SELECTED WORK

Relative Valuation with Machine Learning. 

Journal of Accounting Research 61, no. 1 (2023): 329-376. Geertsema, Paul, and Helen Lu.

Valuation multiples are widely used by both retail and professional investors, often based on subjective judgements including similarity in business, size, profitability and more. Tree-based machine learning algorithms lend themselves well to this peer-based relative valuation task. We train Gradient Boosting Machines to predict valuation multiples. GBMs learn optimal decision rules to divide firms into subgroups with more similar value-relevant fundamentals. GBMs generate more accurate valuation in each month in the test sample. We derive a mathematical formula to show that the GBM prediction is a weighted average of peer firms’ valuation multiples, with more similar (dissimilar) firms carrying higher/lower weights.

↑ Return to top 

Projecting Financial Statements with Chained Machine Learning. 

Geertsema, Paul and Helen, Lu and Ma, Guang (R&R at Journal of Accounting Research.)

The accounting literature so far primarily predicts single line items (for example, earnings per share). Can we leverage the dependency across line items regulated by accounting logic to predict the full financial statements? We develop a chained-learning framework to project the complete set of financial statements, predicting items sequentially in a human-informed order that embeds the interconnections among items and reflects the underlying logic of accounting. Implemented with Gradient Boosting Machines on a large U.S. sample, the chained model reduces out-of-sample forecast errors by 32% compared to a parallel model that predicts each line item independently.

↑ Return to top 

Specifics Matter: An Analysis of Mutual Fund ESG Disclosures. 

Shi, Huayu and Han, Xing and Lee, John B. and Lu, Helen  (R&R at Contemporary Accounting Research.)

Using a large language model to classify ESG statements, we find that funds with more specific disclosures attract greater investor flows, particularly among institutional-oriented funds and during periods of heightened climate concern. Disclosure specificity is also positively associated with the ESG performance of future fund holdings. These findings suggest that specific ESG disclosures serve as credible signals of a fund’s ESG commitment, enabling investors to make more informed, sustainability-conscious decisions. The results offer empirical support for recent regulatory efforts to promote more structured and detailed ESG reporting.

↑ Return to top 

↑ Return to top 

Predicting Performance-Induced CEO Turnovers Out-of-Sample. 

Berkman, Henk, Helen, Lu and Jubin, Zheng (work-in-progress).

↑ Return to top

The Correlation Structure of Anomaly Strategies. 

Geertsema, Paul, and Helen Lu. "The correlation structure of anomaly strategies." Journal of Banking & Finance 119 (2020): 105934. 

We consolidate a large number of mean-significant anomalies into cluster portfolios. More than a third of cluster portfolios remain significant under the Hou et al. (2020) five-factor model — the best performing among six benchmark models tested. A best-first search yields nine factors that subsume all cluster portfolios as well as all significant anomalies, demonstrating the feasibility of a parsimonious description of average realised returns. The expected growth factor (EG) and a cluster portfolio linked to accruals are prominent factors that improve pricing performance. The search-generated model produces a monthly maximum squared Sharpe ratio of 0.51, considerably higher than current benchmark models. 

[Google Scholar] [Github]       [SSRN]   [LinkedIn]   [Curriculum Vitae]