Earnings Season & ML Signal Update

Introduction

As we wrap up 2022, we take one final look at the performance of stocks in and around earnings releases. We also examine how the environment for alpha generation has continued to become more favorable, which should benefit stock pickers into 2023. Lastly, we update our signal that extracts management sentiment from earnings call transcripts through the end of the just-completed earnings season.

Stock Performance and Idiosyncratic Risk During Earnings Season

One phenomenon we have repeatedly pointed out during the past year is the tendency for companies that beat earnings estimates to subsequently outperform the index, and those that miss seeing likely underperformance. The latter of these observations, that stocks missing earnings have tended to underperform, is reflected in the light blue bars in Fig. 1. For the just-completed earnings season, companies missing earnings underperformed the S&P 500 by an average of 3.4% in the 3-day period following announcement.

While this figure is generally in line with performance over the past several earnings seasons, stocks beating earnings have been rewarded to an extent not seen for at least the past three years (see dark blue bars in Fig. 1). Companies beating estimates during the just-completed earnings season outperformed the index by 2.7% on average; this figure is more than triple the average level that outperformers had seen during the previous 11 earnings seasons, dating back to late 2019. Also, the difference between the “reward for beating” and “penalty for missing” also saw its largest value in at least three years.

This result indicates movement toward a phase where idiosyncratic factors are likely to drive stock returns to a greater degree. When idiosyncratic concerns drive stock performance, stock pickers have a better environment in which to generate alpha. We track a model that estimates the degree to which stock performance is driven by idiosyncratic characteristics. It is shown in Fig. 2.

Fig. 2 measures, for each stock, the degree to which the overall market and the sector to which the stock belongs drive overall stock returns. When this figure is low (i.e. “More Macro”) it means that stocks are generally being driven by factors that affect the overall market or sector to a similar degree. In such an environment, differentiating between stocks is difficult, and stock pickers find the sledding tough when it comes to generating alpha.

On the other hand, when this value is high (i.e. “More Stock-Specific”) company-specific concerns are having a relatively larger effect on the returns of individual stocks. In this case, investors can use company research or other means to identify likely outperformers and can more easily generate outperformance.

Fig. 2 shows that after reaching a low in August of 2022, the degree of idiosyncratic risk among stocks has generally increased. While the overall level is still low compared to history, we expect it to continue to rise in the months ahead, meaning stock pickers should face a progressively more beneficial environment in which to generate alpha.

Continued Outperformance for NLP-Based Earnings Call Signal

We have also done work using natural language processing (NLP) to evaluate the sentiment embedded in quarterly earnings call transcripts. Partnering with ProntoNLP, experts in applying NLP techniques to financial data, we developed a signal that evaluates the relative level of optimism or pessimism that managements express when discussing earnings. This signal is like the post-earnings announcement drift anomaly, but instead of focusing on returns conditioned on earnings surprise, we instead focus on how sentiment embedded in an earnings call compares to the level set by management in prior quarters.

In our initial work, we found that companies for which management sentiment improves tended to outperform their peers; likewise, companies where management sentiment had soured tended to underperform. We also found that this signal was distinct from traditional sources of return; as a result, it represents a unique source of alpha. We update the NLP-based sentiment signal to reflect the latest batch of earnings results (see Fig. 3).

From Fig. 3, we see that the basket of stocks for which management sentiment improved the most (dark blue bars) continued to generate outperformance relative to the S&P 500 – this basket of stocks where management’s outlook has turned the most favorable has generated excess return for 4 consecutive quarters. The basket of stocks for which management sentiment soured (gray bars) ALSO outperformed the S&P 500 over Q4, but it generated less excess return than did the positive-sentiment basket.

Considering the spread between the favorable and unfavorable baskets (i.e. constructing a “money-less” position of a long portfolio consisting of the stocks in the favorable basket and a short portfolio made up of the stocks in the unfavorable basket), we see from Fig. 3 that such a strategy generated positive return in Q4. This “long-short” spread strategy has also generated positive performance for four consecutive quarters and six of the past eight quarters.

Conclusion

After a difficult environment for much of 2022, stock pickers may be seeing the light at the end of the tunnel. The market is rewarding earnings “beats” to a degree not seen in several years. Also, the degree to which the market is differentiating between those companies that beat and miss earnings is also at a multi-year high. Going forward, we expect company-specific drivers to play an increasingly important role in determining individual stock return.

As such, stock selection strategies are likely to bear greater fruit in the upcoming months. One such strategy relies on the extraction of management sentiment from earnings call transcripts to identify outperforming stocks. The basket of favored stocks from this strategy has outperformed the market for four consecutive quarters.