We measure earnings distortion using a proprietary human-assisted ML technology featured in a recent paper from Harvard Business School (HBS) and MIT Sloan. This paper empirically shows that Compustat and street earnings estimates are incomplete and less accurate since they do not consistently and accurately adjust for unusual gains/losses buried in footnotes.
We leverage our proprietary dataset of unusual gains/losses to derive Earnings Distortion Scores for ~3,000 stocks. These scores indicate how likely companies are to beat or miss estimates based on how much unusual gains/losses cause unadjusted earnings measures to be over/understated.
The Earnings Distortion Score formula is: Core Earnings Distortion divided by Total Assets
We decile these values and, then, categorize into a 5-tier scoring system:
- Strong Beat – Top decile (Least earnings distortion)
- Beat – Second and third Decile
- In line – Fourth, fifth, sixth and seventh decile
- Miss – Eighth and ninth decile
- Strong Miss – Bottom decile (Most earnings distortion)
We scale core earnings distortion by total assets so large companies don’t dominate the extreme ends of the spectrum, as they are likely to have more earnings distortion simply due to their size. Further, scaling by total assets mirrors the strategy in “Core Earnings: New Data and Evidence”, the paper from professors at Harvard Business School and MIT Sloan that uses the same metric to construct a long/short strategy that delivers abnormal returns of 7-10% annually.
Our Earnings Distortion Scores empower investors to combat management efforts to obfuscate financial performance. For more on how to use our Earnings Distortion Scores, click here.
This article originally published on January 13, 2020.
Disclosure: David Trainer, Kyle Guske II, and Sam McBride receive no compensation to write about any specific stock, style, or theme.