Abstract
Technical analysis includes a huge variety of trading rules. This fact has always been a serious hindrance to the large number of market efficiency studies implemented either to demonstrate the profitability of market-beating systems or to deny their operational feasibility. For evident reasons it is practically impossible and theoretically weak to systematically analyse the entire body of all trading rules. We therefore propose a novel method to form natural classes of trading rules which are found to be robust to changing market scenarios. In particular, groups are formed adopting a similarity measure based on the investing signals of the trading rules. Our clustering methodology adopts a Markov chain bootstrapping technique to generate differentiated scenarios preserving volume and price joint distributional features. An application is developed on a sample of 674 trading rules. Results show that six groups (here identified as trading styles) are sufficient to explain the large portion of the investing signals variance. We also suggest applications of our results to fund performance measurement and the analysis of financial markets.
Acknowledgement
We would like to thank Nicola Doninelli for his helpful suggestions.
Notes
1 Observe that technical analysis comprises many analytical indicators as well as graphical methods. A recent article by Lo et al. (Citation2000), for example, employs pattern recognition techniques known as ‘smoothing estimators’ in an effort to mimic human visual pattern recognition. However, in this study we do not consider visual or graphical methods.
2 Residuals et
have been obtained applying an autoregressive model to the original turnover data vt
with 5 lags:
3 k-means clustering algorithm is a nonhierarchical method of clustering large numbers of observations (the desired number of clusters is to be set in advance). Standard references for these techniques are MacQueen (Citation1967), Anderberg (Citation1973), Hartigan (Citation1975), Späth (Citation1980) and Everitt (Citation1993).