Identifying Stock Similarity Based on Multi-event Episodes

Predicting stock market movements is always difficult. Investors try to guess a stock’s behavior, but it often backfires. Thumb rules and intuition seems to be the major indicator. One approach suggested that instead of trying to predict one particular stock’s movement with respect to the whole market, it may be easier to predict a stock As movement based on another stock Bs movement; the reason being that A may get affected by B after Bs movement, giving the investor invaluable time advantage. Evidently, it would be very useful if a general framework can be introduced that can predict such dependence based on any user defined criterion. A previous paper laid a basic framework for a single event based criterion, but that was not enough where multiple criteria were involved. This paper gives a general framework for multiple events. We show that it is possible to encode a time series as a string, where the final representation depends on the user defined criterion. Then finding string distances between two such encoded time series can effectively measure dependence. We show that this technique is more powerful than the ‘Pairs Trading strategy’ as varied user defined criterion can be handled while detecting similarity. We apply our technique with one practical user defined criterion. To the best of our knowledge, this is the first attempt to find similarity between stock trends based on user defined multiple event criteria.