Although news sentiment is widely acknowledged as a short-horizon alpha, our research suggests news sentiment is predictive of returns even at longer horizons in the China A share market, possibly due to strong participation from retail and HNW investors in Chinese markets.

The signal we backtest is a combination of both news sentiment and volume, derived by applying Natural Language Processing (NLP) on 30,000+ Chinese Mandarin news reports from local media every day. Although news sentiment and volume is available for over 80% of the China A share market going back to 2016, our backtest is restricted to CSI 300 constituents only, as shorting individual stocks beyond the CSI 300 is unrealistic for many investors.  

First, the signal is predictive of both returns and alpha; average annualized net return and alpha increases with signal strength and is monotonic across quintiles. Secondly, we find that the signal’s information coefficient is positively correlated to holding periods of up to 90 days. Therefore in backtesting, we choose to rebalance bi-monthly (60 days) which provides the best tradeoff between signal strength and turnover (and transaction costs). A 60-day rebalanced dollar-neutral portfolio (long/short top/bottom quintile) generates an annualized net return of 18.9% with a Sharpe ratio of 1.46 and Beta of +0.11 from 2016 to 2023.  After removing factor tilts from the signal, it still produces a long/short net return of 8.9% with a Sharpe ratio of 1.06 and Beta of +0.04.