The Hidden Assumptions Behind CAPM, and What Happens Without Them
CAPM's single-line formula hides a long list of simplifying assumptions about how markets and investors behave, assumptions that rarely hold exactly in the real world. Understanding which ones matter most, and what changes when they are relaxed, explains both why the model is criticized and why more sophisticated variants of it are still built on the same foundation.
The full list of assumptions
The original capital asset pricing model rests on a specific set of idealized conditions. Investors are assumed to be rational mean-variance optimizers, caring only about the expected return and standard deviation of their portfolios and nothing else about the shape of the return distribution. All investors are assumed to share identical expectations about future returns, variances, and covariances, an assumption called homogeneous expectations. Markets are assumed frictionless: no taxes, no transaction costs, no restrictions on short selling, and assets that are infinitely divisible. Every investor can borrow and lend unlimited amounts at the same single risk-free rate. All information is available to everyone at the same time and free of cost. And the model is a single-period model, meaning investors choose a portfolio for one holding period and there is no meaningful way to model changing opportunities over time within the basic framework.
Stated all together, the list looks almost absurd, and generations of critics have pointed out, correctly, that no real market satisfies all of them simultaneously. The more interesting and more useful question is not whether the assumptions are literally true, which they are not, but how much the model's core prediction, the security market line, actually depends on each one.
It helps to separate the assumptions into two rough categories: those about investors, and those about markets. The investor-side assumptions, mean-variance preferences, homogeneous expectations, a shared single-period horizon, are really assumptions about psychology and information, and behavioral finance research has spent decades documenting specific, systematic ways real investors depart from them, overconfidence, loss aversion, and disagreement chief among them. The market-side assumptions, frictionless trading, unrestricted borrowing, no taxes, are assumptions about institutional structure, and they are easier to test directly since trading costs, borrowing spreads, and tax rates are all observable numbers rather than unobservable mental states.
Which assumptions matter most
Some assumptions turn out to be largely cosmetic. Relaxing the assumption of costless information, for instance, mostly just adds friction and noise around the same basic equilibrium; it does not change the fundamental shape of the pricing relationship. Other assumptions, however, are load-bearing, and relaxing them produces a genuinely different model rather than a noisier version of the same one. The single most consequential assumption is unrestricted borrowing and lending at a common risk-free rate, because the entire derivation of the security market line depends on every investor being able to lever up or delever their portfolio along the same capital market line. Real investors cannot borrow at the government's risk-free rate; individuals typically pay a meaningfully higher rate to borrow than the rate at which they can lend, and many institutional investors face outright leverage restrictions.
The homogeneous expectations assumption is a close second in importance, and it is worth being precise about what it actually requires. It does not merely require that investors agree on which stocks are good or bad; it requires that every investor computes an identical expected return, an identical variance, and an identical full covariance matrix across every risky asset in the market. Relax this even slightly and different investors will optimally hold different risky portfolios, not just different mixes of the same optimal risky portfolio and the risk-free asset, which breaks the two-fund separation result that makes the security market line derivation work in the first place. In practice, of course, investors disagree constantly, which is precisely why trading volume exists at all; a market where everyone genuinely agreed on every asset's expected return would see almost no voluntary trading beyond routine rebalancing.
The zero-beta CAPM, worked through
Economist Fischer Black showed in 1972 that if unrestricted risk-free borrowing is removed from the model entirely, but short selling of risky assets is still allowed, the security market line still holds, but with a twist: the risk-free rate in the formula is replaced by the expected return on a portfolio of risky assets constructed to have zero correlation with the market portfolio, called the zero-beta portfolio. The zero-beta CAPM equation becomes: E(r) = E(r_z) + beta × (E(r_m) - E(r_z)), where E(r_z) is the expected return on the zero-beta portfolio, which is generally higher than the true risk-free rate because it is a risky portfolio, just one uncorrelated with the market.
Suppose the true risk-free rate is 4%, but because unlimited risk-free borrowing is unavailable, the relevant zero-beta portfolio has an expected return of 5.5% instead. The expected market return remains 10%. Under the standard CAPM, a stock with beta 1.2 would require 4% + 1.2 × (10% - 4%) = 4% + 7.2% = 11.2%. Under the zero-beta version, the same stock requires 5.5% + 1.2 × (10% - 5.5%) = 5.5% + 5.4% = 10.9%. Note the security market line's intercept rises and its slope flattens: the zero-beta model predicts a smaller gap between low-beta and high-beta required returns than the standard model does, which happens to align with a well-documented empirical pattern, that the observed security market line in real data is flatter than plain CAPM predicts.
Other major extensions
Beyond the zero-beta model, several other extensions relax different assumptions. A version incorporating taxes recognizes that dividend income and capital gains have historically often been taxed at different rates, which can shift investor preferences between high-dividend and low-dividend stocks in ways plain CAPM ignores, producing an after-tax security market line where dividend yield becomes a second priced factor alongside beta. A liquidity-augmented version recognizes that some securities are more costly to trade than others, illiquid small-cap stocks compared with heavily traded large-cap names, for example, and that investors demand extra expected return as compensation for that illiquidity, above and beyond compensation for beta risk alone. An intertemporal version, developed by Robert Merton, relaxes the single-period assumption, allowing investors to care not just about next period's wealth but about the risk that investment opportunities themselves might change in the future, which introduces additional priced factors tied to shifts in the investment opportunity set, interest rates, and inflation expectations among them.
A consumption-based version reframes the entire model, arguing that what investors actually care about is not portfolio wealth for its own sake but future consumption, and that an asset's relevant risk is therefore its covariance with aggregate consumption growth rather than its covariance with a stock market index. This reframing is theoretically elegant, connecting asset pricing directly to the economics of consumption smoothing, but it has proven notably difficult to test empirically, in part because consumption data is measured with far more noise and far less frequency than stock returns.
A final, more institutional extension addresses the assumption of unrestricted short selling. Many real-world investors, pension funds and mutual funds bound by their own charters chief among them, are either barred from short selling entirely or face meaningful practical constraints on it. Models that impose these restrictions generally predict that overpriced securities can remain overpriced for longer than the frictionless version of CAPM would allow, since the investors most likely to recognize the mispricing, and best positioned to correct it by selling short, are often exactly the ones prevented from acting on that view. This line of reasoning connects directly to a body of research on short-sale constraints and subsequent stock returns, which has found that stocks with unusually high short-selling costs or restrictions tend, on average, to be more prone to sustained overvaluation than stocks where shorting is cheap and unrestricted.
What the evidence says about the extensions
None of these extensions fully rescue CAPM's empirical performance. The zero-beta model does improve the fit of the security market line's slope in many studies, consistent with the intuition above, but it does not resolve other well-documented anomalies, most notably the tendency of small-capitalization stocks and stocks with low price-to-book ratios to earn higher average returns than their betas alone would predict. The liquidity-augmented version does help explain part of the return premium historically earned by small and thinly traded stocks, since illiquidity and small size are correlated in most markets, but liquidity alone does not explain the full magnitude of the anomaly. These persistent gaps are precisely what motivated the shift toward multifactor models in the 1990s and beyond, which add additional priced risk factors, size, valuation, momentum, and profitability among the most studied, directly into the pricing equation rather than trying to rescue a single-factor model through incremental adjustment.
Why this matters for real portfolios
For most individual investors, the practical lesson from this long list of extensions is not that CAPM should be discarded, but that its output, a single required return number from a single beta, should be treated as a rough first approximation rather than a precise verdict. An investor building a portfolio around a low-cost, broadly diversified index fund is, in effect, sidestepping most of this debate entirely, since a genuinely diversified holding earns its return from bearing the market's aggregate systematic risk however that risk is ultimately priced, whether by plain CAPM, the zero-beta version, or a full multifactor model. The extensions matter most for active decisions: choosing between a value-tilted fund and a market-cap-weighted fund, or judging whether a small, illiquid holding's higher expected return is genuine compensation for risk or simply an artifact of an incomplete pricing model.
Actionable breakdown
- Treat plain CAPM as a floor, not a ceiling, on rigor.
- Use it for quick, first-pass estimates of required return.
- Do not lean on it alone for precise valuation decisions.
- Watch for the assumptions most likely to be violated in your situation.
- Your own borrowing rate is above the risk-free rate; adjust expectations accordingly.
- Illiquid holdings likely deserve a return premium beyond their beta.
- Use multifactor thinking for anything beyond a broad index fund.
- Size, value, and momentum tilts are not explained by beta alone.
- Let a broad index fund do the heavy lifting.
- Diversification sidesteps much of this theoretical debate in practice.
Common pitfalls
A frequent error is applying a single textbook risk-free rate to every investor's situation, when in practice individual borrowing costs, credit card rates or margin loan rates, for instance, are often several points above the government bill rate the formula assumes, which the zero-beta model exists specifically to address. Another pitfall is assuming an extension automatically produces a better real-world forecast simply because it relaxes a more realistic assumption; several extensions improve theoretical elegance without meaningfully improving out-of-sample predictive accuracy. A third pitfall is treating the consumption-based model's elegance as a reason to prefer it practically, when its data and estimation challenges make it far harder to apply reliably than the simpler, if flawed, standard model.
The bottom line
CAPM's formula is simple because its assumptions are strong, and every major extension of the model exists to relax one specific assumption, at the cost of added complexity, in pursuit of a better match to how real markets actually price risk.
Related reading: factor investing, understanding investment risk, the core CAPM formula and mechanism, how academics have tested and challenged CAPM, multifactor models, an overview.