بازدهی بازار و انباشتگی نرخ مبادلات ارز در طول دوره بحران اقتصادی: نگاهی به بحران ارز اروپایی و آسیایی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|24912||2006||20 صفحه PDF||سفارش دهید||9803 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Economics and Business, Volume 58, Issue 4, July–August 2006, Pages 323–342
This study extends the study of foreign exchange market efficiency. It employs several verification testing procedures, rather than using only standard Johansen tests, to re-examine if cointegration among different spot exchange rates is actually present during the 1992–1993 European currency crisis and during the 1997–1998 Asian currency crisis. In contrast to the findings in prior studies, the test results collectively cast strong doubts on the presence of cointegration. Therefore, a cointegration test may not be an appropriate technique to detect and reveal market inefficiency if it in fact transpires during these two crises. Further, this study strongly corroborates empirical evidence that the reliance on Johansen tests can result in spurious findings of cointegration and thus incorrect inferences about efficiency.
Whether or not foreign exchange markets are efficient is of considerable interest to researchers and market participants. Among other econometric techniques, a cointegration analysis has been employed by several recent studies to examine foreign exchange market efficiency. The majority of prior empirical work (e.g., Coleman, 1990; Copeland, 1991; Lajaunie, McManis, & Naka, 1996; Lajaunie & Naka, 1992; MacDonald & Taylor, 1989; Rapp & Sharma, 1999) has found that spot exchange rates for various major currencies generally are not cointegrated during the modern float. The absence of cointegration and thus a cointegrating vector and the error correction model (ECM) (e.g., Engle & Granger, 1987) implies that the current value of one currency cannot be predicted by past values of other currencies. This unpredictability is commonly interpreted by prior studies as evidence of weak-form efficiency in foreign exchange markets. Interpretational ambiguity mainly arises when cointegration among spot exchange rates is detected. Baillie and Bollerslev (1989) interpret the predictability implied by cointegration as a violation of weak-form efficiency or as indirect evidence of a time-varying risk premium. Crowder (1994) finds that a cointegrating vector which is stationary or I(0) by definition does not appear to be a proxy for a time-varying risk premium. This is because the forward premium used to represent the risk premium has a different time series property in that it appears non-stationary or I(1). Baillie and Bollerslev (1994) contend that the forward premium is not a pure I(1) process but fractionally integrated and mean reverting with finite cumulative impulse response weights. On the other hand, Wu and Chen (1998) employ a more powerful unit root test, find that the forward premium is in fact stationary and conclude that foreign exchange markets are efficient even though the presence (or lack thereof) of cointegration is not examined. Jeon and Lee (2002) find that the G-7 countries’ exchange rates are cointegrated during the period between the Plaza Agreement in 1985 and the Louvre Accord in 1987. They conclude that market inefficiency transpires during this period of international policy cooperation to stabilize exchange rates. Further, Haug, Mackinnon, and Michelis (2000) and Rangvid and Sorensen (2002) detect cointegrating relations among exchange rates of the European Union (EU) countries over extended time periods prior to the inception of the European Monetary Union (EMU) in 1999. They however interpret this result as an indication of stability and credibility of the EU exchange rate policy coordination through the Exchange Rate Mechanism (ERM) rather than as evidence of market inefficiency. The Maastricht Treaty (1992) which requires convergence of key economic variables, including exchange rates, among EU nations prior to becoming EMU members can further explain such finding. In relation to voluminous studies using long spans of data, a few studies have performed cointegration tests of spot exchange rates during periods of economic turmoil. Aroskar, Sarkar, and Swanson (2004) find a cointegrating relation among daily spot exchange rates of EU currencies during the European currency crisis of 1992 and 1993. They suggest that weak-form inefficiency exists during the crisis partially because the ECM provides better predictive power for some included currencies than does the random walk model. Further, Aroskar and Swanson (2002) and Jeon and Seo (2003) evidence a cointegrating relation among daily spot exchange rates of Asian currencies during the Asian currency crisis of 1997 and 1998. These two studies conclude that weak-form inefficiency occurs during the crisis as well. Similar to the analyses using long time spans, whether or not cointegration really indicates market inefficiency during these two crises is open to debate. On the one hand, cointegration and its embedded predictability can emerge and entail arbitrage opportunities if foreign exchange markets are truly inefficient during turbulent times. On the other hand, Dwyer and Wallace (1992), Baffes (1994), Engel (1996), Masih and Masih (2001) and Ferre and Hall (2002) indicate that cointegration does not necessarily imply market inefficiency, implying further that cointegration tests may not be appropriate tests of market efficiency for any period. This is because whether or not the predictability derived from cointegration can truly lead to arbitrage opportunities and/or the ability of market participants to earn risk-adjusted excess returns has not been confirmed nor verified. Further, Lence and Falk (2005) indicate that cointegration test results have no implications about market efficiency without additional restrictions on the economy or economies. The upshot is that if cointegration among spot exchange rates during crises (such as the European and Asian crises) is in fact absent in the first place, the possible connection and relevancy of cointegration to inefficiency are mitigated. These include, for instance, whether or not the resultant ECM would provide better predictive ability in relation to the random walk model, whether or not the cointegrating vector could be a proxy for a risk premium, and whether or not cointegration would result in arbitrage opportunities and thus truly imply weak-form inefficiency. Conceptually, in this case, the exchange rates of currencies affected by the crises may simply exhibit considerable volatilities without being cointegrated with one another. It is also possible that the EU exchange rates are cointegrated during the non-crisis period due to the ERM and Maastricht Treaty (1992), but are not cointegrated during the crisis period due to the abandonment of relevant currencies from such exchange rate mandates. Thus, the question is whether or not evidence of cointegration during these two crises has been identified correctly. Hakkio and Rush (1991) indicate that cointegration is a long-run property, and hence extensive time spans, rather than high data frequency used in a short-term crisis period analysis, should be employed to appropriately detect the cointegration evidence. Further, previous studies of spot exchange rates during currency crises (i.e., Aroskar & Swanson, 2002; Aroskar et al., 2004; Jeon & Seo, 2003) base their cointegration findings mainly on the Johansen cointegration methodology (e.g., Johansen, 1988). Sephton and Larson (1991) and Crowder (1996), however, indicate that the statistical power of cointegration tests can be highly suspect. Particularly, the Johansen methodology can result in indeterminate and/or incorrect inferences regarding the presence of cointegration and the number of cointegrating vectors in the system (e.g., Gonzalo & Lee, 1998; Hall, 1991). Given this possibility, the reliance on conventional Johansen tests may have caused spurious findings of cointegration and thus incorrect inferences of inefficiency during the two crises. These misleading findings and inferences can be harmful. For instance, investors in foreign exchange markets might take unnecessarily risky positions in affected currencies and hope to exploit arbitrage opportunities based on the estimate of a cointegrating relation which in fact does not exist. Thus, expanded methodologies are needed to verify the results from any one specific cointegration methodology typically relied upon by past researchers. Given the importance of accuracy in measuring foreign exchange market efficiency, this study re-examines whether or not a cointegrating relation is really present among spot exchange rates of the affected currencies during the 1992–1993 European currency crisis and during the 1997–1998 Asian currency crisis. Several verification testing procedures which have not been considered in conjunction with one another in prior studies are included. First, the conventional Johansen cointegration test (e.g., Johansen, 1988) is performed. Then, its variant which is based on the partial VAR system (e.g., Pesaran, Shin, & Smith, 2000) and provides some econometric advantages over the conventional procedure is implemented. Further, the recursive test of cointegration parameter stability (Hansen & Johansen, 1999); the unit root test of cointegrating vector and common trend estimates based on the Gonzalo–Granger decomposition (Gonzalo & Granger, 1995); the Harris–Inder (HI) cointegration test (Harris & Inder, 1994) under the reversed null hypothesis of cointegration; and the Gregory–Hansen (GH) cointegration test (Gregory & Hansen, 1996) which considers the possibility of an endogenous structural shift in a cointegrating relation are conducted. This battery of tests can potentially add clarity and a new perspective to researchers and foreign exchange market participants concerning evidence of cointegration during the two currency crises which might have been detected spuriously in prior studies based on standard Johansen procedures. This study finds that European exchange rates and Asian exchange rates are cointegrated during the European currency crisis and during the Asian currency crisis, respectively, based on the Johansen tests for the full VAR system and/or for the partial VAR system. However, the cointegration parameters obtained show evidence of instability, especially in the group of Asian exchange rates. The unit root tests also reveal that the estimates of cointegrating vectors and common trends, which conceptually should be stationary and non-stationary, respectively, appear to be identically non-stationary. Further, the null hypothesis of cointegration in the HI tests can be clearly rejected in both European and Asian exchange rate groups. Finally, the null hypothesis of no cointegration in the GH tests cannot be rejected in any exchange rate group. This is true irrespective of the specification for an endogenous structural shift in a cointegrating relation under the alternative hypothesis. These findings collectively cast strong doubts on the validity of cointegration findings detected through Johansen tests and consequently on the usefulness of cointegration tests to reveal market inefficiency even if foreign exchange rate markets are truly inefficient during these two crises. The remainder of this study is organized as follows. Section 2 describes the data and explains the econometric methodology used, with estimation results and related findings set forth in Section 3. Section 4 provides conclusions.
نتیجه گیری انگلیسی
This study extends the study of foreign exchange market efficiency. It employs several verification testing procedures to re-examine if cointegration among different spot exchange rates is actually present during the 1992–1993 European currency crisis and during the 1997–1998 Asian currency crisis. This approach differs from prior studies which have relied primarily on conventional Johansen tests, have detected evidence of cointegration, and consequently, have inferred the existence of market inefficiency during these two crises. The Johansen test results for the full and partial VAR systems collectively indicate that cointegration is present during the two crises, and specifically, that the cointegration evidence appears stronger during the European crisis than during the Asian crisis. However, additional tests cast strong doubts on the validity of these cointegration findings. The estimated cointegrating relations show evidence of instability and non-stationarity. This potentially negates the reliability and usefulness of the resultant ECM to predict the affected exchange rates during these two crises. Further, the reversed null hypothesis of cointegration in the Harris–Inder test can be clearly rejected. Finally, the null hypothesis of no cointegration in the Gregory–Hansen test cannot be rejected irrespective of the specification for a possible endogenous structural shift in a cointegrating relation. The finding that the existence of cointegration during the European and Asian currency crises is in fact doubtful or unlikely has useful implications. First, even if foreign exchange markets are truly inefficient during these two crises, the inefficiency is not revealed through evidence of cointegration and thus a cointegration test does not appear to be useful or relevant in detecting such inefficiency. This conclusion is consistent with prior empirical suggestions that cointegration is a long-run property and long spans of data are needed to appropriately detect a cointegrating relation and that cointegration tests after all are not appropriate tests of market efficiency for any period. In fact, the absence of cointegration during the European crisis may simply be due to the departure of some EU currencies from the exchange rate mandates such as ERM and the Maastricht Treaty (1992). Second, investors in foreign exchange markets should be discouraged from generalizing that cointegration among relevant spot exchange rates usually emerges during periods of economic uncertainty.12 This caution should further prevent them from taking unnecessarily risky currency positions and hoping to gain superior profits based on cointegration and its implied predictability which in fact do not exist. Specifically, because a stationary and stable cointegrating relation is not present during either the European or the Asian crisis even if the possibility for a structural shift is considered, the cointegration-based model is unlikely one of the forecasting models which truly provide investors with arbitrage opportunities and superior returns. Thus whether or not other increasingly sophisticated econometric models (or otherwise basic but overlooked statistical measures such as correlations) can better detect the relationships among currencies and measure market inefficiency during the crisis period should be of interest to researchers. Third, prior empirical evidence that the reliance on the Johansen test can result in spurious findings of cointegration and thus erroneous inferences is strongly supported. While the Johansen test has been found to show reasonable power and size properties, even when it is carefully implemented, additional tests are clearly needed to validate Johansen test results.