Building a Stock Sentiment Signal from Investor Forum Text

Summary

This study constructs a Chinese equity sentiment indicator from posts and comments on an investor forum. After cleaning and tokenizing roughly a year of text, it uses word frequency and visual inspection to assemble a sentiment lexicon. Stronger positive or negative expressions receive larger weights; daily positive and negative scores are combined into a net sentiment measure. The report examines how this series relates to the CSI 300 and describes a rule that smooths the signal with an N-day weighted moving average. Because publication lags by one day, the rule buys at the next open when sentiment is at or above its average, and sells at the next close when it falls below.

Key ideas

• The indicator scores positive and negative forum language with a weighted sentiment dictionary derived from cleaned and tokenized posts.
• The report finds that sentiment and index returns moved together in its sample, with a stated linear correlation of 0.51.
• Weekend and holiday sentiment often had the same direction as next-session sentiment and index returns in the reported observations.
• The trading rule smooths sentiment with a weighted moving average and accounts for a one-day reporting lag.
• The backtest covers March 2020 to March 2021 and reports returns, drawdown, win rate, and payoff ratio, but the historical sample does not establish future performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.