Limitations of CAPM

CAPM is simultaneously the most-used and most-rejected model in finance. You need both halves, stated fairly. An analyst who only knows the formula is naive, and one who only knows the criticisms has no discount rate.

Start with what CAPM gets right. These points survive every attack:

1. The diversification logic. Only non-diversifiable risk commands a premium. This is the deepest idea in asset pricing and no empirical failure touches it.
2. Direction. Diversified low-beta portfolios are less risky than high-beta ones, and equities have out-earned government bonds over long horizons in most markets. The signs are right.
3. Transparency. One equation, three auditable inputs. Everyone can see your Rf, your beta, your ERP, and argue with each. Against "I feel 15% is fair," that's civilisation.

Where does CAPM fail empirically?

Now where it fails empirically, with the record presented straight:

1. The line is too flat. From the earliest large tests (Black, Jensen & Scholes, 1972) to today, low-beta stocks earn more than CAPM predicts and high-beta stocks less. The relationship between beta and average return is far shallower than the SML. This anomaly is strong enough to trade. Frazzini & Pedersen's "Betting Against Beta" (2014) documents returns to buying low-beta and shorting high-beta assets across markets and decades.
2. Beta loses to size and value. Fama & French (1992) showed that once you control for company size and price-to-book, beta adds almost nothing to explaining the cross-section of average US returns (1963–1990). Their 3-factor model (market, small-minus-big, value-minus-growth), later the 5-factor version with profitability and investment added, plus momentum (Jegadeesh & Titman, 1993), describes realised returns far better. Fama and French themselves put it bluntly in their 2004 review: the CAPM's empirical record is "poor enough to invalidate the way it is used in applications."
3. Beta is unstable. Estimate it over 2 years of weekly returns or 5 years of monthly, against NIFTY 50 or NIFTY 500, and you get materially different numbers for the same company. It also drifts as the business and its leverage change. An input this wobbly deserves humility, plus the bottom-up fix you'll meet in the WACC build.
4. Roll's critique (1977). The theory's "market portfolio" is all risky assets: every stock, bond, house, farm, private business, and human career on earth. It's unobservable. Every test uses an index proxy, so nobody has ever tested CAPM itself, only proxies of it. Being unfalsifiable in the strict sense is a serious charge against a scientific model.
5. The equity premium puzzle. Mehra & Prescott (1985) showed that the historical US premium, about 6% arithmetic, is far larger than standard risk-aversion theory can justify. Investors demanded more for equity risk than the model's own economics explains. So the theory doesn't even fit the premium it's built to price.

Why do analysts still use CAPM?

So why do practitioners use it anyway? The defence deserves a fair hearing. Graham & Harvey's survey work (2001) found roughly three-quarters of CFOs use CAPM for the cost of capital, and later surveys of valuation practice say it remains the default. McKinsey's Valuation and Damodaran both keep it. There are four good reasons. (1) The job is different. Fama-French shows beta poorly predicts realised returns, but an appraiser needs an ex-ante required return, a consistent, defensible price of risk to discount with. A return-forecasting machine is a different tool. (2) The alternatives are noisy too. Factor premia are unstable out of sample and the loadings jump around. Swapping one wobbly input for four isn't obviously progress in a valuation. (3) Materiality. In most DCFs your cash-flow and moat judgments move value far more than a half-point of discount-rate refinement, so spend scarce effort where the value is. (4) Fixable inputs. Most of CAPM's practical failure is bad inputs: regression betas, stale premiums. A bottom-up beta and the implied ERP you're about to learn repair those two directly.

Why do Buffett and Munger reject beta as a definition of risk?

Let the dissenters have the last word, because they've earned it. Buffett and Munger reject beta as a definition of risk outright. To a business owner, risk is the probability of permanent loss of purchasing power, and a stock that has fallen 50% below a stable intrinsic value is less risky at the lower price, while its measured beta says nothing of the sort. Both things are true at once. Beta prices co-movement for diversified portfolios. It doesn't measure whether you will lose money in a business. We use CAPM as the course's pricing convention, the grammar in which cost of capital is spoken, stated per the market's logic, and Phases 6 and 9 train the owner's definition of risk alongside it.

Carry both without confusing them. One is a price, and the other is a danger.