What beta is not

Chapter 7 derived beta. This chapter is about the six things it gets taken for, and then about what happens when its central prediction meets data.

It is not volatility

The most common error, and Sharpe corrects it himself: the security market line relates expected returns to market risk "but not, as often believed, to total risk."

Use the decomposition from chapter 7:

$\beta_i = \rho_{iM} \times \frac{\sigma_i}{\sigma_M}$

A low beta can mean two completely different things. Low volatility, or low correlation. A wildly volatile asset that moves independently of the market has a low beta; a placid asset that tracks the market exactly has a beta near 1.

The Indian data shows both components moving separately. Comparing two factor indices against the Nifty 50 over the same five years:

	Standard deviation	Correlation with Nifty	Beta
Nifty 50	13.82%	1.00	1.00
Nifty200 Momentum 30	19.44%	0.84	1.18
Nifty200 Quality 30	13.54%	0.84	0.82

The two factor indices have the same correlation with the market and different betas, because their volatilities differ. Momentum is 41% more volatile than the Nifty but only 18% higher in beta — the lower correlation absorbs the rest.

It is not the risk of losing money

Nothing in chapter 7's derivation mentions loss. Beta is a covariance.

A beta of 0.5 does not mean you can lose half as much. It means that on average, historically, the asset moved about half as far as the market did. A company can have a beta of 0.5 and go to zero on its own account, and the Risk subject's chapter on concentration is full of them.

Low beta is not safety, and the practical consequence is that a "low-beta, defensive" portfolio can be a portfolio of fragile businesses that happen not to co-move.

It is not the whole of a security's risk

Beta tells you about the shared part. What is left over is specific risk — which chapter 3 said diversification removes.

$R^2$ tells you how much is shared. It is the squared correlation:

	Correlation	$R^2$	Share of variance not explained by the market
Nifty200 Momentum 30	0.84	0.706	29%
Nifty200 Quality 30	0.84	0.706	29%

A beta quoted without an $R^2$ is half a number. For a thirty-stock index, 29% of the variance is unexplained by the market. For a single company it is typically far higher — most of a single stock's movement has nothing to do with the index, which is exactly why the beta of one share is such a weak description of it.

It is not stable

Beta is an estimate from a regression, and it depends on everything the regression depends on.

It depends on the window. The Quality index's beta against the Nifty is 0.82 over five years and 0.79 since inception. Its correlation is 0.84 over five years and 0.91 since inception. Same pair, same method, different numbers.

It depends on the frequency. Daily, weekly and monthly returns produce different betas for the same asset over the same period.

It depends on the index chosen. A beta against the Nifty 50 is a beta against fifty large companies. The Corporate finance subject makes this point for a valuation input — it is "a beta against fifty large companies, not against the Indian economy."

And it carries a standard error. The Quantitative methods subject's chapter on simple regression is where that comes from, and the practical upshot is that a beta of 1.08 and a beta of 0.94 are usually not distinguishable. Printing betas to two decimals implies a precision the estimate does not have.

It is not a property of the company

A company does not have a beta. A company's shares have a beta with respect to a chosen index over a chosen window at a chosen frequency. Change any of the three and the number changes, without anything happening to the company.

This is why the Corporate finance subject's method for an unlisted business — take listed comparables, un-lever, average, re-lever — works at all. Beta is being treated as a reusable property of a business's economic exposure, which it approximately is, with a great deal of noise around it.

It is not forward-looking

Every beta you can compute is a beta of the past. A company that has sold a division, changed its leverage, entered a new market or been through a regulatory change is not the company whose returns were regressed.

And leverage changes beta mechanically. A company that doubles its debt raises its equity beta without any change to its business, which is the relation the Corporate finance subject's un-levering is built on.

Now the prediction, tested

Everything above is about reading beta correctly. This is about whether beta does what CAPM says.

CAPM's prediction is sharp: expected return should be a straight line in beta, with the market's excess return as the slope. So a portfolio with a beta below 1 should earn less than the market, and one above 1 should earn more.

Over the five years to 30 September 2026:

	Beta	CAPM-implied return	Actual return	Difference
Nifty 50	1.00	6.38%	6.38%	—
Nifty200 Momentum 30	1.18	6.54%	9.28%	+2.74
Nifty200 Quality 30	0.82	6.22%	6.40%	+0.18

(CAPM-implied return is $r_f + \beta(R_M - r_f)$ with $r_f = 5.5\%$, the current policy repo rate, and $R_M$ the Nifty 50's realised 6.38%.)

The Quality index is the interesting row. It carried 18% less market risk than the Nifty and earned the same return — 6.40% against 6.38%. CAPM says it should have earned 6.22%, which is less than the market, because it took less of the only risk that is paid for.

It did not earn less. It earned fractionally more, with lower volatility too — 13.54% against 13.82%.

This is the shape of the most durable empirical complaint against CAPM. When returns are plotted against beta, the fitted line comes out flatter than the model requires: low-beta portfolios do better than predicted and high-beta portfolios worse. The two rows above are a single small instance of that pattern, in Indian data, over one window.

And the Momentum row is a different problem. Its 2.74 points of return above the CAPM prediction is not explained by beta at all, which is chapter 10's subject — the characteristics that appear to be priced and that CAPM says should not be.

Working the problem

Quality 30: beta 0.82, actual return 6.40%. Nifty 50: 6.38%.

Step 1 — the prediction.

$E(R) = 5.5 + 0.82 \times (6.38 - 5.5) = 5.5 + 0.82 \times 0.88 = 6.22\%$

Step 2 — the gap. Actual 6.40% against 6.22% predicted, so +0.18 percentage points, in the direction the low-beta anomaly predicts.

Step 3 — what this does establish. Within this window, with these indices, holding less market risk did not cost return. That is a genuine observation and it is the right direction to be suspicious in — it matches a pattern found repeatedly in other markets over much longer periods.

Step 4 — what it does not establish, and the list is longer.

Five years is far too short. Expected returns are estimated with enormous error. A gap of 0.18 points over five years, against annual volatility of 13.5%, is statistically indistinguishable from zero — the standard error on a five-year mean return at that volatility is roughly $13.5/\sqrt{5} = 6$ percentage points. The gap is a thirtieth of its own standard error. It is not evidence of anything on its own.

The market proxy is a choice. Both betas are measured against the Nifty 50, which is itself 37.45% financial services. CAPM's market portfolio is all risky assets — every share, bond, property and private business in existence. The test is of CAPM-plus-a-proxy, and a failure could be the proxy's.

The window is unusual. Over these five years the Nifty returned 6.38% against a 5.50% policy rate, an equity risk premium of 0.88 points. When the market's excess return is nearly zero, the security market line is nearly flat by construction and beta barely separates anything. This is a poor window to test a theory whose slope is the market premium.

And the index is selected. The Quality 30 was constructed on quality scores, not chosen at random among low-beta portfolios. Chapter 10 is about the selection and survivorship problems that make factor evidence hard to read.

The honest conclusion. The two rows are consistent with the low-beta anomaly and prove nothing by themselves. The reason to take the anomaly seriously is not this table but its persistence across decades and markets — and the reason to show this table is that it is current, Indian, published by the exchange, and lets you run the test yourself rather than take the finding on authority.

The point

Beta is correlation times the ratio of volatilities, so a low beta can mean low volatility or low correlation and the two are different situations. It is not total risk, not the chance of losing money, not meaningful without an $R^2$, not stable across windows, frequencies or index choices, not a property of a company, and not forward-looking. Its central prediction is that return is linear in beta — and in the five years to September 2026 the Nifty200 Quality 30 carried a beta of 0.82 and returned 6.40% against the Nifty 50's 6.38%, where CAPM predicted 6.22%. One window proves nothing, and the direction matches a flat security market line found repeatedly elsewhere.