مداخلات بانک مرکزی و رژیم گروه های نرخ ارز

This paper presents an endogenous switching regression model for the exchange rate process where the switch is defined by the central bank criteria functions for intervening. We study the signal effect of interventions on the exchange rate using Norwegian daily data on official interventions. We first find that interventions seemed to have been more effective in moving the exchange rate in the expected (‘desired’) direction in the regime when the exchange rate was kept away from the edges of the band. This type of intervention regime also reduces significantly the conditional volatility of the exchange rate. Thus, when the exchange rate was near the weakest edge of the currency band, its conditional variance was significantly larger than when it was moving around its central parity. Finally, we show that in order to obtain consistent estimates, intervention variables cannot enter as exogenous variables in the conditional mean (or conditional variance) of the exchange rate.

There is still considerable scepticism among economists and policy-makers as to whether sterilized interventions can have any significant effect on the exchange rate. Moreover, the effect of sterilized interventions on the exchange rate has not yet been firmly established in the academic environment. In the literature, one distinguishes between sterilized interventions, whose monetary effects are neutralized by offsetting domestic liquidity measures, and nonsterilized interventions, which alter money supplies and therefore involve the joint exercise of monetary policy and exchange-market policy. The issue of how the exchange rate is affected by sterilized interventions is controversial since these interventions are not supposed to alter the countries' relative supplies of domestic and foreign currency in the same way as non-sterilized interventions. In theory, sterilized interventions may affect exchange rates through three hypothetical channels: the portfolio balance1, the noise trading2 and the signalling channels. Here, we study the ‘signalling effect’ of sterilized intervention on an exchange rate in a currency band. The possibility of the ‘signalling effect’ has been suggested by Mussa (1981) who states that interventions (even sterilized) may signal monetary policy objectives. For instance, authorities may need to choose between two options, either to have relatively small variations of the exchange rate within its band, or to allow the exchange rate to move within the entire official band. By intervening frequently inside the band (intramarginally), the central bank may want to signal that it wishes to defend a narrower implicit band than the official one and reduce the volatility of the exchange rate. A different signal is given when interventions occur close to or at the edges of the currency band. The exchange rate is thus allowed to move within the entire band in order to gain some monetary independence (Svensson, 1994).3 Interventions are modelled here as dichotomous choices. The effect of interventions on the exchange rate is measured through hazard functions (weighted probabilities of intervention). We find this approach useful when information about interventions is not (at least not immediately) made public to the market. Instead, one can observe the central bank's participation in the foreign exchange market. In Norway, simultaneous intervention data are not available even in newspapers or similar sources.4 On the other hand, we justify the assumption that the market knows when the central bank intervenes by noting that the Norwegian foreign exchange market is significantly smaller than the markets for larger currencies such as the US dollar, Japanese yen or British pound. It is reasonable to expect that it is harder for the Norwegian central bank to intervene without being noticed by market participants. Hung (1997) introduces the noise trading channel hypothesis in a attempt to explain the negative or positive impact of a secret amount of sterilized intervention on exchange rate volatility. When noise trading is active, authorities may either use volatility-enhancing or volatility-decreasing intervention strategies, depending on how they wish to manage the exchange rate level. Thus, one may conclude secret interventions may undermine rather than strengthen the signalling effect hypothesis since secrecy by definition implies that there is no intention of revealing monetary policy objectives. A couple of observations are in order here. First, in general, it may be easier for central banks to conceal the amount of intervention (especially if they deal with several private market participants) than it is to conceal their participation in the foreign exchange market. This is particularly the case in small foreign exchange markets such as the Norwegian one. Second, on certain occasions, a central bank may want to signal its intended monetary policy and not to go unnoticed by the market when it is active in the foreign exchange market. On other occasions, central banks cannot avoid being noticed by market participants even when they want to. It is difficult then to distinguish between secret and non-secret central bank participation as well as to know what type of participation economic agents believed was occurring. Keeping in mind the above observations, we think that it makes little sense to test the noise trading channel hypothesis since central bank participation in the foreign exchange market, especially the Norwegian market, cannot easily be classified as secret. We can then safely argue that since the Norwegian authorities must have known that the central bank's participation could hardly pass unnoticed, they must have intended to signal their monetary policy objectives when they intervened. Here, we are undoubtedly studying the signal effect of central bank participation on the exchange rate. In this paper, the following issues are considered. First, interventions are modelled as dichotomous endogenous variables that both depend on and affect the exchange rate. Modelling (our qualitative) interventions as endogenous is of importance because under the hypothesis of signalling effects of interventions, exchange market participants actually revise their expectations when they believe that central bank interventions convey new, additional information. It is then necessary to show that there are certain objectives and criteria for intervening (i.e. to defend an implicit narrower band). If this is the case, interventions may possibly affect the exchange rate because they signal monetary policy objectives. Second, we present the differences in the stochastic process for the exchange rate when it depends on interventions occurring mostly close to or at the edges of the currency band (and the exchange rate is allowed to move within the entire official band) and when it depends on interventions occurring mostly inside the band (and the exchange rate moves within a more narrow band than the official one). Third, we investigate whether interventions close to or at the edges of the exchange rate target zone affect the time-variant variance of the exchange rate differently than when there are intramarginal interventions. Fourth, it is shown that the disturbances of the exchange rate mean equation are not only characterized by an autoregressive conditional heteroskedastic (ARCH) process, but also have another source of heteroskedasticity, namely their correlation with the intervention variables. 5 Note that the conclusions drawn concerning the first two issues above (the objectives for intervening and the stochastic process for the exchange rate) should indicate whether or not there have been different regimes for Norwegian central bank interventions. An econometric model that allows us to study the issues described above is presented here. The model is tested using daily data on sterilized spot interventions and the exchange rate for each of two different Norwegian exchange rate policy regimes followed by the Norwegian Central Bank (Norges Bank) between 1986 and 1990. The study covers the period from October 1986 to February 1990, while the regime shift, as announced by the authorities, occurred in mid-June of 1988.67 The first regime allowed the exchange rate to fluctuate between the upper and lower edges of the official currency band with interventions occurring mainly close to and at the edges. In the second regime, the exchange rate was only allowed to move within an unofficial implicit narrower band, within which there have been more frequent interventions by the central bank.8 Note that we do not choose Hamilton's (1989) regime-switching type of model to determine whether a regime switch has occurred as a result of changes in government policies and whether agents anticipated the switch. The introduction of uncertainty about the intervention policy is an interesting topic, but in this paper we would like to test whether the announced policy change did in fact take place and how it has affected the exchange rate. The effect of intervention decisions on the exchange rate variance using the regime-switching methodology is a more difficult task and for the moment is left for future research.9 The study most closely related to this paper is that of Lewis (1995a). The main difference is that she studies floating DM/$ and Yen/$ rates and that the variance is not parameterized to describe the volatility clustering of these exchange rates. It is, however, already very well established that such clustering or long-run persistence is an important feature of this type of financial times series (see Goodhart and O'Hara, 1997). The intervention series in Lewis are also compiled from daily newspaper accounts from the New York Times, the Wall Street Journal, and the London Financial Times. Other related studies include those presented by Baillie and Osterberg, 1997a and Baillie and Osterberg, 1997b. In the first of these, they estimate the mean of the forward premium and generate its conditional variance. They then use a Probit analysis to test whether the generated conditional variance of the forward premium Granger caused intervention. They, therefore, do not estimate jointly the intervention decision and the mean (and variance) of the forward premium. In Baillie and Osterberg (1997b), they conduct a similar analysis, but in this case, they study the log-change of the spot exchange rate instead of using the forward premium. Hung (1997) studies the effect of a secret amount of intervention on both actual and implied exchange rate volatility. The actual volatility is the ex-post volatility realized at the end of period t, calculated as the annualized standard deviation of daily changes in the log of the exchange rate over 10 weekdays during the 2-week period. Interventions are also considered to be exogenous here. The implied (expected) volatility is constructed using currency option prices provided by the Philadelphia Stock Exchange. option prices provided by the Philadelphia Stock Exchange. Loopesko (1984), Dominguez (1989) and Dominguez and Frankel (1990) use data on interventions to study their effect on floating exchange rates using the portfolio choice approach. In these studies, interventions are considered exogenous.10 The model in this paper makes some improvements on previous, related empirical studies of exchange rates. Methodologically, it provides a new model for the process of the exchange rate in a currency band. First, we model the conditional mean and variance of the exchange rate as functions of the intervention decisions. Second, we take into account important statistical properties of the exchange rate, e.g. its longrun volatility persistence and its leptokurtic unconditional distribution. This should improve, compared with previous approaches, the possibilities for studying the effect of the intervention objectives and exchange rate regime on the persistence in volatility. Finally, we model the exchange rate and interventions as determined simultaneously. The paper is organized as follows: In Section 2, we present a model for the exchange rate conditional on two criteria functions for intervening. In Section 3, some methodological problems are explained. In Section 4, we present the estimates of the model for two different types of intervention regimes. Section 5 concludes.

In this paper a particular process for the exchange rate is studied, namely an ARCH or GARCH process that depends on the intervention decisions and on the type of exchange rate policy regime. The conclusions are summarized as follows. First, we have found that only in the second of our two regimes did Norges Bank have the objective of minimizing the variance of the exchange rate, while in the first regime interventions took place mainly to prevent the exchange rate from moving outside the currency band. Second, interventions seem to have been more effective in moving the exchange rate in the expected (‘desired’) direction in any part of the currency band in the regime when the exchange rate was kept away from the edges of the band and a narrower currency band was defended (the second regime). A third conclusion is that, in the second regime, the conditional variance of the exchange rate is significantly smaller than in the first regime. More specifically, central bank interventions in the form of buying and selling foreign currency have reduced the volatility of the exchange rate in the second regime, while interventions selling foreign currency seem to have increased the volatility of the exchange rate in the first regime. One main reason that central bank participation in the foreign exchange market was more successful in reducing the volatility of the exchange rate in the second regime must have been the signaling effect of the objectives of the central bank. These were to keep the exchange rate as close as possible to the central parity or to defend a narrower implicit band. A last conclusion is that if we do not include the intervention decisions as explanatory variables in the conditional mean and variance of the exchange rate, we may obtain biased estimators of the magnitude and persistence of the variance, as can be seen in Figs. 7 and 8. Moreover, the exclusion of this intervention variable as an endogenous explanatory variable in the ARCH or GARCH models can result in heteroskedastic disturbances in the autoregressive exchange rate equation. Disregarding this heteroskedasticity would result in inconsistent estimators. One should also note that most of the studies modelling the exchange rate as an autoregressive (univariate or multivariate) conditional heteroskedastic process have been little concerned with finding the causes of exchange rate volatility. Here, we have tried to study the effects of interventions by the Central Bank of Norway (Norges Bank) not only on the mean but also on the volatility of the Norwegian currency basket. The results strongly indicate the importance of modelling interventions and exchange rates as interdependent variables. Certainly, more detailed information on intervention activity (i.e. when it does take place) would make it possible to build and estimate more appropriately specified models of these relationships.