) period They fd that dustry firm size are not important determants leverage, while pritability firm growth generally are. Toy et al. (1974) also conduct a survey on objective fancial executives different countries. Although ir limited sample prevents from wide generalization, it terestg mention some ir fdgs. In all countries managers thk about capital structure targets book value ( not market value) terms. Furrmore, ir ma goal appears be guaranteeg fancial stability ir company availability funds needed rar than maximizg shareholders' value. All use subject JSTOR Terms Conditions

36 What Do We Know about Capital Structure? 1455 assets (which we call "tangibility"). The rationale underlyg th facr that tangible assets are easy collateralize thus y reduce agency costs debt. Berger Udell (1994) show that firms with close relationships with credirs need provide less collateral. They argue th because relationship ( more formed monirg credirs) substitutes for physical collateral. If so, we should fd tangibility matterg less bank-oriented countries. While cautiong reader about obvious caveats that accompany comparons coefficient estimates across countries, it terestg note that a stard deviation crease tangibility creases book leverage about 20 percent its stard deviation all countries except Japan where it creases leverage 45 percent (we check that coefficient estimate not fluenced outliers). A possible explanation comes from market leverage regressions where importance tangibility Japan not very different from its importance or countries. Perhaps Japanese firms with fixed assets such as l could borrow more over 1980s because collateral value l appreciated ( appreciation was not reflected book value). So on a market bas, firms with a lot fixed assets are not highly levered. But it still puzzlg if tangibility only as important Japan as elsewhere for apparently stronger bank-firm relationships Japan should imply a lesser role for tangibility. D.2. Market--Book The ory predicts that firms with high market--book ratios have higher costs fancial dtress which why we expect a negative correlation. There may be or potential reasons for why market--book ratio negatively correlated with leverage. For stance, shares firms fancial dtress (high leverage) may be dcounted at a higher rate because dtress rk priced (as suggested Fama French (1992)). If th domant explanation, negative correlation should be driven largely firms with low market--book ratios. In fact, negative correlation appears be driven firms with high market--book ratios rar than firms with low market--book ratios. It unlikely that fancial dtress responsible for observed correlation.32 Anor reason for market--book ratio be negatively correlated with book leverage stems from tendency for firms sue sck when ir sck price high relative earngs or book value. Th would imply that correlation between market--book ratio leverage driven firms 32 We estimate a piecewe lear relationship between book leverage market book ratio for firms United States. We estimate different slopes for five different qutiles market book ratio, constrag functional relationship be contuous. The results clearly dicate that negative relationship driven firms with high market book ratios rar than firms with low market book ratios. In fact, for firms with a low market book ratio, relationship between leverage market book weakly positive. Th pattern holds four or six countries with average slope p two qutiles beg lower than slopes botm two qutiles. All use subject JSTOR Terms Conditions

37 1456 The Journal Fance who sue lots equity. We determe amount equity sued a firm (net repurchases) period divide firms quartiles on th bas. We n estimate stard book leverage regression with each quartile. The negative correlation market--book with leverage seems be driven maly large equity suers. In United States magnitude coefficient on market--book ratio thrice as large quartile sug most (f3 = -0.30, t = -7.90) as quartile sug least (f3 = -0.09, t = -1.86) difference stattically significant. Th result not special United States. In Japan, United Kgdom, Canada, market-book more negatively correlated with leverage for firms sug most ( 3Japan = 0.74, t = 4.8, 3UK 0.18, t = 1.83,I3Canada = -0.16, t = -1.28) than for firms sug least (I3Japan = 0.25, t = -1.48, U3UK =-0.14, t = -1.61, I3Canada = -0.12, t = -0.49), though difference coefficients significant only Japan.33 From a oretical stpot, th evidence puzzlg. If market-book ratio proxies for undervestment costs associated with high leverage, n firms with high market--book ratios should have low debt, dependent wher y rae equity ternally via retaed earngs, or externally. An alternative explanation suggested above evidence that firms attempt time market sug equity when ir price ( hence, ir market--book ratio) perceived be high. Thus, se firms have temporarily low leverage. Evaluatg importance each explanation a task for future research. D.3. Size Size may be a proxy for (verse) probability default. If so, it should not be strongly positively related with leverage countries where costs fancial dtress are low. Some economts such as Sheard (1989) Hoshi, Kashyap, Scharfste (199Gb) have suggested that Japanese firms tied a ma bank may face a lower cost fancial dtress because ma bank organizes corporate rescues. Yet size important Japan; a stard deviation crease size creases book leverage 33 percent its stard deviation (compared 23 percent United States).34 Th suggests that size does not simply proxy for a low probability default. Anor argument agast association size with low expected costs fancial dtress that firms tend be liquidated more easily Germany. Under assumption that liquidation very costly, small firms should be especially wary 33 It terestg that we fd se correlations because re a mechanical reason why we should not. The sue equity moves post-sue market book ratio wards one. So for firms sug a lot, we will tend fd bunchg market book ratios less significant correlations. 34 One might argue that size a proxy for wher a firm belongs a ma bank group; but average capitalization group nongroup firms Hoshi, Kashyap, Scharfste (1990b) approximately same, suggestg it not. All use subject JSTOR Terms Conditions

38 What Do We Know about Capital Structure? 1457 debt Germany. However, large firms have substantially less debt than small firms Germany. An alternative argument for size that formational asymmetries between siders a firm capital markets are lower for large firms. So large firms should be more capable sug formationally sensitive securities like equity, should have lower debt. Unfortunately, th neir squares with negative correlation between size leverage observed for most countries, nor it true that large firms sue more. In all four countries for which we have flow funds data, net equity suances firms largest size quartile significantly less over period (as a fraction market value assets 1985) than for firms smallest size quartile. A similar result true when we consider gross equity sues (i.e., without nettg out repurchases). We have conclude that we do not really underst why size correlated with leverage. D.4. Pritability Fally, pritability negatively correlated with leverage. If short run, dividends vestments are fixed, if debt fancg domant mode external fancg, n changes pritability will be negatively correlated with changes leverage. As we have just noted, large firms tend sue less equity. The negative fluence pritability on leverage should become stronger as firm size creases. Th deed case for firms United States. For firms smallest size qutile a unit crease pritability decreases leverage For firms largest qutile, a unit crease pritability decreases leverage -1.09, over 4 times effect as that for smallest qutile ( significantly different). The relationship across qutiles nearly mononic; negative effect earngs on leverage considerably more important for large firms. Of course, as already dcussed, we do not quite underst why large firms are reluctant sue equity. Furrmore, re may be or forces at work, we cannot at present dentangle m. For stance, pritability for small firms may proxy for both amount ternally generated funds quality vestment opportunities, which have opposg effects on dem for external funds (debt). Lookg at or countries, leverage larger firms considerably more negatively correlated with pritability than for small firms Japan, Italy, Canada, while United Kgdom it more positively correlated. There no relationship Germany France. One explanation for why United Kgdom differs so much from United States may be that domant source external fance United Kgdom equity. So firms that are pritable have few vestment opportunities (i.e., large firms) will reduce equity sues drastically. These firms will have a more positive correlation between leverage pritability. By contrast, if pritability also correlated with vestment opportunities All use subject JSTOR Terms Conditions

39 1458 The Journal Fance small firms have, n an crease pritability may lead greater equity suances, reducg correlation between pritability leverage.35 V. Conclusions We fd that, at an aggregate level, firm leverage more similar across G-7 countries than previously thought, differences that ext are not easily explaed stitutional differences previously thought important. The facrs identified previous cross-sectional studies United States be related leverage seem similarly related or countries as well. However, a deeper examation United States foreign evidence suggests that oretical underpngs observed correlations are still largely unresolved. We believe that our work suggests two les for future research. On one h, it necessary strengn relationship between oretical models empirical specifications those models. Th, we believe, will be possible only with more detailed data which will enable us identify more accurate proxies. On or h, a deeper understg effects stitutional differences necessary. These two research sues are related. Only through a better understg actual determants capital structure decions can we thk designg tests uncover possible impact stitutional environment. Conversely, a better understg fluence stitutions can provide us enough ter-country variation so as enable us identify fundamental determants capital structure. REFERENCES Alford, Andrew, Jennifer Jones, Richard Leftwich, Mark Zmijewski, 1993, Relative formativeness accountg dclosure different countries, Journal Accountg Research 31, Aghion, Philippe, Patrick Boln, 1992, An complete contract approach fancial contractg, Review Economic Studies 59, Barclay, Michael, Clifford Smith, 1995, The maturity structure corporate debt, Journal Fance, 50, Bergl, Erik, 1990, Capital structure as a mechanm control: A comparon fancial systems, Masahiko Aoki, Bo Gustafsson, Oliver E. Williamson, Eds.: The Firm as a Nexus Treaties (Sage Publications, London). Berger Alan Gregory Udell, 1994, Relationship lendg les credit small firm fance, Journal Busess 68, Bernanke, Benjam, John Campbell, Toni Whited, 1990, U.S. corporate leverage: developments , Brookgs Papers on Economic Activity I, Borio, C., 1990, Leverage fancg non-fancial companies; an ternational perspective, Economic Papers 27, Bank for International Settlements. Bradley, Michael, Gregg Jarrell, E. Han Kim, 1984, On extence an optimal capital structure: Theory evidence, Journal Fance 39, Firms United Kgdom above median size are much more reluctant sue equity if above median pritability (median equity sue from equity value 1985) than if below (median equity sue from equity value 1985). Conversely, firms below median size are much more willg (able) sue equity if above median pritability (median equity sue from equity value 1985) than if below (median equity sue from equity value 1985). All use subject JSTOR Terms Conditions

40 What Do We Know about Capital Structure? 1459 Coopers & Lybr, 1984, International Tax Summaries. A Guide for Planng Decions (John Wiley & Sons, New York) Derme, Jean, 1990, European Bankg 1990s (Basil Blackwell, Oxford). Diamond, Douglas, 1991, Monirg reputation: The choice between bank loans directly placed debt, Journal Political Economy 99, Edwards, Jeremy, Klaus Fcher, 1993, Banks, Fance Investment Germany (Cambridge University Press, Cambridge). Fama, Eugene, Kenneth French, 1992, The cross-section expected returns, Journal Fance 46, Frankel, Allen, John Montgomery, 1991, Fancial structure: an ternational perspective, Brookgs Papers on Economic Activity 1, Franks, Julian, Col Mayer, 1994, The market for corporate control Germany. Workg paper, London Busess School. Franks, Julian, Walter Torous, 1993, A comparon U.K. U.S. bankruptcy codes, Contental Bank Journal Applied Corporate Fance 6, Harr, Miln, Artur Raviv, 1991, The ory capital structure, Journal Fance 46, Harr, Miln, Artur Raviv, 1992, The design bankruptcy procedures, C.R.S.P. Workg Paper, University Chicago. Hoshi, Takeo, Anil Kashyap David Scharfste, 1990, Bank monirg vestment: Evidence from changg structure Japanese corporate bankg relationships, R. Glenn Hubbard, Ed.: Asymmetric Information, Corporate Fance Investment (University Chicago Press, Chicago). Hoshi, Takeo, Anil Kashyap, David Scharfste, 1991, Corporate structure, liquidity, vestment: Evidence from Japanese panel data, Quarterly Journal Economics 106, International Fancial Stattics, 1992, International Monetary Fund. James, Chrpher, 1994, When do banks take equity? An analys bank loan restructurgs role public debt, Workg paper, University Florida. Jappelli, Tullio, Marco Pagano, 1994, Savgs, growth, liquidity constrats, Quarterly Journal Economics 109, Jensen Michael William Mecklg, 1976, Theory firm: managerial behavior, agency costs capital structure, Journal Fancial Economics 3, Jensen, Michael, 1986, Agency costs free cash flow, corporate fance takeovers, American Economic Review 76, Jensen, Michael, 1989, The eclipse public corporation, Harvard Busess Review 67, Jung, Kooyul, Young-Cheol Kim, Ren6 Stulz, 1994, Investment opportunities, managerial dcretion security sue decion, Workg paper 4907, National Bureau Economic Research. Kaer, Kev, 1994, Corporate restructurg & fancial dtress: An ternational view bankruptcy laws implications for corporations facg fancial dtress, Workg paper, INSEAD. Kester, Carl, 1986, Capital ownership structure: A comparon United States Japanese manufacturg corporations, Fancial Management 15, Kg, Merv, Donald Fullern, 1984, The Taxation Income from Capital, (The University Chicago Press, Chicago). Korajczyk, Robert, Deborah Lucas, Robert McDonald, 1991, The effects formation releases on pricg timg equity sues, Review Fancial Studies 4, Kroszner, Rall, Raghuram Rajan, 1995, Organizational structure credibility: Evidence from commercial bank securities activities before Glass-Steagall act, mimeo, University Chicago. Long, John, Ileen Malitz, 1985, Investment patterns Fancial Leverage, Benjam Friedman, Ed.: Corporate Capital Structure United States (The University Chicago Press, Chicago). Loughran, Timothy, Jay Ritter Krtian Rydqvt, 1994, Initial public fergs: International sights, Pacific Bas Fance Journal 2, All use subject JSTOR Terms Conditions

41 1460 The Journal Fance LoPucki, Lynn, George Triant, 1994, A systems approach comparg U.S. Canadian reorganization fancially dtressed firms, Harvard International Law Journal 35, Mayer, Col, 1990, Fancial systems, corporate fance, economic development, R. Glenn Hubbard, Ed.: Asymmetric Information, Corporate Fance Investment, (The University Chicago Press, Chicago). Mayer, Col, Ian Alexer, 1993, Banks securities markets: Corporate fancg Germany United Kgdom, Dcussion paper, CEPR. Miller, Mern, 1977, Debt taxes, Journal Fance 32, Modigliani, Franco, Mern H. Miller, 1958, The cost capital, corporation fance, ory vestment, American Economic Review 48, Morgan Stanley Capital International Perspective, various sues. Myers, Stewart, 1977, Determants corporate borrowg, Journal Fancial Economics 5, Myers, Stewart, 1984, The capital structure puzzle, Journal Fance 39, Myers, Stewart, Nicholas Majluf, 1984, Corporate fancg vestment decions when firms have formation that vesrs do not have, Journal Fancial Economics 13, Nobes, Chrpher, Robert Parker, 1991, Comparative International Accountg (Prentice Hall, New York). Novaes, Walter, Luigi Zgales, 1995, Capital structure choice when managers are control: Entrenchment versus efficiency. Workg paper, University Chicago. OECD, Fancial Stattics, Part 3: Fancial Statement Non-fancial Enterpres, editions. Pagano, Marco, Ailsa Roell, 1991, Tradg systems European sck exchanges: current performance policy options, Economic Policy 10, Price Waterhouse, 1990, Individual Taxes: A Worldwide Summary (Price Waterhouse, New York) Price Waterhouse, 1991, Corporate Taxes: A Worldwide Summary (Price Waterhouse, New York) Price Waterhouse, various sues, Dog Busess. (Price Waterhouse, New York) Rajan, Raghuram, 1992, Insiders outsiders: choice between relationship arms-length debt, Journal Fance 47, Remmers, Lee, Arthur Snehill, Richard Wright, Theo Beekhuen, 1974, Industry size as debt ratio determants manufacturg ternationally, Fancial Management 3, Roe, Mark, 1994, Strong managers, weak owners: The political costs American corporate fance, (Prcen University Press, Prcen) Rurford, Janette, 1988, An ternational perspective on capital structure puzzle, Joel Stern Donald Chew, Eds: New Developments International Fance. (Basil Blackwell, New York, NY). Salomon Brors International Bond Markets Analys, Sharpe, Steven (1990), Asymmetric formation, bank lendg implicit contracts: A stylized model cusmer relationships, Journal Fance 45, Sheard, Paul, 1989, The ma bank system corporate monirg control Japan, Journal Economic Behaviour Organization 9, Snehill, Arthur, Theo Beckhuen, Richard Wright, Lee Remmers, Norman Toy, Annio Pares, Alan Shapiro, Douglas Egan, Thomas Bates, 1975, Fancial goals debt ratio determants: A survey practice five countries, Fancial Management 4, Teichner, William, 1991, A note on bankruptcy United States, Case note , Harvard Busess School. Toy, Norman, Arthur Snehill, Lee Remmers, Richard Wright, Theo Beekhuen, 1974, A comparative ternational study growth pritability, rk as determants corporate debt ratios manufacturg secr, Journal Fancial Quantitative Analys 9, White, Michelle J., 1993, Corporate bankruptcy: A U.S.-European comparon, Workg paper, University Michigan. Zwiebel, Jeffrey, 1992, Dynamic capital structure under managerial entrenchment. Workg paper, Stanford University. All use subject JSTOR Terms Conditions