Emerging market exchange rate exposure

Emerging market exchange rate exposure

Available online at www.sciencedirect.com Journal of Banking & Finance 32 (2008) 1349–1362 www.elsevier.com/locate/jbf Emerging market exchange rate...

216KB Sizes 2 Downloads 46 Views

Available online at www.sciencedirect.com

Journal of Banking & Finance 32 (2008) 1349–1362 www.elsevier.com/locate/jbf

Emerging market exchange rate exposure Timothy K. Chue a,*, David Cook b,1 b

a School of Accounting and Finance, Faculty of Business, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Department of Economics, School of Business and Management, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong

Received 19 July 2005; accepted 12 November 2007 Available online 4 December 2007

Abstract We estimate the exposure of emerging market companies to fluctuations in their domestic exchange rates. We use an instrumentalvariable approach that identifies the total exposure of a company to exchange rate movements, yet abstracts from the influence of confounding macroeconomic shocks. In the sub-period of 1999–2002, we find that depreciations tend to have a negative impact on emerging market stock returns. In the sub-period of 2002–2006, this tendency has largely disappeared. Since we estimate the exchange rate exposure of firms from different countries with a common set of instruments, we can make coherent, cross-country comparisons of their determinants. We find that the impact of various measures of debt on exchange rate exposure, which is negative and significant in the early sub-period, becomes insignificant and even reverses sign in the recent sub-period. Ó 2007 Elsevier B.V. All rights reserved. JEL classification: F31; F41; G12; G15 Keywords: Exchange rate exposure; Emerging market; International debt

1. Introduction In this paper we estimate the impact of domestic exchange rate movements on the stock market valuations of firms in emerging markets. We are interested in examining exchange rate exposure across different countries as well as the distribution of exposure within countries. We find that emerging market exchange rate exposure has changed within the last few years. In the immediate aftermath of the emerging market crises of the late 1990s (from 1999 to 2002), the share values of most emerging markets firms were negatively affected by exchange rate depreciations. In the more recent years (2002–2006), we find that this negative exposure has disappeared. Previous studies on developed markets typically focus on measuring marginal exchange rate exposure – the *

Corresponding author. Tel.: +852 2766 4995; fax: +852 2330 9845. E-mail addresses: [email protected] (T.K. Chue), [email protected] (D. Cook). 1 Tel.: +852 2358 7614; fax: +852 2358 2084. 0378-4266/$ - see front matter Ó 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.jbankfin.2007.11.005

impact of an exchange rate depreciation on a firm’s stock return, controlling for the return on the national stock index. The inclusion of the national stock index is to control for the confounding effects of macroeconomic shocks, which can lead to simultaneous movements in the local currency and a firm’s stock price. This measure is typically used to assess the exchange rate exposure of certain interesting classes of firms (such as multinationals or exporters) relative to the national average. Yet, for emerging market firms, there are important reasons for us to focus instead on how they are affected by exchange rate fluctuations in absolute terms. First, from a macroeconomic perspective, we are interested in assessing whether emerging market firms as a class are negatively affected by exchange rate devaluations, not just how they perform relative to their respective country averages. Second, Morck et al. (2000) documented that the within-country correlation of stock prices is substantially higher for emerging than for developed markets. Such strong intranational co-movements suggest that there are likely to be important country-specific components in exchange rate

1350

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

exposures. Using measures of total exposure, we can identify any such country-level, macroeconomic determinants of exchange rate exposures; with only measures of marginal exposure, any country-specific effects are already subtracted out and thus unidentifiable. To measure exposure at the national level, we need to distinguish between the direct effects of exchange rate movements on firm value, and the effects of other macroeconomic shocks that simultaneously affect both firm value and exchange rates. Because of the small size of any individual emerging market, its impact on the global financial market is limited. For this reason, we can treat world financial variables as exogenous to the emerging markets’ macroeconomic conditions. Specifically, we use variables such as the US Fed Funds rate, and the yen–dollar and euro– dollar cross-rates as instruments to identify the direct effects of exchange rate movements on firm value. Because we identify the exchange rate exposure for all firms using a common set of instruments, we can make coherent comparisons of the exchange rate exposure of firms in different countries. In a sample of 900 emerging market firms examined over the period 1999 through 2006, we find that the response of stock prices to exchange rate depreciations is mostly negative. On average, a 1% depreciation in the exchange rate is associated with a 0.4% decrease in stock price. There are about twice as many firms that have statistically significant and negative exposure as would be expected if there was no relationship between exchange rates and stock prices. However, these results for the full sample mask the changes in exchange rate exposure that have been occurring over time. Between 1999 and 2002, immediately following the various emerging market crises of the 1990s, we find that a 1% depreciation is associated with a 0.9% decrease in stock price, and about three times as many firms had significantly negative exposure as one would expect from a random sample. This finding is consistent with the phenomenon that numerous authors refer to as ‘‘liability dollarization”.2 By contrast, between 2002 and 2006, we find little evidence of negative exposure. On average, a 1% depreciation is associated with a 0.1% increase in stock prices. Further, there no longer seems to be a negative relationship between exchange rate exposure and debt (international or otherwise) at the firm or national level. In fact, in this recent sub-period, we find that depreciations are more likely to lead to increases in firm value in countries with substantial foreign debt. As we argue below, this finding is consistent with there being a greater dependence of emerging market firms 2 Hausmann et al. (2001) constructed several indicators that measured the ability of a large group of emerging and developed countries to borrow in their own currencies. Caballero and Krishnamurthy (2003) found that financial constraints led the emerging market firms to undervalue insuring against exchange rate depreciations, and take on excessive dollar debt. Calvo (2002) reviewed other reasons for why liability dollarization arose. See also the works of Allen and Gale (2000), Calvo (1996), and Calvo and Guidotti (1990).

on their domestic bond markets, and more opportunities for hedging through better-developed derivatives markets. The paper is organized as follows. Section 2 briefly reviews the current literature. Section 3 estimates the total exchange rate exposure using an instrumental-variable approach. Section 4 investigates the determinants of exposure. Section 5 concludes. 2. Related literature Beginning with the works of Adler and Dumas (1980, 1984), Dumas (1978), and Hodder (1982), there is a long list of studies that examine the exchange rate exposure of companies in developed markets. The main focus of this literature is to estimate exposures at the firm or industry level, and then investigate their determinants. For example, Jorion (1990) and He and Ng (1998) found foreign sales to be an important determinant of exposure for US and Japanese multinationals. Allayannis and Ihrig (2001) and Bodnar et al. (2002) investigated the role of market structure as a determinant of exchange rate exposure. Allayannis and Ihrig established the link between exchange rate exposure and markup, and estimated their model on US manufacturing industries. Bodnar, Dumas, and Marston related an exporting firm’s exchange rate exposure with the extent to which it ‘‘passes through” exchange rate changes to prices, and estimated their model on Japanese export industries. Griffin and Stulz (2001) and Williamson (2001) examined the industry level competitive effects of exchange rate movements. Griffin and Stulz found that these effects were small for a broad set of industries in Canada, France, Germany, Japan, the UK, and the US Williamson found stronger effects for the automotive industry in Japan and the US Doidge et al. (2006) and Dominguez and Tesar (2006) provided more detailed reviews of this literature. In terms of methodology, many studies, including the works of Allayannis and Ofek (2001), Bodnar and Gentry (1993), Jorion (1990, 1991), and Williamson (2001), measure exchange rate exposure using the domestic market index as a control variable. Bodnar and Wong (2003) provided a careful discussion of the motivation for this approach, and further showed that the construction of the market control variable can have a substantial influence on the sign and size of the exposure estimates. A few studies examine the exchange rate exposure of some individual emerging markets at the firm or industry level. Kho and Stulz (2000) studied the currency exposure of the banking sector in five East Asian countries during the Asian financial crisis. They found that currency depreciations had a negative impact on the sector’s stock returns only in Indonesia and the Philippines. Dominguez and Tesar (2006) found that the majority of Thai firms in their sample have a negative exposure to a local currency depreciation. Parsley and Popper (2006) studied how exchange rate pegs influenced the exchange rate exposure of Asia–Pacific firms, and found that countries with a fixed

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

exchange rate against one currency exhibited no less exposure to other currencies. The emerging market financial crises of the 1990s exposed how exchange rate movements can affect emerging market firms through the foreign-currency debt on their balance sheets. Aghion et al. (2001) attributed endogenous currency crises to the presence of foreign-currency debt. Calvo (2002) argued that the prevalence of foreign-currency liabilities in emerging markets limits the desirability of flexible exchange rates. Some empirical work relates international debt with poor firm-level performance during financial crises. Allayannis et al. (2003), using data reported by East Asian non-financial corporations on foreign debt and hedging, found that the performance of East Asian firms around the Asian crisis was negatively affected by three types of debt: domestic-currency debt, foreign-currency debt, and artificial domestic-currency debt (i.e., foreign-currency debt converted into domestic-currency debt through hedging). Among these three types, they found that artificial domestic-currency debt has the most negative impact on firm performance. Aguiar (2005) found that Mexican firms with a large share of their short-term debt denominated in foreign currencies had relatively low levels of investment in years immediately following the peso depreciation of 1994. By contrast, our analysis abstracts from crisis-related devaluations by estimating exchange rate exposures with global financial market variables as instruments. Thus, we can provide, at most, indirect information about the causes or consequences of financial crises. However, our approach allows us to identify companies’ exposure to exogenous exchange rate movements, and the effect of international debt on firm performance specifically through the exchange rate channel. Without this identification, the association between international debt and firm performance during financial crises may be due to a variety of channels related to the internal macroeconomic causes of the crises. 3. Measuring total exchange rate exposure 3.1. Methodology We measure the total impact of depreciations on emerging market firms. For each security j from country x, we estimate the following equation: Rj;t ¼ b0;j þ b1;j Dsx;t þ b2;j RW ;t þ ej;t ;

ð1Þ

where Rj,t is the excess return on stock j, sx,t is the tradeweighted exchange rate of country x,3 and RW,t is the ex3 A number of recent studies, including Parsley and Popper (2006) and Williamson (2001), use multiple bilateral exchange rates in the place of the single trade-weighted exchange rate. This approach has the advantage of allowing the data to select which exchange rate is most significant for an individual firm. However, using multiple bilateral rates would make our second-stage, cross-country analysis of the determinants of exposure difficult to conduct. Therefore, we follow the rest of the literature and use the trade-weighted exchange rate instead.

1351

cess return on the world stock market index. Rj,t is the difference between the local currency return on firm j and the domestic short-term interest rate, and RW,t the world stock market return in US dollars in excess of the 3-month US Tbill rate. We measure the exchange rate as the domesticcurrency price of foreign-currency (so that an increase in sx,t is equivalent to a depreciation). We interpret the coefficient b1,j as the total exchange rate exposure of firm j. Before turning to the estimation results, we address several important questions regarding our approach: First, why do we have to include the term b2,jRW,t in Eq. (1), as estimating the equation without this term seems like an obvious way of measuring total exposure? Second, why do we measure RW,t in US dollars, rather than in local currency? Finally, why can we interpret the coefficient b1,j in Eq. (1) as an estimate of total exchange rate exposure, as the inclusion of the term b2,jRW,t seems to suggest that b1,j is only a measure of marginal exposure? To answer the first question, note that estimating Eq. (1) without the term b2,jRW,t is problematic, even when we use world financial variables as instruments. The reason is for a particular variable to be a valid instrument, it has to be correlated with Dsx,t, yet uncorrelated with ej,t. Even though we can plausibly assume that world financial variables (such as the euro–US dollar exchange rate, or the US short-term interest rate) are exogenous to an emerging market’s local shocks, they can be correlated with global shocks that affect the emerging market. This possibility implies that the correlation between the world instruments and ej,t can be nonzero, violating our identification assumptions. We include the term b2,jRW,t in Eq. (1) as a parsimonious way to ‘‘soak up” any remaining correlation between our world instruments and ej,t. To address the second question, note that in practice, we need to choose a particular currency of denomination for RW,t. Measuring RW,t in the emerging market’s local currency is inappropriate for our purpose, for then, RW,t not only captures the effects of global market stocks, but also the effects of local currency movements. This implies that even when the global market return remains constant, RW,t can still move just because the emerging market’s currency fluctuates. Such local currency-induced movements in RW,t confound with our measurement of exchange rate exposure in b1,jDsx,t. To avoid this problem, we measure RW,t as the world market return in US dollars (from Morgan Stanley Capital International) in excess of the 3-month US T-bill rate. We select the US dollar because the US makes up the largest share of the world market capitalization. Ultimately, whether or not this specification of b2,jRW,t can indeed remove the remaining correlation between our world instruments and ej,t is an empirical question, and we investigate this below using Hansen’s J-statistic. It is important to note that our goal here is to obtain a consistent estimate of the exchange rate exposure coefficient b1,j, rather than maximize the R2 in Eq. (1). Thus, we do not include all factors that can potentially affect firm

1352

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

j’s return in Eq. (1) – as our concern is only whether the term b2,jRW,t is able to adequately capture channels other than the emerging market’s exchange rate that our world instruments can affect firm j’s return. It is also important to note that Eq. (1) is not to be interpreted as an asset pricing model, to which tests of market integration (e.g., Jorion and Schwartz, 1986; Mittoo, 1992) should be applied to decide if it is correctly specified. For our purpose, we only use Eq. (1) to measure the impact of Dsx,t on Rj,t. Whether it can accurately price emerging market assets is not our immediate concern. The third question we need to address is whether we can still interpret the coefficient b1,j as a total exchange rate exposure, after the inclusion of the term b2,jRW,t in Eq. (1). In general, the answer to this question is no: we can only interpret b1,j as a measure of marginal exchange rate exposure, that is, the impact of a given local exchange rate movement on firm j’s stock return, relative to its impact on the world stock market average. But in the case of an emerging market, because of its small size, we can plausibly assume that the impact of an exogenous movement of its exchange rate Dsx,t on the world stock market is negligible. As a result, we can interpret b1,j as the total impact of a given local exchange rate movement on firm j’s stock return. Another way to see this point is by rewriting Eq. (1) as Rj,t  b2,jRW,t = b0,j + b1,jDsx,t + ej,t. Having written in this form, it is clear that in general, b1,j measures the marginal impact of Dsx,t on Rj,t, relative to b2,jRW,t. But in the case when the impact of Dsx,t on RW,t is negligible,4 b1,j actually measures the total impact of Dsx,t on Rj,t. To summarize, our approach begins with the use of world financial variables such as the yen-dollar and the euro-dollar exchange rates as instruments to identify emerging market exchange rate movements that are exogenous to the market’s local conditions. For example, when the emerging market’s currency appreciates against the US dollar at a time when the yen and the euro also appreciates relative to the dollar, it is likely that the emerging market appreciation is exogenous to the market’s local conditions – since the small size of the emerging market makes it unlikely to affect the yen-dollar, and euro-dollar rates. If such a global depreciation of the US dollar (i.e., the dollar depreciates against the emerging market currency, as well as against the yen and the euro) can affect an emerging market firm only through the emerging market’s exchange rate, then the response of the firm’s stock return to this instrumented, exogenous local exchange rate movement is the total exposure that we want. However, a global depreciation of the dollar can potentially affect an emerging market firm not only through the emerging

market’s exchange rate, but also through other channels. We use the firm’s exposure to the global stock market as a parsimonious way to summarize these other channels.5 3.2. Data We obtain weekly common stock returns from the Standard & Poor’s (S&P) emerging market Database (EMDB) for all companies listed in January 2000 from 15 emerging markets: Brazil, Chile, Colombia, India, Indonesia, Korea, Mexico, Morocco, Pakistan, the Philippines, South Africa, Taiwan, Thailand, Turkey, and Venezuela.6 S&P selects companies that are actively traded, with coverage in each country reaching at least 60% of total market capitalization. We assume that firms in the EMDB are the likeliest to have access to international debt markets. Our sample period covers January 1st, 1999 to June 30th, 2006. We choose our sample to begin after the last broad-based emerging market crisis, which ended in 1998. Though there were isolated cases of currency crises during our sample period, we are not aware of any evidence that these crises were large enough to impact global markets. This is important because we use world financial market variables as instruments and reverse causality between large-scale emerging markets crises and global financial markets would undermine their validity. We use trade-weighted exchange rate measures constructed by JP Morgan and obtain short-term (interbank) interest rates from Datastream. We use the world stock market index from Morgan Stanley Capital International, and the 3-month US T-bill rate from the CEIC/DRI Asia database. Table 1 reports the cumulative rate of exchange rate depreciation, the weekly standard deviation, and the largest weekly depreciation for each country over our sample period. We refer to Brazil, Indonesia, Morocco, South Africa, Turkey, and Venezuela as ‘‘crisis countries”, as they have had substantial exchange rate instability (with a weekly standard deviation of at least 2%) over this period. Each of these countries has also experienced weeks with double-digit percentage declines in their trade-weighted exchange rate (except for South Africa, which had a week with a 9% decline). Our global market instruments explain only a small fraction of the exchange rate variance in the crisis countries. Table 1, Column D reports Anderson and Rubin’s (1949) test of the validity of the instruments. We can reject the hypothesis that the instruments are not valid at the 1%

5

4 Even when the impact of Dsx,t on RW,t is zero, the correlation between Dsx,t and RW,t can be nonzero. As we discuss above, certain world financial variables can lead to simultaneous movements in Dsx,t and RW,t, but these co-movements do not imply that Dsx,t causes RW,t.

Whether or not this is a good assumption is examined empirically below, using Hansen’s J-statistic. 6 We exclude the EU countries of Greece and Portugal; transition economies in Eastern Europe and China; Argentina and Malaysia, which had hard pegs with the US dollar over most of our sample period; Egypt, Sri Lanka, Peru, and Zimbabwe, which are countries with insufficient aggregate financial market data.

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

1353

Table 1 Summary statistics for exchange rate movements and other macroeconomic variables FX appreciation rate (%)

Brazil Chile Colombia India Indonesia Korea Mexico Morocco Pakistan Philippines South Africa Taiwan Thailand Turkey Venezuela

Instrument relevance

Macroeconomic indicators

(A) Total change

(B) Weekly standard deviation

(C) Largest weekly change

(D) Anderson– Rubin

(E) Cragg– Donald

(F) Per capita GDP US$

(G) (H) % of GDP

(I)

(J)

Export

Import

Debt

M2

45.3 0.4 26.9 7.5 14.6 23.9 13 2.1 17.2 31.8 5.2 1.9 5.7 147.4 127

2.6 1.1 1.1 0.7 2.2 0.9 1.1 5.1 1.0 0.8 2.0 0.5 0.7 3.1 2.5

18.8 4.9 4.6 2.0 11.6 3.3 4.1 69.5 5.9 4.5 9.2 2.0 3.1 42.2 20

10.42k 32.853| 40.69| 401.70| 8.56 42.780| 38.93| 3.197 174.26| 109.85| 3.314 154.41| 40.10| 7.626 19.10|

2.60 8.44 10.56 174.77 2.13 11.14 10.08 0.79 54.30 31.31 0.82 46.80 10.40 1.90 4.82

$7,128 $9,049 $6,461 $2,245 $2,910 $14,022 $8,466 $3,566 $1,837 $3,775 $9,347 $17,056 $6,011 $6,487 $6,154

8.86 27.43 15.29 10.43 30.71 30.86 24.29 22.43 12.29 38.86 23.43 43.88 43.14 20.00 29.00

9.14 28.00 20.43 11.29 28.43 29.71 26.00 27.71 15.00 46.43 21.14 44.13 44.57 23.71 24.00

27.89 35.44 30.90 28.39 74.37 26.49 40.63 73.32 48.85 62.48 22.92 9.70 59.27 42.14 51.84

35.78 38.55 20.24 46.44 49.80 41.45 30.65 65.37 45.59 51.09 51.05 170.01 83.79 33.04 23.92

24.21 58.75 34.08 18.73 12.59 26.17 14.23 16.81 4.82 15.95 7.05 364.63 37.25 15.87 27.91

$8,009

26.59

27.48

36.76

59.92

60.73

All

(K) Foreign reserves to debt

Columns (A)–(C) of this table report the summary statistics for the percentage appreciation of the trade-weighted foreign-currency index relative to the local currency (Dsx,t) of all the countries in our sample, over the period from January 1st, 1999 to June 30th, 2006. Columns (D) and (E) report two measures of instrument validity and relevance. The Andersen–Rubin statistic tests if the instruments are invalid; statistics that reject the hypothesis of instrument invalidity at the 1% and 5% levels are indicated with | and k, respectively. The Cragg–Donald statistic measures instrument relevance. Columns (F)–(K) report country-level, macroeconomic statistics, including GDP per capita, exports, imports, M2, and external debt, all calculated as a percentage of GDP, and foreign reserves as a percentage of external debt. GDP per capita is measured by its level in 1998, reported in PPP-converted US dollars using 1990 as a base year. The other variables are measured by their time-series averages over 1990–1998.

level for all of the non-crisis countries, at the 5% level for Brazil, at the 10% level for Indonesia, but not for the other crisis countries. Column E reports Cragg and Donald’s (1993) test for instrument relevance. The statistic exceeds Staiger and Stock’s (1997) rule-of-thumb critical value of 10 for rejecting the hypothesis of instrument irrelevance for all of the non-crisis countries, with the exception of Chile, whose Cragg–Donald statistic is above 8. In all of the crisis countries, the statistic is below 5. 3.3. Empirical results We estimate Eq. (1) with instrumental-variables for the 931 firms in our sample and describe the results in Table 2. The value-weighted mean estimate (using market capitalization from 2000 as weights) of b1,j is 0.389, indicating that a 1% depreciation is associated with a 0.4% decline in the stock price (see Column A). The equal-weighted mean in Column E is similar, equaling 0.315. The negative exposures shown in Column A are concentrated in some middle income countries in East Asia and Latin America, such as Korea and Mexico. Slightly more than half of the firms have negative exposure (see Column B). Column C shows the number of firms with b1,j estimates that are significantly different from zero at the 5% critical value. Overall, slightly greater than 7% of all firms are in this category. This number is between 9% and 15% for mid-

dle income East Asia and Latin America. We also report, in Column D, the fraction of firms with negative and significant exposure at the 5% level. Under the null hypothesis of b1,j = 0, the fraction of such firms should constitute approximately 2.5% of the whole population of firms. In our sample, this fraction is slightly over 5%. Most of these cases are again concentrated in the middle income countries, with Korea having the highest fraction, at nearly 14%. In fact, only Korea, Mexico, and Taiwan have more than 7.5% of firms (i.e., three times the number we would expect in a random sample under the null) with negative and significant exposure at the 5% level. We then estimate Eq. (1) with OLS and report the results in Column (F)–(I), to illustrate the importance of using instrumental variables. Not surprisingly, these results show a much stronger relationship between exchange rate depreciations and negative stock price movements. The mean of the coefficient b1,j is about 0.99 and more than 80% of all firms have negative exposure. About 40% of firms have significant exposure and almost all of these significant exposures are also negative exposures. A comparison between the IV results in Columns (A)–(D) and the OLS results in Column (F)–(I) reveals the importance of using instruments to take out the effects of country-level macroeconomic shocks. We also report (in Columns J–L) the IV estimates of the world stock market exposure coefficient b2,j. We find that

1354

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

Table 2 OLS and GMM-IV estimation of exchange rate exposure & world market exposure: full sample results OID test

b2,j

b1,j IV estimates

OLS estimates

IV estimates

(A) (B) Value % Neg. weighted mean

(C) % Sig.

(D) % Neg & Sig.

(E) Equal weighted mean

(F) (G) Value % Neg. weighted mean

(H) % Sig.

(I) % Neg. & Sig.

(J) (K) Value % Pos. weighted mean

(L) % Pos. & Sig.

(M) J-statistics

Brazil Chile Colombia India Indonesia Korea Mexico Morocco Pakistan Philippines South Africa Taiwan Thailand Turkey Venezuela

0.114 0.393 0.276 0.201 0.575 1.045 1.041 0.053 0.168 0.345 0.659

46.67 73.91 72.73 33.57 37.04 73.51 61.40 61.11 46.94 48.28 46.88

10.00 8.70 0.00 6.99 0.00 13.91 14.04 0.00 2.04 0.00 4.69

6.67 6.52 0.00 0.70 0.00 13.91a 8.77a 0.00 2.04 0.00 1.56

0.113 0.312 0.426 0.198 0.301 1.093 0.374 0.081 0.004 0.016 0.637

0.469 0.017 0.050 0.580 0.704 1.521 0.379 0.011 0.049 0.337 0.049

80.00 56.52 50.00 81.82 100.00 91.39 77.19 66.67 42.86 82.76 68.75

46.67a 6.52 0.00 5.59 77.78a 59.60a 33.33a 83.33a 0.00 5.17 31.25a

43.33a 4.35 0.00 5.59 77.78a 59.60a 31.58a 50.00a 0.00 5.17 18.75a

0.835 0.386 0.126 0.578 0.421 0.922 0.835 0.095 0.156 0.299 0.787

96.67 95.65 81.82 96.50 88.89 94.04 94.74 100.00 83.67 91.38 95.31

60.00 58.70 31.82 66.43 31.48 65.56 66.67 11.11 20.41 43.10 71.88

0.00 2.17 0.00 2.80 0.00 1.99 7.02 5.56 0.00 1.72 3.13

0.725 0.126 0.166 1.075

77.36 62.50 75.47 25.00

9.43 9.38 1.89 0.00

8.49a 6.25 1.89 0.00

0.967 0.506 0.707 0.480

1.957 1.604 0.646 0.959

100.00 98.44 94.34 31.25

73.58a 71.88a 79.25a 6.25

73.58a 71.88a 79.25a 0.00

0.654 0.550 0.690 0.232

84.91 93.75 92.45 37.50

43.40 64.06 24.53 6.25

2.83 1.56 0.00 0.00

All Finance

0.389 0.115

57.57 54.55

7.20 3.54

5.16 2.02

0.315 0.128

0.991 0.904

81.95 81.31

40.92a 47.47a

38.99a 45.96a

0.706 0.648

91.41 91.92

52.09 56.57

2.15 0.00

This table reports summaries of OLS and GMM-IV estimation results for the exchange rate coefficient b1,j and the world market coefficient b2,j of the equation

Rj;t ¼ b0;j þ b1;j Dsx;t þ b2;j RW ;t þ ej;t ; where Rj,t is the excess return on stock j, sx,t is the trade-weighted exchange rate of country x, and Rw,t is the return on the world stock index over theperiod between January 1st, 1999 and June 30th, 2006. Columns (A) and (F) report the value weighted mean of the GMM and OLS estimates of b1,j, respectively. Columns (B) and (G) report the respective percentage of the GMM and OLS estimates that are negative. Columns (C) and (H) report the fraction of the GMM and OLS estimates of b1,j that are significant at the 5% critical value, while Columns (D) and (I) report the fraction that are negative and significant. Column (E) reports the equal-weighted mean of the IV estimates of b1,j. Column (J) reports the value-weighted mean of the IV estimates of b2,j. Column (K) reports the fraction of the IV b2,j estimates that are positive, and Column (L) the fraction which are positive and significant. Column (M) reports the fraction of companies for which the J-statistics rejects the overidentification (OID) conditions of the IV specification at the 5% critical value. All standard errors are robust to heteroskedasticity and autocorrelation. The last row of the table reports statistics for financial firms only. a Number of significant observations (at the 5% level) more than three times as many as independent draws on firms with zero exchange rate exposure.

the average coefficient is 0.7, with more than 90% of the coefficients being positive and about half being positive and significant. These findings suggest that a substantial fraction of the emerging market firms we consider has positive and significant world market exposure, and reinforces the reason for why we need to estimate Eq. (1) with RW,t included. In particular, our world instruments can affect an emerging market firm not only through their impact on Dsx,t, but also through the firm’s exposure to RW,t – when b2,j is nonzero. Thus, in this case, failing to include RW,t in the equation would imply that the correlation between the world instruments and ej,t is nonzero, violating our identification assumptions. We also test the overidentifying restrictions for each of our regressions by using Hansen’s J-statistic. We interpret these as tests for potential endogeneity of our instruments and/or misspecifications in Eq. (1), which may be important for several reasons. First, the exchange rate Dsx,t is constructed with common rather than firm-specific trade

weights, leading to potential misspecification. Second, if firm j’s exposure to the world stock index (in US dollars), b2,jRW,t, was a poor measure of global shocks affecting Rj,t, our instruments might violate GMM’s orthogonality conditions. From Table 2, Column (M), we see that only in Mexico and Morocco, there are a larger number of rejections of the overidentifying restrictions than the 5% that would be expected if the null hypothesis of no endogeneity/misspecification were true. And even in these countries, the fraction of rejections is only marginally higher than 5%. We interpret this result as a lack of evidence for endogeneity or misspecification. We find that in the overall sample, exchange rate exposure is mostly negative, especially in the middle income countries of East Asia and Latin America. There are a significant number of firms in Brazil, Chile, Korea, Mexico, Taiwan, and Thailand that have negative and significant exposure. However, an earlier version of this paper finds that negative exposures are much more pre-

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

valent in the period running from January 1st, 1999 through June 30th, 2002. We call this period the ‘‘early sub-period”, and the period running from July 1st, 2002 through May 31st, 2006 the ‘‘recent sub-period”. In Table 3, we report the IV estimation results for the two sub-periods to show how exchange rate exposure has evolved over time. We report a summary of our GMM estimates for the coefficient b1,j for the early sub-period shown in Table 3, Panel I. Approximately 70% of the estimates are negative. The value-weighted mean of b1,j is approximately 0.85 while the equal weighted mean is approximately 0.75. About 10% of the estimates of b1,j are statistically significant at the 5% critical value, and about 9% of the estimates are negative and significant, which is more than three times what we would expect if b1,j were zero. The countries in which there are more than 7.5% of firms with negative and significant exposure are again the middle income countries in East Asia and Latin America, namely, Brazil, Chile, Colombia, Korea, Mexico, Taiwan, and Thailand. South Africa is the only country in which there are a substantial number of firms with positive exposure.

1355

During the early sub-period, there are substantially more firms with negative exchange rate exposure than in the whole period. By contrast, there are substantially fewer firms with negative exposure during the recent sub-period of July 2002–June 2006. The value-weighted and equalweighted estimates of b1,j are both positive and around 0.1. Indeed, only around 40% of the coefficients are negative. About 9% of the coefficients are significantly different from zero, and about 5% are negative and significant. However, the majority of these negatively-exposed firms are concentrated in one country, Korea. A notable number of firms in Taiwan, Thailand, and the Philippines also have significantly negative exposure. On the other hand, there are a significant number of firms in Indonesia, Mexico, Morocco, the Philippines, South Africa, Taiwan, and Thailand with significantly positive exposure. We also report that if we use OLS, the value-weighted means of the exchange rate exposure coefficients are negative and similar in magnitude in both the early and the recent sub-periods (Columns F and L). This result demonstrates the importance of controlling for the endogeneity of exchange rate changes. Our instrumental-variable estimates

Table 3 GMM-IV estimation of exchange rate exposure: sub-period results Panel I. early sub-period (January 1st, 1999 to June 30th, 2002)

Panel II. recent sub-period (July 1st, 2002 to June 30th, 2006)

IV estimates

OLS

IV estimates

(A) Value weighted mean

(B) % Neg.

(C) % Sig.

(D) % Neg. & Sig.

(E) Equal weighted mean

(F) Value weighted mean

(G) Value weighted mean

(H) % Neg.

(I) % Sig.

(J) % Neg. & Sig.

(K) Equal weighted mean

(L) Value weighted mean

OLS

Brazil Chile Colombia India Indonesia Korea Mexico Morocco Pakistan Philippines South Africa Taiwan Thailand Turkey Venezuela

1.272 0.861 1.006 0.333 1.265 1.108 1.868 0.019 0.486 0.220 0.659 0.857 1.164 2.137 0.674

86.67 86.96 86.36 39.16 87.04 80.13 66.67 61.11 44.90 62.07 37.50 90.57 84.38 94.34 68.75

30.00a 19.57a 9.09 7.69 5.56 13.91 26.32a 0.00 2.04 1.72 9.38 9.43 17.19a 1.89 0.00

30.00a 17.39a 9.09a 1.40 5.56 13.91a 22.81a 0.00 2.04 1.72 1.56 9.43a 17.19a 1.89 0.00

1.022 0.599 1.102 0.186 1.163 1.317 0.817 0.045 0.074 0.346 0.798 1.393 1.427 2.294 0.263

0.316 0.180 0.262 0.668 0.676 1.826 0.689 0.011 0.085 0.279 0.240 1.918 1.685 0.519 1.150

1.913 0.408 0.711 0.043 0.752 0.996 0.679 1.416 0.003 0.532 0.497 0.211 1.755 1.348 0.396

5.00 25.00 0.00 42.73 15.79 81.13 31.58 50.00 56.25 24.32 36.96 48.53 27.66 9.68 33.33

0.00 3.13 0.00 5.45 5.26 22.64a 7.89 7.14 0.00 13.51 8.70 8.82 10.64 3.23 0.00

0.00 0.00 0.00 1.82 0.00 22.64a 2.63 0.00 0.00 5.41 0.00 4.41 6.38 0.00 0.00

1.893 0.341 0.637 0.109 1.079 1.987 0.716 1.277 0.165 0.657 0.294 0.120 1.047 1.149 0.287

0.504 0.191 0.039 0.502 0.939 1.237 0.078 0.427 0.029 0.428 0.296 1.586 1.342 1.525 0.038

All Finance

0.851 0.773

69.92 69.19

10.74 8.59

8.92a 8.08a

0.747 0.679

1.093 0.945

0.137 0.587

40.99 35.29

9.01 8.82

5.43 6.62

0.089 0.150

0.751 0.845

This table reports summaries of GMM-IV estimation results for the exchange rate coefficient b1,j of the equation for two sub-periods

Rj;t ¼ b0;j þ b1;j Dsx;t þ b2;j RW ;t þ ej;t ; where Rj,t is the excess return on stock j, sx,t is the trade-weighted exchange rate of country x, andRW,t is the return on the world stock index. Columns (A) and (G) report the value-weighted mean of b1,j for the early and the recent sub-periods, respectively. Columns (B) and (H) show the percentage of firms with negative exposure for each of the sub-periods. Columns (C) and (I) report the percentage of firms with significant b1,j estimates (at the 5% level), while Columns (D) and (J) report the percentage of firms with negative and significant b1,j estimates (at the 5% level). Columns (E) and (K) report the valueweighted mean of b1,j for the two sub-periods. For comparison, Columns (F) and (L) report the value-weighted mean of the OLS estimates of b1,j in each sub-period. a Number of significant observations (at the 5% level) more than three times as many as independent draws on firms with zero exchange rate exposure.

1356

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

picking up a significant change in the exchange rate exposure over time, which is otherwise undetectable in a direct OLS regression. We also examine if our previous results are biased by the presence of financial firms in our sample. Financial firms may have different exchange rate exposure because of their different asset and liability structure, and easier access to hedging instruments. Comparing the last two rows of Tables 2 and 3, we see that the results for financial firms are essentially the same as those for all firms. 4. The determinants of exchange rate exposure This section investigates how various factors affect the total exchange rate exposure of emerging market firms. In particular, we try to identify the reasons behind the change in exchange rate exposure across the two sub-periods. Since we identify exchange rate exposure using a common set of world instruments, it is reasonable to estimate the determinants of exposure pooling firms from different countries. Given the difference in the estimates of exchange rate exposure across countries, we examine the importance of various country-level, macroeconomic variables as determinants of exposure. The study of country-specific effects is possible because we measure total exposure in the first stage. By contrast, if we had measured marginal exposure relative to the country average, any country-specific effects would have been subtracted out and could not be identified. We also examine different firm-level variables, including the measures of international debt. 4.1. The model We regress the exchange rate exposure of a firm on various firm-level and country-level characteristics, and a sector-specific dummy variable. Specifically, for firm j of sector n in country m, we estimate a statistical model of the form b1;j ¼ an þ cXj þ wYm þ /j ;

ð2Þ

where b1,j is the consistent estimate of total exchange rate exposure we obtain from the GMM estimation of Eq. (1), an is a sector-specific intercept, and Xj and Ym are firmand country-level variables, respectively. Specifically, Ym represents long-run country-level data that are measured prior to the particular sub-period used to estimate b1,j. We use firm-level balance sheet and stock market data, Xj, which are measured contemporaneously with the sample period but instrumented with pre-sample lags to control for any potential endogeneity and to eliminate transient measurement error. To increase the accuracy of our estimates, we weight each observation by the inverse of the standard error of b1,j obtained in the first stage. This procedure is suggested by Doidge et al. (2006), so that the betas that are estimated more precisely in the first stage receive a heavier weight in the second stage.

4.2. Data description 4.2.1. Firm-specific variables Our sample consists of 931 companies, which can be classified into nine primary industrial sectors: Agriculture, Forestry, and Fishing; Mining; Construction; Manufacturing; Trade; Transportation and Communication; Finance, Insurance, and Real Estate; Services; and Others. Of these 931 companies, 458 are in Manufacturing, 198 in Finance, Insurance, and Real Estate, 102 in Transportation, and 54 in Trade. All other sectors have fewer than 50 firms represented in the sample. Table 4 reports a breakdown of all firms and all manufacturing and finance firms by their country of origin. Our first set of firm-level explanatory variables is stock market data obtained from the S&P EMDB. The first such variable is a firm’s market capitalization, MARKETCAPj, measured in billions of US dollars. We control for firm size because large firms are more likely to hedge their foreign exchange risks if there are fixed costs of doing so. We also control for INVESTIBILITYj, which is the S&P investibility index ranging from zero to one, and measures the proportion of shares available to foreign investors after adjusting for cross-holdings, government ownerships, and regulations. This variable proxies for foreign and/or institutional ownership in a firm’s share. These owners may induce the firm to pursue different hedging strategies, and may also react to exchange rate movements differently, relative to domestic individual investors. Both MARKETCAPj and INVESTIBILITYj are measured in January, 2000 for the early subperiod, and in January, 2003 for the recent sub-period. We also include the variable TURNOVERj, which is a measure of security j’s liquidity, calculated as the security’s monthly value traded divided by its month-end market capitalization, averaged over a year. TURNOVERj is measured in 1999 for the early sub-period and in 2001 for the recent sub-period. We use turnover as a measure of a stock’s liquidity. When there are international shocks and an emerging market monetary authority allows the local currency to depreciate by keeping interest rates low, stocks with relatively low liquidity tend to benefit more, ceteris paribus. We instrument all these variables with their counterparts from one year earlier, so that the instruments are measured in a period that precedes the time when exchange rate exposure is estimated. We provide an overview of the firm-level variables shown in Table 4, where we report the variables’ means in the early sub-period, broken down by country. The mean firm size in our sample is slightly larger than $1 billion, which is about four times the median. To control for such skewness in size, we estimate Eq. (2) using market capitalization in logs. We also see large differences in the levels of liquidity and investibility across countries. Latin American firms tend to have very low liquidity: the average of TURNOVERj is lower than 5% in all Latin American countries, but is often much higher in other emerging markets. The average investibility of the emerging market firms is slightly higher than 50%. India’s stock market (in 2000) is especially close to

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

1357

Table 4 Descriptive statistics–firm-level variables Panel I. Firm distribution

Panel II. Firm data

Number of firms

MARKETCAP (US$)

% with Intl. debt

IDEBTtoCAP

INVEST IBILITY

TURNOVER

DEBTtoCAP

Total

Mfg.

Finance

Brazil Chile Colombia India Indonesia Korea Mexico Morocco Pakistan Philippines South Africa Taiwan, China Thailand Turkey Venezuela

30 46 22 143 54 151 57 18 49 58 64 106 64 53 16

10 19 8 109 25 81 20 7 24 13 14 74 16 29 9

3 6 9 14 13 33 4 8 14 21 19 17 20 12 5

2102.98 1100.48 357.04 838.79 628.11 1592.92 1929.70 546.36 128.09 484.06 1841.41 2815.04 669.29 1716.28 261.06

60 46 18 35 57 52 63 6 20 43 23 40 67 13 25

0.63 0.37 0.81 0.72 2.01 1.54 1.39 0.19 1.63 2.70 0.27 0.21 1.07 0.06 1.25

0.53 0.73 0.46 0.16 0.62 0.87 0.84 0.56 0.29 0.30 0.91 0.50 0.33 0.71 0.74

0.01 0.01 0.01 0.10 0.08 0.49 0.03 0.01 0.14 0.07 0.04 0.25 0.12 0.20 0.01

1.82 0.48 0.25 1.58 2.70 5.36 1.40 0.09 2.40 2.03 0.20 0.93 1.94 0.10 2.07

Total

931

458

198

1315.10

41

1.14

0.56

0.17

1.94

Panel I of this table provides a breakdown of all firms into different countries. We also report the number of firms that belong to Manufacturing and Finance, the two largest sectors in our sample. Panel II of this table reports country-by-country means of the firm-level variables we use as cross-sectional determinants of exchange rate exposure (b1), which is measured at the end of year 1999. We report variables that are contemporaneous with the measurement of b1. MARKETCAP is a firm’s market capitalization in January 2000 measured in millions of US dollars. INVESTIBILITY is the S&P investibility index of a firm in January 2000. This index ranges between zero and one and measures the proportion of shares available to foreign investors after adjusting for cross-holdings, government ownerships and regulations. TURNOVER is the average over 1999 of the monthly value traded of a security, relative to its month-end market capitalization. IDEBTtoCAP is the ratio of the amount of outstanding international, foreign-currency debt of a firm at the end of 1999 (measured in millions of US dollars), relative to MARKETCAP. We report the percentage of firms in each country with positive levels of IDEBT and reports the mean of IDEBTtoCAP for firms that have positive levels of debt. DEBTtoCAP is the ratio of the total amount of debt outstanding for a firm at the fiscal year-end of 1999 (measured in millions of US dollars), relative to MARKETCAP.

foreign investors, with an average investibility of only 16%. South African firms are especially investible, with 91% of shares open to foreign investors. We are interested in exploring the effect of international, foreign-currency debt on the exchange rate exposure of emerging market firms. We have two measures of debt. Our first measure is obtained from the IFR Platinum database, which reports issues of debt securities and syndicated loans in international markets. This database contains primary-market data that does not depend on the self-reporting of companies in their financial statements. Company j’s outstanding international debt at the end of 1999, IDEBTj, is defined as the sum of the face value of foreign-currency bonds and syndicated bank loans issued before, and with maturity date after December 31st of 1999. Debt issued by subsidiaries that are more than 80% owned by the company is also assigned to its stock of international debt. We measure IDEBTj in billions of US dollars. The main variable we will use to measure exposure to international debt is IDEBTtoCAPj, the ratio of IDEBTj to MARKETCAPj. The instruments for IDEBTj and IDEBTtoCAPj are, respectively, firm j’s outstanding international debt at the end of 1998, and the ratio of firm j’s outstanding international debt at the end of 1998 to the instrument for MARKETCAPj. In Table 4 we report the fraction of firms that have taken on international debt. For the whole sample, this number is approximately 40%. There are large interna-

tional differences though, ranging from 6% in Morocco to 67% in Thailand. The average level of international debt to MARKETCAPj for those firms that have international debt is 1.14. There is a small number of the IDEBTtoCAPj observations with a very large magnitude. Since many firms have a zero level of IDEBTj, we cannot minimize the effects of extreme values by taking logs. Instead, we windsorize the data by setting the highest 2.5% of IDEBTtoCAPj (and its instrument) to their levels at precisely 2.5%. Our second measure of leverage includes all of the firm’s debt, whether domestic or international. We obtain this information from balance sheet data as reported in S&P’s COMPUSTAT Global. DEBTj is the total amount of debt outstanding for firm j, measured in millions of US dollars (converted with the end-of-year exchange rates from the IMF International Financial Statistics). DEBTtoCAPj is the ratio of DEBTj to MARKETCAPj. We measure this ratio at the end of 1999 for the early sub-period, and at the end of 2002 for the recent sub-period. Their corresponding instruments are measured one year earlier. As before, we windsorize this measure by censoring at the 2.5% level. 4.2.2. Country-specific variables We also control for a number of country-level variables. First, we include the variables EXPORTtoGDP and IMPORTtoGDP, the ratios of current-dollar exports and imports to GDP from the United Nations National Accounts

1358

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

database. These variables represent a country’s exposure to international trade and allow us to examine if a country’s exchange rate exposure depends on its export orientation and import dependency. Second, we include the variable M2toGDP, the ratio of broad money (the sum of narrow money and quasi-money) to GDP, obtained from IMF’s International Financial Statistics. This variable represents the depth of a country’s banking system, and is often used to measure a country’s level of financial development. On the one hand, better-developed financial markets should lower firms’ cost of hedging and increase their desire to hedge. On the other hand, the wider range of available financial products may also increase firms’ desire to speculate in the foreign exchange market. The former effect tends to lower, while the latter effect tends to raise, the magnitude of a firm’s exchange rate exposure. We also include the variables DEBTtoGDP, the ratio of external debt to GDP, and FXtoDEBT, the ratio of foreign exchange reserves to external debt, both obtained from Lane and Milesi-Feretti’s (2001) database. DEBTtoGDP represents exposure to international debt at the country-level, and FXtoDEBT represents a government’s ability to stabilize the exchange rate in the face of a crisis. We also include PC_GDP, a country’s per capita GDP measured in PPP-converted US dollars, obtained from the Penn World Tables. This variable proxies for a country’s overall level of development, and different authors have documented that it is related to various measures of local bond market development (see Burger and Warnock, 2006, for example). A country with a better-developed local bond market should rely less on foreign-currency debt. We view the first five variables (EXPORTtoGDP, IMPORTtoGDP, M2toGDP, DEBTtoGDP, and FXtoDEBT) as naturally stationary, and it is more appropriate to measure them with time-series averages (rather than with observations at a particular point in time). For this reason, we take averages of these variables over the nine years prior to the beginning of each of the two sub-periods (i.e., 1990– 1998 for the early sub-period, and 1993–2001 for the recent sub-period). By contrast, PC_GDP is not naturally stationary and we just use the variable’s value for the year that immediately precedes the estimation period for b1 (i.e., 1998 for the early sub-period and 2001 for the recent sub-period). Columns (F) through (K) on Table 1 report descriptive statistics for these variables. The debt-to-GDP ratios range between 10% and 75%, with an average level of about 40%. Foreign exchange reserves as a share of debt ranges from 4% to over 350%. The M2-to-GDP ratios lie between 20% and 170% of GDP, with a mean value of about 60%. In terms of 1990 US dollars, the emerging markets we consider have GDP per capita that lies between $2,000 and $17,000, with an average of approximately $8,000. 4.3. Empirical results We report the weighted-IV estimates of c and w shown in Table 5. All standard errors are heteroskedasticity-consistent. If c is negative for a particular firm-level explanatory

variable, an increase in this variable will make an exchange rate depreciation more damaging for the firm’s value. On the other hand, w, the coefficient on a country-level variable, captures the marginal impact of an increase of the countrylevel variable on the (cross-sectional) conditional mean exposure of that country. For each sub-period, we estimate three specifications. First, we use country-level data only, and report results for the early sub-period in Column A1, and results for the recent sub-period in Column A2. Second, we estimate the model with country-specific variables from the S&P EMDB and our first measure of indebtedness, IDEBTtoCAP. These results for the early and recent sub-periods are reported in Columns B1 and B2, respectively. Third, we add a second measure of indebtedness, DEBTtoCAP, which is obtained from S&P Compustat Global. This third set of firms excludes financial firms, and other firms that appear in the EMDB but not in Compustat Global. For this reason, the sample size for this exercise is smaller, with results reported in Columns C1 and C2. We find a number of commonalities across the two subperiods, in both the firm-level and the country-level coefficient estimates. At the country-level, we find that exchange rate exposure occurs through the channel of international trade. We find that a higher ratio of exports to GDP tends to have a positive effect on exchange rate exposure, while imports tend to have a negative effect. This result is consistent with the usual intuition that exporting firms gain and importing firms lose when the domestic currency depreciates. In the early sub-period, both coefficients are significant at the 10% level, except when the smaller sample, as reported in Column C1, is used. In the second sub-period, all the coefficients are significant at the 1% level. We also find that the coefficients on M2toGDP and FXtoDEBT are generally insignificant, with the exception of the coefficient on M2toGDP for the recent sub-period when only country-level variables are included (see Column A2). As we discuss above, M2toGDP may serve as a measure of a country’s level of financial development and may be related to the magnitude (but not the sign) of exchange rate exposure. In results not reported, we regress the absolute value of exposure on M2toGDP, and find that the coefficient is insignificant in the early sub-period, but becomes positive and significant (at the 1% level) in the recent sub-period. At the firm-level, we find that large firms tend to have less negative exchange rate exposure. But as we argue above, firm size should more plausibly be related to the magnitude, rather than the sign, of exchange rate exposure, as large firms are more likely to hedge and have a smaller exposure coefficient in absolute value. Indeed, in results not reported, when we regress the absolute value of exposure on ln(MARKETCAPj), we find the coefficient to be significantly negative at the 1% and 5% levels, respectively, in the early and recent sub-periods. The coefficient on TURNOVERj is negative and significant at the 1% critical value. This result implies that the less liquid firms are more favorably affected by an exchange rate depreciation. Liquidity risk is a well-known issue in emerg-

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

1359

Table 5 Exchange rate exposure: Firm- and country-level determinants Panel I. Early Sub-period (January 1st, 1999 to June 30th, 2002)

Panel II. Recent Sub-period (July 1st, 2002 to June 30th, 2006)

(A1)

(A2)

Ln(MARKETCAP) INVESTIBILITY TURNOVER IDEBTtoCAP

(B1)

(C1)

0.084| (.03) 0.011 (.182) 1.007| (.327) 0.279| (.067)

0.053 (.042) 0.199 (.226) 0.454 (.334) 0.225k (.094) 0.094| (.034) 0.984| (.225) 0.018| (.006) 0.054 (.029) 0.039 (.022) 0.208 (.569) 0.002 (.002) .262 605

DEBTtoCAP Per Capita GDP Foreign Debt GDP Exports GDP Imports GDP M2 GDP Foreign Reserves Foreign Debt

Centered R2 Sample size

0.877| (.154) 0.022| (.005) 0.042} (.023) 0.035} (.018) 0.143 (.446) 0.002 (.002) .144 931

0.757| (.148) 0.012| (.004) 0.029} (.018) 0.027} (.016) 0.531 (.365) 0.003 (.002) .175 917

0.428k (.168) 0.024| (.005) 0.088| (.024) 0.142| (.019) 0.013k (.006) 0.003 (.003) .176 644

(B2)

(C2)

0.088k (.039) 0.212 (.248) 0.732| (.196) 0.010 (.064)

0.058 (.049) 0.349 (.277) 0.409| (.158) 0.023 (.089) 0.032 (.038) 0.271 (.199) 0.022| (.005) 0.086| (.028) 0.120| (.022) 0.006 (.006) 0.002 (.003) .257 447

0.011 (.165) 0.021| (.004) 0.076| (.02) 0.105| (.016) 0.002 (.005) 0.001 (.002) .238 634

This table shows the weighted instrumental-variable estimates of the regression

b1;j ¼ an þ cXj þ wYm þ /j ; where b1,j is our consistent estimate of exchange rate exposure of firm j from the GMM-IV estimate of Eq. (1), an is a sector-specific intercept, Xj are the firm-level variables, and Ym are the country-level variables. Both Xj (in block letters) and Ym (in italics) are listed on the first column of the table. Regressions (A1)–(C1) are based on the early sub-period. Regressions (A2)–(C2) are based on the recent sub-period. We weight each observation by the inverse of the standard error of b1,j in the first-stage estimation. Heteroskedasticity-consistent standard errors are in parentheses. |,k, and } denote statistical significance at the 1%, 5%, and 10% levels, respectively.

ing markets.7 As noted above, an exchange rate movement is a combination of a shock and a monetary-policy response. An exchange rate depreciation may be avoided when the domestic monetary authority responds to an external shock by raising the domestic interest rate. The ability of market makers to provide liquidity to domestic markets depends on the cost of domestic funds (see Brunnermeier and Pedersen, 2007, for a full model, and Chordia et al., 2005; Coughenour and Saad, 2004, for empirical evidence on the importance of funding liquidity for market liquidity). Therefore, illiquid shares that depend more on these market makers may be hurt the most by an increase in the domestic interest rate, and benefit the most from the alternative, an exchange rate depreciation. This channel explains why turnover and exchange rate exposure is negatively related. Finally, we find that INVESTIBILITYj is not significantly associated with exchange rate exposure in either sub-period, and TURNOVERj remains significant regardless of whether INVESTIBILITYj is included.

Notwithstanding the commonalities we document above, there are important differences between the two sub-periods, with respect to the impact of external debt on exchange rate exposure. During the early sub-period, firms with more international debt and firms in countries with a larger amount of external debt have more negative exchange rate exposure. In the recent sub-period, this effect disappears and even reverses. In Column B1, we can see that the coefficient on IDEBTtoCAPj is negative and statistically significant at the 1% level. Quantitatively, a onestandard deviation increase in IDEBTtoCAPj in 1999 makes exchange rate exposure more negative by approximately 0.25. This magnitude is quantitatively significant, since the value-weighted exchange rate exposure is 0.85. This finding is consistent with the idea that, holding the value of assets constant, a depreciation damages the balance sheets of firms with foreign-currency debt.8 We also find that, at the national level, external debt is associated 8

7

For a set of emerging market firms similar to those we consider here and using ‘‘zero return” as a measure of non-trading, Bekaert et al. (2007) found that the proportion of daily zero returns out of all daily return observations equaled 48.5%.

In a previous version of the paper, we focus on the early sub-period and the negative impact of international debt on exchange rate exposure. We show that this negative impact is robust to the inclusion of such firmlevel balance sheet variables as the share of debt that is short-term, the current ratio, and the fraction of assets that is physical capital.

1360

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

with negative exchange rate exposure. The coefficient on DEBTtoGDP is negative and significant at the 1% critical value. If exchange rate depreciations have negative effects on firms with foreign debt, the negative macroeconomic effects of an exchange rate depreciation in a country with many such firms should spill-over onto other firms, regardless of their own debt levels. However, during the recent sub-period, the attitude of equity markets toward both firm-level and country-level debt changes dramatically. We find that the coefficients on IDEBTtoCAPj and DEBTtoCAPj are insignificant at any reasonable critical value, while the coefficient on DEBTtoGDP becomes positive and significant at the 1% level. This latter result implies that firms in countries with a large amount of external debt benefit from exchange rate depreciations. It should be noted that IDEBTtoCAPj is measured during the early sub-period and so may more accurately reflect the level of debt holdings during that period. To alleviate this concern, we have experimented with alternative measures, such as dividing IDEBT by MARKETCAP from the recent sub-period or using an ordinal ranking of IDEBTtoCAP. Both alternative measures generate the same results as the original measure. Comparing the results reported in Columns C1 and C2, we see that our second firm-level measure of leverage, DEBTtoCAP, points to a similar conclusion. In the early sub-period, the coefficient on DEBTtoCAP is negative and significant at the 1% critical value. In the recent sub-period, this coefficient becomes insignificant. Without more detailed information of the currency denomination of each firm’s and each country’s outstanding debt, and the fraction of which that is hedged, we are unable to pin down exactly why the effect of debt on exchange rate exposure changed the way it did across the two sub-periods. However, we argue that our findings are consistent with the fact that both the domesticcurrency bond markets and the foreign exchange derivatives markets become better developed in recent years. With better-developed derivatives markets, foreign-currency liabilities can be more effectively hedged. When firms borrow more externally in the domestic currency, an emerging market’s monetary authority is more inclined to keep interest rates low (and let the domestic currency depreciate) in response to world market shocks. We also find that, in the early sub-period, exchange rate exposure is most negative among the richest countries – the coefficient on the log of per capita GDP is negative and significant at the 1% critical value. In the recent sub-period, the coefficient is positive and significant when we include country-level variables only, and becomes insignificant after we control for firm-level characteristics. This finding is consistent with studies which report that, in recent years, richer emerging markets tend to have better-developed domestic bond markets (see Burger and Warnock, 2006). Faced with world financial market shocks, countries with more domestic-currency bonds outstanding may find it more beneficial to keep interest rates low and allow the local currency to depreciate.

We also test whether the exchange rate exposure coefficients are different across sectors. Using the specification with firm- and country-level variables for the early sub-period (reported in Column B1), we fail to reject the hypothesis that all sectoral dummies being equal at a 10% critical value. For the recent sub-period (reported in Column B2), we are able to reject the hypothesis at the 1% critical value. This latter result is driven by the fact that in this sub-period, agricultural firms have significantly more negative, and mining firms have significantly more positive, exchange rate exposure. We are unable to reject the hypothesis that the sector dummies are the same at the 10% level for the remaining sectors. There are fewer than 30 mining firms and fewer than 10 agricultural firms in our sample. Firm-level variables and country-level variables each explain substantial amounts of the variation in exposure. A simple OLS regression of b1,j on sector dummies, a simple OLS regression of b1,j on the country-specific variables, and a reduced-form regression of b1,j on the firm-level instruments have adjusted R2 of 0.012, 0.159, and 0.090, respectively. Including all variables, a simple OLS regression has an adjusted R2 of 0.229. These results show that country-level variables capture the bulk of all explained variations in exposure, but firm-level variables have significant contributions as well. A significant fraction of firms in our early sample does not survive into the recent sub-period (see Table 5). To make sure that our different results across the two sub-periods are not driven by these firms, we repeat our first-period analysis, focusing only on those firms that survive into the second period. The results of this exercise are essentially the same as those we report above for the full sample of firms. In particular, in the first sub-period, the value-weighted exchange rate exposure of these firms is 0.75%, and 8.7% of them have negative and significant exposure at the 5% level. (From Table 3, Panel I, we see that the corresponding numbers for the full sample of firms are 0.85% and 8.92%, respectively.) At the same time, international debt at both the national and firm-level remains significant determinants of exposure at the 1% critical value, with more indebted firms and firms located in countries with more international debt having more negative exposure.9

5. Conclusion In this paper, we use an instrumental-variable technique to identify the total exposure of emerging market equities to exchange rate changes. This total exposure includes both 9

Also, there may be reasons to suspect that the results for Korea and Taiwan are different from those for the rest of the countries in our sample, as Korea and Taiwan can arguably be classified as ‘‘developed”. To address this concern, we repeat our second-stage analysis, but with all Korean and Taiwanese firms excluded. We find that our results regarding the determinants of exchange rate exposure are not sensitive to the inclusion of Korea and Taiwan.

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362 Table 6 The changing debt market Foreign-currency bonds to domestic-currency bonds

Brazil Chile Colombia India Indonesia Korea Mexico Pakistan Philippines South Africa Taiwan Thailand Turkey Venezuela

March (1999)

June (2006)

0.25 0.18 0.54 0.07 0.94 0.65 2.92 0.05 0.21 0.02 0.05 0.42 0.35 0.92

0.12 0.37 0.37 0.05 0.12 0.30 2.07 0.05 0.32 0.02 0.09 0.12 0.23 0.16

The table reports the ratio of international debt securities to domestic debt securities in March 1999 and June 2006, as reported by the Bank for International Settlements (2007).

the exchange rate exposure at the national level and the firm-level differences within a country. Using this instrumental-variable approach, we detect changes in the exchange rate exposure of emerging market stock returns over time. In the early sub-period of January 1999–June 2002, immediately following the emerging market currency crises of the 1990s, emerging market firms are mostly negatively exposed to exchange rate changes. In the recent subperiod of July 2002–June 2006, this negative exposure disappears. The traditional approach of estimating marginal exposure (by controlling for the national stock market return) would not be able to discover this broad, marketwide change over time. We go on to investigate the country-level and firm-level determinants of exposure. We find that in the early subperiod, there is a negative relationship between a company’s exchange rate exposure and (1) its level of international, foreign-currency debt; (2) its level of total debt; and (3) the level of external debt of the country in which the company is located. By contrast, in the recent sub-period, the relationship between debt and negative exchange rate exposure, at both the firm and country level, disappears and even reverses sign. The standard methodology used by previous studies would not have revealed the changing role of national debt levels in determining exchange rate exposure. This decline in negative exchange rate exposure coincides with a changing structure in many emerging markets’ debt. Eichengreen et al. (2006) outlined some of the structural changes being made in the East Asian and Latin American bond markets. In many countries, we observe very rapid growth in the market for domestic securities. McCauley and Jiang (2004) reported that, since the Asian crisis, Asia’s local currency bond markets have grown larger than its foreign-currency counterpart. At the same time, as documented by Jeanneau and Tovar (2006), there

1361

is a rapid growth in the Latin American domestic debt markets, which also exhibit lengthened maturity and a lower prevalence of currency- and inflation-indexed debt.10 Table 6 presents the ratio of outstanding foreign-currency bonds to domestic-currency bonds for all countries in our sample (except Morocco), in the first quarter of 1999 and in the second quarter of 2006. In 9 out of the 14 countries, there is a noticeable decline in the size of the international bond market relative to the domestic bond market. Only in three countries is there a substantial increase in the relative size of the foreign bond market. More broadly, data from the Bank for International Settlements (BIS) show that, in June 2002, the notional value of OTC exchange rate derivatives in currencies outside the top 13 largest currencies was equivalent to 2.2 trillion. By June 2006, this sum had increased to 6.7 trillion. Over the same period, interest-rate securities in currencies outside the top 12 currencies increased from 2.3 trillion US dollars to more than 8 trillion. Although we cannot break down this data further by currency, the rapid growth in derivatives markets may have increased the opportunities for emerging market firms to hedge their exchange rate risks. Acknowledgements We thank two anonymous referees, Kalok Chan, Lewis Chan, Leonard Cheng, Gregor Smith, Giorgio Szego¨ (the editor), and various seminar participants for helpful comments, and Anu Chavali for valuable research assistance. Cook acknowledges the research support from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. HKUST6213/00H). This paper was reviewed and accepted while Prof. Giorgio Szego¨ was the Managing Editor of The Journal of Banking and Finance and by the past Editorial Board. References Adler, M., Dumas, B., 1980. The exposure of long-term foreign currency bonds. Journal of Financial and Quantitative Analysis 15, 973–994. Adler, M., Dumas, B., 1984. Exposure to currency risk: Definition and measurement. Financial Management 13, 41–50. Aghion, P., Bacchetta, P., Bannerjee, A., 2001. Currency crises and monetary policy in an economy with credit constraints. European Economic Review 45, 1121–1150. Aguiar, M., 2005. Investment, devaluation, and foreign currency exposure: The case of Mexico. Journal of Development Economics 78, 95– 113. Allayannis, G., Ofek, E., 2001. Exchange rates exposure, hedging, and the use of foreign currency derivatives. Journal of International Money and Finance 20, 273–296. Allayannis, G., Brown, G., Klapper, L., 2003. Capital structure and financial risk: Evidence from foreign debt use in East Asia. Journal of Finance 58, 2667–2709. Allen, F., Gale, D., 2000. Optimal currency crises. Carnegie-Rochester Conference Series on Public Policy 53, 177–230. 10

Eichengreen et al. (2006), McCauley and Jiang (2004), and Jeanneau and Tovar (2006) also noted a continuing lack of liquidity in developing countries’ bond markets.

1362

T.K. Chue, D. Cook / Journal of Banking & Finance 32 (2008) 1349–1362

Anderson, T.W., Rubin, H., 1949. Estimation of the parameters of a single equation in a complete system of stochastic equations. Annals of Mathematical Statistics 20, 46–63. Bank for International Settlements, 2007. Statistical tables, BIS Quarterly Review, September. Bekaert, G., Harvey, C.R., Lundblad, C., 2007. Liquidity and expected returns: Lessons from emerging markets. Review of Financial Studies 20, 1783–1831. Bodnar, G., Dumas, B., Marston, R.C., 2002. Pass-through and exposure. Journal of Finance 57, 199–231. Bodnar, G., Gentry, W., 1993. Exchange rate exposure and industry characteristics: Evidence from Canada, Japan, and the US. Journal of International Money and Finance 12, 29–45. Bodnar, G., Wong, F., 2003. Estimating exchange rate exposures: Issues in model structure. Financial Management 32, 35–67. Brunnermeier, M.K., Pedersen, L.H. 2007. Market liquidity and funding liquidity. Review of Financial Studies, forthcoming. Burger, J.D., Warnock, F.E., 2006. Local currency bond markets. IMF Staff Papers 53, 133–146. Caballero, R.J., Krishnamurthy, A., 2003. Excessive dollar debt: Financial development and underinsurance. Journal of Finance 58, 867–893. Calvo, G., 1996. Money, exchange rates, and output. MIT Press, Cambridge. Calvo, G., 2002. On dollarization. Economics of Transition 10, 393–403. Calvo, G., Guidotti, P., 1990. Indexation and maturity of government bonds: An exploratory model. In: Dornbusch, R., Draghi, M. (Eds.), Capital Markets and Debt Management. Cambridge University Press, Cambridge. Chordia, T., Sarkar, A., Subrahmanyam, A., 2005. An empirical analysis of stock and bond market liquidity. Review of Financial Studies 18, 85–129. Coughenour, J.F., Saad, M.M., 2004. Common market makers and commonality in liquidity. Journal of Financial Economics 73, 37–69. Cragg, J.G., Donald, S.G., 1993. Testing identifiability and specification in instrumental variables models. Econometric Theory 9, 222–240. Doidge, C., Griffin, J.M., Williamson, R., 2006. Measuring the economic importance of exchange rate exposure. Journal of Empirical Finance 13, 550–576. Dominguez, K.M.E., Tesar, L.L., 2006. Exchange rate exposure. Journal of International Economics 68, 188–218. Dumas, B., 1978. The theory of the trading firm revisited. Journal of Finance 33, 1019–1030.

Eichengreen, B., Borensztein, E., Panizza, U. 2006. A tale of two markets: Bond market development in East Asia and Latin America. Hong Kong Institute for Monetary Research, Occasional Paper. Griffin, J.M., Stulz, R.M., 2001. International competition and exchange rate shocks: A cross-country industry analysis of stock returns. Review of Financial Studies 14, 215–241. Hausmann, R., Panizza, U., Stein, E., 2001. Why do countries float the way they float? Journal of Development Economics 66, 387–414. He, J., Ng, L., 1998. Foreign exchange exposure, risk, and the Japanese stock market. Journal of Finance 53, 733–753. Hodder, J., 1982. Exposure to exchange rate movements. Journal of International Economics 29, 217–236. Jeanneau, S., Tovar, C.E., 2006. Domestic bond markets in Latin America: Achievements and challenges. BIS Quarterly Review, 51–64. Jorion, P., 1990. The exchange rate exposure of the US multinationals. Journal of Business 63, 331–345. Jorion, P., 1991. The pricing of exchange rate risk in the stock market. Journal of Financial and Quantitative Analysis 26, 363–376. Jorion, P., Schwartz, E., 1986. Integration versus segmentation in the Canadian stock market. Journal of Finance 41, 603–613. Kho, B., Stulz, R.M., 2000. Banks, the IMF, and the Asian crisis. PacificBasin Finance Journal 8, 177–216. Lane, P., Milesi-Feretti, G., 2001. The external wealth of nations: Measures of foreign assets and liabilities for industrial and developing nations. Journal of International Economics 55, 263–294. McCauley, R., Jiang, G., 2004. Diversifying with Asian local currency bonds. BIS Quarterly Review, 51–66. Mittoo, U., 1992. Additional evidence on integration in the Canadian stock market. Journal of Finance 47, 2035–2054. Morck, R., Yeung, B., Yu, W., 2000. The information content of stock markets: Why do emerging markets have synchronous stock price movements? Journal of Financial Economics 58, 215–260. Parsley, D., Popper, H., 2006. Exchange rate pegs and exchange rate exposure in East and South East Asia. Journal of International Money and Finance 25, 992–1009. Staiger, D., Stock, J.H., 1997. Instrumental variables regression with weak instruments. Econometrica 65, 557–586. Williamson, R., 2001. Exchange rate exposure and competition: Evidence from the automotive industry. Journal of Financial Economics 59, 441–475.