There are many studies predicting a stock market using social media. Suwa et al. (2017) proposed a VI index prediction model. They assumed that changes sentiments of investors are topics change posted on social media. However, in their prediction model, the data to use verification is included in the data to use to develop their topic model. Hence, their model might be overfitting. Therefore, we propose a prediction model of VI index avoiding overfitting. We developed a program that applies new posting messages to topic models of a learning period. We created data for verification using this program. As a result, we found that a logistic regression using time series topics on the past seven trading days may predict a rise in the VI index.
