Why Beta Alone Explains So Little of a Stock's Return
Investors are routinely told that a stock's beta measures its risk, full stop. The single-index model makes that claim precise and testable, and when researchers actually ran the numbers, beta turned out to explain a much smaller slice of any individual stock's return than the popular story suggests. Understanding why matters directly for how many holdings you need before diversification actually works.
The single-factor idea
The single-index model splits a stock's return into two pieces: the part that moves with the overall market, and everything else. Formally, return on stock i = alpha + (beta × market return) + residual, where alpha is the stock's average return unrelated to the market, beta measures how sensitively the stock swings when the market swings, and the residual captures whatever is left over: earnings surprises, a lawsuit, a product launch, a management change, anything specific to that company. The residual is assumed uncorrelated with the market and, crucially, uncorrelated across different stocks' residuals as well. That second assumption is what makes the model powerful: if firm-specific shocks really are independent of each other, then holding many stocks should cause those shocks to cancel out, leaving a portfolio that behaves almost purely like a scaled-down version of the market.
This is not just a theoretical convenience. It is a direct, testable claim about how much of any stock's variance comes from a single common source (the market) versus firm-specific noise. Researchers have been able to check that claim against real price data for close to a century of U.S. trading history, and the answer is more nuanced than either die-hard indexers or stock pickers usually admit.
The single-index model has a theoretical cousin worth naming explicitly: single-factor arbitrage pricing theory, which arrives at almost the identical equation from the opposite direction. Rather than fitting a regression line to historical data, single-factor APT starts from a no-arbitrage argument: if returns really are driven by exposure to one common factor plus independent noise, then any pricing pattern that deviated from a straight line relating factor exposure to expected return would create a risk-free arbitrage opportunity, which well-functioning markets should compete away almost instantly. The index model is the empirical, data-fitting half of this story; single-factor APT is the theoretical, no-arbitrage half. When they agree, as they largely do for the broad shape of the risk-return relationship, that agreement is itself informative: it means the empirically estimated betas are not just a statistical curiosity but are consistent with a coherent economic argument about why prices should behave that way in a competitive market.
Testing it with real numbers
Estimating beta is a straightforward regression: beta equals the covariance between a stock's returns and the market's returns, divided by the variance of the market's returns, or beta = Cov(stock, market) / Var(market). Suppose a stock's monthly return covaries with the market at a covariance of 12 (in percent-squared units), and the market's monthly return variance is 16 (equivalent to a monthly standard deviation of 4%). Then beta = 12 / 16 = 0.75. This stock, on average, moves three-quarters as much as the market: a 4% market month tends to come with a 3% move in the stock, all else equal.
Now bring in alpha. Say the market's average monthly excess return (over the risk-free rate) is 0.8%, and the stock's own average excess return over the same period is 1.1%. The single-index model predicts an excess return of beta × market excess return = 0.75 × 0.8% = 0.60%. The stock actually returned 1.1%, so its estimated alpha is 1.1% minus 0.60%, or 0.50% per month, roughly 6% a year above what its market exposure alone would predict. Whether that 0.50% is a genuine skill signal or statistical noise depends entirely on how precisely it is estimated, which is where sample size and standard errors come in, and why a handful of good months proves almost nothing.
The second worked example gets at the diversification claim directly, using variance decomposition. Total variance splits into a systematic piece and a residual piece: total variance = beta2 × market variance + residual variance. Take a stock with a total monthly return variance of 64 (an 8% monthly standard deviation, fairly typical for an individual mid-cap stock) and the same beta of 0.75 against a market variance of 16. Systematic variance = 0.752 × 16 = 0.5625 × 16 = 9. Residual variance is then 64 minus 9, or 55. The R-squared of this regression, the fraction of total variance explained by the market factor, is 9 / 64 = 0.14, or about 14%. In plain terms: the market explains only 14% of this stock's month-to-month variance. The other 86% is company-specific noise that a single-stock investor is carrying for no compensation, since diversifiable risk earns no risk premium in any standard asset pricing framework.
What decades of data show
Large-scale regressions of individual U.S. stock returns against a broad market index have repeatedly found R-squared values clustered in the 10% to 35% range for the typical stock, with plenty of names below 10%. Utilities and large, stable, market-weight-heavy companies tend to sit at the higher end; small, newer, or single-product companies sit at the lower end. This is the empirical face of the same result shown in the worked example above: for any one stock, firm-specific noise usually dominates market-driven variance, often by a wide margin.
A second, related empirical finding concerns the shape of the relationship between beta and average return. If the single-factor model held perfectly and investors were only compensated for bearing market risk, then average returns sorted by beta should trace a straight line with a slope equal to the market risk premium. Decades of cross-sectional tests instead find that line is flatter than theory predicts: low-beta stocks earn more than the model says they should, and high-beta stocks earn less. This low-beta pattern has itself become one of the most studied anomalies in empirical finance, and it is the direct descendant of testing the single-index model rather than simply assuming it.
A third finding concerns portfolios rather than single stocks. As you combine more securities whose residuals are genuinely close to independent, the portfolio's R-squared against the market rises steadily, because the idiosyncratic pieces increasingly cancel while the common market exposure does not. Empirical studies of randomly constructed portfolios find that most of the reduction in unsystematic risk happens within the first 20 to 30 holdings, after which the marginal benefit of adding one more name shrinks fast. That is the single-index model's diversification prediction, confirmed directly in the data rather than assumed from theory.
A fourth finding, easy to overlook, concerns stability. Betas estimated from one five-year window and re-estimated from the next five-year window for the same company are correlated but far from identical; a stock with an estimated beta of 0.75 in one period might show 0.60 or 0.90 in the next, simply from sampling variation and genuine shifts in the company's business mix, leverage, or competitive position. This instability is one reason professional risk models rarely use a raw historical beta unadjusted; many blend the historical estimate toward 1.0, the market average, on the reasoning that an extreme estimated beta is more likely to be partly noise than a stable, permanent company characteristic.
Using this in a real portfolio
The practical lesson translates almost directly into portfolio construction. If you hold five stocks, you are still carrying a meaningful amount of avoidable, uncompensated firm-specific risk, because five residuals do not cancel each other out very effectively. If you hold 30 to 50 stocks spread across different industries, the residual risk has mostly washed out and your portfolio's return behavior is dominated by its aggregate beta, meaning your results will track the market closely, for better and worse. This is precisely why broad index funds, which effectively hold thousands of names, exhibit R-squared values against their benchmark north of 0.98: virtually all their variance is systematic by construction.
The model also clarifies what active stock picking is actually betting on. A concentrated portfolio is a bet that your handful of alphas are real and large enough to outweigh the extra unsystematic risk you are voluntarily taking on. Given how noisy alpha estimates are (the 0.50% monthly alpha in the earlier example needs years of data before you can distinguish it from zero with any confidence), concentrated bets require either genuine informational or analytical edge, or a willingness to accept return volatility that has nothing to do with market conditions and everything to do with company-specific outcomes you did not diversify away.
There is also a subtler, practitioner-level lesson buried in the beta-instability finding above: forecasting a future beta is a genuinely harder problem than measuring a past one. A risk model built purely on trailing five-year betas will systematically overreact to companies that happened to have an unusually volatile or unusually calm recent history, then get surprised when that company's beta reverts toward the market average in the following period. This is the statistical basis for shrinkage adjustments used throughout the investment industry, and it is worth remembering any time a fund fact sheet or brokerage screen presents a single beta number as if it were a fixed, known physical constant rather than a noisy estimate with its own margin of error.
Actionable breakdown
- Reading a stock's beta and R-squared
- Treat a low R-squared as normal, not a red flag.
- Recognize beta measures market sensitivity, not total risk.
- Check the estimation period; betas drift over time.
- Building a diversified core
- Hold 25 to 30 names minimum for single-stock strategies.
- Spread holdings across unrelated industries, not just tickers.
- Prefer index funds when you want pure systematic exposure.
- Evaluating claimed alpha
- Demand years of data before trusting a positive alpha.
- Ask what benchmark was used to compute beta.
- Compare alpha's size to its statistical uncertainty.
Common pitfalls
Treating beta as a complete risk measure: beta only captures systematic exposure; a low-beta stock can still carry enormous idiosyncratic risk that shows up in a single bad earnings call.
Assuming residuals are actually independent: in real markets, firm-specific shocks correlate more than the model assumes during sector-wide events, such as a regulatory crackdown hitting every company in an industry at once, which is why sector concentration undermines diversification even with many holdings.
Chasing a short-window alpha: a 0.50% monthly outperformance over six months is statistically indistinguishable from luck; the standard error on alpha estimated from small samples is typically larger than the alpha itself.
Using a single stale beta indefinitely: a beta estimated years ago for a company that has since changed its business mix, leverage, or product lineup no longer describes its current risk; refresh the estimate rather than anchoring on an old number.
The bottom line
The single-index model's core prediction, that most of an individual stock's variance is diversifiable noise rather than market risk, holds up remarkably well in the data, which is the strongest empirical argument for owning dozens of stocks or a broad fund rather than a handful of favorites.
The single-index model · The Capital Asset Pricing Model · Tests of the multifactor CAPM and APT · Factor investing guide · Understanding risk