Predicting Firm Profit Using Machine Learning Techniques

In a previous post, we presented an article on using an econometric model for predicting the P/E ratio. In this post, we will discuss a different approach for predicting a firm’s financials.

Reference [1] utilized the Gradient boosting method for predicting a firm’s profitability. Gradient boosting is a method that belongs to the family of Machine Learning techniques. It allows us to treat a large number of factors and build a predictive model,

Gradient boosting is a machine learning technique for regression, classification and other tasks, which produces a prediction model in the form of an ensemble of weak prediction models, typically decision trees. When a decision tree is the weak learner, the resulting algorithm is called gradient boosted trees, which usually outperforms random forest. It builds the model in a stage-wise fashion like other boosting methods do, and it generalizes them by allowing optimization of an arbitrary differentiable loss function. Read more

The authors used the Gradient boosting method to predict firm profit, and they compared results to those predicted by Fama-MacBeth regressions,

This paper compares firm profit predictions based on Fama-MacBeth regressions to predictions based on gradient boosting. Gradient boosting provides higher quality predictions due to their ability to include many more factors. The predictions are evaluated directly and also in three test settings; one from behavioral finance, one from corporate finance, and one from asset pricing.

They found that,

…the gradient boosting approach (denoted GBRT), due to Friedman (2002) produces better firm protfit predictions than does the Fama and MacBeth (1973) approach (denoted FM). This is true both in-sample and out-of-sample. It is primarily due to the ability to include many more factors without over-fitting the data.

… we find that large firms and investment grade firm profits are more predictable than average firms. Firms with high R&D, market-to-book, and cash flow volatility have less predictable profits than average firms. Among publicly traded firms that exit, unprofitable firms tend to be liquidated or bankrupt; while protable firms tend to be involved in an acquisition, a merger, an LBO, or to become a private firm. During the financial crisis of 2007-2009 and during NBER recessions, firm profits become less predictable. The reduced predictability during bad times affects average firms much more than it affects investment grade firms.

In short, using the Gradient boosting method, the authors were able to predict firm profit. We found it impressive that they used 140 factors to build a predictive model without overfitting.

References

[1] MZ. Frank, K. Yang, Predicting Firm Profits: From Fama-MacBeth to Gradient Boosting, 2021. https://ssrn.com/abstract=3919194

Originally Published Here: Predicting Firm Profit Using Machine Learning Techniques