Modelling the Relationship Between the RTS Index And Activity of Investors on the Internet

<p class="p1">This paper investigates the possibility of predicting the price movement change, based on the information factors that characterize the activity of investors on the Internet. Under the activity on the internet here means activity, which can used to describe their mood and interests.</p><p class="p1">In this study, the frequency of queries of company names, which are included the RTS index, in the Google search engine will be used,&nbsp; as an indicator of investors&#39; attention to these companies and sentiment index, derived using data from Twitter, as an indicator of investor&#39; sentiment.</p><p class="p1">The relationship between these factors was investigated for two different methods of data representation: for absolute and increment values. In the analysis of absolute values, we tried to find cointegrating connections in the source data, but it has not been successful. Cointegration was found for only two of the twenty-two companies, but even for these companies we could not build a good model.</p><p class="p1">In the case of the increments, all time series became stationary, and was applied Granger causality test. As a result of this test is failed to obtain the desired result. Despite this, we constructed linear autoregressive and distributed lag model, which quite accurately describe the dynamics of these series: matching percentage in price movement direction with the model never fell below 61% for the training set and was 70% in the average for all companies in all training periods. With regard to this indicator for the test period, there were also achieved acceptable results. Average success of prediction of the price movement direction for all companies in the last year was 63 %, which is enough to make a profit.</p><p class="p1">According to the results of forecasts, constructed models can to predict the direction of the RTS index in 57% of cases, using the average, and 60 % when using a weighting.</p><p class="p1">Furthermore, we found that the quality of the autoregressive model is significantly reduced in the absence of the information indicators. This suggests the importance of these factors to predict stock prices and their high predictive power.</p>