| This dissertation consists of three empirical studies on the potential nonlinearity in a central bank's foreign exchange intervention and in the effects of intervention on the exchange rate. One interesting result is that despite general acceptance in the literature, a friction model which assumes a specific type of nonlinearity may not be better than a linear model in explaining intervention behavior. However, a more flexible nonlinear model, i.e. a threshold model, explains intervention better than a linear model. A threshold model is also found to be useful in characterizing the conditions on which intervention becomes effective in countering excessive exchange rate movements.; Under a floating exchange rate system, central banks do not intervene most of the time although exchange rates are fluctuating continuously. A friction model for intervention is based on a hypothesis that intervention occurs if the exchange rate is highly unstable and does not occur otherwise. While previous studies accept this hypothesis, without testing, as an appropriate explanation for the infrequency of intervention, the hypothesis is tested in this study with official daily data on intervention by US and German central banks. If the underlying hypothesis is true, then an economic model built on it (a friction model) must explain the actual intervention better than a model without such information (a linear model). As reported in Chapter 1, however, the explanatory power of the friction model is lower than that of a linear model in terms of the degree of correlation (R2) between the observed amount of intervention and the fitted values. This result implies that the core assumption of the friction model may not be true at least on a daily basis.; In Chapter 2, the test is about a similar hypothesis that a central bank's reaction to a low degree of instability is different from its reaction to a high degree of instability. This assumption is weaker than the friction hypothesis in that a central bank is allowed to react to small values, as well as large values, of the explanatory variables. The result is that a model allowing this type of regime-switching (a threshold model) explains intervention significantly better than a linear model. The relative frequency and average size of intervention are both larger on average in a high instability regime than in a low instability regime.; The empirical results in Chapter 3 indicate that intervention can be effective under certain conditions although not effective on average. First, when the size of intervention is large enough to exceed a threshold, such intervention tends to be effective. Secondly, intervention is effective when the short-run upward or downward trend is not too strong to lean against. Finally, if central banks wait for the right timing to break a trend driven by chartists or noise-traders, who make short-term trading decisions based on technical trading rules rather than economic fundamentals, such strategic intervention becomes effective. |
