Stock price forecasting taking into account the popularity of search queries

Abstract

The article proposes to study the impact of behavioral factors on forecasting the stock price, and presents the model of the above-mentioned forecast. The factors that are supposed to be taken into account are used as indicators of the popularity of search queries within the certain topic. Two companies (Apple and Royal Dutch Shell) were selected for research, because they have different development histories. The Glossary based on papers on this subject has been created; it consists of 67 terms of economic, social and political meaning reflecting sentiment-oriented behaviors of traders, and the criteria for search of high popularity queries have been determined. We assume that the queries characterized by high popularity index make a great impact on the dynamics of the stock price. Besides, the article addresses the issue of neural networks as upon researching we observe that they might serve as indicators of the popularity of search queries. Furthermore, we introduced the forecast for two companies from different sectors of the economy. In addition, forecasting was made with various combinations of these search queries, which were combined according to the semantic load. During the experiments it was revealed that for the information-sector company the quality of the model increased significantly due to the inclusion of behavioral factors, while for the processing-sector company the data did not significantly improve the forecast. This follows from the specificity of the study. The results can indicate the adequacy of the constructed models and confirm the feasibility of using the popularity indicators of search queries for forecasting the stock price.