Predicting anomalies - It is possible

In our last post, "Anomalies offer opportunities after they are reported", we discussed the timing of anomalies. You can make money by fast reaction to new information anomalies Now, we want to discuss research that states that anomalies can be predicted, or you can act early on anomalies. 

This conclusion comes from the paper, "Predicting Anomalies" from the same set of authors. It is a nice companion piece. If we can make judgments on what happens after anomalies, we should be able to test whether there is some drift or systematic behavior prior to the anomaly. Well, what do you know but there is drift before these anomaly trading signals. You can make money acting before these information events; however, these signals are often centered around anomaly trading signals that are hard to forecast. 

The researchers find that a simple martingale model will work best. Acting on the signal from the prior quarter is a good indicator of what will happen next year based on the current quarter. We can say there is trend in anomalies. There are predictable patterns prior to the release of annual financial statements. Accounting-based anomaly signals based on financial accounting events are predictable with even the worst prediction models have F1 scores above 60%. Of course, the signal strength will vary with the anomaly. For example, earnings and revenue show less persistence. This should be expected since those are the accounting numbers that investors spend the most time studying. Some financial accounting data are very persistence and predictable while others are not.

This is an important paper that requires careful review for any equity quant manager.