Efthymios G. Pavlidis, Ivan Paya and David A. Peel 1073- A regime switching behavior of the exchange rate can be attributed to factors such as
the heterogeneity of opinions among agents, the presence of transaction costs, the inter-
action of chartists and fundamentalists, the peso problem, different monetary and fiscal
policies between countries, as well as the implications of the dirty floating exchange rate
regime (see Engel and Hamilton, 1990; De Grauwe and Vansteenkiste, 2001; Lee and Chen,
2006). - The use of MS is also motivated by parameter instability in empirical exchange rate models
(see Rossi, 2006), which may be attributed to “swings” in expectations about future values
of the exchange rate (Frankel, 1996), as well as by rational expectations models of exchange
rate determination in which the weight attached to fundamentals by practitioners changes
over time (Bacchetta and van Wincoop, 2004). - In a related study, Sarno and Valente (2005) examine the evolution of the relationship
between fundamentals and the exchange rate by employing the recursive procedure of
Pesaran and Timmermann (1995) to real-time data. The authors use a broad set of fun-
damentals for five major US dollar exchange rates over the post-Bretton Woods era. In
the preliminary analysis, a “virtual search” is conducted over all possible models and the
optimal combination of fundamentals is determined period by period. In the cases where
the best model outperforms the RW, a real-time forecasting exercise is implemented. The
main implication of the experiment is that fundamentals contain predictive power for the
movements of the exchange rate. However, the importance of each of the fundamental
variables changes over time. Furthermore, conventional model selection criteria cannot
identify the “correct” model to beat the RW in real time. - Engel and West (2005) note that estimates of the interest semi-elasticity of money
demandλtypically range from 29 to 60, which implies that in the monetary modelb
is between 0.97 and 0.98. - However, Sarno and Taylor (1998) note that panel unit root tests tend to reject the null
of a unit root even if a single series is stationary. Moreover, Rapach and Wohar (2004)
emphasize the importance of the homogeneity assumption. If such an assumption is
not empirically supported then pooling data across countries may result in false infer-
ence. - For example, Frankel and Rose (1996) and Lothian (1997), among others, show that real
exchange rates mean revert in the long run by using pooled data; Groen (2000) and Groen
and Kleibergen (2003) find evidence of cointegration between monetary fundamentals
and exchange rates. - Bootstrap samples are generated according to the fitted VAR by sampling from the esti-
mated residuals with replacement. Next, the slope coefficient is estimated by panel
dynamic OLS and the corresponding t-ratio is computed. The above procedure is
repeated 2,000 times so as to obtain bootstrap distributions andp-values. - The bootstrap procedure is the same as before, with the exception that the second equation
of the previous null DGP becomes:
zi,t=μiz+γizi,t− 1 +∑qij= 1φi1,jsi,t−j+∑qij= 1φi2,jzi,t−j+εiz,t.Cointegration requires− 2 <γi<0.- Groen (2005) emphasizes the importance of expectations on the validity of the monetary
model. He examines the European Union exchanges rates of Canada, the US and Japan
for the period from 1975 to 2000. The cointegration framework adopted is an extension
of the Johansen method for a panel of VECM models which allows heterogeneous short-
run dynamics (see Groen and Kleibergen, 2003). Overall, exchange rates appear to be
predictable at medium- to long-term horizons, i.e., one to four years. However, the results
are sensitive to the cointegrating parameters restriction.