مداخله بانک مرکزی و پویایی نرخ ارز : منطقی برای روند رژیم سوئیچینگ نرخ ارز

By proposing a stochastic intervention model of exchange rate determination, this paper provides an alternative rationale for the success of the Markov-switching model in explaining exchange rate dynamics. One extreme case is a pure floating rate model while the other extreme one is a driftless random walk model. The relation between the exchange rate and the future fundamentals under a non-intervention state is looser than the one under a pure floating exchange regime. This article also provides a method for detecting a central bank's interventions when intervention data are not available. Applying the stochastic intervention model to the monthly NT$/US$ exchange rates in 1989M1–2004M6, we find that it outperforms both the pure floating rate model and the random walk model in terms of the likelihood value and the diagnostic test of heteroscedasticity. In addition, with the constructed intervention state index in this article, the estimation of the stochastic intervention model is found to be consistent with the hypothesis that the regime switches of exchange rates are due to a central bank's (non-)interventions. J. Japanese Int. Economies21 (1) (2007) 64–77.

The most popular exchange rate arrangement nowadays is a regime between undisputed floats and undisputed pegs. In order to capture reality, this paper first proposes a simple model of short-run exchange rate determination with stochastic interventions from a central bank. Under stochastic intervention, the exchange rate sometimes is endogenously determined by market fundamentals, while it sometimes is manipulated by the central bank. The implied state-dependent exchange rate adjustment of a theoretical model can then be shown to conform to the actual exchange rate process for emerging market countries such as Taiwan. Due most likely to “fear of floating” and “fear of pegging,” many emerging market countries that claim they are floating actually manage their exchange rates (Calvo and Reinhart, 2002, and Levy-Yeyati and Sturzenegger, 2005). The stylized facts are the following. Unlike monetary authorities in the European Monetary System or those in several major industrialized countries, the central banks in emerging market countries do not coordinate with other countries to undertake joint currency interventions. Moreover, the central banks in these emerging market countries neither publicly announce the target of their exchange rate policy nor promise under what situation that they will undertake intervention operations. Even if they announce it ex ante, they can undo it ex post. In addition, intervention data are seldom revealed by the central banks of emerging market countries.1 In constructing the analytical framework, we consider that the central bank undertakes intervention operations on a case-by-case basis, but the exact timing and the exact magnitude of the intervention are not known by market participants ex ante. These characters make our model distinct from those in Hsieh (1992), Natividad-Carlos (1994), Lewis (1995), and other target-zone models. In these previous papers the central bank follows an explicit target rate or intervention rule, through which the central bank affects the exchange rate by changing monetary aggregates or reserves. The stochastic intervention model in this paper implies a state-dependent adjustment of the exchange rate. In the case of non-intervention, market fundamentals and the probability of the central bank's continual non-intervention shall determine the exchange rate. On the other hand, in the case of intervention the exchange rate is assumed to follow a driftless random walk process as the monetary authority usually pursues a stable value of its home currency in practice. Because the process of exchange rates displays non-linearity, many empirical studies employ regime-switching models to fit exchange rate data or to forecast future exchange rates (e.g., Engel and Hamilton, 1990, Engel, 1994 and Engel and Kim, 1999; Bollen et al., 2000 and Dewachter, 2001; and Clarida et al., 2003). The research works in Sill and Wrase (1999) and Sarno et al. (2004) appear to be the first approach that provides a rationale for using Markov-switching models in explaining exchange rate dynamics; that is, market fundamentals themselves are regime switching.2 This article provides an alternative theoretical base for using regime-switching models. In a stochastic intervention model, the exchange rate is sometimes pegged by the central bank and at other times determined by market fundamentals. Therefore, the exchange rate follows a regime-switching process even though the market fundamentals do not. Furthermore, the relation between the exchange rate and the fundamentals under a non-intervention period is different from the one under a pure floating exchange regime. Specifically, the relation between the exchange rate and the future fundamentals under a non-intervention state is looser than the one under a pure floating exchange rate regime while that between the exchange rate and the contemporary fundamentals under a non-intervention state is closer than the one under a pure floating exchange rate regime. This feature has never been embedded in empirical studies. In order to find our model's empirical usefulness, we apply our stochastic intervention model to Taiwan data. Specifically, this article derives the theoretically-implied dynamic process of an exchange rate, which is state-dependent, for the empirical time-series analysis. This makes our article distinct from pure empirical models. The (non-)intervention periods for the exchange rate of the New Taiwan dollar (hereafter, the NT dollar) against the US dollar can be estimated from the smoothing probability after estimating the Markov-switching model. Findings from our empirical investigation are consistent with the practice that Taiwan's central bank conducts stochastic intervention in the foreign exchange market. By comparing our stochastic intervention model with two extreme cases, a driftless random walk model and a pure floating rate model, we find that the stochastic intervention model has the highest likelihood value and removes the autoregressive conditional heteroscedasticity effects in the residuals for the depreciation rate of the NT dollar. In addition, the stochastic intervention model has a statistically equal out-of-sample forecasting power as its random walk counterpart. Finally, the process of a constructed intervention state index in this paper is found to be consistent with the non-intervention periods estimated by the smooth probability with the stochastic intervention model. The remainder of the paper is organized as follows. Section 2 describes the stochastic intervention model. After solving the model, we show that parameters of the depreciation rate are state-dependent and hence provide a rationale for using the Markov-switching model in estimating the exchange rate process. Section 3 begins with the data's description. The state-dependent depreciation rates, together with exogenous market fundamentals, are then examined empirically. Some comparisons among the stochastic intervention model, a random walk model, and a pure floating model are conducted. Finally, a comparison between the estimated intervention periods of the stochastic intervention model and a constructed intervention state index is conducted, too. Conclusions are summarized in the last section.

In order to model the exchange rate process in a small-open economy with frequent central bank interventions, this paper provides a short-run, central bank stochastic intervention model to explain the state-dependent dynamics of the depreciation rate. One extreme case of the model is a pegged exchange rate regime in which the exchange rate follows a zero-drift random walk process. Another extreme case of the model is a pure floating rate regime in which the exchange rate is determined by market fundamentals. The theoretical implication is that stochastic intervention behaviors of a central bank change the public's exchange rate expectations and thus make the effects of fundamentals on the exchange rate depend on the probability of future intervention. Specifically, under a non-intervention state, the relation between the exchange rate and the future fundamentals is looser than under a floating exchange rate regime while that between the exchange rate and the contemporary fundamentals is closer than under a floating exchange rate regime. This is a feature that the article embeds in the empirical estimation. With the guide of the theoretical model this article provides a method in detecting a central bank's intervention without any intervention data. That is, we use a regime-switching model for estimating the unobservable intervention and non-intervention states. This is especially important since most of the emerging market countries do not publish their intervention data. Applying Taiwan's data to our stochastic intervention model, we find that our model outperforms both the random walk and the pure floating rate models in terms of the likelihood value of the NT$/US$ depreciation rate and diagnostic tests. In the estimated pegged period, our stochastic intervention model has a forecasting power statistically equal to its random walk counterpart in terms of Diebold and Mariano's (1995) statistics. In addition, with the constructed intervention state index in this article, the estimation of the stochastic intervention model is found to be consistent with the hypothesis that regime switches of exchange rates are due to a central bank's (non-)interventions. What this article has done is the first step to closely relate time-series analysis to an economic model. Empirical results including the failure of the diagnostic tests for the serial correlation and the possibility of a lack of robustness in the estimated non-intervention period during the 1997 Asian financial crisis have to be improved. Sarno and Taylor's (2001) marvelous work focuses on the effectiveness of exchange rate intervention—one important topic not addressed by us. In fact, in this article we implicitly assume that all central bank interventions are successful. The exchange rate process under ineffective intervention together with non-intervention is attributed to market fundamentals. However, the process of exchange rates governed by market forces and the process of exchange rates under ineffective intervention (e.g., during the 1997 Asian financial crisis) may not be the same. An extension for distinguishing the two cases constitutes the subject of future research.