Short-Term Prediction Of The Closing Price Of Financial Series Using A Epsilon-Machine Model

This research addresses the problem of the short-term prediction of the closing prices of financial series. As a method of solution, an epsilon-machine model is proposed, which is constructed with a probabilistic finite state machine with an input function and an output function. The definition of the alphabet of the epsilon-machine is generated by transforming the values of the returns of the financial series into integers, using the central limit theorem. The transition between the source and destination states is determined by three rules, based on the calculation of Shannon's entropy. To validate the model, we used values from the American financial series S&P500, Nasdaq, Nyse-Amex, Nyse-Composite, IPC-MMX, and the Asian companies Hang Seng and Nikkei n225. The results obtained were contrasted with the predictions using the Naive strategy financial model. The results are very encouraging, since in all cases the results of our model were closer to the real values than those obtained with Naive strategy. It is noteworthy that in four series the change of sign was predicted with an approximation of close to 60%. (C) 2019 Elsevier B.V. All rights reserved.