A groundbreaking new study challenges conventional wisdom about how we look at investing factors. The paper, “Does Peer-Reviewed Research Help Predict Stock
Returns”, written by Andrew Chen of the Federal Reserve Board, Alejandro Lopez-Lira of the University of Florida, and Tom Zimmermann of the University of Cologne, asks a provocative question: Does peer-reviewed academic research actually help predict stock returns better than simply mining data for patterns?

The Key Finding: Data Mining Works Nearly As Well As Research

The study’s headline result is both surprising and humbling for the academic finance community: Simply scanning through accounting ratios looking for statistical patterns produces investment returns nearly identical to strategies that had behavioral or risk-based explanations. The researchers found that data-mined strategies retain about 51% of their returns out-of-sample (after the original testing period), while peer-reviewed academic strategies retain about 53% – a mere 2 percentage point difference.

How They Tested This

The researchers analyzed about 200 published stock return predictors from academic papers and compared them to strategies created by mining through 29,000 different accounting ratios. For each published academic strategy, they created matching data-mined strategies that had similar statistical significance in the original sample period.

The data mining approach was remarkably simple: they just looked for accounting ratios that had strong statistical relationships with future stock returns, without trying to explain why those relationships existed. They then tracked how well both the academic and data-mined strategies performed after their initial discovery period.

Important Implications for Investors

1. Simple Patterns Matter: The study suggests that simple statistical relationships, even without deep theoretical explanations, can identify profitable trading strategies. This validates quantitative approaches that focus on “what works” rather than “why it works.”
2. Academic Theory May Not Add Much Value: Perhaps most surprisingly, strategies backed by sophisticated academic theories and risk-based explanations actually performed slightly worse out-of-sample than strategies discovered through pure data mining. This challenges the common belief that theoretical understanding helps identify more robust investment approaches.
3. Early Discovery of Major Themes: The data mining approach identified many major investment themes years or decades before they were published in academic journals. For instance, it picked up on the importance of factors like:

• Investment rates
• Share issuance
• Accruals
• Inventory growth
• Earnings surprises

These would later become well-known academic factors, but simple data analysis spotted them first.

4. Diversification Benefits: The researchers found that data mining uncovered many strategies with low correlations to each other. This suggests potential diversification benefits from combining multiple factors, even if they’re discovered through statistical analysis rather than theory.

The “Bitter Lesson” for Asset Pricing

The researchers draw a parallel to artificial intelligence research, citing what’s known as the “bitter lesson”: that simple methods with lots of data often outperform hand-crafted solutions based on human expertise. They suggest that investment research may face a similar reality – that scanning large datasets for patterns might be more productive than developing complex theories.

This doesn’t mean theory is worthless. Rather, the authors suggest that data mining could help identify empirical patterns that theory can then try to explain. This might be more productive than starting with theory and then looking for supporting evidence.

Practical Takeaways for Investors

1. Don’t Overthink It: Simple quantitative strategies based on clear statistical relationships can work well, even without deep theoretical justification.
2. Focus on Data Quality: Having good, clean data may be more important than having sophisticated theories about why certain strategies should work.
3. Diversification Matters: Both academic and data-mined strategies tend to weaken over time. Combining multiple approaches can help manage this decay.
4. Be Open to New Sources: Good investment ideas can come from both rigorous academic research and simpler statistical analysis. Don’t dismiss either approach.
5. Watch for Decay: Both academic and data-mined strategies tend to produce lower returns after they’re discovered. Plan for this by:

• Having a margin of safety in your analysis
• Combining multiple strategies
• Continuously researching new approaches

The Big Picture

This research suggests that the investment community might benefit from being more humble about our ability to understand exactly why certain strategies work. Simple statistical relationships, properly tested and carefully implemented, might be just as useful as complex theories.

For investors, this validates quantitative approaches that focus on what the data shows rather than requiring detailed explanations for every pattern. While theory and understanding are valuable, they may not be necessary prerequisites for developing effective investment strategies.

The study also highlights the enduring challenge of finding sustainable edges in markets. Whether discovered through academic research or data mining, investment advantages tend to weaken over time as more investors attempt to exploit them. This reinforces the importance of continuous research and adaptation in investment management.

For the broader investment community, these findings suggest we might want to devote more resources to exploring and analyzing data, rather than primarily focusing on developing theoretical explanations. While both activities have value, the empirical results suggest that simple data analysis might be surprisingly powerful relative to more theory-driven approaches.