گرفتن پویایی زمان مداخله بانک مرکزی

We estimate central bank reaction functions using the autoregressive conditional hazard model and the autoregressive conditional binomial model. We find that the Federal Reserve and Bundesbank intervened when the market was calmer, and the Bundesbank intervened in response to exchange rates being out-of-line with fundamentals. Japan intervened in response to changes in the nominal exchange rate, and intervention differed before and after Eisuke Sakakibara became Director General of the International Finance Bureau of the Ministry of Finance in Japan. We argue that these results are consistent with central bank policy goals and the effect of intervention on the exchange rate.

Sterilized intervention is an intriguing policy tool utilized by central banks.1 Even though interventions are small relative to total foreign exchange market activity (the largest intervention in the DM/$ market in the 1980s was $1.3 billion compared to total daily market activity of $1300 billion, and the Federal Reserve intervened at times in amounts as little as $50 million), there appears to be a consensus amongst central bankers that intervention is an effective policy tool. All of the central bankers surveyed by Neely (2001) believe that intervention is able to alter the exchange rate. Additionally, in surveying the academic literature on intervention’s effectiveness, Neely (2005) finds that there is a consensus amongst a key group of papers that “interventions successfully move exchange rates, at least in the very short run” (p. 11). A related question is whether central bank intervention behavior is consistent with stated policy goals. These policy goals are commonly thought of as being to: (a) move exchange rates into line with long-run fundamentals, (b) resist undesirable exchange rate changes, and/or (c) remove excess volatility from the foreign exchange market (Neely, 2001). Studies that estimate central bank reaction functions (i.e., functions that determine when a central bank will intervene) obtain mixed results as to what motivations trigger an intervention. Baillie and Osterberg (1997a) find evidence that the U.S. Federal Reserve and German Bundesbank intervened in response to the DM/$ exchange rate moving away from its target, but not in response to excess foreign exchange market volatility. The authors also found mixed evidence for the Bank of Japan intervening in response to excess market volatility and in response to the exchange rate moving away from its target. That is, Baillie and Osterberg (1997a) found that the Bank of Japan intervened in response to excess volatility by buying and selling dollars, while they found that the Bank of Japan intervened in response to the exchange rate moving away from its target only by selling dollars. Almekinders and Eijffinger, 1994 and Almekinders and Eijffinger, 1996 find evidence that the Federal Reserve and Bundesbank have intervened in order to resist undesirable exchange rate changes (i.e. “leaning against the wind”) and in response to increases in exchange rate volatility (to “calm disorderly markets”). Ito (2002) finds evidence for Japan “leaning against the wind” from April 1991 to March 2001, but this behavior is only statistically significant in the second subsample of his data (after Eisuke Sakakibara became Director General of the International Finance Bureau of the Ministry of Finance). However, the institutional background given in Ito (2002) suggests that interventions prior to Sakakibara are “leaning against the wind”, whereas interventions after Sakakibara are “leaning with the wind” (intervening to further support exchange rate changes if those changes are in the desired direction). Thus, there is a consensus as to the effects intervention has on the foreign exchange market. However, it is less clear if central banks consistently follow their policy goals when deciding whether or not to intervene. The goal of the present paper is to investigate what motivations trigger a central bank intervention, in order to determine whether or not intervention behavior is consistent with stated policy goals. Similar to monetary policy actions, there is considerable time dependence in intervention decisions, with successive periods of intervention being followed by successive periods without intervention (see Table 1). That is, when something causes the central bank to decide to undertake an intervention, the central bank tends to intervene on successive days. Thus, the serial correlation of intervention decisions must be controlled for if motivations for intervention are to be teased out of the data. For example, suppose there is an overnight appreciation in the nominal exchange rate, and the central bank decides to intervene to counter this movement. And, suppose this appreciation in the nominal exchange rate was a one-time event, but the intervention in response to it happened on successive days (including days where no change in the nominal exchange rate took place). In this case, a model that fails to account for the dynamic nature of intervention behavior will not find statistical evidence that a change in the nominal exchange rate triggers intervention, even though such a change in fact does. This is relevant because in his survey of central bankers, Neely (2001) finds that 95% of central bankers either “sometimes” or “always” determine the size of interventions based on the market’s reaction to initial interventions. That is, central bankers intervene in response to a particular event, observe the market’s reaction to it, and then further intervene if the market’s reaction was not consistent with the central bank’s objective. Table 1. Comparison of the time dynamics in weekly monetary policy data and daily intervention data. Monetary policy Intervention Date Event uN(t)uN(t) Date Event uN(t)uN(t) 6/1/1984 0 1 11/27/1987 0 1 6/8/1984 0 1 11/30/1987 1 11 6/15/1984 1 10 12/1/1987 1 1 6/22/1984 1 1 12/2/1987 1 1 6/29/1984 0 1 12/3/1987 0 1 7/6/1984 0 1 12/4/1987 0 1 7/13/1984 0 1 12/7/1987 1 3 7/20/1984 1 4 12/8/1987 0 3 7/27/1984 0 4 12/9/1987 0 3 8/3/1984 0 4 12/10/1987 0 3 8/10/1984 1 3 12/11/1987 1 4 8/17/1984 0 3 12/14/1987 1 1 8/24/1984 0 3 12/15/1987 0 1 8/31/1984 1 3 12/16/1987 0 1 9/7/1984 0 3 12/17/1987 0 1 9/14/1984 0 3 12/18/1987 1 4 9/21/1984 1 3 12/21/1987 0 4 9/28/1984 1 1 12/22/1987 0 4 10/5/1984 1 1 12/23/1987 0 4 10/12/1984 1 1 12/24/1987 0 4 10/19/1984 1 1 12/28/1987 0 4 10/26/1984 0 1 12/29/1987 1 6 11/2/1984 0 1 12/30/1987 1 1 11/9/1984 1 3 12/31/1987 1 1 11/16/1984 0 3 1/4/1988 1 1 11/23/1984 1 2 1/5/1988 1 1 11/30/1984 0 2 1/6/1988 1 1 12/7/1984 1 2 1/7/1988 0 1 Notes : The monetary policy data is a subset of the data set used by Hamilton and Jordà (2002). The intervention data is a subset of the Federal Reserve intervention data set used by the present paper. The variable “Event” takes the value of 1 if a Federal funds target change or intervention takes place on a given day, and 0 otherwise. uN(t)uN(t) is the duration between the nn th and (n−1)(n−1) th events. Table options We add to the intervention literature by capturing the dynamic nature of intervention decisions by estimating two new econometric models designed to address this issue, the autoregressive conditional hazard (ACH) model and the autoregressive conditional binomial (ACB) model. The ACH model of Hamilton and Jordà (2002) is an extension of the autoregressive conditional duration (ACD) model of Engle and Russell (1998). The ACD estimates the time between successive events. That is, the ACD answers the following question: if an intervention took place today, how long will it take until the next one? The ACH extends the ACD to estimate the probability of observing an intervention on a given day, conditional on the information known up to that point. This information includes both the expected duration, or time, between successive interventions and information related to stated policy goals. Hamilton and Jordà (2002) have success in applying the ACH to monetary policy data. That is, using an ACH model and weekly data spanning 1984–2001, the authors are able to outperform a standard vector autoregression in terms of mean-squared error in predicting federal funds target rate changes. Given that monetary policy and intervention data have similar time dynamics (see Table 1), this suggests that the ACH may be well-suited to examine motivations responsible for central bank intervention. On the other hand, the ACB, which is a binary extension of the autoregressive conditional multinomial model of Russell and Engle (2005), allows the probability of an intervention occurring on a given day to depend on the past history of interventions and on the past response probabilities of intervention, along with information related to intervention’s policy goals. Additionally, the ACB model nests a probit model, allowing us to directly test the relevance of the time dynamics in the central bank’s intervention decision. Thus, in addition to determining whether or not intervention behavior is consistent with policy goals, a secondary goal of this paper is to compare and contrast the performance of these two models, as they are designed to capture a common characteristic of the data, which is the successive nature by which interventions take place, once the decision to intervene has been made. Using data for interventions by the United States Federal Reserve and the German Bundesbank spanning January 5th, 1987 to January 22nd, 1993 and for interventions by the Bank of Japan spanning April 1st, 1991 to February 28th, 2001, we find that the ACB outperforms the ACH for all central banks in terms of the Schwarz Bayesian Criterion of Schwarz (1978), the pseudo R2R2 ( McFadden, 1974), and in-sample and out-of-sample predictability. 2 The Rivers and Vuong (2002) test of non-nested likelihoods rejects the ACH in favor of the ACB for all central banks. Thus, using the ACB to test motivations for intervention implied by the Plaza Agreement and Louvre Accord (and emphasized by Neely, 2001 and other studies), we find that the Bundesbank intervened in order to move the exchange rate into line with long-run fundamentals, but the Federal Reserve did not. Interestingly, we find that both the Federal Reserve and Bundesbank preferred to intervene when foreign exchange market volatility was low, rather than intervening when volatility was high with the goal of reducing it. As far as we know, this is the first time that a negative relationship has been found in a reaction function between foreign exchange market volatility and the probability of intervention. Papers such as Dominguez, 2002 and Dominguez, 2006 find that market volatility increases following an intervention. Thus, our results suggest that the Federal Reserve and Bundesbank did not want to make a volatile market more volatile through intervention. We find strong evidence that the Bank of Japan has intervened in response to day-by-day changes in the nominal ¥¥/$ exchange rate. In contrast with Ito (2002), we find statistical evidence that the direction of this response is different following Eisuke Sakakibara becoming the Director General of the International Finance Bureau of the Japanese Ministry of Finance on June 1st, 1995. 3 Intervention behavior by the Bank of Japan is found to be “leaning against the wind” prior to Sakakibara’s appointment and found to be “leaning with the wind” following his appointment. We argue that these results are consistent with central bank policy goals, evidence in the academic literature as to the effects intervention has on the foreign exchange market, and the institutional environment at the time. The remainder of this paper is organized as follows. Section 2 describes the modeling of the probability of intervention using the ACH and ACB. Section 3 presents the results. Section 4 offers a comparison of the two econometric methods, and section 5 concludes.

What motivations trigger a central bank intervention? In this paper, we utilized two dynamic, binary choice models to test motivations for why intervention in the foreign exchange market might occur. The ACB results (which we argue are the preferred results) suggest that the Federal Reserve did not intervene in response to a deviation of the exchange rate from fundamentals and that the Federal Reserve preferred to intervene when the market was calmer. As far as we know, this is the first time a negative relationship between intervention and volatility has been shown in a reaction function. We found that the Bundesbank intervened in response to the exchange rate deviating from the one implied by fundamentals and, like the Federal Reserve, that the Bundesbank intervened when the market was calmer. In contrast with Ito (2002), we found that the Bank of Japan was leaning against the wind during the time period before Sakakibara, whereas it leaned with the wind during Sakakibara. We also found little evidence for the Bank of Japan intervening in response to excess market volatility. As discussed in Section 3.3, we believe these results are consistent with central bank policy goals, evidence in the academic literature as to the effects of intervention on the foreign exchange market, and the institutional environment at the time. Additionally, our results suggest that the form of the time dynamics assumed in a binary model matters both for global and local model performance and for the sign and significance of exogenous variables. By capturing the time dynamics in a straightforward manner using the ACB model, we were able to not only increase model performance, but also to find new and novel results using daily intervention data. Thus, the manner by which the time dynamics are captured is important not only for this application, but future applications as well.