Title

 

Modeling and Forecasting of Financial Time Series Data Using Statistical Methods, Neural Networks and Genetic Algorithms

Abstract

 

Financial time series like stock prices and exchange rate are non linear and non trivial in nature. The series is stochastic which makes it difficulty for modeling and prediction.
Traditionally statistical methods like statistical clustering and regression analysis have been used for modeling the series. However most of models are more suitable for linear processes and their application to nonlinear series have generally shown less satisfactory results. Since the later half of the last decade, developments in the field of artificial intelligence and soft computing have made possible the use of neural networks for financial forecasting. Neural Networks, inspired from human neural system have the ability to approximate non linear functions. A further development in the field of AI has been the evolutionary regression or genetic programming method. Designing neural networks and genetic programming for robust financial prediction is a subject of on going research. This paper employs neural networks, genetic programming and regression based methods for modeling exchange rate series. Experimentation has been attempted with the input output set and the design of neural networks to achieve accurate modeling. This paper discusses the experimentation methods and the modeling techniques, which have been used, and compares the results that have been obtained from them. The results show that radial basis networks are the most suitable for forecasting the series and this model leads to very accurate prediction.

 

This document is the report of BSc. Senior Project conducted by Group 316 during Winter and Spring Quarter 2002-03 at Lahore University of Management Sciences. The report is being submitted to the Department of Computer Sciences for review and approval.

The project aimed at testing the hypothesis that is financial time series predictable. Financial time series data exhibits a non linear and non trivial nature. The stochastic nature of the time series makes it difficult to model. In this project exchange rate series was used for the study.

According to the Efficient Market Hypothesis it is impossible to predict the future value of the series due to the possibility of arbitrage. For example if everyone knows that the price of dollar will fall in the time period t+1 then everyone will try to buy dollar at time t+1 putting pressure for the price to rise back again. Hence the prediction which is realized by everyone in the market will not turn true. EMH assumes that every investor has perfect and equal information and hence no one can predict the rate profitably. However the financial analysts and those who make millions in the market believe that it is not true. If EMH is true then investing in speculative trade in the financial currency market is no better than gambling where the odds are equally likely. If the hypothesis is not true then, as many financial analyst claim, the market can be predicted based on better knowledge and superior prediction. The traditional methods of modeling time series were based on Box Jekins and ARMA models. However these models suffer from the limitation to capture non linearity (ACM 1994). Neural Networks have been found to show superior function approximation and modeling. Various models and structures of neural networks like feed forward and recurrent have been used to model time series data.

In this project various methods and techniques for modeling time series were applied to the problem of financial modeling and their results were compared to reach the most optimal prediction. The project used econometric and stochastic methods like ARMA, evolutionary regression like genetic programming and AI based models like neural networks.

Earlier works on the subject have either only focused on AI based methods or used statistical techniques; very few studies have used such a broad range of methods. Also no published work exists for modeling the data set under study, the exchange rate of Pakistani Rupee and US Dollar has never been used for modeling by so many different techniques. This project extends
the breadth of existing stock of literature and applies it to the Pakistani currency market.

This report is divided into four sections. The first section studies the data. It contains the results of data analysis and statistical tests which have been applied. This is followed by a review of the various techniques of modeling time series which have been used in this project. The third section describes the methodology and implementation specifications and the fourth section contains the results.

The project showed that neural network gave superior prediction of exchange rate. The prediction was good enough to generate a profitable investment by using a simple buy and sell strategy. Among the neural network models Radial basis networks proved to be most optimal. This project tried to keep the focus of the work academic and research based. However as the results show, the neural networks and other system developed for prediction during this project can be used by any financial analyst and investment firm for making forecast for the exchange rate series. If the significance of a project is to be measured by its productivity, then this work would definitely do wonders simply because it can sell.

Faculty Advisors

Dr. M.M. Awais and  Dr. Asim Kareem