Cooperative optimization for efficient financial time series forecasting

The highly dynamic nonlinear and volatile nature of stock market has remained a challenging issue for the researchers in mathematical economics as well as in financial engineering. Despite the existence of a number of statistical and soft computing methodologies for stock market forecasting, there is still need for an efficient and accurate forecasting model for this purpose. Although a wide range of nature inspired evolutionary algorithms have been developed and applied successfully in the domain of stock market forecasting, their performance may vary significantly from one stock market to another. Therefore selection of an algorithm involves an inherit risk associated with it. In this paper, instead of employing a single algorithm and investing the total time budget in it, we construct a cooperative algorithm framework, which takes two algorithms as its constituent algorithms. Particularly two population-based algorithms such as genetic algorithm (GA) and chemical reaction optimization (CRO) have been chosen as the constituent algorithms. A multilayer perceptron (MLP) architecture have been used as the forecasting model. The cooperative algorithm framework executes each constituent algorithm with a part of the entire computation time budget and encourages interaction between them, so that they can benefit from one other. The cooperative algorithms have evaluated on five fast growing real stock market data set consisting daily closing prices. Empirical results have shown the superiority of the cooperative algorithms over individuals in terms of prediction accuracies.