J.P.Morgan’s Quantitative & Derivatives Strategy team, headed up by Marko Kolanovic, has found a significant signal in news sentiment data. The signal, however, seems to decay rapidly after about two days.
Using sentiment data provided by Alexandria Technology, the team carried out the following analysis:
1) For each stock in the Russell 3000, a daily net sentiment score was calculated for the period 2000 to 2019. The score was based on the number of positive/negative news items from Alexandria Technology’s Dow Jones News only source within a 24 hour period (between 4:00 PM on date t-1 and 3:30 PM on date t).
2) They then tested the sentiment factor in isolation. Chart 1 shows how there is a significant decay in returns as the lag increases. (The “Sentiment 1 Lagged 1 Day” curve refers to using sentiment data from the previous 24 hours as described above). The strategy, however, involves daily rebalancing which requires high turnover and is unfeasible for most investors.
3) They then tested the sentiment factor with other traditional style factors (e.g. Momentum, Value, Growth, Quality, Low vol) to see whether it is possible to combine sentiment with a lower frequency signal to improve performance after transaction costs. Two methods for combining sentiment with another factor were tested: a) a weighted method that averages the signal from both factors and b) a method that combines the signals from both factors. Their analysis shows that combining signals rather than combining weights produces a lower turnover while preserving the returns.
The J.P.Morgan team conclude that sentiment data from Alexandria Technology has significant predictive power as a short-term signal. The signal, however, seems to decay rapidly after about two days. Moreover, because of the high turnover of sentiment signals, the team found that the rebalancing frequency, the signal smoothing lookback period and the method for combining the signal with other signals plays a significant role in strategy performance after transaction costs.
Please contact us if you would like to request access to the full paper.