Evolutionary Based ARIMA Models for Stock Price Forecasting

Abstract

Time series prediction is mostly based on computing future values by the time set past behavior. If the prediction like this is met with a reality, we can say that the time set has a memory, otherwise the new values of time set are not affected by its past values. In the second case we can say, there is no memory in the time set and it is pure randomness. In a faith of ”market memory”, the stock prices are often studied, analyzed and forecasted by a statistic, an econometric, a computer science... In this article the econometric ARIMA model is taken for previously mentioned purpose and its constructing and estimation is modified by evolution algorithms. The algorithms are genetic algorithm (GA) and particle swarm optimization PSO.