Stock Exchange Prediction

The nature of stock markets has a tendency to change its appearance during the course of time. The related data sequence measured typically in successive times, say the daily closing price of a stock, is named a time series; a term from statistics and signal processing. For these irregular (neither periodic nor quasi periodic) time series, we are concerned in if there is any structure in data that can account for the behavior and evolution of them. By the word structure, we mean any correlation between the individual sampled values. We believe that there is a nonlinear correlation in the stochastic time series at hand and so we hope that they are predictable. Of course, there are many unknown parameters effecting the nonlinear correlation, which may be regarded as an additive white noise. This, along with the complex dynamics and the high frequency components of these time series, made the analysis and so the prediction task very difficult. Furthermore, the stock time series are changing on a wide variety of time scales. Therefore, we need a very powerful tool, which could manipulate the data in a multi-resolution manner, learn the local or global structures in the data adaptively, and use the generated knowledge to do one or multi step ahead forecasts truly. The poor forecasting results of other techniques, e.g. technical analysis, had disappointed the investors. Due to the lack of a dominant way to challenge the problem, a very radical theory was introduced known as efficient market hypothesis in the late 1960s. Strong-form efficiency implies that in a financial market, no one can earn excess return. Whether or not the market is always efficient, we believe that it is too complex to be analyzed by simple methods. We believe that such a facility only is available in the fields of signal processing, artificial intelligence, machine learning, and computer science.

Here, at NDL in SRRF, we have employed NARX recurrent neural networks and MARS models to generate one step and multi step ahead forecast for NYSE and Tehran Stock Exchange time series. We have Backtested our algorithms in NYSE with some stocks e. g. CATERPILLAR, DELL, and GENERAL MOTORS and we have got wonderful results. You can contatc us to learn more about the results.