When low beta stocks stop working
I have long been a fan of the low beta anomaly. The low beta anomaly describes the observation that minimum variance portfolios and low beta stocks in general outperform high beta stocks and the market overall. This goes against all theories that state that returns should be compensation for systematic risks, so systematically riskier stocks should have higher returns in the long run.
Imagine my surprise, then, when I came across a new paper by Zhiqi Cao, Wenfeng Wu and Youchang Wu who claim that the low beta anomaly works, but not when you need it most. They looked at the capital market line in the CAPM model (I know, the CAPM has been violated hundreds of times, but people still use it for some reason), but split it between times of high and low economic uncertainty. The chart below shows the result.
Capital Market Line in times of high and low uncertainty
Source: Cao et al. (2025)
The fascinating insight of this research is that low beta stocks outperform high beta stocks when economic uncertainty is low. But when economic uncertainty is high, high beta stocks outperform. The authors claim that in times of low uncertainty, many investors are more comfortable owning stocks with high volatility and high beta. The result is that in these times, the price of high beta stocks gets bid up and subsequent returns decline. But when economic uncertainty is high, investors retreat to safer stocks, and the low beta anomaly disappears.
Asked Gemini if this is a fake study... :-).. (Only provided the last paragraph as input, no name or source anything...)
While the claim seems logical at first glance, it largely reverses established financial findings regarding the low-beta anomaly. Generally, the opposite is true.
The text you provided appears to be a direct translation of a specific research perspective, such as that discussed by Joachim Klement or in the paper "Uncertainty and the Beta Anomaly".
Analysis of the Claim
The Low-Beta Anomaly: Historically, low-beta stocks often achieve better risk-adjusted returns than high-beta stocks. This is a paradox because, according to the Capital Asset Pricing Model (CAPM), higher risk (beta) should be rewarded with higher returns.
Behavior During High Uncertainty: In times of high economic uncertainty (e.g., recessions), investors typically flee to "safe havens"—meaning low-beta stocks like utilities or consumer staples. This usually causes low-beta stocks to outperform high-beta stocks during downturns.
Behavior During Low Uncertainty: High-beta stocks typically shine when uncertainty is low and the market is optimistic, as they amplify market gains.
Why the Claim Might Be "True" (as a Specific Theory)
The claim you cited argues for a reversal of the anomaly based on investor sentiment:
Low Uncertainty: Investors are overconfident and bid up high-beta stocks, making them overvalued and leading to poor subsequent returns (hence, low-beta outperforms).
High Uncertainty: Investors become risk-averse and flock to low-beta stocks, bidding up their prices and making them overvalued. This allows high-beta stocks—now neglected and "cheap"—to potentially offer better expected returns.
Conclusion: The claim is not necessarily a "fake," but it represents a specific, counter-intuitive research hypothesis that challenges the standard view of how low-beta strategies work. Most market data still suggests that high-beta is the primary loser during periods of high stress.
Interesting article. Here’s a quant’s take on what might really be behind the schizophrenic behaviour of beta: finance textbooks write this as
beta= var(stk)/covar(stk,mkt)
But if we unpack this a bit we can rewrite this as
beta= sd(stk)/sd(mkt) *corr(stk,mkt)
This means low beta can have two sources — low relative volatility and low correlation to the market. Low correlation to the market isn’t necessarily good. When General Public Utilities had its three mile island problem, the stock’s vol spiked but its beta went to zero because its correlation with the market went to zero.
By the way, rewriting the formula for beta as I’ve done here reveals another serious flaw with its calculation: correlation, the most important variable in the formula, is the square root of r-squared, the quantity that expresses the statistical usefulness of beta. In other words, the lower the correlation, the lower the reliability of your beta calculation. Does that strike you as problematic?