Stock Market Prediction Using Aritificial Neural Networks

Only available on StudyMode
  • Download(s) : 71
  • Published : May 13, 2012
Open Document
Text Preview
Stock Market Prediction Using Artificial Neural Networks
Tariq Waheed in supervision of Dr. Xiang Cheng Department of Electrical and Computer Engineering, National University of Singapore Engineering Drive 3 Singapore 117576, Email: tariq@nus.edu.sg Abstract— Stock market is a very dynamic field whose prediction still remains a very challenging task for scholars and veteran traders alike. The study presented in this paper is an attempt to predict the daily and weekly rates of returns of the stock market and compare the results to the return generated by the naïve buy-and-hold strategy. The first part of the study explores the various auto-regressive and neural network models in predicting the daily and weekly S&P 500 index returns solely using the historical index data. It is observed that neural networks perform better than auto-regressive models in predicting the stock market and hence, S&P 500 Index is not a perfect linear time series. The study shows that three-model approach can help to better perform in the task of stock market prediction. The second part of this study uses technical indicators and three-model approach to see if adding more information and trending the data with technical indicators can help to achieve higher returns in stock market as compared to the naïve buy-and-hold strategy. It is observed that Moving Averages are better at giving buy and sell signals in market than Relative Strength Index. A significant observation made from both parts of the study is that weekly returns can be predicted in more confidence than daily returns. The ex-ante testing of the models is done and evaluated after considering the commission costs and using the short-selling strategy which is found to be a profitable strategy. There is an observation that adding more information to the neural network regarding the historical price data can lead to better prediction and hence, higher profits. Index Terms—Hierarchy systems, artificial neural networks, stock market prediction, trend classification, moving averages, relative strength index.

movement that it has become practically impossible to find out the future of stock prices. However, if certain factors like economic and political environment remain constant, then the more veteran trader are able to predict the direction of stock market with a considerable amount of confidence. However, since everyone is a not a veteran trader, people have tried to use computer systems to forecast the price of stock based on current situation. Currently, the work that has been done to predict the stock market has had limited success as most of the well performing systems did well only when commission costs were ignored. However, commission costs are very practical and hence cannot be ignored. II. CHALLENGING THE EFFICIENT MARKET HYPOTHESIS The Efficient Market Hypothesis (EMH) states that it is not always possible to outperform the market, adjusted for risk, by using any kind of information that is already known by the market. Any new information which arises will be quickly and efficiently absorbed into the price of the stock. The scope of information with regards to EMH encompasses past prices of a stock, fundamental analysis, and public or private information given out by company. The EMH exists in three forms [1] namely: (i) The “Weak” form according to which the past stock prices cannot be used in predicting future stock prices, (ii) The “Semi-strong” form according to which any publicly announced information cannot be used to outperform the market, and (iii) The “Strong” form according to which nothing can be used to earn excess returns over the market. (A) Some Evidences in Support of EMH: A study was conducted by Fama [3] in which he used serial correlations on the changes in the natural log of the price of thirty stocks in the Dow Jones Industrial Average. Fama concluded from his study that there is no strong evidence of linear dependence between lagged price changes or...
tracking img