One of the most frequent questions we receive regarding ChinaScope's NLP service - SmarTag centers on the decay rate of signals generated from news sentiment. This short article will look at the change of information coefficient of a sentiment-based factor across a tracking period of 20 days. We are examining Mainland Chinese news impact on China A-share stocks.
First, we construct a sentiment based factor based on the following parameters:
- Aggregate daily news from 3:00 pm of the previous trading close to 3:00 pm of current day’s trading close
- Taken the average sentiment score for this period as the Sentiment Factor: sentiment score = Average (sentiment category [-1,0,1] * sentiment weighting * 100 )
IC Decay of Positive Sentiment Score
Positive sentiment has a limited positive IC on one day lag, but immediately reverses starting from the second day. This is quite intuitive, since liquidity on the A-share market is quite robust and the internet is highly integrated in China's society, so positive news are generally digested very quickly. However, the curious part of this is that from day two, the IC doesn't just diminish, it actually reverses into a rather significant negative territory, and remains there rather steadily for a prolonged period of time. So, from an absolute IC perspective, there is quite a bit of longevity in Positive Sentiment.
IC Decay of Negative Sentiment Score
Negative sentiment's IC decays more slowly than positive sentiment, although it vacillates a bit throughout the tracking period of 20 days. The lingering predicative power of negative sentiment can be in part explained by the fact that the A-share market is heavily participated by retail investors and quasi-professionals who do not have access to shorting mechanisms, which means that the market cannot readily digest negative information as efficiently as positive information. This provides a unique alpha opportunity for institutional investors who have access to share borrow or have a broad base stock inventory which they can use to effect short positions.
Written by Tom Liu