Constructıng Tradıng Strategıes Usıng Artıfıcıal Intellıgence Based Models: An Applıcatıon For Borsa Istanbul

Abstract

The aim of this thesis is to test the Efficient Market Hypothesis using artificial intelligence-based techniques. In this regard, we utilize artificial intelligence based models that have both deep and shallow architectures which are Long Short Term Memory (LSTM) Networks and Support Vector Regression (SVR) to predict the next day's close price of the selected stocks from BIST30 Index. Next, we construct trading strategies by making use of the predictions produced by the forecasting models. We feed these models using a comprehensive dataset including technical analysis indicators and investor sentiment variables. Thus, we predict the following day's close prices both by using historical price data which is accessible without any costs and the investor sentiment containing market's non-rational components. In order to proxy investor sentiment, we use Bloomberg's news sentiment data which is developed to imitate a human in processing financial news. We show the superior performance of our trading strategies that are constructed using both LSTM and SVR models compared to simply buy and hold market index in terms of all performance metrics. Moreover, we reach similar results when transaction costs are considered. Our findings reveal that successful predictions can be made and trading strategies can be built using publicly available information and artificial intelligence-based models. Moreover, investing with these strategies, above-average risk-adjusted return can be yielded. Thus, we provide contradictory evidence to EMH's negatory arguments about the asset price predictability.