آیا نوسانات نرخ ارز بیش از حد است؟روش ARCH و AR
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|8237||2004||33 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : کاهش ارزش، اثرات ترازنامه، ترکیب ارز از بدهی, Volume 44, Issue 1, February 2004, Pages 122–154
Since the implementation of the floating exchange rate system in 1973, world leaders, policy makers and researchers have been seriously concerned with the volatility of foreign exchange rates. Financial crises in Mexico, Russia, and Asia renewed the debate about whether increased volatility may suppress international trade and what can be done to discourage currency speculation. The earlier literature using unconditional dispersion measures strongly suggested that exchange rate changes are less volatile than those of other asset prices. This paper shows that this result holds up under a battery of tests that try to elicit the statistically appropriate measure of conditional variance. While the discussion of the econometric issues is a major part of the paper, the readers are reminded that the issue is not that the hypothesized variance processes differ dramatically; rather the issue is that, even after applying the battery of tests, the conditional and unconditional results suggest similar conclusions.
Since the change by most industrialized nations from the Bretton Woods system of fixed exchange rates to a flexible-rate regime, there has been serious, and as yet unabated concern over the volatility of floating exchange rates. Those who advocate a return to a fixed rate system—politicians, bureaucrats and academics from around the world—argue that the increased risk generated by the volatility of flexible rates causes a severe dampening of the level of international trade and investment, thus lowering the welfare of residents of all trading nations. While admitting that theoretical possibility in imperfect markets, proponents of flexible rates argue that the case is overstated: foreign exchange markets for major currencies are at least as efficient as those for any other asset, and the volatility that is observed in such markets is not “excessive,” but an efficient mechanism for exchanging information, certainly preferable to the enforced disequilibrium of a fixed-rate system in a rapidly changing global economy. The 1992 pound crisis and the 1997 Asian financial crisis illustrate that the fixed exchange rate system is very difficult, if not impossible, to maintain, especially within a more integrated global market in which information and monetary assets can be transferred within several seconds. The 1992 pound crisis not only shook the currency markets but also triggered extremely large devaluations of French franc and Italian lira (The Wall Street Journal, September 17, 1992). While the crisis was directly caused by the British government’s announcement to withdraw from the Exchange Rate Mechanism (ERM) in Europe, it was indirectly set off by the German reunification. It is well known that Germany suffered significant expenses and inflation following reunification. As a result, the German central bank tightened its monetary policy to reduce the inflation rate. Interest rates rose significantly because of the new policy. The higher interest rate caused capital to flow into Germany and out of other European Monetary System countries, such as U.K. and Italy. Great Britain found that it could no longer afford to raise interest rates high enough to keep the capital at home, and therefore withdrew from the ERM, a fixed exchange rate system. The financial crisis in Asia that forced the Thai baht to devalue more than 20%, and the Malaysian ringgit, the Phillipine peso, and the Indonesia rupiah to devalue almost 10% within 1 or 2 weeks, illustrates that trade deficits, budget deficits, inflation rates, and volatility of foreign exchange rates are closely related. However, without a ruler or a target for comparison, one cannot judge whether these changes are indeed excessive. Tobin (1978) argued that exchange rates are too variable because financial markets are “excessively efficient”—that capital sloshed back and forth among countries in response to trivial disturbances. Tobin (1978) and Dornbusch (1976) recommended that a tax on foreign exchange transactions would reduce volatility. Asking a question similar to this study, Frankel and Meese (1987) used several techniques to assess the out-of-the-sample forecasting properties of typical exchange-rate models. They found that the models were not satisfactory even after the consideration of rational expectations, risk premia, “peso-problems,” bubbles, etc. Flood (1987, p. 157) recommended asking ourselves the following question: “Have flexible exchange rates in other episodes been inexplicably more variable than other asset prices?” This is the criterion that will be used to assess “excessiveness” in this study. Early work on the volatility of exchange rates as an asset price (e.g., Frenkel, 1981 and Bergstrand, 1983) used measures of unconditional variability and monthly data. More recently, researchers (Lastrapes, 1989; Engle, Ito, & Lin, 1990) have phrased the problem in terms of a variance process over time (autoregressive conditional heteroskedasticity, ARCH) and have estimated volatility of weekly and daily series. The first approach led to the conclusion that exchange rates are probably no more volatile than other asset prices. Using the second, modeling variance as a process rather than a number, Lastrapes (p. 73) found the possibility of greater persistence of exchange rate shocks, especially with varying-in-parameter ARCH models, than with measures used in earlier work. But the work to date has been incomplete in scope and depth. There has been no comparison of the behavior of exchange rate series with other asset prices in the ARCH framework. The crucial question of the frequency of data used—more formally, the level of temporal aggregation—has not been systematically addressed. Many previous studies of foreign exchange rate behavior, even those with ARCH modeling, have assumed away any problems of an autoregression (AR) of exchange rate shocks. The non-nested hypotheses of ARCH and AR specification have not been fully and explicitly tested. Bera, Higgins, and Lee (1992), Bollerslev, Engle, and Nelson (1994), Diebold (1986), Bollerslev (1987), and Stambaugh (1993), to name a few, have pointed out the importance of modeling both AR in mean and ARCH in variance. Engle, 1982 and Engle, 1983 in fact included lags of macroeconomic variables in the mean equations. This study provides empirical evidence relevant to these questions and, thus, addresses some of the questions of the seriousness of the degree of exchange rate volatility with floating rates. It uses both monthly and weekly data on spot and forward rates, short-term interest rates, long-term government bond yields and stock market indices in the United States, the United Kingdom, Switzerland, Japan, Germany, France and Canada to investigate exchange rate and other asset price paths over time. Note that the asset prices included in this study are variables of many monetary models (see more discussion in Section 3). In the process, three hypotheses are tested. Two of them are focused on the parameters of conditional distributions and their stability across asset prices and measurement frequency: (1) The volatility of exchange rates, as characterized by the conditional variance process, is no different for exchange rates than for other asset prices. (2) The volatility/variance process of exchange rates (and other asset prices) is unaffected by the level of temporal aggregation of measurement. The third addresses the very nature of the generating process distributions and what we can learn of them from sample evidence. (3) The time path of shocks to exchange rates and other asset prices is characterized neither by an ARCH process nor by persistence of shock in mean (AR). The results from these tests have significant implications for economic theories and public policies. For example, it is increasingly apparent that standard economic theories and models fail to explain most foreign exchange rate movements. Messe and Rogoff (1983) present very strong evidence to support the notion that a “naı̈ve” random-walk model performs much better than the flexible-price monetary model (Frenkel, 1976 and Bilson, 1981) and the sticky price model (Dornbusch, 1976 and Frankel, 1979). The finding of any ARCH processes indicates that when non-linearity is incorporated into these models, their forecast errors may be uncorrelated but not independent: a nonrandom walk (see Diebold, 1986 for similar discussion). Bera et al. (1992) demonstrated that nonlinearity might have a significant impact on tests for ARCH. The observed ARCH effects are consistent with the leptokurtosis in exchange rate changes that has been documented by numerous studies (Westerfield, 1977; McFarland, Pettit, & Sang, 1982; So, 1987). They also reflect the clustering of volatility—large changes followed by large changes (of either sign) and small changes followed by small changes—and therefore can be used to formalize the contiguous periods of volatility and stability discovered by Mussa (1979). Moreover, since volatility will affect imports and exports, which in turn may have impacts on domestic economic variables such as wages, prices, output, and employment, evaluation of the effectiveness of government intervention in the money markets or the foreign exchange markets will not be an easy task. The next section reviews briefly some of the previous literature on which the work is based. Section 3 presents the theoretical model of asset price time path used for the investigation and describes the data sets and estimating equations. Section 4 presents sample evidence that justifies a search for autoregressive processes in conditional mean or variance, as well as results of OLS and ARCH estimation. The basis for and sample results of robust tests for AR and ARCH are also presented. In the final section, comparisons are made about relative volatility across asset prices and across levels of temporal aggregation. A summary and suggested directions for future work are also given.
نتیجه گیری انگلیسی
This study has examined unconditional and conditional measures of volatility in weekly and monthly series of exchange rates and other asset prices for seven countries. Although in such a broad empirical investigation there are bound to be differences in patterns, at the risk of oversimplification, we can generalize to a few conclusions about our hypotheses of interest. We find both differences and similarities between the conditional volatility of exchange rates and other asset prices. As is clear in Table 5, the conditional volatility of exchange rates (spot-lag) is decidedly less than that of either the “forecast errors” (spot-forward) or the conditional volatility of other asset prices. This mirrors the relationship of the three categories seen in the unconditional measures of Table 1 and Table 2 and in previous work. But the exchange rates, like other asset prices, seem far more likely to be characterized by the ARCH (conditional) variance process in a weekly model than in a monthly one. This difference in variance by model frequency was one of the marked differences between the monthly and weekly series. However, the most striking and important conclusion from this work comes in our investigation of the third hypothesis about the form of asset price distributions. Although conventional tests appeared to indicate an extensive occurrence of both ARCH and AR in the asset price series, use of tests that are robust to distributional misspecification painted a markedly different picture. With the exception of weekly Canadian price data and monthly U.S. asset prices and the precious metals series, these distributions are not characterized by AR processes. The behavior of most, moreover, shows little evidence of autoregressive conditional volatility, with a frequency of significant results nowhere near as high as the conventional tests had suggested. We must conclude that not only is it inappropriate to characterize all price series similarly (as ARCH, or as GARCH, or not), but any that testing to investigate such processes may result in erroneous conclusions unless robust tests are used. We have attempted to seek greater empirical insight into our data series by treating them not as a cross-sectional group of 15-year series, but as a set of individual time series of “news” or “innovations” to the determination of speculative prices. We looked at the exchange-rate series as composed of multiple potentially structurally different models connected either by unknown or by “known” break points defined by macroeconomic events or policy shifts in an approach such as that used by Lastrapes (1989). The decidedly mixed results indicate that there is no simple answer, and that the entire topic needs more work on theoretical and empirical modeling in later research