Empirical evaluation of an automated intraday stock recommendation system incorporating both market data and textual news

In this study we evaluate the effectiveness of augmenting numerical market data with textual-news data, using data mining methods, for forecasting stock returns in intraday trading. Integrating these two sources of data not only enriches the information available for the forecasting model, but it can potentially capture joint patterns that may not otherwise be identified when each data source is employed separately. We start with market data and then gradually add various textual data representations, going from simple representations, such as word counts, to more advanced representations involving sentiment analysis. To find the incremental value of each data representation, we build an end-to-end recommendation process including data preprocessing, modeling, validation, trade recommendations and economic evaluation. Each component of the modeling process is optimized to remove human bias and to allow us to impartially compare the results of the various models. Additionally, we experiment with several forecasting algorithms to find the one that yields the best results according to a variety of performance criteria. We employ data representation procedures and modeling improvements beyond those used in previous related studies. The economic evaluation of the results is conducted using a simulation procedure that inherently accounts for transaction costs and eliminates biases that have potentially affected previous related data-mining studies. This research is one of the largest-scale data-mining studies for evaluating the effectiveness of integrating market data with textual news data for the purpose of stock investment recommendations. The results of our study are promising in that they show that augmenting market data with advanced textual data representation significantly improves stock purchase decisions. Best results are achieved when the approach is implemented with a nonlinear neural network forecasting algorithm. (C) 2013 Elsevier B.V. All rights reserved.