The predictive performance of technical analysis has been extensively examined using different evaluation measures. On the one hand, Diebold and Nason (1990) show that a wide range of univariate models, both linear and nonlinear, fail to beat the no-change prediction measured by out-of-sample MSE.
On the other hand, empirical investigations using sign prediction or profitability measures repeatedly revealed superior performance of technical analysis relative to the random walk benchmark; see, for example, Levich and Thomas (1993) for exchange rates or Alexander (1961) for early findings for stock markets. While these results suggest that hidden structures exist that are impossible to detect by means of standard univariate models, attempts to identify these regularities have been unsuccessful (Dewachter, 1997).
The logic of technical analysis presented above, however, suggests that its forecasting success will be state dependent because it only predicts the direction of future exchange rate changes accurately when regime shifts alter the time series properties of unobservable fundamentals. However, the exchange rate is certainly not always driven by the dynamics of hidden fundamentals, implying that periods of technical forecast dominance are followed by periods of standard fundamental analysis prevalence.
In order to provide empirical evidence on the forecasting success of technical analysis, the econometric approach should be able to capture the switching between different kinds of exchange rate expectations. Following these considerations, we apply a Markov regime-switching procedure originally suggested by Engel and Hamilton (1990) and developed further by, among others, Dueker and Neely (2001), Engel (1994), Vigfusson (1997) and Dewachter (2001).
Prof. Stefan Reitz
Next: Model specification
Summary: Index