A threshold model for the French franc/deutschmark exchange rate

Abstract

The behaviour of the French franc/deutschmark exchange rate is examined in this paper. During the time period studied, these currencies were constrained to lie within prescribed bands relative to one another and the usual random walk explanation of the exchange rate may not be appropriate. The data are examined for evidence of non-linear structure and it is shown that a piecewise linear SETAR model provides a better explanation and superior forecasting performance than a random walk.

The performance of non‐linear exchange rate models: a forecasting comparison

“…The study in Chappell et al (1996) differs from those presented in Kräger and Kugler (1993) and in Peel and Speight (1994) since it is focused on the forecasting performance of non-linear models fitted to the levels, rather than the changes, of some bilateral ERM exchange rates considered at daily frequency. It is important to stress that if the forecast assessment (for more than one step ahead) is carried out on the basis of criteria such as the MSFE, the choice of data transformations is not neutral as shown by Hendry (1993, 1995): evaluation in differences is penalising relative to evaluation in levels.…”

“…The issue of whether to evaluate the forecasting performance for the differences or the levels of the series is distinct from the issue of whether to estimate a model in the differences rather than the levels. According to Chappell et al (1996), the inherent design of the ERM, based on the existence of a band in which the exchange rate is allowed to fluctuate without intervention by the central banks, could be the rationale for the presence of at least one threshold. Thus, the exchange rate follows a random walk process within the band but stationary autoregressive processes in the proximity of the ceiling or the floor such that the whole process exhibits mean-reverting features.…”

“…3 In contrast with most of the studies in which it is documented that the forecasting superiority of the non-linear models is often confined to the one-step-ahead horizon, the SETAR models estimated by Chappell et al (1996) yield noticeable gains outperforming the random walk and the linear model at horizons as long as five and ten steps ahead. Brooks (1996Brooks ( , 1997 analyses the daily British pound/US dollar exchange rate returns for the period 1974.…”

Stock Index Modeling Using Hierarchical Radial Basis Function Networks

“…Exchange rate series exhibit high volatility, complexity and noise that result from an elusive market mechanism generating daily observations [1]. Much research effort has been devoted to exploring the nonlinearity of exchange rate data and to develop specific nonlinear models to improve exchange rate forecasting, i.e., the Autoregressive Random Variance (ARV) model [2], Autoregressive Conditional Heteroscedasticity (ARCH) [3], self-exciting threshold autoregressive models [4]. There has been growing interest in the adoption of neural networks, fuzzy inference systems and statistical approaches for exchange rate forecasting problem [5][13] [14].…”

Forecasting Exchange Rates Using Neural Networks for Technical Trading Rules

“…Given this evidence, the natural question often is whether any nonlinearity can be exploited for improved forecasting. The most commonly considered nonlinear models are the threshold model and the bilinear model [see Chappell et al (1996) and Brooks (1997) among others], and the artificial neural network (ANN) model [see, for example, Kuan and Liu (1995)]. The nonlinear models reported in the literature almost always include lagged returns on the exchange rates themselves as explanatory variables.…”