top of page
  • Leapday Quantitative Research Team

How does REACTION stack up against an earnings drift strategy?


Summary


This report compares REACTION signals to a PEAD strategy. While REACTION is inspired by observations related to PEAD, it is inherently different in its construction and objective and this report sets out to study these differences. Between 2016-2021H1 and inclusive of costs, a portfolio using a PEAD strategy based on the earnings surprise returns -3.8% per year with a Sharpe ratio of -0.40 and max drawdown of -35.8%. A portfolio using a PEAD strategy based on the initial price momentum returns 4.3% per year with a Sharpe ratio of 0.55 and max drawdown of -11.3%. Comparably, a portfolio using REACTION signals returns 13.2% per year with a Sharpe ratio of 1.78 and max drawdown of -5.2%. A portfolio constructed using REACTION signals has a low correlation to a portfolio constructed from either earnings surprise or momentum signals. These results indicate that REACTION signals strongly outperform PEAD signals, are uncorrelated to PEAD signals, and exemplify the distinction between REACTION and the PEAD signals.



Read as PDF

Leapday - How does REACTION stack up against an earnings drift strategy_
.pdf
Download PDF • 226KB






Introduction


In 1968 Ball & Brown published a seminal study, “An Empirical Evaluation of Accounting Income Numbers”, in which they introduced an empirical observation of how stock prices change after earnings announcements, an effect known as post-earnings announcement drift (PEAD). This market anomaly describes the tendency for stock prices to rise/drop for a period of time after earnings announcements for companies that release good/bad results. Since this initial study, thousands of articles have been published on this topic. It continues to be an area of active research in the finance community.


The Leapday REACTION model was inspired by observations related to the PEAD. Our model has some similarities to the PEAD in terms of its focus on activity shortly after earnings announcements. However, the REACTION model is inherently different in its construction and objective. In this report we focus on answering two questions with respect to how REACTION relates to the PEAD: (1) how does REACTION performance compare to a PEAD strategy; (2) how frequently does REACTION trade with the direction of price momentum? With this motivation, in this report we set out to compare PEAD vs. REACTION performance and examine to what extent REACTION signals differ from a strategy based on the initial price momentum.


We will proceed by creating a PEAD signal, stating the portfolio assumptions used in testing the PEAD signal vs. the REACTION signal, presenting the portfolio performances, investigating the degree to which REACTION signals follow the initial price momentum, and discussing the findings.



Signals


When a company reports a positive/negative earnings surprise it will usually have an immediate positive/negative price move. However, in some cases the opposite occurs and despite a positive/negative earnings surprise a company’s stock will have an immediate negative/positive price move. To account for both of these cases, we create three types of PEAD signals.

  1. Earnings Surprise PEAD - Measurement of the percent error of the reported earnings per share compared to the market’s consensus estimate. We rank the scores -100 to +100.

  2. Market Adjusted Jump Momentum PEAD - Measurement of the market beta adjusted return of the stock from the close before the earnings announcement to the price 15 minutes before the first close after the earnings announcement. We rank the returns -100 to +100.

  3. Market and Volatility Adjusted Jump Momentum PEAD - Measurement of the market beta adjusted return of the stock from the close before the earnings announcement to the price 15 minutes before the first close after the earnings announcement scaled by the stocks trailing 60 day volatility. We rank the returns -100 to +100.


Leapday’s REACTION signals are available for 3 entry points after an event day, defined as the first day the market is open subsequent to a company’s earnings announcement. These 3 entry points are: (1) 60 minutes after the open on the event day (t0 open + 60min); (2) just before the market close on the event day (t0 close); (3) before the market open on the next trading day after the event day (t1 open). For this report, we will focus on the t0 close signal since the jump momentum PEAD would not be tradeable until the t0 close.


Our previous portfolio construction reports using REACTION signals used a 3d hold for the signals and we will use a 3d hold for the PEAD signals to make an equal comparison. Like the PEAD signals, the REACTION signals range from -100 to +100 indicating the direction and magnitude of the signal. Also like our previous portfolio construction reports, we use approximately the top one-third of signals ordered by signal magnitude for both the REACTION and PEAD signals. This approach provides a balance between using the strongest signals and having enough trading activity to utilize available capital. Table 1 below shows the number of signals these settings correspond to between 2016-2021H1 for REACTION, Earnings Surprise PEAD, Market Adjusted Jump Momentum PEAD, and Market and Volatility Adjusted Jump Momentum PEAD.




Portfolio Construction


As outlined in the previous section, we will use the strongest one-third of Earnings Surprise PEAD signals, Market Adjusted Jump Momentum PEAD signals, Market and Volatility Adjusted Jump Momentum PEAD signals, and t0 close REACTION signals to build simulated portfolios. We will hold trades for 3 days for each signal to keep the study comparable. Below are all the assumptions used in our portfolio construction.


  • Enter a trade at the market close on the first day the market is open subsequent to a company’s earnings announcement (t0 close) and exit the trade on the close 3 days later (t3 close).

  • Create a market-neutral portfolio by hedging trades using an opposite side trade in SPY with a beta-adjusted notional value that offsets the beta risk of the event trade.

  • Starting capital of $50M and assuming constant capital throughout the portfolio (i.e., we do not reinvest profits).

  • Portfolio leverage used during periods of high activity.

  • Trade no more than 6 percent of the starting capital per position.

  • Trade no more than 2 percent of a company’s historical average trade value per position. (The historical average trade value is calculated using the trailing 20 day average traded value.)

  • Beyond the two constraints on trade size stated above, trade equal weight.

  • Round-trip trading costs are based on breakpoints for liquidity group as follows:

    • Low (<$3M historical average trade value): 20 bps

    • Mid ($3M-$25M historical average trade value): 10 bps

    • High (>$25M historical average trade value): 5 bps

    • Costs are allocated half on entry and half on exit

    • In the metrics presented below, we assume $0 in hedge costs (for the SPY hedge) and the hedge does not use any equity or margin



Results


Below we present metrics from a portfolio that trades the top one-third of the Earnings Surprise PEAD signals, Market Adjusted Jump Momentum PEAD signals, Market and Volatility Adjusted Jump Momentum PEAD signals, and t0 close REACTION signals to build simulated portfolios and holds those trades for 3 days using the portfolio assumptions stated above. Metrics below are inclusive of costs.




The metrics associated with the chart above are as follows.



Below in Table 4 are additional metrics for the portfolio between 2016-2021H1 including maximum drawdown and correlation to SPY.



As shown above, the REACTION portfolio strongly outperforms all three PEAD signals with regards to absolute returns, Sharpe ratio, and max drawdown. It should be noted that the steep drop for the Earnings Surprise PEAD signals in 2020 reflects the backwards looking nature of an earnings surprise signal, which was counterproductive during the peak of uncertainty in the COVID-19 pandemic.


Finally, we look at the correlation of returns between the REACTION signals and the PEAD signals in Table 5 below.



As shown above, returns from a portfolio constructed using REACTION signals are uncorrelated to returns from portfolios built using PEAD signals. This result exemplifies the distinction between REACTION and the PEAD signals.



Momentum vs Reversion


The REACTION signals we use in this report are available just before the market close on the event day. Therefore we can observe the initial market reaction to an event and include that information in our signal. To what extent do the REACTION signals relate to the initial market reaction? How often does the REACTION signal predict the initial market reaction will continue versus revert over the coming days? In this section we explore these questions.


To analyze the dependence on the initial market reaction for the REACTION signals, we will examine the relationship between the signals that are traded in a REACTION portfolio (i.e., the top one-third of REACTION signals) and those traded in the Jump Momentum portfolio (i.e, the top one-third of momentum moves between the t-1 close and the t0 close). We construct Jump Momentum portfolios using both market beta adjusted moves and the market beta adjusted moves scaled by the stock’s volatility.


Table 6 below shows summary statistics relating the REACTION signals to the initial market reactions. We examine the overlap in signals in a REACTION portfolio with signals in the Jump Momentum portfolios.



The table above indicates that the REACTION model is not relying heavily on momentum to create forecasts. Only approximately 30% of momentum moves that would be traded in a Jump Momentum portfolio are traded by the REACTION model. In other words, 70% of the REACTION signals would be ignored by a PEAD strategy that was based on momentum. Extending the analysis above, we observe that about 70% of REACTION signals in the portfolio have the same direction as the momentum move leading up to the signal (i.e., buy signals are preceded by an up move in price, and sell signals are preceded by a down move in price). This result indicates that 30% of REACTION signals are contrary to the prevailing initial market reaction direction.


Next, we examine the correlation between the signals in a REACTION portfolio with the corresponding signals from the Jump Momentum portfolios. Table 7 below shows this correlation for all stocks and also shows the correlation when we limit trading to the most liquid 500 and 1,000 stocks.



With a correlation of 0.46 between the signals traded in a REACTION portfolio and their corresponding rank for the preceding momentum, the REACTION signals are fairly uncorrelated to a simple momentum signal. When we limit trading to the most liquid 1,000 stocks the correlation drops to 0.30 and when we limit trading to the most liquid 500 stocks the correlation drops further to 0.26. This result indicates REACTION signals are especially uncorrelated to momentum moves for higher liquidity stocks. In prior research we’ve shown that higher liquidity stocks drive returns when using large portfolio sizes. Therefore, the fact that REACTION has relatively low correlation to a momentum signal for higher liquidity stocks indicates that REACTION adds value for predicting stocks that tend to be the most price efficient.



Discussion


While the REACTION signals overlap with PEAD strategies with respect to trade timing, the trades generated from each strategy differ significantly. While the Earnings Surprise PEAD strategy was historically an effective strategy, over the past six years it would have lost money. We believe the degradation of this simple strategy is a result of improvements in market efficiency and dissemination of information to all groups of investors. During this same period a PEAD strategy based on the initial market jump would have been slightly profitable albeit with some potentially large drawdowns. Leapday’s REACTION signals provide a much more accurate and consistent signal for predicting the PEAD due to their unique construction which incorporates multiple dimensions of information that relate to earnings events.



Conclusion


This report answers two questions: (1) how do REACTION signals compare to a PEAD strategy; (2) does REACTION always trade with the direction of price momentum? The research indicates that a portfolio using REACTION signals strongly outperforms portfolios using various PEAD strategies. Between 2016-2021H1 and inclusive of costs, a portfolio using an Earnings Surprise PEAD returns -3.8% per year with a Sharpe ratio of -0.40 and max drawdown of -35.8%. A portfolio using a Jump Momentum PEAD returns 4.3% with a Sharpe ratio of 0.55 and max drawdown of -11.3%. Finally, a portfolio using REACTION signals returns 13.2% with a Sharpe ratio of 1.78 and max drawdown of -5.2%.


The correlation between returns from a portfolio constructed using REACTION signals and Earnings Surprise PEAD signals is 0.03 and the correlation of portfolio returns using REACTION signals versus Jump Momentum PEAD signals is 0.16 indicating REACTION is uncorrelated to PEAD. Further analysis indicates that only 30% of signals that would be traded by a Jump Momentum strategy would be traded by the REACTION model, indicating that REACTION does not always trade with momentum and exemplifying the distinction between REACTION and PEAD signals.



 


Disclaimer


This material is solely for informational purposes and is not an offer or solicitation for the purchase or sale of any security, nor is it to be construed as legal or tax advice. References to securities and strategies are for illustrative purposes only and do not constitute buy or sell recommendations. The information in this report should not be used as the basis for any investment decisions. We make no representation or warranty as to the accuracy or completeness of the information contained in this report, including third-party data sources. The views expressed are as of the publication date and subject to change at any time. Hypothetical performance has many significant limitations and no representation is being made that such performance is achievable in the future. Past performance is no guarantee of future performance.

Comments


bottom of page