Why Factor Models Do Not Fully Explain Buffett's Historical Returns

Hou and coauthors find that q-factor models do not fully explain Berkshire’s historical excess returns. This concerns model explanatory power and does not directly evaluate every quantitative fund’s actual performance.

Source dates and applicability

The main source is The Economics of Security Analysis, published in 2024; its Berkshire test covers February 1968–December 2020. The 2017 working paper supplies theoretical background. As of October 2026, the historical residual alpha remains a research question. No updated sample is tested here, so it establishes neither present replication performance nor the cause of the residual.

1. Quantitative Researchers Have Already Tried Systematic Replication

A paper by Hou, Mo, Xue, and Zhang points out that a quantitative strategy selecting only stocks with low valuations cannot fully represent Buffett’s investment approach.

Starting from the security analysis framework used by Buffett and his teachers Graham and Dodd, the researchers attempt to express this approach as a more formal theory of asset pricing.

2. The Researchers’ Theoretical Foundation

Their underlying theory is called the investment CAPM, or investment-based asset pricing.

The name sounds complicated, but the intuition is simple: stock prices must ultimately connect to a company’s investment decisions, returns on investment, and the return required by the market.

This framework connects stock analysis with business operations.

They identify three core dimensions of pricing that should sound familiar to anyone who knows Buffett:

1. Companies with strong profitability are generally worth more.
2. But if their share prices are still low, those companies are cheap.
3. Good companies bought cheaply often offer higher expected returns.

The third point is precisely the proposition the paper seeks to formalize. Across stocks, those with high expected returns tend to combine better quality with modest valuations. This logic does not require assuming market inefficiency first. The researchers argue that it can coexist with an efficient-market framework.

The model permits links between fundamentals and expected returns without first assuming market inefficiency. It does not guarantee excess returns from careful analysis.

3. What the Historical Model Leaves Unexplained

After analyzing more than half a century of US stock data and the historical performance of thousands of actively managed funds, the researchers find that their model still cannot fully explain Buffett’s excess returns.

Specifically, section 4.2 and Table 11 report that from February 1968 through December 2020, Berkshire’s monthly excess return averaged 1.41%, with a q-factor-adjusted monthly alpha of 0.59%.

Another surprising result concerns expected growth, which the researchers had assumed Buffett would value and included in the q-factor model. In the Berkshire sample, its loading has the opposite sign. However, this q5 expected-growth loading of -0.23 (t = -1.30) is statistically insignificant.

The paper discusses limitations of accounting measurement, including how R&D, advertising and employee training affect profitability and investment measures. Extending that discussion, this article suggests that management quality, long-term capital allocation and organizational capabilities may also contribute to the gap between the model and Buffett’s decisions; these explanations are not directly tested by the paper.

The q / q5 models explain part of Berkshire’s historical returns. The unexplained component remains a model residual; identifying it as judgment or another ability would require further evidence.

4. References

1. Kewei Hou, Haitao Mo, Chen Xue, and Lu Zhang (2024), The Economics of Security Analysis, Management Science 70(1): 164–186; author-hosted published PDF. Section 4.2 and Table 11 support the February 1968–December 2020 sample, 1.41% average monthly excess return, and 0.59% q-factor alpha. The q5 expected-growth loading is -0.23 (t = -1.30), statistically insignificant. Section 5 discusses accounting measurement limits, including expensed R&D, advertising, and training. The management-quality and organizational-capability examples are this article’s interpretation, rather than variables directly tested in Table 11.
2. Kewei Hou, Haitao Mo, Chen Xue, and Lu Zhang (2017), The Economics of Value Investing, NBER Working Paper 23563, June 2017. Supports the theoretical background connecting investment CAPM and value investing with efficient markets. This earlier working paper cannot support empirical results ending in 2020.