Technical analysis indicators are tools that use current and past stock prices to come up with buy/sell recommendations. Baffa and Ciarlini show that the efficiency of these tools can be drastically improved if they are combined with partially observable Markov decision process (POMDP) techniques.
Specifically, the authors assume that the state of the market at any moment in time can be described by its past trends and its future trend. The future trend is not directly observable; we assume that a technical analysis indicator provides partial information about it. The Markov character means that we assume that the probability of each state at the next moment in time depends only on the current state--it does not directly depend on past states. Baffa and Ciarlini then apply the known POMDP techniques to the training data, in order to determine the corresponding transition probabilities. The resulting model can predict the probability of different future states, and therefore provide buy/sell recommendations based on these predictions.
To empirically test these recommendations, the authors compare the performance of this methodology when applied to the 2008-2009 Brazilian stock prices, with the prices from 2000 to 2007 as the training data. Both periods include boom and crisis subperiods. For most indicators, combining the indicator with the POMDP techniques leads to much better performance than the indicator by itself. During the crisis period, the new recommendations would even enable users to gain a significant profit, while all known recommendations would lead to losses.