Improving Moving Average Trading Rules with Boosting and Statistical Learning Methods, страница 6

We focused our empirical work on combining the predictive information provided by variablelength moving average rules (VMA), perhaps the simplest and most widely used technical rules among practising technicians, which have generated intriguing results in many academic studies (Brock et al., 1992). Paying special attention to VMA rules usually refl ects the fact that in fi nancial markets profi table trading is infl uenced more by signalling the correct direction of change than the amount of change. Thus models with small forecast errors may be less profi table than models better able to forecast the direction of price movements. Nevertheless, the boosting methodology may be extended to any kind of rules and applied to combine any sort of price predictions.

The data were collected over the period 4 January 1993 to 31 December 2002, consisting of 2253 observations during 10 consecutive annual periods.

During these periods we have considered the set of moving average trading rules [a, b, c], where a = 1, 2, 5, 10 represents the length of the short moving average, b = 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200 represents the length of the long moving average, and c = 0%, 1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, 10% represents a band introduced in order to avoid deceptive signals. This wide set of rules contains the most usual ones employed by technicians (see Brock et al., 1992).

Therefore, a total of 704 moving average trading rules have been considered in which the statistical learning methods (Boosting, Bayesian model averaging, and Committee methods), all described above, have been applied. In every case, out-of-sample technical trading rules are obtained using the statistical learning methods with algorithms that have a training period of 100 days.

Table I. Several statistical and economical fi tness measures for moving averages and statistical learning methods. Cost = 0.2%. Full sample: from 4 January 1993 to 31 December 2002