Abnormal effects of weekdays on forecasting Stock prices by Neural Networks

Abstract

Artificial Neural Network (ANN) is different from classical models because of its learning and generalizing capabilities. ANN can asses lacking or mistaken data and it can produce a solution for complex problems. Recently, ANN has been utilized especially for the resolution of financial problems that emerged in fluctuating and uncertain environments which has been difficult to categorize. ANN is generally used to estimate financial failures, crisis periods, exchange rates and indexes in Turkey by comparing the results with those pertaining to classical models. There is no study carried out about estimating share prices for every day of the week and quantifying the effects of days on share price indices. The aim of this study is to predict share price indices for every day of the week and determine possible performance differences (measured by forecast errors) among the days of the week. For this purpose, BANVT, SKRPLC, TATKS and TUKAS which are traded in Istanbul Stock Exchange are used to estimate friday's closing prices by ANN with %0,73 forecast error. Then, share price indexes for monday, tuesday, wednesday and thursday are estimated and the prediction of tuesday's closing price is found to be more difficult to estimate compared to the other days of the week. © EuroJournals Publishing, Inc. 2012.