In this paper, J.P. Morgan looks to neutralize unintended factor exposures in short-term reversals. They find that stocks with lower news volume or with exhibit stronger reversals, and anticipation of earnings news leads to stronger subsequent reversals.
Specifically, they create four tradable baskets that systematically long liquid stocks within TOPIX 500 with weekly rebalancing.
- Purified Reversal: Buy past week losers based on sector-neutral residual returns
- News Enhanced Trading Reversals: Purified Reversal without stocks with medium to high news volume
- News and Sentiment Enhanced Trading Reversals: News Enhanced Trading Reversals without stocks with poor earnings sentiment
- Tactically Weighted Enhanced Trading Reversals: News and Sentiment Enhanced Reversals, and tactically tilt towards stocks that will report earnings in the next week
Using RavenPack’s sentiment scoring and (earnings) event detection capabilities significantly improves a reversal trading strategy (across the baskets 2, 3, and 4), both in terms of Information Ratio and Returns. See the charts below for details:
Trading Reversals: Methodology
For more details, our Chief Data Scientist gives an overview of the 4 weekly strategies: In this White Paper the quants at JP Morgan [JPM] are back with a new stab at using RavenPack data to boost the performance of traditional strategies.
Coming off the back of two inspiring reports, (1) Big Data and AI Strategies and (2) Value Strategies based on Machine Learning, the new JPM report centers on the Japanese stock market and demonstrate how news-driven reversal strategies exhibit improved predictability when exposure to traditional risk factors have been eliminated. In particular, since 2006, they were able to achieve an Information Ratio of 0.91, which is an improvement of 0.29 over a standard reversal strategy (without a news overlay).
Below, we present the main results of the JP Morgan report in an easy-to-follow step-by-step guide to their strategies. In total, four weekly strategies of increasing complexity are introduced. The four strategies take long equal-weighted positions in select constituents of the Japanese equity index, the TOPIX 500, while the index itself is shorted in equal amount. To make the strategies more scalable, a set of liquidity filters are applied, resulting in a final investment universe of between 150-400 stocks. Performance is calculated imposing transaction costs of 4 basis points and weekly rebalancing.
Strategy 1: Purified Reversal
Firstly, to come up with a strategy benchmark, JPM creates a simple reversal signal that buys past losers on a weekly basis, evaluated using Fama-French factor-neutral returns. As an additional twist, they also remove any sector-specific exposures. The hypothesis being that stocks with the largest negative residual return will see the largest reversal during the following week and hence outperform. The figure below summarizes the results of the purified reversal strategy and highlights the importance of factor-neutralizing the returns. In particular, the Information Ratio (IR) increases from 0.52 to 0.62 when sector-neutralization is included - primarily driven by lower volatility.
Strategy 2: News Enhanced Reversal
An obvious issue with the strategy above is that it cannot distinguish whether a particular price move was driven by news or simply “noise”. If a large drop in the weekly price of a company is due to an actual deterioration in company fundamentals — as revealed by news — there is no reason to expect a rebound in the stock price the following week.
JP Morgan combats this by using the RavenPack Analytics (RPA) suite to exclude high-news volume companies on the grounds that if a company has a lot of news coverage the stock movement is more likely to be justified by new information about the company’s fundamentals. Strategy 2 thus builds on the results by buying only those “weekly losers” which have news volume in the bottom three deciles (30%). This condition results in an IR of 0.80 on an annualized return of 5.4% - both marked improvements on the simple reversal strategy.
Strategy 3: News and Sentiment Enhanced Reversal
News volume is just one of the ways in which RavenPack Analytics can be used as an overlay in a reversal strategy. In particular, news sentiment can also be expected to play a role. For instance, a stock which has rallied strongly during the past week, without news sentiment to support this move, is more prone to experience a downward correction the following week.
An additional filter is added using RPA’s event sentiment score to calculate weekly sentiment averages for earnings-related news. This is done for each company in the TOPIX 500, and allows only those companies with earnings sentiment in the top 40% of the portfolio. This step further boosts the annualized return to 5.6% from 5.4% implying a modest increase in IR to 0.81 from 0.80.
Strategy 4: Tactically Weighted Enhanced Reversal
Now, having demonstrated that a sizable increase in risk-adjusted and absolute returns is possible, when overlaying a simple weekly reversal strategy with news volume and earnings sentiment, the quants at JP Morgan have one last trick up their sleeves, i.e. tactically weighting the portfolio of stocks in strategy 3 based on future earnings announcements.
Since earnings announcements generally see stronger reversals, the JP Morgan quants create a strategy which gives higher weights to stocks which are close to the earnings announcement date. They achieve this by doubling the weight of a stock if it is set to announce earnings results in the coming week. This yields an IR of 0.91 compared with 0.81 above. The annualized return is 6.3%, corresponding to an increase of 0.7 percentage points on top of the News and Sentiment Enhanced Reversal strategy, which is a marked improvement on the most simple strategy with an annualized return of 3.6% since 2006.
Utilizing the state-of-the-art news detection capabilities in RavenPack Analytics, the quantitative team at JP Morgan has shown how to improve a simple weekly reversal strategy by buying only those weekly losers with news sentiment in the top 40%, news volume in the bottom 30%, and an upcoming earnings announcement. This ensures a 47% improvement in Information Ratio and a 75% increase in annualized return over a simple reversal strategy (without the RavenPack news layer).
Rationable for Short-Term Reversals
Since the seminal work of Jegadeesh (1990) on the reversal patterns in monthly stock returns1, “contrarian” strategies that buy losers and sell winners in the recent past have become popular, especially among quantitative hedge funds that attempt to profit from temporary price dislocations.
There are three major competing explanations for the phenomenon of short-term reversals:
- Sentiment based (i.e. behavioral): - Schiller (1984) suggested that dividends and stock prices may be driven by social optimism or pessimism.
- Liquidity based (i.e. market microstructure):
- Cross-correlations based (i.e. non-synchronous trading):
- “Purifying” reversal returns by removing its systematic exposures to conventional risk factors help to enhance the returns whilst significantly reduce risk
- Stocks without news coverage show a stronger reversal pattern than stocks with news coverage, especially for recent losers. This suggests that investors overreact to spurious price movements
- Earnings-related sentiment is a more promising signal than...
- Subrahmanyam (2015) found that the reversal of monthly returns is related to market overreaction, rather than inventory effects or order flow innovations.
- De Bondt and Thaler (1985) applied statistical tests to quantify market overreaction, and found that the effect of overreaction is larger for losers than for winners, leading to reversal effects.
In this school of thought, short-term reversals are driven by price pressure generated from the demand for liquidity, which will be reverted when liquidity providers react to profit opportunities arise from price deviations from fundamentals.
Jegadeesh and Titman (1995) provided evidence that return reversals are explained by dealers' inventory imbalances and the microstructure dynamics of bid-ask spreads. Avramov et al (2006) showed that illiquid stocks exhibit stronger reversals than liquid stocks, as proxied by the Amihud measure.
Kaul and Nimalendran (1990) looked at NASDAQ stock returns and claimed that short-term price reversals is due to bid-ask spread, rather than market overreaction.
Lo and MacKinlay (1990) pointed out that whilst negative serial correlations of stock returns leads to a profitable reversal strategy, it is not the only condition. In fact, even if prices are not negatively auto-correlated, a reversal strategy can still have positive returns if cross-auto-covariances are positive, i.e. a higher return of stock A today implies a higher return for stock B tomorrow, i.e. stock A leads stock B.
In this note, we look at weekly reversals in Japan. The Japanese market tends to exhibit a more mean-reverting behavior, rather than being momentum-driven. Whilst simple reversal strategies have worked decently in Japan, we attempt to understand drivers of such reversal patterns. In particular, we study how news volume, news sentiment and earnings announcements could affect reversal patterns. We obtain news sentiment data from a vendor called RavenPack, which we introduced the details in “Value Strategies based on Machine Learning”
What have we learnt about Reversals?
Our major findings on reversal patterns are as follows:
Click here to continue reading the "Enhancing Reversal Trading with News and Neutralization" White Paper.Request White Paper