عدم قطعیت بحران ارز: شواهدی از بررسی داده ها

How does uncertainty about fundamentals affect speculation in the foreign exchange markets? This paper studies empirically the role of uncertainty in currency crises. Uncertainty, which is measured using the dispersion of survey forecasts of key macroeconomic variables, is found to have a non-monotonic effect on exchange rate pressures: it heightens speculative pressures when expected fundamentals are good and eases them when they are bad. This prediction is consistent with a broad class of currency crisis theories, ranging from first-generation to global-game models. The proposed empirical strategy remains valid in the presence of forecasters with strategic objectives and addresses potential endogeneity bias by building a novel set of instrumental variables.

Recent theoretical contributions to the currency crisis literature have turned the spotlight on the role of uncertainty during crises, showing that the same fundamentals may or may not lead to a speculative attack depending on the precision of information about them. The matter is also critical for policy purposes. For example, if greater uncertainty increases the probability of a speculative attack, then exchange rate regimes will be more vulnerable in periods of high uncertainty and policy-makers should adjust their policies accordingly. There has been little debate, however, about the empirical significance of uncertainty about fundamentals during currency crises. This paper analyzes this question by proceeding in two stages. First, we show that a broad class of currency crisis theories with a unique equilibrium—ranging from first-generation models with public information (such as those pioneered by Krugman, 1979 and Flood and Garber, 1984) to global games with public and private information (Morris and Shin, 1998 and Hellwig, 2002)—predict that the effect of uncertainty on exchange rate pressures is non-monotonic and varies with expected fundamentals. 1 Specifically, these models predict that a reduction in information precision (i.e. an increase in uncertainty) raises the share of speculators attacking the currency when expected fundamentals are “good” and has the opposite effect when expected fundamentals are “bad”. This broad-based prediction—which previous studies have overlooked—has a very intuitive interpretation: as information about fundamentals becomes less precise, speculators rely less on it in order to decide whether to attack the currency. Thus, as information about good fundamentals becomes less reliable, speculators lose confidence in the good state of the economy and augment exchange rate pressures. By the same token, as information about bad fundamentals becomes less reliable, speculators wobble about the success of a speculative attack and diminish exchange rate pressures. Second, we study empirically how speculation in the foreign exchange markets is affected by uncertainty, by measuring the latter with the dispersion of survey forecasts of key macroeconomic variables in six Asian countries.2 Survey forecasts are an appealing source of information to test models in which exchange rate pressures depend on agents beliefs about fundamentals. Since 1995, for the six Asian countries in our sample Consensus Economics has gathered individual forecasts from a panel of analysts including both international financial firms and domestic enterprises at a monthly frequency—a higher frequency than that of some key macroeconomic variables, such as GDP growth. Survey forecasts are also inherently forward looking, just like the exchange rate pressures that the currency crisis literature tries to explain. Moreover, it is possible to relate the mean and variance of the individual forecasts to, respectively, expected fundamentals and the precision of information, which are the key parameters of currency crisis models. The main challenge in the empirical studies of the effect of uncertainty on exchange rate pressures is that causality could go both ways. While currency crisis theories predict a causal effect running from the mean and variance of the forecasts to speculative pressures, shocks to unobservable determinants of exchange rate pressures could also affect the distribution of the forecasts. In this paper, this complex interaction is teased out by building a novel set of instrumental variables. These instruments include the level and dispersion of forecasts of consumption growth and the unemployment rate in the United States, as well as alternative composite instruments computed as the trade-weighted version of the same macroeconomic indicators for the G-7 countries excluding Japan (G-6 countries, hereafter). Are these instruments valid for GDP growth forecasts of Asian countries? The ideal instrument is an external source of variation that randomly changes the mean and variance of the GDP growth forecasts of Asian countries, in a manner that is uncorrelated with the unobservable determinants of exchange rate pressures. Forecasts of domestic demand conditions in the United States (or in G-6 countries) satisfy the requirement of being positively correlated with GDP growth forecasts of Asian countries, because the export-oriented economies of the latter depend on cyclical conditions in the former. Moreover, the proposed instruments are exogenous to speculative pressures in Asian countries since they reflect mostly domestic factors in the United States, on which exchange rate developments in Asian countries have a negligible effect. These two characteristics, however, are not sufficient to fully justify the use of forecasts of domestic demand conditions in the United States as instruments for GDP growth forecasts of Asian countries. The additional identifying assumption which is required is that these instruments have no reason to be included as regressors in the second stage of the IV estimation.3 In other words, it is necessary to assume that forecasts of domestic demand conditions in the United States only affect exchange rate pressures in Asian countries indirectly, i.e. by modifying these countries’ GDP growth forecasts. We justify this exclusion restriction by including forecasts of U.S. interest rates, as well as their interaction with lagged levels of short-term debt in each Asian country. These variables aim at capturing changes in financial market conditions in the United States that might affect exchange rate pressures in Asian countries—for example, through higher expected costs of servicing their external debt—and that might be proxied by the proposed instruments. Then, the identifying assumption maintained in this paper is that, after controlling for U.S. interest rates and their interaction with Asian countries’ short-term external debt, forecasts of domestic demand conditions in the United States do not have any residual direct effect on Asian countries’ exchange rate pressures. Using IV estimates is crucial also to address another problem that could potentially affect the empirical analysis. The mean forecast is only an imperfect measure of expected fundamentals (think, for instance, at the sample error due to the fact that only a finite number of forecasters is interviewed). Therefore, in a multivariate specification like the one that is predicted by the theory, OLS would provide biased estimates of all coefficients (in a direction that is difficult to assess), including also the coefficient related to uncertainty, which is the main variable of interest. Instrumental variables, then, are the main weapons to address this second problem as well. Table 1 presents some preliminary evidence about the effect of uncertainty on exchange rate pressures. The north-west panel of the table presents that, when GDP growth forecasts are bad, high uncertainty is associated with low exchange rate pressures.4 Conversely, the north-east panel of the table presents that, when GDP growth forecasts are good, high uncertainty is associated with high exchange rate pressures. Both findings are consistent with the non-monotonic effect predicted by the theory that uncertainty heightens speculative pressures when expected fundamentals are good and eases them when they are bad. These results, which are confirmed by simple OLS regressions, are corroborated by the IV estimates, as well as by an extensive set of robustness checks.There are relatively few studies in the empirical literature on currency crises that have focused on the role of uncertainty. Early exceptions are Hodrik (1989), who unsuccessfully used estimated conditional variances of money supply, industrial production, and consumer prices, to account for the dynamics of the forward exchange-rate premium; and Kaminsky and Peruga (1990), who estimated a GARCH-in-mean restricted VAR model. Our paper differs from these studies for the novel testable prediction derived from a broad class of currency crisis theories and the use of survey data. Our paper is also closely related to the voluminous theoretical literature on first-generation and global-game models of currency crises. It should be clear that in this paper no attempt is made in order to discriminate between these two types of models, but the role of uncertainty is studied by encompassing different models. On the former matter, this paper suggests, if anything, that to discriminate between these models, one cannot look at their implications about the role of uncertainty, which are essentially the same. On the other hand, our findings do have some implications for the relevance of different theories. Some important models, in fact, predict that higher uncertainty always encourages speculative attacks. For instance, Flood and Marion (2000) build a model with multiple equilibria in which higher uncertainty about a real shock raises the risk premium on asset holdings, causing expected fundamentals to deteriorate and moving the economy to a region where a self-fulfilling attack may occur. Uncertainty appears to have a negative effect also in Flood and Garber (1984), because their model assumes an exponential distribution of fundamentals in which the effect of higher uncertainty cannot be distinguished from the effect of deteriorating expected fundamentals.5 Our empirical results, then, cast doubts on theories in which uncertainty always heightens speculation. Interestingly, Bloom et al. (2009) have recently documented that uncertainty, as measured by the volatility of several micro and macro economic variables, is countercyclical. These findings, coupled with ours, imply that changes in uncertainty contribute to make exchange rate pressures less cyclical. Thus, an intriguing result from this paper is that, for what concerns speculative pressures, the effect of uncertainty is not as bad as it is generally thought. The rest of the paper is organized as follows. Section 2 presents the testable implications on the role of uncertainty and expected fundamentals that are derived from both first-generation and global-game models. Section 3 describes the data. Section 4 illustrates the empirical results. Section 5 concludes. The proofs of the theoretical results, a detailed discussion of the link between model parameters and their empirical counterparts, and additional robustness checks of the empirical results can be found in the online supplementary material.

Does uncertainty heighten or ease speculation in the foreign exchange markets? Could the effect of uncertainty have different signs depending on the economic outlook? To analyze these questions, we develop an empirical framework based on a broad class of currency crisis theories with a unique equilibrium. As a first contribution, we prove that both first-generation models with public information and global games with public and private information predict that uncertainty has a non-monotonic effect on speculative pressures; namely, uncertainty heightens speculative pressures when expected fundamentals are good and eases them when they are bad. By measuring expected fundamentals and uncertainty with the mean and the dispersion of survey forecasts, we apply the proposed empirical framework to six Asian countries, finding strong evidence of the non-monotonic effect of uncertainty predicted by the theory. Instrumental variable estimates and several robustness tests confirm this result. How does this paper relate to previous explanations of the Asian crisis? While our focus on the role of uncertainty is unique, our findings are broadly consistent with the prevailing idea that the Asian crisis was rooted in the financial conditions of the corporate and banking sectors. Indeed, this is probably the reason why GDP growth forecasts turn out to be the best measure of exchange rate fundamentals, proving to be empirically superior to forecasts of other macroeconomic variables (such as inflation or the current account) that have played a key role in other crisis episodes. The statistically significant coefficient of the interaction between U.S. interest rates and short-term external debt, which results from the estimates, is consistent with the role of internationally illiquid banks emphasized by Chang and Velasco (2002) as well as with that of foreign currency borrowing by domestic firms in the model of Aghion et al. (2004). Our paper, however, cannot be seen as a direct empirical test of these models, which are characterized by multiple equilibria. Developing a framework for testing the role of uncertainty in models with multiple equilibria is a topic for future research, possibly using regime switching econometric techniques as in Jeanne and Masson (2000). Future empirical research is also needed to verify whether data on other well-known currency crises in Latin America and Europe confirm the statistical significance of uncertainty about fundamentals. In addition, it might be worthwhile exploring whether these variables can enhance the predictive power of early warning systems, which are often based on past fundamentals. Indeed, drawing on a previous version of this paper, Bannier (2006) applies our empirical framework to Mexican data obtaining qualitatively similar results and Köhler-Geib (2006) uses it to explore the role of uncertainty in explaining sudden stops.