Probabilistic Analysis of Market Impact of Analysts’ Recommendation Revisions

In this paper, we propose to model the short-term impact of recommendation revisions on stock price movements in probabilistic framework. Through the Bayesian models, we can consolidate the information of all analysts' recommendations on stocks and predict the post-recommendation price drifts of the underlying stocks. In addition, typically there are only a small number of recommendation revisions on specific stocks on each day. It implies that other analysts' views to a specific stock are unobservable. With the advantage of generative models, missing observations in the data of recommendation revisions can be accommodated easily. Secondly, we perform an empirical investigation of the Hong Kong equity market and find that the impact of recommendation revisions on stock prices is significant in the 1-day horizon, but insignificant in the longer time horizon when taking account of the transaction costs. Also, we propose trading strategies derived from the posterior probability of directions of price drifts based on the Bayesian models. In our experiment, during the out-of-sample period (1 Jan, 2019 to 21 Feb, 2020), the trading strategies gain double-digit profit returns after transaction costs.