ABSTRACT
We examine the relation between news arrival intensity, volatility and volume at an intraday frequency using a global dataset. The analysis is based on news analytics platforms that use natural language processing to perform entity recognition, classification by topic and sentiment analysis. We introduce our own news arrival intensity metric, which is simple and intuitive, and present compelling evidence that intraday volume and volatility forecasts can be improved using these metrics.
For stocks traded in the USA and Europe, we use Refinitiv's news analytics dataset based on the news written in English. We present strong and robust out-of-sample performance of our model in these markets. The results for the USA and Europe suggest that it is possible to extract stronger signals from news articles written in languages that are native to each market. For this reason, we extend our model to stocks traded in Japan using a news analytics dataset provided by FTRI/Alexandria. This dataset is based on articles in Japanese published by Nikkei. Our model successfully harnesses markedly strong predictive power of news in this application. In particular, our out-of-sample analysis for Japanese stocks shows that about 80% ∼ 90% of the stocks would have benefitted from the use of our news arrival intensity metrics.
Our results also suggest a spillover effect within sectors: Volatility in stock i tends to increase if other companies in the same sector experience an increase in news arrival intensity. We demonstrate that the output of the model is economically and statistically significant and remains robust over time even in the presence of outlying data points. Our model can be applied to optimal trade execution both at the stock and at the portfolio level.
