Devaluations and emerging stock market returns

Devaluations and emerging stock market returns

Emerging Markets Review 3 (2002) 409–428 Devaluations and emerging stock market returns Jack Glen* International Finance Corporation, 2121 Pennsylvan...

181KB Sizes 2 Downloads 93 Views

Emerging Markets Review 3 (2002) 409–428

Devaluations and emerging stock market returns Jack Glen* International Finance Corporation, 2121 Pennsylvania Avenue, NW, Washington, DC 20433, USA Received 31 January 2002; accepted 30 June 2002

Abstract Stock returns over the 2 years surrounding 24 currency devaluations are examined. Using bootstrapped distributions, returns preceding the devaluation are shown to be significantly below normal, in both dollar and local currency terms. Most of the downturn, however, occurs well before the month of the devaluation. Returns following a devaluation are normal. While industry and company specific effects appear to influence return behavior, only country effects and leverage levels are statistically significant. At the country level, both aggregate economic activity (GDP) and the size of the devaluation are important in explaining return behavior. The stock of foreign debt has little impact on returns. Finally, even though returns appear to anticipate devaluations, they are not statistically significant at predicting the size of the devaluation. 䊚 2002 Published by Elsevier Science B.V. Keywords: Stock market; Currency

1. Introduction Large and discrete currency movements are not uncommon events in many emerging markets.1 When and why such events occur has been hotly debated among monetary and macroeconomists for many years.2 Exactly what impact these events have on the real and financial sectors of the underlying economies is less well studied. *Tel.: q1-202-473-8641; fax: q1-202-974-4367. E-mail address: [email protected] (J. Glen). 1 IMF (2002, Box 3.3, page 113) presents an overview of emerging market crises, including both currency and banking crises, and provides ample references to the literature on these events. 2 Dornbusch (2001) provides a non-technical overview of emerging market crises. Kamin et al. (2001) look at crises empirically to identify the factors, both internal and external, that predict when crises occur. 1566-0141/02/$ - see front matter 䊚 2002 Published by Elsevier Science B.V. PII: S 1 5 6 6 - 0 1 4 1 Ž 0 2 . 0 0 0 4 4 - 4

410

J. Glen / Emerging Markets Review 3 (2002) 409–428

Conceptually, a devaluation should both increase inflation and stimulate economic activity through exports. In addition, by relaxing the constraints imposed on the monetary authorities, interest rates can be relaxed and liquidity increased. But these results are not always the case and devaluations are often followed by high interest rates and poor economic performance. Over the last two decades a number of emerging stock markets have developed as viable investment alternatives for international investors. While one can debate the various merits of these markets, one thing many of them have in common is exposure to the risk of a currency devaluation. For the foreign investor this risk is obvious because stocks are priced in local currency and any exchange rate change will pass through to stock returns. However, stock prices themselves reflect the value of the real assets underlying the shares and the impact of currency changes on these real assets is uncertain.3 This paper looks at stock market performance over a sample of 24 devaluation events covering 18 emerging market countries over the period 1980–1999. The analysis compares stock market performance before and after the devaluation event with the general distribution of returns for these markets. The paper also examines several factors that differentiate the performance of the markets. The findings suggest that devaluations have a significant negative impact on returns, both in dollar and local currency terms, in the months preceding the event. This impact is statistically significant on average, but there is also considerable variation across events and countries. The analysis suggests that market reaction following an event is closely tied to economic recovery and a more relaxed monetary environment. The remainder of the paper is organized as follows: Section 2 describes the country event data; Section 3 looks at the behavior of individual stocks during these events and Section 4 examines the links between country market behavior and the macroeconomy. Sections 5 and 6 examine the ability of stock markets to predict devaluation size and longer-term behavior of markets prior to currency events, respectively. Conclusions and a summary are contained in the final section. 2. Country market indices The data used in this study consist of stock market returns for 18 countries from the Emerging Markets Database over most of the life of the database: 1980–2001. The database also includes exchange rate data and from that data 24 devaluation events were identified, where an event is defined as a significant change in value relative to the US dollar. Not all of these events involved moving from a fixed peg to a new higher fixed value. In some cases the currencies were maintained in a controlled float, often with a predetermined rate of depreciation. Also, in some cases the currency regime that followed the devaluation was one of market determination. 3 Research on the impact of currency exposure on stock returns is well developed, but addresses a different topic than is addressed here. For the impact of hedging currency exposure in stock portfolios see Glen and Jorion (1993); for measuring currency exposure see Bodnar and Franco Wong (2000).

J. Glen / Emerging Markets Review 3 (2002) 409–428

411

Table 1 Summary statistics

Argentina Brazil Chile Greece India Indonesia Jordan Korea Korea Malaysia Mexico Mexico Nigeria Pakistan Philippines Russia Thailand Thailand Thailand Turkey Turkey Venezuela Venezuela Zimbabwe

Event date

XC

LC STK

USD STK

January 1991 January 1999 June 1982 November 1980 July 1991 July 1997 October 1988 January 1980 October 1997 July 1997 December 1982 December 1994 March 1995 June 1998 July 1997 August 1998 July 1981 November 1984 July 1997 February 1991 April 1994 December 1985 November 1995 December 1982

(0.40) (0.41) (0.16) (0.05) (0.18) (0.07) (0.15) (0.17) (0.05) (0.04) (0.48) (0.30) (0.73) (0.04) (0.09) (0.37) (0.09) (0.15) (0.19) (0.09) (0.35) (0.48) (0.52) (0.17)

0.72 0.19 0.06 (0.03) 0.26 (0.01) 0.15 0.02 0.26) (0.05) 0.01 (0.07) 0.08 (0.23) (0.06) (0.41) (0.00) 0.06 0.33 0.13 0.05 (0.04) 0.13 0.02

0.04 (0.30) (0.11) (0.08) 0.03 (0.08) (0.02) (0.15) (0.30) (0.09) (0.47) (0.35) (0.71) (0.26) (0.15) (0.62) (0.09) (0.10) 0.08 0.03 (0.32) (0.50) (0.46) (0.15)

Event data is the date on which the devaluation event is considered to have taken place and is the basis for the return calculations. XC is the percentage change in the value of the local currency relative to the dollar. LC STK is the change in the local currency stock price index during the month of the event, expressed in decimal form. USD STK is the change in the dollar stock price index during the month of the event, expressed in decimal form. Figures in parentheses are negative.

For these reasons, precisely defining the events is difficult, but the events are obvious from the data. A list of the 24 events is presented in Table 1. Most of the countries in the sample experienced a single event, but four countries experienced two events and one country—Thailand—experienced three devaluations. The magnitude of the events, which is measured relative to the dollar, has great variation, ranging from a low of 4% (Malaysia 1997) to a high of 73% (Nigeria 1995). Note, however, that these single period change percentages can be deceiving. In the case of Indonesia (1997), although the change during the defined event month was only 7%, the cumulative currency depreciation from the 6 months preceding the event to 18 months following the event was 73%. For that reason, looking only at a single point in time may be misleading. In what follows, the analysis considers a longer period of time from the period 6 months before the defined event month to a full 18 months after the event. This longer time frame allows one to consider the full impact of the event on both the currency and the stock market.

412

J. Glen / Emerging Markets Review 3 (2002) 409–428

Much of what follows will consider the average performance of the group of markets presented in Table 1. As a first example, consider Fig. 1, which presents the average currency movement over the 24-month window surrounding the devaluation events. In this case and in all analysis that follows, time 0 is taken to be the month for each country event presented in Table 1. Returns and currency values are normalized to 100 for that date and the analysis considers changes in the normalized index for returns over the 24-month window. In Fig. 1 we see that the average currency value across this group of countries declined in the 5 months preceding the devaluation event, and then dropped by approximately 30% during the month of the event. Following the event the currencies continued to depreciate, on average losing an additional 26% in the subsequent 18 months. Note that these are nominal values. The remainder of the paper examines the stock market reaction to these events.4 Table 1 also presents 1-month stock index changes over the month of the event. It provides a first glimpse of how varied market reactions are to currency events. Looking at the local currency stock price effect, only 10 of the 24 events are accompanied by a negative price impact in local currency terms in the stock market. Moreover, in some cases the stock price reaction is both large and positive; e.g. in Argentina in January 1991 the stock price index increased by 72% during the same month that the currency was devalued by 40% reflecting the fact that the devaluation was part of a program of broad economic reforms that ultimately led to several years of vibrant economic growth. Of course, for international investors the combined impact of the currency change and the price impact is important. In terms of US dollars, Table 1 shows that the stock returns are negative in 20 out of 24 events. Importantly, given the evidence from the local currency returns in the table, most of the negative dollar returns reflect currency movements only, suggesting that any stock price movements associated with the currency event occurred before the event dates on which the table is based, which we will see in the next figure. Fig. 2 presents a longer-term view of the stock market reaction to events surrounding devaluation events. The figure presents the average total return index (across the 24 events) for the 24-month window surrounding each event. Return indices are presented in both dollars and local currency. The difference between the two indices is striking. The local currency index declines in the 6 months preceding the event, but most of that decline is in the second and third month prior to the event; returns are positive (nearly 1%) in the month of the event after having dropped nearly 13% in the previous 5 months (the price index dropped by over 14%). Also, despite a small drop in the first month following the event, returns then turn up in all subsequent months. In contrast, dollar returns experience a sharp drop in the month of the event—the currency effect—but they too turn up after a single month and continue positive for nearly all of the subsequent 18 months. One important difference between the two is in the final level of the index. For the local 4 There is a massive academic literature on exchange rate dynamics that would help one to understand the behavior exhibited in Fig. 1. A good place to start is Levich (1988).

J. Glen / Emerging Markets Review 3 (2002) 409–428

Fig. 1. Mean currency value across countries. Months before and after devaluation, total of 24 events.

413

414

J. Glen / Emerging Markets Review 3 (2002) 409–428

Fig. 2. Cross-country average, total return stock index. Local currency and US$, months before and after event, normalized to 100 on date of devaluation, 24 total events.

currency return index, the average level at month tq18 is 220, more than 100% above its level on the event date month and also 94% above its level 6 months prior to the event. For the dollar return index, the ending value of the index is 54% above the level on the month of the event, but it remains 6% below its level of 6 months prior to the event.5 The reasons behind the movements displayed in Fig. 2 are something we need to try to understand better. First, why do the markets turn down well before the devaluation event? Most likely, this represents the ability of the market to respond quickly to the unfavorable economic environment—as characterized by slowing growth and tight monetary policy—that precedes a devaluation event. This is not the same as forecasting the time and size of a devaluation, but does reflect the importance of the economic environment on stock prices and, perhaps, the uncertainty associated with the currency. Second, the market reaction to the devaluation suggests that, on average, the event is associated with an improvement in the economic environment, perhaps either a return to growth or a relaxation of monetary policy, or both. More formal measurement of economic performance before and 5 The local currency returns in Fig. 2 are nominal and adjusting for inflation makes a significant difference. Using annual rates of inflation, which are admittedly crude, and monthly compounding, the real local currency return index starts the period at a value of 126, which is slightly above the nominal value. Normalized to 100 at ts0, the real local currency index climbs to an end value of only 138, compared to 220 for the nominal index. Overall, the pattern of returns is the same for the nominal and real local currency indices, but inflation eats away much of the returns reported in local currency.

J. Glen / Emerging Markets Review 3 (2002) 409–428

415

Table 2 Returns: months before and after devaluation Local currency

Dollar

Mean

S.D.

Mean

S.D.

Price index y6 months q6 months q12 months q18 months y6 to q18 months

y0.010*** 0.283 0.639 1.086 1.051

0.316 0.620 1.308 1.967 1.885

0.303*** 0.046 0.264 0.454 y0.090***

0.225 0.534 1.082 1.121 0.769

Total return index y6 months q6 months q12 months q18 months y6 to q18 months

0.019*** 0.316 0.703 1.196 1.224

0.322 0.635 1.325 1.987 1.986

y0.282*** 0.075 0.318 0.541 y0.009***

0.233 0.554 1.101 1.156 0.784

Mean is the average across the 24 events; S.D. is the standard deviation of the average returns across the 24 events. *** Below the 1% level of the empirical distribution of returns.

after currency events would be a welcome addition to our knowledge on this subject and would help us to understand better the reasons behind the stock price movements displayed in the figure. By presenting only the average across events, Fig. 2 hides the considerable variation observed in the 1-month returns in Table 1. Table 2 reveals this crosssectional variation in the form of the cross-sectional standard deviation for the set of country event returns that will be used henceforth to examine market reactions to these events. The table contains local currency and dollar returns for five different holding periods: 6 months prior to and including an event, 6 months following an event, 12 months following an event, 18 months following an event and the 24month window that encompasses these other holding periods. The table provides returns based on both price indices and total return indices to allow examination of both price and dividend effects. The table formalizes what is observed in Fig. 2. Mean returns are negative for the 6 months preceding and including an event (except for the local currency total return index). These are statistically significant at the 1% level in each case, where significance is tested using an empirical distribution of returns for each of the holding periods.6 In the case of the 6-month returns preceding the events, those returns are in the lowest 1 percentile of the empirical distribution, suggesting that they are unusually low returns. Except for the dollar total returns for 24 months, 6 The empirical distribution is generated by randomly sampling from the 16 country return distributions over their entire sample (which varies across countries and is determined by the source: EMDB), then taking the average across the 24 random samples. This process is repeated 10 000 times using the statistical software S-PLUS. The resulting empirical distribution of returns is then used to calculate return percentiles, which identify which event returns are unusual.

416

J. Glen / Emerging Markets Review 3 (2002) 409–428

which are also low, all other returns are not extreme values in the empirical distribution. The dollar 24-month returns are significantly below what one would expect given the normal distribution of returns, but at less than y10% they are not as significant economically as one might expect. Moreover, one can see that, except for the period preceding the event, all other returns are positive. This may provide little comfort, however, upon consideration of the large cross-sectional variation, as evidenced by the standard deviation in the table. This variation, which increases with the holding period, is considerable and suggests that without a better understanding of what drives returns during the period surrounding these events, investing either before or after an event involves considerable risk. The next two sections examine factors that explain this cross-sectional variation. 3. Company stock returns There are a number of reasons to expect that the stock returns of different companies within a country would behave differently following a devaluation event. Exporters would benefit from the depreciated currency, as would import competitors. Companies with significant levels of foreign-currency debt would suffer from the revaluation of this debt in local-currency terms. And local service providers could suffer if the event were accompanied by a general economic downturn. This section attempts to explain the cross-sectional variation observed in individual stock returns during devaluation events using company-level stock-market data. From the countries and event dates in Table 1, data were compiled for a total of 955 companies that underlie the country indices used in the previous section. Those company stock returns are examined in what follows.7 Figs. 3 and 4 begin the process by looking at the average performance of indices created by combining individual company dollar price indices across countries after having sorted the companies by industry and size. Fig. 3 presents a set of three (randomly chosen) industry-based indices chosen to reflect the differences across different industries. In both figures, the price indices are dollar-based and so one common element is the sharp drop that occurs during the month of the event. Apart from that drop, however, there are notable differences in the behavior of the indices. In Fig. 3 mining recovers very rapidly from the event, even though it never does fully recover the initial level it had. In contrast, finance and services either remain flat or continue to decline for several months before any recovery is observed. Even after 18 months the finance index has just barely recovered to its event level, something the service index never achieves in the sample. Apparently, industry effects can be important, reflecting the exposure each of them has to the general 7 These companies represent all of the companies contained in the EMDB indices for the relevant time periods.

J. Glen / Emerging Markets Review 3 (2002) 409–428

417

Fig. 3. Devaluation impact: sectors. Average price index across companies within sector, months before and following devaluation.

economic environment that characterizes a currency event and to the currency event itself.8 Fig. 4 presents dollar price indices where companies have been sorted into three groups on the basis of market capitalization. Companies totaling less than $100 million are considered as small, companies of (more than $100 million and) less than $500 million are considered as medium, and all other companies are considered as large.9 Again all three indices exhibit the characteristic drop in dollar prices at the time of the event, but recovery differs depending on size. Small firms begin to recover quickly and end the period well above the initial price. Large firms continue down for several months before beginning a long and slow recovery. Overall, company size appears to influence return behavior. Perhaps this is because smaller firms have less currency exposure than medium- and large-size firms. Alternatively, small firm may simply be more nimble and therefore more able to respond quickly to these events. Finally, it is possible that this apparent size effect reflects correlations with country and industry, but that is not obvious from the data. 8 I examined a total of eight industries, which account for the bulk of the companies. Among those I present only three in the figure for ease of presentation. Among those not presented, manufacturing and infrastructure had patterns most similar to mining (although infrastructure had a much larger decline preceding the event), agriculture behaved in a similar manner to services, whereas the trade and construction sectors looked more like finance. 9 These cut-off points were chosen to more or less divide the sample into three equal subsamples. Each subsample has slightly more than 300 companies.

418

J. Glen / Emerging Markets Review 3 (2002) 409–428

Fig. 4. Devaluations and MCAP. Small -$100 m; medium -$500 m, US$ total return index, months before and after devaluation.

Both of these figures present interesting pictures of the possible importance of firm characteristics on return behavior before and after devaluation events. But neither of these pictures provides a very convincing statistical analysis of the issue because they do not test for statistical significance of the effects, nor do they control for other possible contributing factors. To attempt to address these issues, Table 3 presents results from a regression of individual company returns over the holding periods described in Table 2. The independent variables in the regression are log(market capitalization) (MCAP), trading volumeymarket capitalization (volumey MCAP) and priceybook value (PBV). These are three readily available company variables that have been used widely to explain stock return behavior.10 In addition, the regression includes dummy variables for industries and countries.11 The results in Table 3 are striking. For the local currency returns only a single coefficient on the three independent variables is significant at the 5% level. There is somewhat more success with the dollar returns, but these three factors have little to say about how returns react to devaluation events. Also note that industry effects are relatively unimportant; only one or two of the industry dummies out of five included in each regression are statistically significant. Instead, what the regressions reveal is the importance of country effects. Not only are many of the country 10 Fama and French (1992) provide an analysis of the US stock market that employs both size and marketybook ratios to explain the behavior of stock returns; they also provide numerous references to the literature. Amihud and Mendelson (1991) examine the implications of liquidity (volume) on returns. 11 The industry dummies included are: agriculture, financial, infrastructure, manufacturing and services.

J. Glen / Emerging Markets Review 3 (2002) 409–428

419

Table 3 Company-level return regressions

Local currency y6 months q6 months q12 months q18 months 24 months USD y6 months q6 months q12 months q18 months 24 months

Intercept

MCAP

Volume

PBV

R2

Country

Industry

0.383** (0.096) 1.227** (0.189) 2.949** (0.429) 2.970** (0.650) 4.684** (0.617)

y0.009 (0.042) 0.098 (0.083) y0.366* (0.189) y0.319 (0.286) y0.357 (0.272)

y0.297** (0.107) y0.204 (0.209) y0.309 (0.475) y.0937 (0.719) y1.195* (0.683)

0.002* (0.001) 0.001 (0.002) y0.001 (0.005) y0.000 (0.007) 0.000 (0.007)

0.51

12

1

0.57

12

2

0.59

14

0

0.50

12

0

0.50

13

0

y0.330** (0.060) 0.732** (0.139) 1.859** (0.326) 1.853** (0.412) 0.985** (0.265)

y0.007 (0.006) 0.017 (0.013) y0.058* (0.033) y0.063 (0.041) y0.049* (0.027)

y0.259** (0.079) y0.078 (0.170) y0.211 (0.431) y0.775 (0.546) y0.804** (0.351)

0.002** (0.001) 0.000 (0.002) y0.001 (0.004) y0.000 (0.005) 0.001 (0.004)

0.51

10

1

0.43

13

2

0.61

13

0

0.47

13

0

0.48

13

1

This table reports regression results for company-level data. The independent variable is either localcurrency or USD returns. The independent variables include size (Market capitalization (MCAP)), liquidity (volume tradedyMCAP (Volume)) and price book value ratio (PBV). Country event and sector dummy variables are included, those dummies are not reported, but the number of significant (5%) coefficients is reported. * Significant at the 10% level. ** Significant at the 5% level.

dummy coefficients significant in each regression, but the level of significance is generally at the 1% level or less. Moreover, the country dummies account for the bulk of the explained variation in the regressions, accounting for 47% of the total of 48% of the R-squared in the dollar regression for a 24-month holding period. Missing from the model in Table 3 is any measure of the leverage of the individual companies. Leverage can have devastating effects during crises both through the high interest rates that usually prevail, as well as through any currency impact if the debt is foreign-currency denominated. Unfortunately, company level data for these markets and time periods are not readily available. Using Worldscope data, however, I was able to compile leverage ratios (total liabilitiesytotal assets) for a set of companies in Brazil, Malaysia and Thailand (1997). The results in Table 4 include this measure of leverage in the regression.12 12

23.

The number of companies in each of the 3 countries are: Brazil—21; Malaysia—85; Thailand—

420

J. Glen / Emerging Markets Review 3 (2002) 409–428

Table 4 Company-level regressions

Local currency y6 months q6 months q12 months q18 months 24 months USD y6 months q6 months q12 months q18 months 24 months

Intercept

MCAP

Volume

PBV

Leverage

R2

y0.068 (0.123) 0.137 (0.109) y0.235 (0.173) 0.010 (0.241) y0.256 (0.216)

0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000)

0.128 (0.404) y0.376 (0.359) y0.430 (0.569) y0.297 (0.793) y0.359 (0.713)

0.001 (0.001) 0.000 (0.001) y0.000 (0.002) y0.000 (0.003) y0.000 (0.002)

0.003 (0.146) y0.459*** (0.130) y0.339 (0.206) y0.450 (0.287) y0.433 (0.258)

0.09

y0.252*** (0.095) y0.307*** (0.089) y0.417** (0.185) y0.117 (0.273) y0.331** (0.142)

0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000)

0.094 (0.314) y0.171 (0.296) y0.220 (0.601) y0.129 (0.900) y0.236 (0.469)

0.002 (0.001) y0.000 (0.001) y0.000 (0.002) y0.000 (0.003) y0.000 (0.002)

0.022 (0.114) y0.312*** (0.107) y0.225 (0.221) y0.385 (0.325) y0.283* (0.170)

0.28

0.73 0.80 0.71 0.59

0.85 0.83 0.75 0.58

This table reports regression results for company-level data for Brazil, Malaysia and Thailand (1997) only. The independent variable is either local currency or dollar returns. The dependent variables include size (Market capitalization (MCAP)), liquidity (volume tradedyMCAP (Volume)), price book value ratio (PBV) and leverage (Total liabilitiesytotal assets). Country and sector dummy variables are included, those dummies are not reported. * Significant at the 10% level. ** Significant at the 5% level. *** Significant at the 1% level.

Comparisons between Tables 3 and 4 are difficult because of the relatively small number of companies included in Table 4. What stands out in Table 4 is the significance of the leverage variable in explaining returns during the 6 months following a currency crisis. Companies with higher leverage have lower stock returns for 6 months, but the returns in all other periods are unaffected by leverage, although the values of the coefficients are similar across the various holding periods. These results suggest that the macroeconomic effects of a devaluation event are central to understanding stock return reactions. The next section returns to country indices to examine these effects in more detail. 4. Macroeconomics and the stock market Devaluation events both reflect the environment in which they occur, as well as set the stage for what is about to take place. In many cases, devaluations are the

J. Glen / Emerging Markets Review 3 (2002) 409–428

421

Fig. 5. Argentina: 1991:01. Stock market return indices, US$ and local currency.

culmination of a series of poor policy decisions and represent the only option for an embattled central bank. Without a change in the underlying policies, however, devaluations by themselves are unlikely to solve the problem. More likely, inflation and economic stagnation will follow. When devaluation is accompanied by improved policies, however, one can see miraculous results. Consider, e.g. Fig. 5, which presents the stock return index for Argentina during the period leading up to and following its January 1991 devaluation. That event was part of a broader package that included privatization, civil service reform and trade liberalization. Consequently, the economy performed remarkably during the following 2 years, which is reflected in the stock return index. In contrast consider Fig. 6, which presents the case of Indonesia and its devaluation event of 1997. With poor economic performance and considerable political uncertainty following the event, the stock market produced poor returns. This link between business conditions and market performance has also been explored by Fama and French, 1990 who report strong evidence that expected returns in both stocks and bonds are correlated with the business cycle. To understand the link between stock returns at the time of a devaluation event and the underlying economy better, consider the regression results presented in Table 5. In this case, the dependent variables are the holding period returns described in Table 2. The independent variables are the change in the GDP growth rate during the year in which the event occurs (dGDP0), as well as the GDP growth rate in the year of the event (GDP0) and the following year (GDP1). These three variables summarize the economy by looking at the extent of any collapse leading up to the event, as well as any recovery that follows. Not all of these are legitimate forecasting

422

J. Glen / Emerging Markets Review 3 (2002) 409–428

Fig. 6. Indonesia: 1997:07. Stock market return indices, US$ and local currency.

variables as they are not predetermined, but they do summarize the state of economy and this is clearly not a forecasting exercise. The regression results are notable for at least two reasons. First, with the exception of the 6-month returns prior to the event, the explained variation of the regressions (R 2) is quite large, generally on the order of 50% or more, which is relatively high for regressions involving stock returns. Second, again with the exception of the 6month prior returns, most of the regression coefficients are statistically significant. Interpretation of the coefficients presents a challenge. Generally the coefficient on the GDP1 variable is positive. For example, in the case of local currency returns at the 24-month holding period, a 1% increase in GDP in the year following an event results in a 9% increase in the 24-month holding period return. Similar interpretations can be given to the other two coefficients. In the case of GDP0 the coefficient is negative, suggesting that countries with the worst initial conditions in terms of growth subsequently produced the best returns, but the impact of that effect diminishes with holding period and disappears completely at the 24-month horizon for dollar returns. Conversely, the direction of change in GDP at the time of the event has a positive impact on returns, suggesting that countries which were starting to improve at the time of an event produced higher returns. While the results in Table 5 suggest that economic performance explains much of the cross-sectional behavior of stock returns during the period surrounding a devaluation, the results in Table 6 challenge that interpretation of events. In Table 6 the regressions of Table 5 are augmented by including the percentage change in

J. Glen / Emerging Markets Review 3 (2002) 409–428

423

Table 5 Local currency and USD returns regressed on GDP and change in GDP

Local currency y6 months q6 months q12 months q18 months 24 months USD y6 months q6 months q12 months q18 months 24 months

Intercept

GDP0

GDP1

dGDP0

R2

0.049 (0.102) 0.669** (0.139) 1.380** (0.284) 2.232** (0.489) 2.171** (0.419)

y0.008 (0.023) y0.114** (0.031) y0.187** (0.063) y0.335** (0.108) y0.260** (0.092)

0.005 (0.010) 0.033** (0.014) 0.078** (0.028) 0.090* (0.048) 0.089** (0.041)

0.018 (0.017) 0.085** (0.024) 0.184** (0.048) 0.254** (0.083) 0.285** (0.071)

0.11

y0.383** (0.069) 0.301** (0.111) 0.771** (0.213) 1.135** (0.209) 0.222 (0.162)

0.029* (0.015) y0.072** (0.024) y0.105** (0.047) y0.159** (0.046) y0.044 (0.036)

y0.006 (0.007) 0.041** (0.011) 0.074** (0.021) 0.067** (0.021) 0.040** (0.016)

y0.015 (0.012) 0.062** (0.019) 0.143** (0.036) 0.176** (0.036) 0.098** (0.028)

0.20

0.57 0.60 0.47 0.58

0.63 0.67 0.65 0.62

Regression coefficients with standard errors in parentheses. Returns are prior to (y) or following (q) the devaluation, except for 24 months, which is the 2 year period from 6 months prior to 18 months following the devaluation. GDP0, GDP growth rate in the year of the devaluation; GDP1, GDP growth rate in the year following the devaluation; dGDP0, change in GDP growth rate in the year of the devaluation from the year prior to the devaluation. * Significant at the 10% level. ** Significant at the 5% level.

the exchange rate during the month of the devaluation event. While this is not a predetermined variable for the 6-month returns, it is for all other holding periods. The results for the growth measures in Table 6 stand in stark contrast to those of Table 5. First, the level of GDP growth in the year of the event (GDP0) is no longer statistically significant, regardless of the holding period. GDP1 retains some significance, more for the dollar returns than for local currency return. Similarly, the change in GDP growth over the two years (dGDP0) loses some significance, especially for the local currency returns. In contrast to the growth measures, the change in the exchange rate (dLC) is significant at high levels in 6 of the 10 regressions and appears to work equally well for both local currency and dollar returns. Close examination of the results, however, reveals that the manner in which the size of the devaluation explains stock returns is complicated. First, note that all coefficients (except for USD—6 months) are negative. Given that the devaluation variable is defined as the percentage change

424

J. Glen / Emerging Markets Review 3 (2002) 409–428

Table 6 Local currency and USD returns regressed on GDP, change in GDP and size of devaluation

Local currency y6 months q6 months q12 months q18 months 24 months USD y6 months q6 months q12 months q18 months 24 months

Intercept

GDP0

GDP1

dGDP0

dLC

R2

y0.316 (0.195) y0.040 (0.229) 0.419 (0.548) 1.064 (0.991) y0.080 (0.665)

0.027 (0.026) y0.047 (0.031) y0.096 (0.074) y0.224 (0.134) y0.045 (0.090)

y0.002 (0.010) 0.020* (0.012) 0.060** (0.027) 0.069 (0.050) 0.048 (0.033)

y0.010 (0.021) 0.030 (0.025) 0.109* (0.059) 0.163 (0.106) 0.110 (0.071)

y1.002** (0.468) y1.947*** (0.551) y2.640* (1.317) y3.210 (2.383) y6.185*** (1.599)

0.28

y0.238 (0.142) y0.366** (0.158) y0.052 (0.396) 0.124 (0.355) y0.055 (0.337)

0.015 (0.019) y0.008 (0.021) y0.026 (0.054) y0.062 (0.048) y0.018 (0.046)

y0.003 (0.007) 0.029*** (0.008) 0.059*** (0.020) 0.049** (0.018) 0.035** (0.017)

y0.003 (0.015) 0.010 (0.017) 0.079* (0.042) 0.097** (0.038) 0.076** (0.036)

0.398 (0.341) y1.832*** (0.381) y2.262** (0.952) y2.778*** (0.854) y0.761 (0.810)

0.74 0.67 0.52 0.76

0.25 0.83 0.75 0.81 0.64

Regression coefficients with standard errors in parentheses. Returns are prior to (y) or following (q ) the devaluation, except for 24 months, which is the 2 year period from 6 months prior to 18 months following the devaluation. GDP0, GDP growth rate in the year of the devaluation; GDP1, GDP growth rate in the year following the devaluation; dGDP0, change in GDP growth rate in the year of the devaluation from the prior year; dLC, percentage change in the exchange rate in the month of devaluation. * Significant at the 10% level. ** Significant at the 5% level. *** Significant at the 1% level.

in the dollar value of a unit of local currency, the negative coefficients suggest that, in the case of the 6-month local currency returns, a 1% increase in the rate of devaluation is related to a similar level of increase in the local currency returns; large devaluations are associated with higher returns. One interpretation of this result is that larger devaluations provide more relief for companies, allowing for more relaxed monetary conditions and stronger economic conditions. This result suggests that we need to understand better the link between devaluation size and subsequent recovery. Also note that the amount of explained variation increases compared to what is reported in Table 5. The lack of statistical significance on the GDP growth measures should not be taken as evidence that they provide no value in the regressions in Table 6. Estimating the regressions with devaluation size (dLC) as the only independent variable results in a sharp drop in the amount of variation explained. For example, the R 2 on the regression for 24-month dollar returns drops to 27%, compared to 76% when the

J. Glen / Emerging Markets Review 3 (2002) 409–428

425

GDP measures are included in the regression, even though none of the GDP measures are statistically significant. The results in Table 6 suggest that model specification is important. The results in the earlier section using company-level data also suggested that leverage played an important role in stock performance following a currency crisis. At the macro level, external debt levels could also have a significant impact on stock returns around devaluation events. To examine that, I estimated a specification of the model in Table 6, but included various measures of external debt. With only a very few exceptions, none of those variables played a significant role in explaining stock market behavior.13 These regression results suggest that predicting the size of a forthcoming devaluation is an important part in making asset allocation decisions, as addressed in the following section. Interestingly, the size of a devaluation is statistically more important than economic growth in explaining stock returns and countries with high rates of devaluation have experienced higher levels of both local- and dollardenominated stock returns subsequent to the devaluation. 5. Stock markets as predictors of currency and stock returns The analysis so far has considered the magnitude of devaluations and the reaction of stock markets to those events. The aggregate behavior of markets reported in Fig. 2, however, suggests that the stock markets in these countries actually anticipate the currency event, turning down well before the major loss in currency value. That raises the question of the extent to which movements in stock prices signal the upcoming devaluation. To address this question consider a regression where the dependent variable is the change in the value of the currency in the month of the devaluation and the independent variable is the stock market return generated in the 5 months up to the month prior to the devaluation. This independent variable is predetermined at the time of the devaluation and is therefore a legitimate forecasting variable. The results of this regression are reported in Table 7, which includes a regression with returns calculated using price change only, as well as one calculated using the total return index. The results in the table provide no evidence that stock returns in the months preceding the devaluation have any predictive power. The explained variation (R 2) is very low and the regression coefficient on the lagged stock return is not statistically different from zero. Apparently, even though stock markets turn down well in advance of currency events, returns provide little information on what to expect in future currency values. 6. How far in advance do markets turn down? The analysis in the preceding sections examines only the 6 months preceding a devaluation. While that may seem to be a long time for a market to foresee a 13 Measures used were external debtyGDP, debt serviceyexports, private external debtytotal external debt. These results are not reported to conserve space.

426

J. Glen / Emerging Markets Review 3 (2002) 409–428

Table 7 One-month change in currency values relative to US dollar during month of devaluation event regressed on stock market returns over the period ty6 to ty1, where stock returns are expressed either as price index changes or total return index changes Price index

Total return

Constant

y0.246 (0.038)

y0.242 (0.038)

Stock return

y0.291 (0.225)

y0.300 (0.225)

R2

0.07

0.07

Regression coefficients with standard errors in parentheses.

currency event, some evidence suggests that markets turn down much earlier.14 Fig. 7 provides evidence on this by presenting both 3- and 12-month moving averages of the dollar price indices averaged across the 24 currency events for the 24-month period preceding a devaluation (as well as the 12 months following the event). The figure shows that the 3-month moving average crosses the 12-month average from above at a point 18 months prior to the devaluation. From this point on the market continues its downward path until the time of the event, at which point it promptly turns up. The figure obviously hides much of the variation discussed in the preceding sections, but it also illustrates the extent to which stock markets, on average, are very perceptive about these events. Note, however, that the bulk of the downturn preceding the currency event occurs in the 6-month period immediately preceding the event, lending some support to the bulk of the analysis in this paper, which has concentrated on that period of time. 7. Conclusions Large and relatively discrete currency movements are not uncommon in emerging markets. Despite this, we know little about the impact such currency movements have on stock markets. This paper examined 24 devaluation events and the longterm performance of stock returns. On average, while stock returns are reduced in the period leading up to a devaluation event, the period following these events is characterized by normal return behavior. There is considerable variation across events, however, and much of this variation can be explained by economic growth, the size of the devaluation, and the industry and country in which the event occurs. The real failure of this analysis is that it is unable to explain the reaction of markets prior to a devaluation event. What the evidence here shows is that the only significant reaction of stock markets to devaluations takes place in the 6 months leading up to the event, with all of that reaction having already concluded a month 14 Goldstein et al. (2000) find that annual stock market returns measured 18 months prior to a devaluation provide an important signal of pending devaluation.

J. Glen / Emerging Markets Review 3 (2002) 409–428

Fig. 7. Stock price average across 24 events. US$ price index, 12- and 3-month moving average.

427

428

J. Glen / Emerging Markets Review 3 (2002) 409–428

before the event month. The reaction is negative and none of the analysis presented here is very successful at explaining the cross-sectional variation observed in those reactions. For academics, that should be considered an invitation to further work. For practitioners, the lessons from this work are clear: know the country environment in which you invest. The evidence suggests that when the country environment is good, stock returns will also be good in spite of a currency event. Conversely, when the country environment is poor, trying to sit out the impact of a devaluation can be a long and painful process. References Y. Amihud, H. Mendelson, 1991 Liquidity, asset prices and financial policy, Financial Analy. J., November–December pp. 55–66. G. Bodnar, M.H. Franco Wong, Estimating exchange rate exposures: some ‘weighty’ issues, National Bureau of Economic Research. Working Paper Series (US); No. 7497:1–44, January 2000. R. Dornbusch, A primer on emerging market crises, NBER Working Paper Number 8326, 2001. Fama, E., French, K., 1990. Business conditions and expected returns on stocks and bonds. J. Financial Econ. 25, 23–49. E. Fama, K. French, 1992 The cross-section of expected stock returns, Journal of Finance, June pp. 427–465. J. Glen, P. Jorion, 1993 Currency hedging for international portfolios, Journal of Finance, XLVIII, pp. 1865–1886. M. Goldstein, G. Kaminsky, C. Reinhart, 2000 Assessing financial vulnerability, Institute for International Economics. IMF, World Economic Outlook, April 2002. S. Kamin, J. Schindler, S. Samuel, The contribution of domestic and external factors to emerging market devaluation crises: An early warning systems approach, International Finance Discussion Papers, Number 711, Board of Governors of the Federal Reserve System, September 2001. Levich, R.M., 1988. Empirical studies of exchange rates: price behavior, rate determination and market efficiency. In: Jones, R., Kenen, P. (Eds.), Handbook of International Economics. North-Holland, .