Forecasting non-stationary financial time series through genetic algorithm

TL;DR

This paper presents a novel approach combining genetic algorithms and discrete wavelets to accurately forecast trends in non-stationary financial time series, including major indices like NASDAQ and BSE.

Contribution

It introduces a new method integrating wavelet analysis with genetic algorithms for trend prediction in complex financial data.

Findings

Accurate trend forecasts for NASDAQ and BSE indices.

Identification of self-affine behavior in financial time series.

The derived analytic equation effectively models index trends.

Abstract

We utilize a recently developed genetic algorithm, in conjunction with discrete wavelets, for carrying out successful forecasts of the trend in financial time series, that includes the NASDAQ composite index. Discrete wavelets isolate the local, small scale variations in these non-stationary time series, after which the genetic algorithm's predictions are found to be quite accurate. The power law behavior in Fourier domain reveals an underlying self-affine dynamical behavior, well captured by the algorithm, in the form of an analytic equation. Remarkably, the same equation captures the trend of the Bombay stock exchange composite index quite well.