بررسی بحران ارز: رویکرد A بیزی مارکوف سوئیچینگ

In this paper we examine the nature of a currency crisis. We do so by employing an out-of-sample forecasting exercise to analyse the Mexican crisis in 1994. Forecast evaluation was based on modern econometric techniques concerning the shape of forecaster’s loss function. We also extend the empirical framework suggested by Jeanne and Masson [Jeanne, O., Masson, P., 2000. Currency crises and Markov-switching regimes. Journal of International Economics 50, 327–350] to test for the hypothesis that the currency crisis was driven by sunspots. To this end we contribute to the existing literature by comparing Markov regime switching model with a time-varying transition probabilities with two alternative models. The first is a Markov regime switching model with constant transition probabilities. The second is a linear benchmark model. Empirical results show that the proxy for the probability of devaluation is an important factor explaining the nature of currency crisis. More concretely, when the expectation market pressure was used as a proxy of probability of devaluation, forecast evaluation supports the view that currency crisis was driven by market expectation unrelated to fundamentals. Alternatively, when interest rate differential is used as a proxy for probability of devaluation, currency crisis was due to predictable deterioration of fundamentals.

The currency crises of the EMS in 1992–1993, of Mexico in 1994 and the Asian crises in 1997 have been accompanied by considerable controversy over their causes. There are two main theoretical models, that explain currency crises. The first generation of currency crises models was determined by monetary and fiscal policy that are inconsistent with maintaining the fixed currency peg. The failure of the first generation model to explain the EMS currency crises led to the second generation model. More concretely, although expansionary monetary and fiscal policy may have been an issue in some countries such as Italy and Spain this was not the case in some others such as UK and France. In the second generation model the central bank acts as an optimizer where the decision to devalue or not seems to be motivated by the desire to avoid adverse macroeconomic consequences of maintaining the peg. The second generation model adds two new elements to the first generation model. The first new element concerns the notion of macroeconomic fundamentals which, in the second generation models, include any variables that might affect the loss function of the central bank. In addition, the second generation model emphasizes the role that market expectations can have on the monetary authority’s decision to devalue or not. This leads to the second new element of the second generation model. Specifically, the model provides the theoretical framework of self-fulfilling speculation and multiple equilibria. The logic of self-fulfilling crises is based on the idea that devaluation expectation increases the cost of retaining a peg and therefore the desire of the policy-maker to devalue. One way to defend a currency peg is by raising the ex ante nominal interest rate, which affects economic growth negatively. Under such circumstances, the policy maker might prefer to devalue rather than to maintain high interest rates. Therefore, the decision to devalue or not is affected by market expectations regarding changes in monetary policy. The disconnection of fundamentals from market expectations is the main property that differentiates the first generation model from the second generation model. Jeanne, 1997 and Jeanne, 2000 in the so-called escape clause model provided the theoretical framework which reconciles both models. More concretely, the escape-clause model of currency crises views the fixed exchange rate regime as a conditional commitment. Jeanne (2000, p. 6), argues that “the main message of the escape-clause model is that currency crises should be analysed in the context of a conflict among contradicting policy objectives. In the limit any type of currency crises can be analysed in the escape-clause perspective”. This can be shown by endogenizing monetary and fiscal policy in the first generation model. Although the first generation model shows that currency crises are the consequence of the monetary and fiscal policies followed by government, it has not addressed the question of why these policies were pursued. Jeanne (2000) shows that there is a level of interest rate that authorities in the first generation model could adopt and defend the fixed exchange rate. If raising interest rates were not costly then currency crises would not occur. Therefore, once the role of interest rates is introduced into the first generation model, the logic of the first generation model is the same as the logic of the second generation model.1 Jeanne (2000) emphasizes that the escape-clause model provides a political compromise between fundamentalist and the proponent of the self-fulfilling view.2Jeanne (2000, p. 5) also argues that “self-fulfilling and fundamentalist are not mutually exclusive. For a currency to be vulnerable to self-fulfilling speculation, the fundamentals must first put it in a state of fragility. The occurrence and the precise time of crises may be impossible to predict solely on the basis of fundamentals, but the latter play a crucial role”. Empirical literature shows that expectation of devaluations are subject to abrupt shifts unrelated to fundamentals. The regime shifts on market expectations can be interpreted as jumps between multiple equilibria. More concretely, Jeanne and Masson (2000) show that strategic complementaries between market expectations, about the intended policy rule and the policy actually adopted, produces multiple equilibria.3 It is important to observe that the presence of multiple equilibria is due to speculators sharing a common knowledge of the information set.4 In Jeanne and Masson (op. cit.) and in Jeanne (1997), what coordinates the public’s expectations and leads the economy across the different equilibria is a sunspot (waves of optimism or pessimism). Jovanovic (1989) shows that, if a sunspot is independent from fundamentals, then it is necessary to distinguish the dynamic of the fundamentals process from the sunspot process. A Markov regime-switching (MRS) model provides a framework that satisfies the distinction between the two processes. In particular, the data generating process of a MRS model consists of two components: the first component gives rise to the autoregressive dynamic of fundamentals, and the second component describes the dynamics of an unobserved state variable which follows a Markov process. The second component represents the sunspot. According to Jeanne and Masson (2000), what defines a sunspot in the context of MRS model is the assumption of a constant transition probability matrix.5 This implies that switches of the unobserved state variable from one equilibrium to the other is independent of fundamentals. Jeanne and Masson, 2000, Piard, 1997 and Psaradakis et al., 1997 show that MRS models do a better job in describing the speculative attack on the French franc in 1993 than the simple linear models. The same methodology is applied by Gonzales-Garcia and Jesus (1999) to the 1994 Mexican crises and Cera and Saxena (1999) to the 1997 Indonesian currency crisis. However, empirical support on sunspots by these studies was based on an in-sample forecast comparison of MRS with a linear model. Evidence of a better in-sample fit of MRS than the in-sample fit of linear models led these studies to conclude that speculative activity jumps up and down driven by a sunspot. However, although nonlinear models outperform linear models in an in-sample forecasting exercise this was not the case in out-of-sample forecasting exercise where linear models were found to perform often better than nonlinear models (Clements and Smith, 1999 and Diebold and Nason, 1990). An out of sample forecast comparison of the MRS model with the linear model is required in order to provide empirical support to a currency crises model driven by sunspots. Furthermore, we argue that the definition of the sunspot given by Jeanne and Masson (2000) in the context of the MRS model is rather restrictive. We allow the switch in market expectations to be a function of fundamentals. We do so by using a MRS model with time-varying transition probabilities. Statistically significant effects of fundamentals on transition probabilities would lead to the conclusion that speculative attacks are driven not only by an external uncertainty but also by fundamentals. This is consistent with the argument of Jeanne (2000) that the first and second generation models are not mutually exclusive. The main contribution of the paper is that it provides the empirical framework that needs to be used to explain the nature of currency crises. We do so by employing an out-of-sample forecasting exercise to analyse the Mexican crisis in 1994. An out-of-sample forecasting comparison of a duration independent MRS6 model with a MRS model with time-varying transition probabilities indicates whether market expectation are subject to abrupt shifts unrelated to fundamentals. Evidence that the model with time-varying transition probabilities outperforms the model with fixed transition probabilities indicates that market expectations change gradually up to a certain point on the basis of fundamentals and then jump suddenly due to an external uncertainty independent of fundamentals. Furthermore, evidence that the later MRS model outperforms a linear benchmark model provides empirical support to the currency crisis model with multiple equilibria. We evaluate the forecast performance of linear and nonlinear model by taking into account the forecaster’s loss function. This is so because recent literature in economics and finance shows that policy makers have an asymmetric loss function and macroeconomic variables follows a nonlinear process. Under such circumstances, traditional test statistic such as the RMSFE is not appropriate to select the best model.7 To this end, we follow Patton and Timmermann (2006) who suggest a test statistic to select the best forecasting model regardless forecasters’ loss function. Forecast evaluation proceeds into two steps. First, we assume that forecasters have a quadratic loss function and we select an optimal forecast on the basis of the RMSFE criterion. Second, we adopt a more realistic assumption that forecasters have an unknown asymmetric loss function and we use a quantile type test to select a rational forecast.8 This paper proceeds by introducing the empirical methodology utilized in this paper. Section 3 presents the econometric methodology adopted to bring the model to the data. Section 4 explains data and empirical results from an application to the Mexican crises in 1994. The final section summarises and concludes

The aim of this paper to test whether currency crisis can be explained by the escape-clause model suggested by Jeanne (1997). The key element of the escape-clause element is that both fundamentals and external uncertainty play a significant role in the genesis and occurrence of currency crisis. To bring model to data we extend the Markov regime switching model suggested by Jeanne and Masson (2000) by allowing transition probabilities to be time-varying. We employ an out-of-sample forecasting exercise to compare MRS model with time-varying transition probability both with two alternative MRS models where transition probabilities are constant and with a linear benchmark model. We evaluate model’s forecast performance based on modern econometric techniques. More concretely, we select an optimal forecast by taking into account forecasters’ loss Function. In line with recent literature in monetary economics, forecasters’ might have an unknown asymmetric loss function. Under, such circumstances the traditional RMSFE criterion is not appropriate to select an optimal forecast. We follow the suggestion of Patton and Timmermann (2006) and employ a quantile test to select an optimal forecast under a general loss function and a general DGP. Forecast evaluation is implemented in two steps. First, we assume that forecasters have a quadratic loss function and we select an optimal forecast on the basis of RMSFE criterion. Empirical results shows that in the case where EMP was used as proxy for a probability of devaluation the MRS FTPM2 outperforms both MRS TVP and AR(1) model. Alternatively, in the case of IRD, the AR(1) produces the best forecast among the three models. Under such circumstances currency crisis, depends on the proxy for devaluation probability, is driven either by animal spirits or by predictable deterioration of fundamentals. This implies that currency crisis can be explained either by the first or second generation model but not by their reconciliation induced by the escape-clause model. Evidence that the proxy for probability of devaluation is important factor explaining currency crisis is not satisfactory. In the second step, we adopt a more realistic assumption concerning the shape of forecasters’ loss function than the QLF assumed in the first step. We assume that forecasters have an unknown asymmetric loss function. Evidence show that for a part of our sample the AR(1) model found to be rational