Hong Kong-based independent research firm ExtractAlpha recently published a white paper on IRP and data provider New Constructs’ proprietary Core Earnings metric, and identified a trading signal from this data that produces a 10.1% annualized return.
New Constructs is an independent research and fundamental data provider that leverages AI tools to extract hard to find metrics from financial filings in order to provide unique insights about public companies’ real economic earnings. New Constructs uses this data to adjust corporate earnings for activities that are not central to a company’s business activities. This adjusted value is called “Core Earnings”.
According to an academic study conducted in 2019 by professors Ethan Rouen and Charles C.Y. Wang of Harvard Business School, and Eric So of MIT’s Sloan School of Management, New Constructs’ Core Earnings metric is a more accurate measure of a firm’s current and future performance than is provided by any other traditional fundamental data provider.
In its recently published white paper, ExtractAlpha found that New Constructs’ Core Earnings metric is a more persistent measure than Net income, as its autocorrelation with next year’s value is 48%, compared to 31% for net income. Core Earnings also displays higher autocorrelation than Net Income the further out into the future you go.
While this is interesting, it is not the real point of ExtractAlpha’s study. ExtractAlpha discovered that the difference between Core Earnings and reported Net Income, what it terms the distortion, has strong predictive value for future net income.
In fact, ExtractAlpha’s analysis found that over the period 2015 – 2021, if you create a top/bottom decile monthly rebalanced long/short portfolio where you go long the stocks with the lowest distortion and short the stocks with the highest distortion, you would generate annualized returns of 10.1% with a Sharpe ratio of 1.44.
Even more impressive, ExtractAlpha found that using a Fama-French return attribution analysis, that 9.3% of the 10.1% total return would be idiosyncratic alpha versus portfolio performance due to factor or sector exposures. In other words, the bulk of the returns associated with this trading strategy would not be found in other traditional trading strategies (value, momentum, short-term reversal, sector exposure, etc.).
The recent study published by Hong Kong-based ExtractAlpha on New Constructs’ Core Earnings data, the trading signals created from this data, and the resulting trading strategy it produced should be extremely helpful to New Constructs as it tries to market its fundamental data to buy-side investors.
It is interesting that New Constructs felt it needed to engaged ExtractAlpha to produce this study in order to prove that their proprietary Core Earnings metric is a more accurate and persistent measure than traditional profitability measures. In our minds, this finding was made in 2019 when professors Ethan Rouen and Charles C.Y. Wang of Harvard Business School, and Eric So of MIT’s Sloan School of Management published their academic study called Core Earning: New Data and Evidence. However, the new ExtractAlpha study takes the next step, in helping the buy-side identify predictive signals and develop a viable trading strategy from the New Constructs data.
In our minds, what this shows is that many buy-side firms don’t have the bandwidth to ingest, prepare and analyze new datasets in a timely fashion – particularly given the explosion of new datasets that have entered the marketplace in recent years. Consequently, a growing number of data vendors are finding that they need to do much of this work themselves by identifying trading signals from their data and creating investable strategies based on these signals. Data vendors like New Constructs believe this will make their data easier to consume, and therefore reduce the buy-side’s time to purchase their data.
It is this very pressure that has caused data vendors to seek out analytical support from a growing number of firms like ExtractAlpha, Neuravest, AltHub / Invisage, and CloudQuant, to name but a few. We suspect that the demand data vendors will have for help from these third-party data analytics providers is likely to expand in the future, particularly as the growth in the number of data providers continues to outstrip the buy-side’s ability to integrate and analyze these datasets themselves.