Efthymios G. Pavlidis, Ivan Paya and David A. Peel 1065bootstrapp-values indicate that the null hypothesis of no cointegration is rejected
at the 5% level. In turn, Mark and Sul (2001) test for short-horizon in-sample pre-
dictability in the presence of cointegration.^78 The results provide strong evidence
in favor of predictability based on both monetary and PPP fundamentals. Finally,
the authors examine whether macroeconomic fundamentals contain power in
forecasting exchange rates at horizonsk=1, 16. They use Theil’sU-statistic and
construct bootstrap critical values under the assumption of cointegration. Over-
all, forecasts based on monetary fundamentals dominate forecasts based on the
PPP fundamentals and the RW with a drift for the majority of countries when
the US dollar or the Swiss franc is the numeraire currency, but not in the case of
the Japanese yen.^79
In this section we have shown how, more than 30 years after the breakdown
of Bretton Woods, the difficulty of forecasting exchanges rates using economic
fundamentals has become a stylized fact in international finance. Although the
availability of longer datasets on modern floating rates and the application of recent
sophisticated econometric techniques regarding panel data, nonlinear models, as
well as forecast evaluation measures, are promising, researchers and practitioners
are still faced with the problem of deriving models with robust behavior in terms
of out-of-sample forecasting across exchange rates and time periods.
22.7 Conclusions
We have provided a selective overview of a few of the key relationships which will
play critical roles in determining the behavior of the exchange rate and an eval-
uation of the efficacy of forecasting methods. The major change in their analysis
over the last decade has been the application of more sophisticated time series tech-
niques motivated by theoretical considerations such as the limits to arbitrage and
the microstructure of the exchange market. Nonlinear models seem able to provide
some explanation of the PPP puzzle and the forward bias problem. However, issues
of how data sampled at a different frequency to the economic decision impacts on
the nonlinear models needs further investigation. Also, the recent finding of joint
long memory and nonlinearity in these relationships is a new puzzle. The possibil-
ity that the exchange market can exhibit bubbles is of long-standing interest. The
new analysis of Phillipset al.(2006) would appear to be an exciting development
and might provide new insights on this issue over the next few years.
The nonlinear methods appear to offer scope for improved forecasts than the
previous linear models. However, researchers and practitioners are still faced with
the problem of deriving models with robust behavior in terms of out-of-sample
forecasting across exchange rates and time periods, even when the model employed
for estimation or forecasting is appropriate for the policy regime in operation (or
anticipated).
Notes
- In fact, the first example of such arrangements was apparently Austria-Hungary between
1896 and 1914 (Flandreau and Komlos, 2003). Other important examples where the