How the D&B Data Cloud and Machine Learning Help Predict Stock Returns
Dun & Bradstreet has the largest global commercial database on the planet. At the core is proprietary trade payment data collected from thousands of trade suppliers, which is closely related to account payables on company’s balance sheet. The Credit Score Archive Database (CSAD, 2004-2018) and Detailed Trade Risk Index (DTRI, 2011 – 2018), the engines that help power D&B Credit, are directly derived from the raw trade payment data, and they describe companies’ payment behavior and various risk assessments.
Including further mathematical transforms, 400+ attributes are extracted from CSAD and DTRI databases to form the foundation of the Dun & Bradstreet US Equity Alpha Factor Library. Through rigorous statistical analysis and machine learning, I and my colleagues in Analytics Innovation demonstrate their values in forecasting future stock returns.
I’ve authored several whitepapers that detail our work studying capital markets against payment data stock returns. Overall, we believe Dun & Bradstreet data is a unique data source that, with complete coverage on public stocks in the US, provides extra information not available in other alternative data sources.
Click on any of the papers below to get an in-depth look at our stock market insights.
D&B US Equity Alpha Factor Library
Here we formally introduce the alpha factor library including data sources, factor constructions, and test methodologies. We systematically test 400+ Dun & Bradstreet trade factors for a high degree of statistical significance in forecasting future one-month excess stock returns (or alphas). Such alphas are not explained by Carhart’s Four-Factor Model (i.e., the Fama-French 3-Factor Model plus Momentum). Test results include single factor alpha (residual returns), exposure to Carhart’s 4 factors and t-statistics, as well as historical performance separation between top and bottom deciles. The complete list of attributes identified, with test statistics, is available upon request.
Using Dun & Bradstreet US Equity Alpha Factor Library and Machine Learning to Improve Stock Portfolio Returns
We applied machine learning to model equity future beta adjusted returns using Dun & Bradstreet data attributes. With 400+ factors, we can pick the worst-performing stocks with reasonable success, over 16 quarters (2014 – 2017). The results compare favorably with benchmarks using single factors (both public and proprietary).
High Relative Trade Credit Underperforms
Here we found that stocks with high relative trade credit (to sales) underperform those with low relative trade credit. The underperformance is statistically significant after adjusting for the Fama-French 3-factor model + MOM (aka FF3+MOM).
The Hidden Cost of Growing Trade Supplier Networks Too Fast
Dun & Bradstreet found that stocks with the fastest-growing number of trade suppliers year-over-year underperform those with slowest-growing trade suppliers. The underperformance is statistically significant after adjusting for the Fama-French 3-factor model + MOM (aka FF3+MOM).
When Slow or Negative Payment Experiences Accelerate
Dun & Bradstreet found that stocks with the fastest increase in the number of slow or negative payment experiences month-over-month underperform those with slowest increase. The underperformance is statistically significant after adjusting for the Fama-French 3-factor model + MOM (aka FF3+MOM).