Forecasting price intervals in the Brazilian stock market using discrete-time markov chains

Abstract

Discrete-time Markov Chains have been preferentially used to predict the trend of stock and index prices using three states (price rise, price fall, price stay steady) and steady state analysis. While the prediction of price intervals has been little explored. In this work, we implement three different ways of building the transition probability matrix for forecasting price intervals, this prediction is compared with actual data and the hits percentage is measured. In addition, the relationship between the hits percentage and the period of construction of the probability transition matrix is calculated, as well as the relationship between the percentage of hits and the number of price intervals or states. The analysis was performed using 10 random stocks from São Paulo stock exchange with data from 2010 to 2019. One of the methods evaluated, which consisted of fixed-length intervals, using a transition probability matrix of 12 months and 5 intervals, was the one that presented the best performance, obtaining a total average of hits percentage above 81%. Furthermore, four investment strategies were implemented taking into account the results of this method, showing that it is possible to increase investments with the results of the method.