The Asian Financial Crisis, Uphill Flow of Capital, and Global Imbalances: Evidence from a Micro Study

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 942 August 2008 The Asian Financial Crisis, Uphill Flow of Capital, and Global Imbalances: Evidence from a Micro Study Brahima Coulibaly and Jonathan N. Millar NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at http://www.ssrn.com/.

The Asian Financial Crisis, Uphill Flow of Capital, and Global Imbalances: Evidence from A Micro Study Brahima Coulibaly and Jonathan Millar (cid:3) Board of Governors of the Federal Reserve System August 21, 2008 Abstract ThisstudyassessestheroleoftheAsian(cid:133)nancialcrisisofthelate1990sintheemergenceand persistence of the large current account surpluses across non-China emerging Asia, which have beenasigni(cid:133)cantcounterparttotheU.S.currentaccountde(cid:133)cit. Usingpaneldataencompassing nearly 3,750 (cid:133)rms, we trace the current account surpluses to a marked and broad-based decline in corporate expenditures on (cid:133)xed investment in the aftermath of the crisis that cuts across a wide spectrum of countries, industries, and (cid:133)rms. The lower corporate spending in turn depressedaggregateinvestmentrates,widenedthesaving-investmentgap,andallowedtheregion to turn into a net exporter of capital. We then consider the factors behind this reduction in postcrisis corporate investment. While weaker (cid:133)rm-level fundamentals in the postcrisis period seemtoexplainpartofthedropininvestmentrates,ongoingre-structuringowingtolargedebts accumulated and excess investment undertaken in the run-up to the crisis has been the main sourceofrestraintpostcrisiscorporateinvestment. Theresultssuggestthatevenafteradecade, thee⁄ectofthe(cid:133)nancialcrisisisstilla⁄ectingcorporateinvestmentdecisionsinemergingAsia, andthatastherestructuringcompletesitscourse, investmentrateswilllikelyrisetocontribute to a gradual reduction in the region(cid:146)s current account surpluses. Keywords: Global Imbalance, Emerging Asia, Current Account, Investment JEL classi(cid:133)cations: F3, F21 CoulibalyisEconomistintheDivisionofInternationalFinanceoftheFederalReserveSystem. Jonathan (cid:3) Millar is Economist in the Division of Research and Statistics. Mailing address: Division of International Finance, Board of Governors, Federal Reserve System, Mail Stop 24, Washington D.C. 20551,USA; email: brahima.coulibaly@frb.gov. Tel.: (202)-452-2609;fax: (202)-736-5638. WethankJaneHaltmaierandShaghil Ahmedforhelpfulcomments,andJoshuaMazenandRossKnippenbergforexcellentresearchassistanceship. The views expressed in the paper are those of the authors and do not necessarily re(cid:135)ect those of the Board of Governors or the Federal Reserve System. 1

1 Introduction IntheaftermathoftheEastAsian(cid:133)nancialcrisisof1997and1998,theratioofaggregateinvestment toGrossDomesticProduct(GDP)inemergingAsiaexcludingChinafellfromanaverageofaround 33 percent to about 25 percent, and has remained at about this level in subsequent years. At the same time, aggregate savings as percent of GDP in these countries have declined only slightly, leading to a swing in the current account balances from slight de(cid:133)cits in the period leading up to thecrisistosubstantialsurplusesinthepostcrisisera. Thesesurplusesenabledtheregiontobecome an exporter capital in de(cid:133)ance of theory suggesting that capital should (cid:135)ow from capital-abundant to capital-scarce countries where returns on capital are higher. Indeed, data on the patterns of global current account imbalances indicate that the wider saving-investment gap for the region has been a signi(cid:133)cant counterpart to the large current account de(cid:133)cits in the United States since 1997, suggesting a possible role of the Asian (cid:133)nancial crisis in the emergence of global imbalances. Globalimbalances, thegrowingcurrentaccountde(cid:133)citoftheUnitedStatesandthecorresponding current account surpluses and accumulation of foreign exchange reserves in others countries(cid:150) mainly in East Asia and, more recently, in oil-exporting economies(cid:150)have been portrayed as perhaps the most important risk to the global economy. Chief among the risks is the possibility that the imbalances could unwind abruptly, with sharp contractions in assets prices (including the U.S. dollar), paving the way for a global (cid:133)nancial and economic crisis. Concerns of this nature have been voiced by Obstfeld and Rogo⁄[2000], Blanchard et al. [2005], Mussa [2004], and others.1 The quest to understand the causes of these imbalances and how they might unwind has generated a considerable amount of research that has tended to emphasize four broad explanations: di⁄erences in stages of demographic transitions (Feroli, 2003; Ferrero, 2002), di⁄erences in economic growth (Engel and Rogers, 2005), heterogeneity in stages of (cid:133)nancial market development (Caballero et. al, 2006; Mendoza et al., 2007), and emerging market (cid:133)nancial crises (Bernanke, 2005; Kamin and Gruber, 2007).2 The contribution of this paper is in spirit of the fourth explanation. Bernanke [2005] was among the early advocates of the view that (cid:133)nancial crises in emerging markets contributed to the emergence of global imbalances. He argues that the global imbalances owe to the availability of excess saving (or a savings glut) from overseas that has (cid:133)nanced the U.S. current account de(cid:133)cit. Bernanke notes that the global excess saving has mainly originated 1Additional references include Mann [2004], and Roubini and Setser [2004]. 2See also, Hubbard (2006), Prasad et al. (2006), Ju and Wei (2006), Obstfeld and Rogo⁄(2005) and others. 2

in emerging market economies, a development that he attributes to the series of (cid:133)nancial crises, including the Asian (cid:133)nancial crises in the 1990s. Gruber and Kamin [2007] more formally test this hypothesis for emerging Asia and con(cid:133)rm the predominant role of the Asian (cid:133)nancial crisis as an explanation for the patterns of global imbalances. Using aggregate data and a panel regression model similar to the approach in Chinn and Prasad [2003], they (cid:133)nd that none of the standard fundamental determinants of current accounts can explain either the large surpluses in emerging Asia or the large U.S. current account de(cid:133)cit unless the model is augmented to account for the Asian (cid:133)nancial crisis of the late 1990s. They conclude that the Asian (cid:133)nancial crisis played a key role in promoting current account surpluses for the economies in the region. Our study extends this line of inquiry by attempting to uncover the mechanism that links the decade old (cid:133)nancial crisis to the current account surpluses in non-China emerging Asia. While the link between the crisis and current account surpluses has been established by previous research, the mechanisms through which it might be occurring remain a open question. There are two main channels through which the (cid:133)nancial crisis could have caused the region to run current account surpluses. The (cid:133)rst (more direct) channel suggests that the postcrisis current account surpluses are the result of private optimizing behavior in the aftermath of the crisis. For example, the crisis couldhavedisrupted(cid:133)nancialintermediationwithintheeconomyresultinginacreditcrunchorthe crisis could have weakened the balances sheets of (cid:133)rms prompting prolonged cut backs in corporate investment spending. The second (indirect) channel suggests that the current account surpluses could be the result of shifts in government policies in the aftermath of the crisis such as keeping exchangeratesundervaluedtopromoteexport-ledgrowth,andtohelpaccumulateforeignexchange reserves as a bu⁄er against future crises (see for example Mann [2004]). Disentangling the source of the postcrisis investment drag has important implications for the future adjustment of these imbalances. If the surpluses are the consequence of optimal private behavior, one might expect the imbalances between saving and investment to narrow as the e⁄ect of the (cid:133)nancial crisis fades. If, as advocated by some studies, the surpluses are the result of deliberate government policies to promote economic development through export-led growth, they could persist for the foreseeable future. To better understand the link between the (cid:133)nancial crisis and current account surpluses, we use a large cross-country paneldata setof 3,750publicly traded (cid:133)rms in eight emerging Asian countries (Hong Kong, Indonesia, Malaysia, the Philippines, Singapore, South Korea, Taiwan, and Thailand). Using (cid:133)rm level data a⁄ords the unique opportunity to con- 3

ductagranularassessmentofthedeterminantsofinvestment, andtostudythemechanismthrough which the (cid:133)nancial crisis could be a⁄ecting investment dynamics in the region. To our knowledge, this is the (cid:133)rst comprehensive micro study on the determinants of emerging Asia(cid:146)s current account surpluses and on the unique role of the (cid:133)nancial crisis. The results from the study con(cid:133)rm the predominant role of the (cid:133)nancial crisis in generating the current account surpluses in the region since 1998. We (cid:133)nd that the shortfall in the region(cid:146)s aggregate investment that generated the current account surpluses owes to a marked and broadbased decline in corporate spending on (cid:133)xed investment in the aftermath of the crisis that cuts across a wide spectrum of countries, industries, and (cid:133)rms. We then consider the factors behind the postcrisis lower investment. The analysis indicates that weaker postcrisis fundamentals (valuation, pro(cid:133)tability etc.) account for part of the lower corporate investment spending, but more importantly, ongoing re-structuring owing to large debts accumulated and the excess investment undertaken in the period leading up to the crisis appear to be the main factors weighing down the postcrisis investment. These (cid:133)ndings support the hypothesis that the region(cid:146)s current account surpluses are a direct result of the (cid:133)nancial crisis, and suggest that as restructuring completes its course and excess capacity wanes, investment rates could rise to reduce the current account surpluses. The remainder of the study is structured as follows: in the next section, we review the pattern of global imbalances. In section 3, we describe the (cid:133)rm-level data used for the analysis. Section 4 shows the e⁄ect of the (cid:133)nancial crisis on balance sheets of (cid:133)rms and corporate investment. Section 5 estimates an econometric model of (cid:133)rm investment and presents the results. In Section 6 we further analyze the unique role of excess debt on corporate investment, discuss the implications for the path of current account surpluses in Section 7, and conclude in Section 8. 2 Pattern of Global Imbalances Table 1 presents the patterns of the global current account balances and highlights the importance of emerging Asia. The growing de(cid:133)cit in the United States (U.S.), particularly since the Asian (cid:133)nancial crisis, mirrors the growing surplus in emerging Asia. Adjusting the current accounts to exclude oil imports and exports paints a clear picture of U.S. de(cid:133)cits almost totally o⁄set by 4

emerging Asia(cid:146)s surpluses as indicated in Table 2.3 At the eve of the crisis in 1996, emerging Asia excludingChinaregistereda$6billioncurrentaccountde(cid:133)cit. Thede(cid:133)citreversedtoa$110billion surplus in 1998 that widened further to over $200 billion in 2005. During the same period, China(cid:146)s current account surplus rose from $11 billion in 1996 to $214 billion in 2005. Emerging Asia as a whole remains the single largest counterpart to the U.S. current account balance, tallying a $416 billion surplus to the U.S. $556 billion de(cid:133)cit, excluding oil, in 2005. This surplus is split nearly down the middle between China and the combined surplus in the Asia-8 region: Hong Kong, Indonesia, Korea, Malaysia, the Philippines, Singapore, Taiwan, and Thailand. These patterns of global current account balances popularized the conundrum that developing economies have turned into net exporters of capital, which contradicts theory suggesting that capital should (cid:135)ow from capital-abundant advanced countries to capital-scarce developing countries where returns on capital are higher. As indicated in Figure 1, the current account of the emerging Asia-8 economies switched from slight de(cid:133)cits prior to the crisis to sustained surpluses averaging over 5 percent of GDP since 1998(cid:150)around the time when the U.S. current account began to deteriorate. Figure 2 shows the current account surpluses for the Asia-8 economies in terms of the excess of national saving over investment. Saving rates on average have declined only slightly on balance, but investment rates dropped sharply from an average of about 33 percent between 1990 and 1997 to 25 percent in 1998, and have stayed at around this level ever since.4 Figure 3 breaks down investment into private and public sector components. Nearly all of the decline in the aggregate investment rate can be attributed to private investment behavior, as the public sector(cid:146)s investment rate held steady at about 5 percent of GDP since 1991. The private investment rate on the other hand dropped signi(cid:133)cantly during the crisis period, from over 27 percent of GDP in 1996 to below 20 percent in 1999. The drop in the private sector investment rate (even as the saving rates remained high) accounts for the swing in the current account from de(cid:133)cits to surpluses among the Asia-8 economies since 1998, and appears to have played a large 3We exclude oil trade because the portion of the global imbalances that owes to oil trade is well understood. It surfaced in tandem with the runup in oil prices, and all else equal, it will most likely fade if oil prices recede and/or the demand for oil falls. The portion of the imbalances from the non-oil trade on the other hand, has been around much longer, and it is much less understood. 4The analysis does not include China in part because it was not a⁄ected by the (cid:133)nancial crisis, which occurred at a time when China maintained restrictions on capital (cid:135)ows, and in part because China(cid:146)s investment and saving appears to be driven by di⁄erent dynamics. 5

role in the emergence of current account surpluses. 3 Description of Firm Level Data Forouranalysis, we use an unbalanced panel of annual(cid:133)rm-leveldatafrom 1991to 2005.The (cid:133)rmleveldatawereconstructedusinginformationfromtheWorldscopeDatabaseandincludedatafrom 3,750 publicly listed companies in eight countries that were more a⁄ected by the (cid:133)nancial crisis. Althoughdataareavailableforsome(cid:133)rmspriorto1991,thecoverageisgenerallyquitethinpriorto 1991 so we limit our sample to the period from 1991 to 2005. The number of (cid:133)rms available in the Worldscope dataset grows substantially over the course of the sample period, which suggests that changes in sample composition may be an important issue. For this reason, we focus exclusively on regression estimates in the within dimension and control for identi(cid:133)able (cid:133)rm characteristics in order to limit the e⁄ect on our estimates of changing sample composition. We construct the following variables: The investment rate (I =K ), Tobin(cid:146)s Q (Q ), the it it it rate of cash (cid:135)ow (CF =K ), the (cid:135)ow of external (cid:133)nancing (XF =K ), the debt-to-equity ratio it it it it (Debt =Equity ), debt-to-capital ratio (Debt =K ), and the ratio of short-term cash assets to it it it it capital (Cash =K ). In addition, in order to control for the e⁄ect of (cid:133)rm size on behavior, we it it constructed a binary variable for each (cid:133)rm in each year using quartiles of their capital replacement values: Firms with capital holdings in the lowest quartile in any given year were de(cid:133)ned as small (Dsm = 1), while (cid:133)rms with holdings in the largest quartile are de(cid:133)ned as large (Dlg = 1). To it it remove the e⁄ect of outliers, we drop observations for any variable that are in the extreme tails (below the 1=4 percentile and above the 99 3=4 percentile) of the their cross-sectional distribution (cid:0) in any given year. Appendix B provides a detailed description of the Worldscope variables used and the data construction process. 4 Corporate Investment Before and After the Crisis Theguidingprincipleofouranalysisisthataggregatevariationsininvestmentcanonlybeexplained by changes in fundamentals that cut across all (cid:133)rms. For this reason, we begin by looking at the systematiccomponentsofinvestmentandfundamentalsbeforedivingintoamoredetailedanalysis. Yearlyvariationsininvestmentandeachrelevantfundamentalcanbedecomposedinto(cid:133)xede⁄ects, aggregate e⁄ects, group e⁄ects for small and large (cid:133)rms, and idiosyncratic components: 6

Iit f I=K b I=K b I=K Dsm a I=K e I=K Kit = i + S L it + t + it ; (1) " x it # " f i x # " bx S bx L #" D i lg t # " ax t # " ex it # where x is a vector of relevant investment fundamentals for i. This equation decomposes it I=K I=K the investment rate for a given (cid:133)rm i into three broad components: f the (cid:133)xed e⁄ect, a i t the aggregate component at t = 1991;:::;2005, and e I=K the idiosyncratic component; fx, ax and it i t ex are the corresponding components of the vector x .5 Dummies for small and large (cid:133)rms are it it included to control for group e⁄ects related to the relative size of the (cid:133)rm, which many studies have shown to a⁄ect investment behavior.6 We use panel regressions to estimate decompositions in (1). For each variable we normalize the year e⁄ect to be zero in our base year of 1996(cid:150)the year that immediately preceded the crisis.7 The estimated year e⁄ects in all other years capture the total e⁄ect of latent aggregate factors relative to their e⁄ect in this base year. Figure 4 shows the time path of our estimated year e⁄ects for the investment rate, along with aggregate e⁄ects for three commonly cited fundamentals: Tobin(cid:146)s Q, internal cash (cid:135)ow, and the rate of return on assets (ROA). The estimated time path for the (cid:133)rm-level investment rate(cid:150)depicted in the top left panel of Figure 4(cid:150)shows a distinct pattern that closelyresemblesthetrajectoryofaggregateinvestmentshowninFigure2.Atthetimeofthecrisis, the investment rate fell noticeably and then remained persistently low through the remainder of our sample period. The postcrisis investment rate is about 12 percentage points below the precrisis average, though the drag appears to attenuate late in the sample. The aggregate component of Q follows a pattern over our sample that broadly resembles that of investment, including a large decline during the crisis years and little sign of postcrisis recovery. This hints that less-favorable investment prospects may have played some role in the postcrisis investment slump. Return on assets(ROA)andcash(cid:135)owdeterioratedconsistentlyintheyearspriortocrisisthrough1998. Since then, both ROA and cash (cid:135)ow have improved steadily, peaking in 2004 at levels only somewhat below their precrisis norms. In Figure 5, we plot the time path of estimated year e⁄ects for external funding (cid:135)ows and for 5The (cid:133)rm-level (cid:133)xed e⁄ect also controls for a number of e⁄ects that cannot be separately identi(cid:133)ed, including (cid:133)xed (cid:133)rm characteistics, country e⁄ects, industry e⁄ects, and an aggregate e⁄ect for our baseline year of 1996. 6These controls are warranted, even though we include (cid:133)xed e⁄ects, because the dataset is su¢ ciently long that the relative size of the incumbent (cid:133)rms in our sample tends to increase over the course of the sample period. 7By including controls for (cid:133)rm size, we remove from the aggregate component the portion attributable to shifts in(cid:133)rmsbetweensizecategoriesfrom yeartoyear. Thisiswarrantedbecauseourtimeseriesislongenoughthatthe size of a given (cid:133)rm could change substantially within the sample. 7

some other selected (cid:133)nancial indicators that may in(cid:135)uence (cid:133)rms(cid:146)access to external funding: the debt-to-equity ratio, and ratios of debt obligations and cash holdings to capital. These indicators suggest that, on balance, (cid:133)rms relied extensively on external (cid:133)nancing in the leadup to the crisis, which resulted in a substantial buildup of debt relative to capital and equity on the eve of the turmoil even as pro(cid:133)tability declined and cash holdings deteriorated. After the crisis, external (cid:133)nancing dropped signi(cid:133)cantly and debt levels moved down steadily. This preliminary analysis suggests that the marked downturn in investment at the time of the crisis coincided with a broad deterioration in (cid:133)rms(cid:146)investment fundamentals and (cid:133)nancial health. But since the crisis, debt levels have gradually fallen and most fundamentals have shown signs of recovery that have not yet fed through to investment spending. One notable exception to this pattern is Tobin(cid:146)s Q, which has shown little signs of improvement. Theory suggests that Tobin(cid:146)s Q should, under ideal conditions, summarize all information that is pertinent for the current rate of investment. Taken at face value, the lack of meaningful improvement in Tobin(cid:146)s Q provides some rationale for the drop in investment over the postcrisis period. We explore more formally whether investment fundamentals are behind the drop in the postcrisis investment in the next section. Theintensityofthepostcrisisinvestmentdragappearstobesimilaracrossindustries,countries, and (cid:133)rm sizes. Figure 6 considers the country dimension, showing estimated year e⁄ects from our panel of (cid:133)rms for each of the Asia-8 countries. We obtain these estimates using a regression of the form shown in Equation (1) but with the year dummies interacted with separate dummy variables for the eight countries in our panel. The bottom panel of the (cid:133)gure shows a somewhat simpler cut of the data. We replace the full set of year dummies in our panel regression with each country dummy interacted with a "postcrisis" dummy D97+ that is set to one from 1997 t onward. Results using this speci(cid:133)cation indicate that all the countries in our sample experienced a postcrisis investment drag, though the e⁄ect appears to be less pronounced for Hong Kong and Taiwan. Figure 7 repeats the same exercise, but with separate postcrisis dummies for nine broad industry categories, where industries are categorized according to the (cid:133)rst digit of the (cid:133)rm(cid:146)s SIC (Standard Industrial Classi(cid:133)cation) code. There do not appear to be signi(cid:133)cant di⁄erences in the postcrisis investment drag across industries. Though estimated postcrisis e⁄ect for two of the industries(cid:150)industry 0 (Agriculture, Forestry and Fishing) and industry 6 (Finance, Insurance and Real Estate)(cid:151)are not distinguishable from zero at standard signi(cid:133)cance levels, formal tests (not shown) cannot distinguish the magnitude of the postcrisis e⁄ect across these nine industries. 8

Figure 8 shows the results from a similar exercise using dummies for (cid:133)rm size. Firms in all three size categories have experienced postcrisis investment declines, and the e⁄ect appears to have been stronger for smaller (cid:133)rms. In sum, the shortfall in region(cid:146)s aggregate investment cuts across a wide spectrum of countries, industries, and (cid:133)rms. 5 Determinants of Drag on Postcrisis Corporate Investment Our econometric speci(cid:133)cation is motivated by a standard value maximization problem for a competitive (cid:133)rm that faces adjustment costs for capital (see Appendix A for details). The speci(cid:133)cation relates the (cid:133)rm(cid:146)s rate of investment Iit to its current value of Q, which is a valid summary of Kit relevant investment fundamentals for(cid:16)a (cid:133)r(cid:17)m that faces no (cid:133)nancial constraints or costs that limit its ability to raise funding for investment and is small enough that it treats all prices as given.8 We augment this speci(cid:133)cation by including the same set of controls used in equation (1), along with additional (cid:133)rm-speci(cid:133)c variables intended to capture the e⁄ect of restricted access to outside funding and other factors. Though these additional variables shouldn(cid:146)t matter under the idealized conditions set out above, empirical studies using (cid:133)rm-level data provide ample reason to believe that internal cash (cid:135)ows, non-price credit rationing, capital structure, and other factors in(cid:135)uence investment even after controlling for Q.9 For now, rather than estimating a full set of year e⁄ects to capture unexplained aggregate variation, we restrict the time pattern of these e⁄ects somewhat by simply including a single dummy variable for the crisis and postcrisis period (D97+). We also t interact this postcrisis e⁄ect with some of the (cid:133)rm characteristics identi(cid:133)ed above in order to allow the postcrisis e⁄ect to vary for (cid:133)rms with di⁄erent selected characteristics: I K it = f i +b S D i s t m+b L D i lg t +a p D t 97++c0x it +aS p D i s t mD t 97+ +aL p D i lg t D t 97+ (2) it (cid:16) (cid:17) N (cid:0) (cid:1) +aHST DHSTD97+ + an DnD97+ +e ; p i t p i t it n=1; (cid:0) (cid:1) n X =n (cid:0) (cid:1) 6 8It also assumes that the homogeneity conditions described by Hayashi [1982] hold, so the measure Tobin(cid:146)s Q is a good proxy variable for the shadow value of capital. 9Prominent among numerous examples are Summers [1981], Gilchrist and Himmelberg [1995], and Erickson and Whited [2000]. 9

where DHST is a dummy that indicates whether (cid:133)rm i is located in Hong Kong, Singapore or Taii wan, and Dn is a dummy variable that controls for all industries n = 1;:::;N, with the exception i of our baseline industry n.10 For practical reasons, we chose our baseline industry to be manufacturing, mainly because this category accounts for about three-(cid:133)fths of the (cid:133)rms in our sample.11 Given this speci(cid:133)cation, the coe¢ cient a on the postcrisis dummy D97+ can be interpreted as the p t unexplained aggregate component of investment over the postcrisis period for our baseline (cid:133)rm: A medium-sized manufacturing entity located in Indonesia, Malaysia, South Korea, Thailand, or the Philippines. The coe¢ cients on the various interaction terms show the incremental e⁄ect on investment from latent aggregate factors for (cid:133)rms with that speci(cid:133)c attribute. For instance, the coe¢ cient aS on DsmD97+ represents the additional postcrisis e⁄ect on investment for small p it t (cid:133)rms, over and ab(cid:0)ove the bas(cid:1)eline e⁄ect for medium-sized (cid:133)rms. Results using variations of this speci(cid:133)cation are shown in Table 3. As a basis of comparison for subsequent estimates, the (cid:133)rst column of the table shows results with no controls other than the postcrisisdummyandsizee⁄ects, whiletheestimatesshowninthesecondcolumnincludeallofthe controls in equation (2) except the fundamentals x . According to these estimates, the investment it rate for our baseline (cid:133)rm declined by about 12 percentage points in the postcrisis period, with a very narrow con(cid:133)dence interval. This postcrisis drag did not di⁄er in a statistically meaningful way for small and large (cid:133)rms, but was about 71 percentage points less intense (but signi(cid:133)cant 2 nonetheless) for the countries in our sample that appear to have been less a⁄ected by the crisis (Taiwan, Hong Kong, and Singapore).12 Inthenextstepsoftheanalysis,weincludeothervariablesthatcouldberelevantforinvestment in order to assess how much of the 12 percentage point postcrisis decline in the investment rate can be explained by fundamentals. As mentioned earlier, a voluminous empirical literature suggests that Q does not always summarize all the factors that are relevant for determining investment in practice. For this reason, we view Q as an imperfect proxy of a (cid:133)rm(cid:146)s perceived investment prospects, and estimate alternative speci(cid:133)cations that include additional controls for other funda- 10We include controls for the postcrisis e⁄ect in Hong Kong, Singapore and Taiwan because we felt the Asian (cid:133)nancial crisis had less of an e⁄ect on these countries, given a priori considerations. The results shown in Figure 6 suggest that South Korea might also be included in this group. Our results are not sensitive to this change in speci(cid:133)cation. 11More speci(cid:133)cally, we include in our baseline all (cid:133)rms whose SIC code has a (cid:133)rst digit of either 2 or 3. These industry controls have almost no e⁄ect on our results. 12Whenoneaddsthepostcrisise⁄ecttotheincrementale⁄ectforHTScountries,thecombinede⁄ectis41 percent, 2 with a standard error of about 0.1. 10

mentals such as internal cash (cid:135)ows, external funding, and the debt-to-equity ratio.13 Results using these speci(cid:133)cations are presented in columns (2) through (5). When we simply include Tobin(cid:146)s Q in the regression as a proxy for time-series variation in investment opportunities, this reduces the contribution in the postcrisis period of latent aggregate factors to about 91 percentage points; the 4 other variables explain even less.14 To allow for the possibility that independent variations in these fundamentals may collectively explain the postcrisis investment drag, columns (6) through (8) include combinations of these variables. Column (6) shows results from a speci(cid:133)cation that includes both Tobin(cid:146)s Q and internal cash(cid:135)owasmeasuresoffundamentals. Bothofthesevariablesentersigni(cid:133)cantlyintotheregression with high signi(cid:133)cance. Even so, the postcrisis drag is trimmed to about 81 percent(cid:151)down just 31 2 2 percentage points from the baseline speci(cid:133)cation(cid:151)and it remains highly signi(cid:133)cant. Column (7) includes the same variables in the third column, along with the debt-to-equity ratio, while column (8) adds both the debt-to-equity ratio and external funds. The estimates show that the debt-toequity ratio adds almost no additional explanatory power for the aggregate e⁄ect (or investment in general) over and above Q and cash (cid:135)ow. And, though we have strong reservations about whether the external funding (cid:135)ow can be plausibly regarded as exogenous, including this information only seems to explain another percentage point of the aggregate e⁄ect, leaving a still-substantial 71 2 percentage points unexplained. The last column of the table considers whether the sensitivity of investment to any of these fundamentals has changed over the postcrisis period, which might give insights about the nature of the latent aggregate factors that appear to have held back capital spending. These estimates suggest no statistically signi(cid:133)cant interaction between the postcrisis aggregate e⁄ect and Tobin(cid:146)s Q or the debt-to-equity ratio. For external (cid:133)nancing, the interaction is negative and statistically signi(cid:133)cant, suggesting (subject to our caveat about endogeneity) that the e⁄ect of the aggregate shock has been more intense for (cid:133)rms that were more reliant on external (cid:133)nancing prior to the crisis. In any case, although these interactions are intriguing, they explain little or none of the postcrisis e⁄ect. Even the most favorable speci(cid:133)cation shown in column (9) leaves unexplained 13Weincludethedebttoequityratiobecausethe(cid:133)rm(cid:146)srequiredrateofreturnoncapital(cid:150)animportantinvestment determinant(cid:150)is generally a function of its debt-to-equity ratio, except under the special conditions described by Modigliani and Miller [1958]. 14Though not a focus in this context, the estimated coe¢ cient on Q (0.023) is similar in magnitude compared to results from previous empirical work. The coe¢ cient on internal cash (cid:135)ow is also in line with estimates in other studies. 11

about one-half of the postcrisis drag on aggregate investment. Table 4 examines whether the unexplained portion of the aggregate drag on investment over the postcrisis period has been accentuated by cross-sectional di⁄erences in (cid:133)rms(cid:146)balance sheets on the eve of the crisis. For this purpose, we interact our postcrisis dummy with readings for selected (cid:133)rm-speci(cid:133)cvariablesin1996(cid:151)theyearthatimmediatelyprecededthecrisis. Weconsiderprecrisis values of Tobin(cid:146)s Q, external (cid:133)nancing, cash (cid:135)ow, debt-to-equity, and debt-to-capital. Columns (1) through (4) indicate that variations across (cid:133)rms(cid:146)Tobin(cid:146)s Q, external (cid:133)nancing, cash (cid:135)ow, and the debt-to-equity ratio on the eve of the crisis do not help explain the postcrisis drag. However, when weconditionthepostcrisise⁄ectontheincludethecontrolforthedebt-to-capitalratioattheeveof the crisis, we (cid:133)nd(cid:151)quite stunningly(cid:151)that this interaction term essentially explains the remainder ofthedropinthepostcrisisinvestmentrate. Forthiscase, thepostcrisisinvestmentdragisreduced to about 2 percent, but it is not statistically signi(cid:133)cant from zero. This suggests that the precrisis debt level a⁄ected the magnitude of the postcrisis investment drag above and beyond what can be justi(cid:133)ed by (cid:133)rms(cid:146)investment fundamentals. Figure9showsestimatesofhoweachofthesefourfactors(listthefourfactorshere)contributed tothetotaldropincapitalspendingovertime. Theseresultsareobtainedbyestimatingyeare⁄ects after controlling for various factors. The bottom solid line shows the entire set of aggregate e⁄ects (relative to their 1996 value) without conditioning for any fundamentals. The area between each chart and the bottom solid line captures the portion of the postcrisis investment drag that is explained away by each of the variables shown. For example, controlling for Tobin(cid:146)s Q, reduces the postcrisis investment drag by about 3 percentage points on average. Cash (cid:135)ow and external (cid:133)nancing, takentogether, explainanevensmallerfractionofthepostcrisisinvestmentdrag-around 1 percentage point. However, when we allow the year e⁄ects to interact with the eve-of-the-crisis debt-to-capital ratio, the postcrisis investment drag is almost entirely accounted for. Taken at face value, these estimates suggest that while poorer fundamentals contributed to the drop in capital spending after the crisis(cid:150)cross-sectional, variations in the debt-to-capital ratio on the eve of the crisis appear to be the single most important factor behind the postcrisis investment drag. According to the estimates in this (cid:133)gure, this debt hangover e⁄ect has been attenuating in recent years. In the following section, we further explore this (cid:133)nding in more detail. 12

6 Precrisis Excessive Debt and Investment, and postcrisis Investment Drag To further understand the apparent debt hangover e⁄ect identi(cid:133)ed in the previous section, we plot in Figures 10 and 11 some characteristics of (cid:133)rms grouped by their debt-to-capital distribution in 1996 (top quartile, mid quartiles, and bottom quartile). Firms in the top quartile of the 1996 debt-to-capital distribution accumulated sizeable debt obligations in the years leading up to the crisis. This is consistent with the substantial amount of capital that (cid:135)owed into the region over this period, which(cid:151)it is widely believed(cid:150)re(cid:135)ected the abundant credit availability for many (cid:133)rms in this region.15 In particular, the bulk of this debt build up appears to have been concentrated in (cid:133)rms in the top quartile, whose debt levels exceeded both their total capital holding and the value of their equity. Indeed, for the rest of the (cid:133)rms in our sample, debt levels remained relatively steady throughout the sample period. This debt accumulation by these high-debt (cid:133)rms appears to have gone hand in hand with very high levels of capital spending: investment rates for these (cid:133)rms rose to roughly 20 percent in 1996. This suggests that the debt-to-equity e⁄ect identi(cid:133)ed in the previous section might also be described as a capital overhang. To assess this conjecture, for excess precrisis investment by constructing a binary dummy variable that takes a value of one for (cid:133)rms whose average investment rate in 1995 and 1996 was at least two standard deviations above the yearly cross-sectional mean. Interestingly, when we reestimate our regression with the debtoverhang variable replaced with an interaction between this crude proxy for overinvestment and the postcrisis dummy, the postcrisis investment drag also becomes insigni(cid:133)cant (Column(6)). This suggests that these two variables largely capture the same e⁄ect, which supporting the argument that(cid:133)rmsborrowedheavilytoinvestexcessively.16 Onbalance, thishighinvestmentoccurredamid a backdrop of weakening fundamentals. For example, (cid:133)rms in the high-debt group had lower values of Tobin(cid:146)s Q prior to the crisis than other (cid:133)rms, and these values were dropping rapidly in 1995 and 1996. At the eve of the crisis, returns-on-assets and cash (cid:135)ow had also declined sharply for the median (cid:133)rm in this high-debt group. As indicated earlier, investment declined the most during the crisis and the postcrisis drag was 15See for example Ito [1999] or Calvo and Reinhart [1999] for extended discussions. 16The result is preserved when we control for the interaction betweeen the postcrisis dummy and Debti96, and the Ki96 coe¢ cientsforbothOverInvestmentand Debti96 arebothstatisticallysigni(cid:133)cant. Theresultisalsopreservedwhen Ki96 we control for the interaction between the postcrisis dummy and Q XF96, and CFi96 interacted with the postcrisis 96 Ki96 Ki96 dummy. 13

more pronounced for the (cid:133)rms in the top quartile of the 1996 debt-to-capital ratio distribution. Tobin(cid:146)s Q has improved somewhat since the crisis, notably for (cid:133)rms in the lower-debt group, but less so for (cid:133)rms in the high-debt category. Nonetheless, Tobin(cid:146)s Q remains below the standard benchmark of "1", above which, theory suggests that (cid:133)rms should resume investment spending. Pro(cid:133)tability and cash (cid:135)ow have improved signi(cid:133)cantly for all (cid:133)rms including those in the high-debt group. External (cid:133)nancing has also declined substantially for this group, turning negative since 1998, suggesting that they have been using generated cash (cid:135)ows to repay debt accumulated during the runup to the crisis. Consistent with this observation, debt-to-capital and debt-to-equity ratios have fallen signi(cid:133)cantly to levels comparable to those of other (cid:133)rms. For (cid:133)rms with lower debt, part of the cash (cid:135)ow has been allocated to dividend payments, and investment has improved a bit for these (cid:133)rms in the recent years. All told, the evidence suggests the postcrisis investment drag is indeed a direct e⁄ect of the (cid:133)nancial crisis, rather than a consequence of deliberate government policies to boost the current account. Since the crisis, however, investment rates have remained low despite signi(cid:133)cant improvements in fundamentals such as pro(cid:133)tability and cash (cid:135)ow, consistent with a scenario where excess capacity inherited from the precrisis period reduced the need for additional investment spending in the postcrisis period as indicated by lower values of Tobin(cid:146)s Q. The (cid:133)nancial crisis e⁄ect is most pronounced for high-debt (cid:133)rms that, taken together, accounted for about 25 percent of total investment in 1996. These (cid:133)rms apparently used high levels of debt (cid:133)nancing to maintain excessive levels of capital spending during the runup to the crisis. In theory, under ideal conditions, Tobin(cid:146)s Q should summarize all information that is relevant for a (cid:133)rm(cid:146)s investment, including the e⁄ect of large debts or excess capacity of capital. Under this ideal scenario, the level of debt prior to the crisis should not o⁄er additional information that is relevant for the postcrisis investment behavior. Many empirical studies have documented that Tobin(cid:146)sQfallswellshortofthisstandardinpractice. Weinterpretourresultasre(cid:135)ectingviolations oftheassumptionsthatsupporttheQ-theoryofinvestment. Indeed, thedebtcana⁄ectinvestment by raising, for example, the agency cost of external (cid:133)nancing. Whited [1992] (cid:133)nds that including the e⁄ect of a debt constraint in a standard (cid:133)rm investment model greatly improves the model(cid:146)s performance, suggesting an important role for debt levels in investment behavior.17 The drop in postcrisis investment that we document in this study could be rationalized along 17See also Myers [1977] or Myers and Majluf (1984) for additional discussion. 14

two dimensions. In the (cid:133)rst scenario, the excessive debt accumulated by many (cid:133)rms prior to the crisis raised the perceived riskiness associated with providing capital to these (cid:133)rms, boosting their postcrisiscostofcapitalandtherebypushingdowntheirinvestment. Inthesecondscenario,current investmentlevels,thoughlower,areconsistentwithdesiredlevelsofinvestmentbythese(cid:133)rmsgiven their perceived cost of capital. In this view, (cid:133)rms have been e¢ ciently allocating their (cid:133)nancial resources over the postcrisis period to pay down debts, pay out dividends, or to accumulate liquid assets that can be used to fund investment when solid prospects arise. Due to data limitations, we are unable to analyze whether the drop in capital spending since the crisis has been associated with higher costs of external funding. Instead, we test whether the level of debt in 1996 a⁄ected (cid:133)rms(cid:146)postcrisis investment response to changes in the cash (cid:135)ow. The rationale behind this test is that (cid:133)rms that have more access to internal funds should be less a⁄ected by funding limitations imposed from external sources. As such, the postcrisis investment dragforthese(cid:133)rmsshouldbelessintensefor(cid:133)rmswithhighercash(cid:135)owsand/ormoreampleliquid asset holdings.18 Similarly, investment should be more sensitive to cash (cid:135)ow for (cid:133)rms that have accumulated more debt. We test both of these conjectures and report the results in columns (7) and (8). For both of these cases, the results show no evidence that investment in the aftermath of the crisiswasmoresensitivetocash(cid:135)ow,evenwhenwerestrictourregressionsampleto(cid:133)rmsinthetop quartile of the 1996 debt distribution-column (8). In column (9), we test the second conjecture by interacting three variables: the postcrisis dummy, cash (cid:135)ow, and the 1996 debt-to-capital variable. This interaction term captures di⁄erences in the sensitivity of the postcrisis investment to cash (cid:135)ow for (cid:133)rms with greater debt holdings on the eve of the crisis. If high debt inherited from the crisis is restraining investment by raising the cost of external funds, we would expect the postcrisis investment by (cid:133)rms with higher precrisis debt holdings to be more sensitive to cash (cid:135)ow than (cid:133)rms with lower precrisis debt. The coe¢ cient on this interaction term is negative and statistically signi(cid:133)cant. At best, the result suggests that the postcrisis investment was less (not more) sensitive to cash (cid:135)ow for (cid:133)rms with high precrisis debt levels. In sum, the evidence suggests that higher costs of external funding were not behind the drop in capital expenditures since crisis. Instead, it appears that investment prospects were not strong enough to encourage higher investment spending beyond the capacity inherited from the precrisis 18For examples, see Fazzari, Hubbard and Petersen [1988b], Gilchrist and Himmelberg [1995], and others. 15

period. As a result, (cid:133)rms allocated internal funds to accumulate liquid assets, make dividend payments and, in the case of high-debt (cid:133)rms, repay debt. In the next section, we review possible implications for the future adjustment of the current accounts balances. 7 Discussion and Implications for Adjustment of Current Accounts Though it seems hard to believe that the adverse e⁄ects of the Asian (cid:133)nancial crisis can still a⁄ect investment decisions after a decade, the results from this study indicate that the e⁄ect of the 1997 (cid:133)nancial crisis continues to be a drag on (cid:133)rm investment. This drag in investment has played a big role in generating the current account surpluses in emerging Asia, and has enabled the region to export capital as the anemic domestic investment is unable to absorb the region(cid:146)s savings. The most important factor weighing on investment is the e⁄ect of excessive debt and excess investment that occurred in the years leading up to the crisis. As restructuring completes its course and excess capacity wanes, it is conceivable that corporate investment will continue to improve. We believe, however, that investment rates would likely not rise back up to the highs seen in the period leading up to the crisis. Since roughly one-third of the postcrisis investment drag can be attributed to fundamentals, it seems reasonable to expect investment rates to recoup only one-third of their declines after the e⁄ect of the crisis dissipates. For aggregate investment rates, this would imply increases in the vicinity of 3 percentage points, contributing to an equivalent reduction in current account surpluses all else equal. Theresultsfromthisstudyhaveimplicationsthatpresentchallengesforsomealternativeexplanations for emerging Asia(cid:146)s current account surpluses. Recent papers by Dooley, Folkerts(cid:150)Landau and Garber argue that the current account surpluses in emerging market economies are the result of deliberate government policies to promote export led-growth. According to this explanation, investment from domestic sources is ine¢ cient for fostering growth and emerging market economies require foreign direct investment to reach their development objectives. This argument suggests that current account surpluses are necessary to accumulate assets that are in turn invested in developed economies as potential collateral to induce inward foreign direct investment (FDI) from those developed countries (see Dooley, Folkerts-Landau and Garber, 2003, 2004). A similar line of argument by Mann [2004] suggests that the current account surpluses in Asia are the result of a 16

shiftinpolicytopromoteexport-ledgrowthafterdomestic-ledgrowthintheearly1990sresultedin the Asian (cid:133)nancial crisis. Yet another popular explanation attributes emerging Asia(cid:146)s surpluses to underdeveloped (cid:133)nancial systems that are unable to intermediate domestic saving (Prasad, Rajan, and Subramanian [2006]) or to ful(cid:133)ll local residents(cid:146)needs for high quality foreign (cid:133)nancial assets (Mendoza, Quadrini, and Rios-Rull [2007]). An implication for these various explanations is that Asia(cid:146)s current account surpluses would persist for the foreseeable future. While the results from this study do not directly refute these explanations,theyrationalizethecurrentaccountsurplusesinawaythatsuggeststhatthesurpluses need not persist. The study documents the predominant e⁄ect of the role of the (cid:133)nancial crisis on current account surpluses through its adverse e⁄ect on (cid:133)rms(cid:146)balance sheets. As (cid:133)rms complete the restructuring of their balance sheets and prospects strengthen, investment could rise to reduce the current account surpluses. However, it remains unlikely that investment rates would rise to the highsseenintheyearsleadinguptothecrisisastheseratesseemtohavebeenfueledbyinvestment beyondlevelssupportedbyfundamentals. Eliminatingtheregion(cid:146)scurrentaccountsurpluseswould require reductions in the region(cid:146)s savings rates which are among the highest in the world. 8 Conclusion This study reviewed the role of emerging Asia(cid:146)s current account surpluses in global imbalances and assessed the unique role of the (cid:133)nancial crisis using a cross-country data set of 3,750 (cid:133)rms. The results indicated that the current account surpluses in the region are a direct result of the e⁄ect of the (cid:133)nancial crisis. In the aftermath of the crisis, corporate expenditures on (cid:133)xed investment declined signi(cid:133)cantly, contributing to lower aggregate investment rates. The shortfall in corporate spending generated investment that fell short of saving, turning the region into a net exporter of capital. We then considered the factors behind the postcrisis lower investment rates. Our analysis indicated that weaker postcrisis fundamentals account for a portion of the lower investment rates, but ongoing re-structuring owing to large debts accumulated and the excess investment undertaken in the period leading up to the crisis are the main factors weighing on the postcrisis corporate investment. These results support the hypothesis that the region(cid:146)s current account surpluses are a direct result of private restructuring behavior in response to the (cid:133)nancial crisis. As this restructuring completes it course, investment rates will likely rise to reduce the 17

region(cid:146)s current account surpluses, contrary to alternative explanations for Asia(cid:146)s current account surpluses that imply that they will persist for the foreseeable future. We do not, however, expect investment rates to rise back up to the unsustainable levels that prevailed at the eve of the crisis. A full adjustment of the surpluses would require reductions in the region(cid:146)s saving rates. Future research on the determinants and prospect of the region(cid:146)s saving rates is well indicated. 18

Appendix A A Neoclassical Model of Corporate Investment Our theoretical framework for (cid:133)rm investment is motivated by a fairly standard neoclassical qtheory of investment where (cid:133)rms face adjustment costs for adjusting their capital stock. Brie(cid:135)y summarizing this framework, we assume that markets are perfectly competitive, that all market participants share the same costless information, and that (cid:133)rms face no internal adjustment costs other than those for capital. The neoclassical (cid:133)rm chooses an investment rate that maximizes the market value of its future cash (cid:135)ows from capital, which is represented by the value function (see also Hubbard [1998]): V (K ;(cid:23) ;" ) = max (cid:5)(K ;(cid:23) ) p [I +(cid:8)(I ;K ;" )]+(cid:26) E [V (K ;(cid:23) ;" )] i;t i;t it i;t i;t t i;t i;t i;t it i t i;t+1 i;t+1 it+1 Iit f (cid:0) g (3) where K is given by the following capital accumulation condition: i;t+1 K = (1 (cid:14) )K +I : (4) i;t+1 i i;t i;t (cid:0) In this formulation, i and t denote the (cid:133)rm and time period respectively, and (cid:26) is the relevant discount factor for future cash (cid:135)ows. (cid:5)( ) is the (cid:133)rm(cid:146)s (gross) pro(cid:133)t function, which, after opti- (cid:1) mizing out variable production factors, is a function of its current capital stock K and a random i;t variable (cid:23) that captures changes in productivity and/or the market price of variable inputs. The i;t (cid:133)rm treats (cid:23) as given. (cid:8)( ) is a function that captures internal capital adjustment costs, I i;t i;t (cid:1) investment, p is the relative price of capital goods net of the capitalized value of future tax shields. t The random variable " is an adjustment cost shock that is observed by the (cid:133)rm but not by econoit metricians,and(cid:14) istherateofdepreciationofcapitalfor(cid:133)rmi. E [ ]isanexpectationconditional i t (cid:1) on information available at time t. The (cid:133)rst-order condition for value maximization provides the following familiar investment equation: @(cid:8)(I ;K ;" ) i;t i;t it 1+ = q ; (5) i;t @I i:t where q is marginal q: The shadow value to the (cid:133)rm of an incremental unit of capital in the i;t 19

following period, reckoned in terms of capital. In turn, this shadow value is the present value of anticipated cash (cid:135)ows that the (cid:133)rm expects from a marginal increase in next period(cid:146)s capital stock, in units of capital: q (cid:26) i E @V (K i;t+1 ;(cid:23) i;t+1 ) = (cid:26) i 1 (cid:26)s(1 (cid:14) )sE @(cid:5)(K i;:t+s ;(cid:23) i;t+s ) @(cid:8)(I i;t+s ;K i;t+s ) : i;t (cid:17) p t @K p i (cid:0) i t @K (cid:0) @K it i;t+1 it i:t+s i:t+s (cid:20) (cid:21) s=1 (cid:20) (cid:21) X Equation (5) shows that (cid:151) given the form of the adjustment cost function, its capital stock, and the adjustment cost shock " (cid:151) marginal q is su¢ cient to determine the (cid:133)rm(cid:146)s current investment it (cid:135)ow. To obtain an econometric model, we assume that the adjustment cost function takes the following quadratic form: ’ I 2 it (cid:8)(I ;K ;" ) = " K (6) i;t i;t it it it 2 K (cid:0) it (cid:20) (cid:21) which is linearly homogeneous in capital and investment. Substituting equation (6) into equation (5) provides the following structural equation: I it = ’ 1+’ 1q +" : (7) (cid:0) (cid:0) i;t i;t K (cid:0) it Assuming that, as in Hayashi [1982], the (cid:133)rm is a price taker in all markets, its pro(cid:133)t function is linear in capital (which requires that the production function be linearly homogenous in all inputs), and (cid:133)nancing and investment decisions are independent, the shadow value of capital in equation (7) can be replaced with the average value of capital Q , where the value of the (cid:133)rm is measured it excluding the current dividend. This yields the following reduced-form speci(cid:133)cation: I it = f +a +bQ +" (8) i t i;t it K it where we have assumed that each (cid:133)rm(cid:146)s adjustment cost shock " is composed of three separate i;t components: A (cid:133)rm-level (cid:133)xed e⁄ect f , a latent e⁄ect a that is common to all (cid:133)rms, and an i t idiosyncratic e⁄ect " that varies randomly over time. it 20

B Detailed Data Description For each (cid:133)rm in our panel, annual values of each variable are determined as follows: Thereplacement value of capital (p K ) foreach(cid:133)rminagivenyearis determinedbytaking it it (cid:15) the (cid:133)rm(cid:146)s total asset value less the value of its current assets, where variables are as recorded by Worldscope at the end of the preceding year. For a few (cid:133)rms there were gaps in the book value data from Worldscope. In these cases, we (cid:133)lled in these missing data for the nominal capital stock in these years by assuming that the real stock grew at a constant rate su¢ cient to reconcile the available capital stocks at the beginning and endpoints of the gap. In the process of making this calculation, we converted nominal capital stocks to real (and vice versa) using yearly values of the aggregate investment de(cid:135)ator for the country where the (cid:133)rm was located. The market value of capital (p V ) in a given year is the sum of the market value of the it it (cid:15) (cid:133)rms(cid:146)equity (share price times the number of common shares outstanding) plus the book valueofitsdebtminusthebookvalueof itscurrentassets, asrecordedinWorldscopebalance sheet information for the end of the preceding year. Investment rates Iit for each (cid:133)rm in each year are determined by taking from the (cid:133)rm(cid:146)s (cid:15) Kit Worldscope cash (cid:135)(cid:16)ow (cid:17)statement its uses of cash to acquire (cid:133)xed assets, netting out sources of cash from sales of property, plant and equipment, and then dividing this net total by the replacement value of the (cid:133)rm(cid:146)s capital. Cash (cid:135)ow CashFlowit isthe(cid:133)rm(cid:146)scash(cid:135)owfromoperations(asrecordedintheWorldscope (cid:15) Kit cash (cid:135)ow s(cid:16)tatement) d(cid:17)ivided by the replacement value of its capital. Tobin(cid:146)s Q (Q ) is the total market value of the (cid:133)rm(cid:146)s capital divided by the replacement it (cid:15) value of the (cid:133)rm(cid:146)s capital. Debt-to-equity ratio Debtit of a (cid:133)rm is the current value of its equity divided by the book (cid:15) Equityit value of its debt, wh(cid:16)ere the c(cid:17)alculation of both variables are as described above. External(cid:133)nancing ExtFinit iscalculatedbydividingthe(cid:133)rm(cid:146)s(cid:135)owoffundingfromexternal (cid:15) Kit sources (as recorded(cid:16)in the W(cid:17)orldscope cash (cid:135)ow statement) by the replacement value of the 21

(cid:133)rm(cid:146)s capital. Debt-to-capital ratio Debtit of a (cid:133)rm is the book value of the (cid:133)rm(cid:146)s debt in the current year (cid:15) Kit divided by the replac(cid:16)ement(cid:17)value of its capital. Cash-to-capital ratio Cashit is the value of the (cid:133)rm(cid:146)s cash investments in the current year (cid:15) Kit (as recorded in World(cid:16)scope b(cid:17)alance sheet information) divided by the replacement value of its capital. Return on assets is the Worldscope estimate for the current year, calculated as current net (cid:15) income before preferred dividends plus current after tax interest expenses, all divided by the total book value of assets in the previous year. 22

References Blanchard, Olivier, Giavazzi, Francesco and Sa, Filipa (2005), "The U.S. Current Account and the Dollar." NBER Working Paper 11137. Bernanke, Ben S. (2005), "The Global Saving Glut and the U.S. Current Account De(cid:133)cit," Homer Jones Lecture, St. Louis, Missouri, April 14. Bordo, Michael D. (2005), "Historical Perspective on Global Imbalances," NBER Working Paper #11383. Calvo, Guillermo A. and Carmen M. Reinhart (1999) "When Capital In(cid:135)ows Come to a Sudden Stop: Consequences and Policy Options" in Emerging Capital Markets in Turmoil: Bad Luck or Bad Policy?, Cambridge, MA: MIT Press 2005. Caballero, R., E. Farhi and P. Gourinchas (2006) "An Equilibrium Model of (cid:146)Global Imbalances(cid:146) and Low Interest Rates," Working Paper, MIT. Chinn, M.D, and Prasad, E.S. (2003). "Medium-term Determinants of Current Accounts in Industrial and Developing Countries: An Empirical Exploration," Journal of International Economics 59, 47-76. Dooley, M.P., Folkerts-LandauD., andGarber, P.(2003)"AnEssayontheRevisedBrettonWoods System" NBER Working Paper 9971. Dooley, M.P., Folkerts-Landau D., and Garber, P. (2004) "Direct Investment, Rising Real Wages and the Absorption of Excess Labor in the Periphery" NBER Working Paper 10626. Engel, C. and Rogers J. (2005), "The U.S. Current Account De(cid:133)cit and the Expected Share of World Output," Working Paper, University of Wisconsin. Erickson,TimothyandWhited,ToniM.(2000),"MeasurementErrorandtheRelationshipbetween Investment and q," Journal of Political Economy, 108(2000), 1027-1057. Fazzari, Steven and Michael Athey (1987), "Asymmetric Information, Financing Constraints and Investment", Review of Economics and Statistics 69, 481-487. 23

Fazzari, Steven M.; Hubbard, R. Glenn and Petersen, Bruce C. (1988a) "Financing Constraints and Corporate Investment," Brookings Papers on Economic Activity, 1988(1), 141-206. Fazzari, Steven M., Hubbard, R. Glenn and Petersen, Bruce C. (1988b) "Investment, Financing Decisions, and Tax Policy," American Economic Review, May, 78 (2), 200-205. Feroli, M. (2003) "Capital Flows Among the G-7 Nations: A Demographic Perspective," Finance and Economics Discussion Series No. 54. Board of Governors of the Federal Reserve System. Ferrero, A. (2005) "Demographic Trends, Fiscal Policy and Trade De(cid:133)cits," Working Paper, New York University. Hausmann, Ricardo and Federico Sturzenegger (2005), "Dark Matter Makes the U.S. De(cid:133)cit Disappear," Financial Times (8 December), A15. Hayashi,Fumio(1982),"Tobin(cid:146)sMarginalqandAverageQ:ANeoclassicalInterpretation,"Econometrica, 50 (1), 213-224. Hubbard, R. Glenn (1998) "Capital-Market Imperfections and Investment", Journal of Economic Literature, Vol. 36, No. 1, 193-225. Gilchrist, Simon and Himmelberg, Charles P. (1995), "Evidence on the Role of Cash Flow for Investment," Journal of Monetary Economics, 36, 531-572. Gruber, Joseph W. and Kamin, Steven B. (2007), "Explaining the Global Pattern of Current Account Imbalances," Journal of International Money and Finance, 26, 500-522. Kamin, Steven B. (2007), "The Revived Bretton Woods System: Does It Explain Developments in Non-China Developing Asia?," Paper presented at the Federal Reserve Bank of San Francisco conference, "The Revived Bretton Woods System: A New Paradigm for Asian Development?" Available from: <http://www.frbsf.org/economics/conferences/0502/kamin.pdf>. Takatoshi, Ito (1999) "Capital Flows in Asia," NBER Working paper #7134. Kaplan, Steven N. and Zingales, Luigi (1997), "Do Investment-Cash Flow Sensitivities Provide Useful Measures of Financing Constraints," Quarterly Journal of Economics, 112(1), 169-215. 24

Mussa, Michael (2004), "Exchange Rate Adjustments Needed to Reduce Global Payments Imbalances," in C. Fred Bergsten and John Williamson eds, Dollar Adjustment: How far? Against What? Washington, D.C.: Institute for International Economics, 113-138. Mann, Catherine (2004), "Managing Exchange Rates: Achievement of Global Re-balancing or EvidenceofGlobalCo-Dependence?"unpublishedmanuscript,InstituteforInternationalEconomics (July). Mendoza, E.G., Quadrini, V., and Rios-Rull, J.S. (2007), "Financial Integration, Financial Deepness, and Global Imbalances" NBER Working Paper 12909. Modigliani, Franco and Miller, Merton (1959), "The Cost of Capital, Corporation Finance and the Theory of Investment," American Economic Review, 48(June), 261-297. Myers, Stewart C. (1977). "Determinants of corporate borrowing," Journal of Financial Economics 5, 147-175. Myers, Stewart C. and Nicholas S. Majluf, (1984), "Corporate (cid:133)nancing decisions when (cid:133)rms have investment information that investors do not," Journal of Financial Economics 13, 187-220. Modigliani, Franco and Miller, Merton (1963), "Corporate income taxes and the cost of capital: a correction." American Economic Review, 53(3), 433-443. Obstfeld,MauriceandRogo⁄,Kenneth(2000),"PerspectivesonOECDCapitalMarketIntegration: Implications for U.S. Current Account Adjustment," in Global Economic Integration: Opportunities and Challenges, Kansas City: Federal Reserve Bank of Kansas City, 169-208. Obstfeld, Maurice and Rogo⁄, Kenneth (2004), "The Unsustainable U.S. Current Account Position Revisited," NBER Working Paper no. 10869 (October). Obstfeld,MauriceandRogo⁄,Kenneth(2005),"GlobalCurrentAccountImbalancesandExchange Rate Adjustments," Working Paper, UC Berkeley. Prasad, E.S., Rajan, R., and Subramanian, A. (2006) "Foreign Capital and Economic Growth" NBER Working Paper 13619. Roubini, Nouriel and Setser, Brad (2004), "The US as Net Debtor: The Sustainability of the US External Imbalances," New York University (mimeo), November 2004. 25

Summers, Lawrence H. (1981), "Taxation and Corporate Investment: A q-Theory Approach," Brookings Papers on Economic Activity, 1981(1), 67-127. Tobin, James (1969), "A General Equilibrium Approach to Monetary Theory," Journal of Money, Credit and Banking, 1(1), 15-29. Whited, T. (1992), "Liquidity Constraints and Corporate Investment: Evidence from Panel Data," Journal of Finance, 47(4), 1425-1460. 26

PDG fo tnecreP 01 5 0 5 01 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 Year Asia 8 China United States Figure 1: AnnualcurrentaccountbalancesaspercentofGrossDomesticProductfortheUnitedStates,China,and the Asia-8 region, 1980 to 2005. 27

PDG fo tnecreP 53 03 52 02 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 Year Investment Saving Figure 2: Aggregate national saving and private investment as percent of Gross Domestic Product for the Asia-8 region, 1980 to 2005. PDG fo tnecreP 03 52 02 51 01 5 1991 1993 1995 1997 1999 2001 2003 2005 Year Private Public Figure 3: AggregateprivateandpublicinvestmentaspercentofGrossDomesticProductfortheAsia-8region,1991 to 2005. 28

50. 0 50. 1. 51. Fixed Investment / Capital 1991 1993 1995 1997 1999 2001 2003 2005 year 5.1 1 5. 0 5. 1 Tobin's Q 1991 1993 1995 1997 1999 2001 2003 2005 year 2. 1. 0 1. Cash Flow / Capital 1991 1993 1995 1997 1999 2001 2003 2005 year 4 2 0 2 4 6 ROA (%) 1991 1993 1995 1997 1999 2001 2003 2005 year Note: For all variables, the time effect for 1996 is set to zero. Figure 4: Estimated year e⁄ects for selected (cid:133)rm-level variables from a regression using data from our full sample of Asian (cid:133)rms, 1991 to 2005, with 95 percent con(cid:133)dence interval. Regressions control for (cid:133)xed e⁄ects and (cid:133)rm size. 1. 0 1. 2. External Financing / Capital 1991 1993 1995 1997 1999 2001 2003 2005 year 4 3 2 1 0 1 Debt / Equity 1991 1993 1995 1997 1999 2001 2003 2005 year 1. 0 1. 2. 3. Debt / Capital 1991 1993 1995 1997 1999 2001 2003 2005 year 2. 1. 0 1. 2. Cash / Capital 1991 1993 1995 1997 1999 2001 2003 2005 year Note: For all variables, the time effect for 1996 is set to zero. Figure 5: Estimated year e⁄ects for selected (cid:133)rm-level variables from a regression using data from our full sample of Asian (cid:133)rms, 1991 to 2005, with 95 percent con(cid:133)dence interval. Regressions control for (cid:133)xed e⁄ects and (cid:133)rm size. 29

3. 2.K/I1 n.o tc0effE1. 2. 1991 1993 1995 1997 1999 2001 2003 2005 year HKG IDN KOR MYS PHL SGP THA TWN Note: Values in 1996 are set to zero. K/I no tceffE sisirctsoP 3. 2. 1. 0 1. 2. HKG IDN KOR MYS PHL SGP THA TWN Country Figure 6: TopPanel: Estimatedaggregatetimee⁄ectsoninvestmentrateforeachcountryfromaregressionusing our panel of (cid:133)rms in Asia-8 countries, 1991 to 2005. Bottom Panel: Estimated mean postcrisis drag on investment rate for each country, with 95 percent con(cid:133)dence interval. Both regressions also control for (cid:133)xed e⁄ects. 30

K/I no tceffE 4. 2. 0 2. 4. 1991 1993 1995 1997 1999 2001 2003 2005 year SIC 0 SIC 1 SIC 2 SIC 3 SIC 4 SIC 5 SIC 6 SIC 7 SIC 8 K/I no tceffE sisirctsoP 4. 2. 0 2. 4. 0 2 4 6 8 SIC Code: Digit 1 Figure 7: Top Panel: Estimated aggregate time e⁄ects on investment rate for each one-digit industry level from a regression using our panel of (cid:133)rms in Asia-8 countries, 1991 to 2005. Bottom Panel: Estimated mean postcrisis drag on investmentrate foreach one-digitindustry level, with 95 percentcon(cid:133)dence interval. Both regressionsalso control for (cid:133)xed e⁄ects. 31

K/I no tceffE 1. 0 1. 2. 1991 1993 1995 1997 1999 2001 2003 2005 year Small Medium Large K/I no tceffE sisirctsoP 1. 0 1. 2. small medium large Firm Size Figure 8: Top Panel: Estimated aggregate time e⁄ects on the rate of investment of small, medium and large (cid:133)rms from a regression using our panel of (cid:133)rms in Asia-8 countries, 1991 to 2005. Bottom Panel: Estimated mean postcrisis drag on investment rate for small, medium and large Asia-8 (cid:133)rms, with 95 percent con(cid:133)dence interval. Both regressions also control for (cid:133)xed e⁄ects. 32

50. 0 50. 1. 51. 1991 1993 1995 1997 1999 2001 2003 2005 year None Q Q, CF/K Q, CF/K, XF/K Q, CF/K, XF/K, D/K(96) Figure 9: Investment year e⁄ects after controlling for fundamentals and precrisis conditions shown. E⁄ects are estimated using our full sample of Asian (cid:133)rms from 1991 to 2005, and control for (cid:133)xed e⁄ects and (cid:133)rm size. 33

5.1 1 5. 0 Debt / Capital 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile 4 3 2 1 0 Debt / Equity 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile 52. 2. 51. 1. 50. 0 External Financing / Capital 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile 3. 2. 1. 0 Investment / Capital 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile Figure 10: Median (cid:133)rm characteristics by debt-to-capital distribution in 1996. 34

5.2 2 5.1 1 5. Tobin's Q 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile 52. 2. 51. 1. 50. 0 Cash Flow / Capital 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile 01 8 6 4 2 ROA (percent) 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile 80. 60. 40. 20. 0 Dividends / Capital 1992 1995 1998 2001 2004 Year Bottom quartile Mid quartiles Top quartile Figure 11: Median (cid:133)rm characteristics by debt-to-capital distribution in 1996. 35

secnalabmI tnuoccA tnerruC labolG fo noitubirtsiD cihpargoeG :1 elbaT 5002 0002 6991 5002 0002 6991 PDG fo tnecrep srallod .S.U snoillib 8.1- 3.1- 1.0 375- 603- 03 lairtsudnI 4.6- 2.4- 6.1- 297- 514- 521setatS detinU 6.3 6.2 4.1 661 021 66 napaJ 0.0 7.0- 0.1 3- 14- 47 *aerA oruE 0.1 9.0 4.0 65 03 51 rehtO 3.4 6.1 2.1- 115 911 38gnipoleveD 7.4 3.2 3.1- 252 77 04aisA 2.7 7.1 8.0 161 12 7 anihC 0.5 0.4 9.1- 601 36 23- 8-aisA 3.0 6.1 8.3- 51- 6- 51aisA rehtO 5.81 7.01 9.2 381 76 51 tsaE elddiM 3.2 6.1 2.1- 81 7 6acirfA 4.1 5.2- 1.2- 43 94- 93aciremA nitaL 1.1 6.1 3.1- 42 61 41- RSSU remroF dna eporuE nretsaE .a.n .a.n .a.n 26- 881- 35ycnapercsiD lacitsitatS ni sa seirtnuoc fo tes emas eht morf atad gnisu detaluclac saw 6991 ni aera oruE eht rof tnuocca tnerruc eht ,ycnetsisnoc niatniam oT .6002 rebmetpeS kooltuO cimonocE dlroW dnuF yratenoM lanoitanretnI :ecruoS .elbacilppa ton setoned ".a.n" .0002 36

seirtnuoC 8-aisA dna ,anihC ,setatS detinU eht rof secnalaB tnuoccA tnerruC fo noitulovE :2 elbaT edarT liO gnidulcxe tnuoccA tnerruC edarT liO gnidulcni tnuoccA tnerruC 5002 0002 6991 5002 0002 6991 5002 0002 6991 5002 0002 6991 PDG fo tnecrep srallod SU snoillib PDG fo tnecrep srallod SU snoillib 5.4- 0.3- 7.0- 655- 792- 35- 4.6- 2.4- 6.1- 508- 614- 521setatS detinU 6.9 8.2 3.1 412 43 11 2.7 7.1 8.0 161 12 7 anihC 5.9 7.6 4.0- 202 401 6- 1.5 0.4 9.1- 901 26 23- 8-aisA 5.41 6.5 7.0- 8.52 4.9 2.1- 1.11 1.4 6.1- 7.91 0.7 6.2gnoK gnoH 8.2 1.3 6.5- 9.7 2.5 7.21- 6.1 8.4 4.3- 5.4 0.8 7.7aisenodnI 1.7 6.6 4.1- 5.55 6.33 0.8- 1.2 4.2 1.4- 6.61 3.21 1.32aeroK htuoS 1.11 0.6 9.6- 5.41 4.5 0.7- 2.51 4.9 4.4- 9.91 5.8 5.4aisyalaM 4.8 7.1 2.2- 2.8 3.1 8.1- 4.2 9.2- 4.5- 4.2 2.2- 5.4senippilihP 6.04 9.81 4.91 3.74 5.71 0.81 5.82 6.11 0.51 2.33 7.01 9.31 eropagniS 9.9 3.5 6.5 3.43 2.71 2.61 7.4 8.2 8.3 4.61 9.8 9.01 nawiaT 4.4 0.21 3.5- 8.7 7.41 7.9- 1.2- 6.7 9.7- 7.3- 3.9 4.41dnaliahT lanoitanretnI :ecruoS .ecnalab llarevo eht morf lio ni edart ten gnitcartxe yb detaluclac era lio gnidulcxe secnalab tnuocca tnerruC .6002 rebmetpeS kooltuO cimonocE dnuF yratenoM 37

slatnemadnuF leveL-mriF :etaR tnemtsevnI level-mriF fo noissergeR lenaP :3 elbaT tiI tiK )9( )8( )7( )6( )5( )4( )3( )2( )1( )0( 860: 870: 680: 780: 611: 001: 990: 690: 911: 821: (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) +79 )720:( )010:( )010:( )010:( )010:( )010:( )010:( )010:( )010:( )700:( 500: 810: 210: 210: 020: Q )300:( )200:( )200:( )300:( )400:( ti 442: 442: 172: 172: 103: tiFC )831:( )440:( )740:( )740:( )050:( tiK 602: 221: 351: tiFX )640:( )810:( )320:( tiK 500: 2000: 2000: 6000: (cid:0) (cid:0) (cid:0) tiD )500:( )2000:( )2000:( )2000:( tiE 460: 170: 480: 480: 861: 431: 490: 931: 761: msD )020:( )020:( )220:( )220:( )620:( )320:( )120:( )420:( )620:( ti 220: 120: 720: 720: 750: 140: 630: 830: 650: (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) glD )110:( )110:( )110:( )110:( )110:( )010:( )110:( )110:( )110:( ti :htiw detcaretni ymmud +79 640: 150: 850: 850: 270: 060: 860: 750: 470: STHD )110:( )110:( )210:( )210:( )210:( )110:( )210:( )210:( )210:( ti 610: 320: 800: 800: 230: 740: 400: 430: 130: (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) msD )910:( )020:( )220:( )220:( )420:( )220:( )220:( )320:( )420:( ti 410: 810: 020: 020: 900: 800: 910: 310: 900: (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) glD )110:( )900:( )010:( )010:( )010:( )900:( )010:( )010:( )010:( ti 400: Q )400:( ti 300: (cid:0) tiFC )041:( tiK 690: (cid:0) tiFX )050:( tiK 500: (cid:0) tiD )500:( tiE 581: 281: 631: 531: 730: 311: 421: 960: 630: 910: 2R 00912 00912 00912 00912 00912 00912 00912 00912 00912 00912 N yrtsudnidnaymmudsisirctsopehtneewtebsnoitcaretnidnaezismr(cid:133)roflortnoc"0"noitac(cid:133)icepstublladna,stce⁄edex(cid:133)edulcnisnoissergerllA .noitalerroc laires dna yticitsadeksoreteh ot tsubor era srorre dradnatS .)nwohs ton( ymmud 38

snoitidnoC sisircerP ot ytivitisneS :wolF tnemtsevnI level-mriF fo noissergeR :4 elbaT scitsiretcarahC mriF sisircerp tiI tiK )9( )8( )7( )6( )5( )4( )3( )2( )1( 120: 050: 310: 610: 120: 580: 980: 170: 590: (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) (cid:0) +79 )920:( )421:( )930:( )010:( )030:( )030:( )410:( )210:( )910:( 700: 110: 700: 700: 700: 700: 700: 700: 600: Q )300:( )400:( )300:( )300:( )300:( )200:( )300:( )300:( )300:( ti 482: 692: 972: 562: 552: 272: 282: 872: 792: tiFC )870:( )381:( )401:( )770:( )860:( )070:( )670:( )280:( )180:( tiK :htiw detcaretni ymmud +79 100: Q )620:( 69 451: (cid:0) 69FX )750:( 69iK 910: (cid:0) 69iFC )330:( 69iK 700: (cid:0) 69itbeD )200:( 69iytiuqE 390: 621: 890: 990: (cid:0) (cid:0) (cid:0) (cid:0) 69itbeD )430:( )180:( )530:( )630:( 69iK 671: 650: (cid:0) (cid:0) tiFC )081:( )590:( tiK 251: (cid:0) tnemtsevnIrevO )210:( 240: (cid:0) tihsaC 69itbeD )910:( tiK (cid:3) 69iK 442: 363: 932: 542: 832: 702: 322: 342: 332: 2R 58901 8042 58901 47901 58901 36801 29901 15901 64301 N .)nwohs ton( ymmud yrtsudni dna ymmud sisirctsop eht neewteb snoitcaretni dna ,ezis mr(cid:133) ,stce⁄e dex(cid:133) edulcni snoisserger llA .noitalerroc laires dna yticitsadeksoreteh ot tsubor era srorre dradnatS 39