Beta in Investing Is a Slope, Not a Risk Score

LSEG, the company that owns Refinitiv and the London Stock Exchange, publishes five different betas for the same security. Ninety-day and 180-day betas off daily closing prices. Two-year and three-year betas off weekly prices, built from 104 and 156 data points. A five-year beta off 60 monthly points.^[1] All five are correct. All five are different numbers. Nothing about the company changed between them.

That should tell you something about what beta is. It is not a property of a business the way a debt load or a gross margin is. It is the output of a regression, and if you change the regression you change the answer. I keep running into people who treat the beta on a stock quote page as a risk rating, somewhere between a credit score and a speed limit. It isn't one. It is a slope.

Here's what makes this worth 2,000 words instead of a tweet: once you understand what the slope actually measures, you find out that a stock with a beta of 0.5 can be more volatile than a stock with a beta of 1.8. Not in a contrived edge case. As a matter of arithmetic that follows directly from the definition.

Beta is a slope, and that is the entire definition

Take a stock's returns for some past window. Take a market index's returns over the exact same periods. Subtract the risk-free rate from both so you are working with excess returns. Plot the pairs on a scatter chart with the index on the horizontal axis and the stock on the vertical, then fit the line that minimizes the squared distance to the points. Beta is the slope of that line. In closed form, it is the covariance between the two return series divided by the variance of the index's returns.^[2]

Because the index is being regressed against itself, the index's own beta is exactly 1.00 by construction. Everything else is read relative to that. Morningstar states the interpretation plainly in its methodology paper: a beta of 1.10 “has tended to have an excess return that is 10% higher than that of the index in up markets and 10% lower in down markets,” and a beta of 0.85 indicates performance “15% worse than the index in up markets and 15% better in down markets.”^[2] Push it further and a beta of 1.8 says: for a 10% move in the index, expect roughly 18% out of the part of this stock the index drives.

Read that last clause again, because everything in this article hangs on it. The part of this stock the index drives. Beta says nothing whatsoever about the rest.

R-squared is the number that decides whether the beta means anything

The same regression that spits out beta spits out R-squared, and almost nobody looks at it. R-squared is the share of the stock's movement that the index explains. Morningstar's methodology paper is blunt about it: “An R-squared measure of 35%, for example, means that only 35% of the fund's movements can be explained by movements in the index,” and then, in the sentence I wish every brokerage app printed under its beta field, “R-squared can be used to ascertain the significance of a particular beta estimate. Generally, a high R-squared will indicate a more reliable beta figure.”^[2]

An index fund tracking the S&P 500 has an R-squared near 100%, so its beta is a genuinely informative number. A biotech with one drug in phase three trials might have an R-squared of 5%, which means 95% of what happens to that stock has nothing to do with the market at all. Its beta is still printed to two decimal places. It is still nearly meaningless.

The arithmetic nobody shows you: low beta can mean high volatility

Here is the part that changed how I read a stock screener. Beta and R-squared are not independent. Both come from the same two ingredients, so you can rearrange them into a single expression for how volatile a stock actually is:

Total volatility = (beta / square root of R-squared) x the market's volatility.

The market's volatility appears on both sides of any comparison, so it cancels. What's left is beta divided by the square root of R-squared, and that ratio is the honest ranking of how much a stock moves. Run two stocks through it.

This is not a hypothetical the industry is unaware of. Morningstar spells it out in the same document: “A low beta does not imply that the fund has a low level of volatility; rather, a low beta means only that the fund's index-related risk is low.” Their example is a gold fund, which “will usually have a low beta (and a low R-squared)” and yet “might fluctuate wildly because of rapid changes in gold prices.”^[2] That warning has been sitting in a public methodology PDF since 2015 and it still does not make it onto the quote pages people actually read.

If what you want is a measure of how much a holding moves, you want standard deviation, and it is worth being precise about what volatility actually means as a statistic before you start ranking positions by it.

Two providers, two betas, one company

Three choices go into a beta and every provider makes them differently: the lookback window, the return interval, and the index. Morningstar runs a trailing 36 months of monthly returns against the S&P 500 for US stocks.^[2] LSEG offers the five windows I opened with, with the benchmark set by the security's primary exchange.^[1] Bloomberg's default is a two-year window on weekly prices, and Bloomberg and FactSet both regress weekly rather than monthly.^[3] Capital IQ's standard tearsheet figure is a five-year beta.^[4]

So when someone tells you a stock's beta is 1.3, the only correct response is: measured how, against what, over which window? Babson's finance library guide, which exists precisely because students kept getting different answers from different terminals, concludes that “there is no right answer” and that you have to know each source's methodology to pick one.^[3]

Beta is silent about the risk that actually kills a position

The Capital Asset Pricing Model splits a security's excess return into two pieces: the systematic part, proportional to the market, and the unsystematic or idiosyncratic part. Beta measures the first. And the model's central claim, as Morningstar summarizes it, is that “there are no rewards for taking on unsystematic risk,” because a diversified investor can make it disappear.^[2]

That is a fine assumption if you hold the whole market. It is a terrible assumption if you hold eleven stocks. The theory says idiosyncratic risk is unpriced because it is diversifiable, which is a statement about a portfolio, not a statement about your portfolio. If you are concentrated, that risk is not diversified away, it is simply uncompensated, and it is the exact category that produces the outcomes people actually lose money on. Accounting fraud is idiosyncratic. A phase three failure is idiosyncratic. A recall is idiosyncratic. None of them are in beta. It is worth reading what diversification does and does not remove alongside this, because the two ideas are the same idea viewed from opposite ends, and the choice between an index fund and single names is largely a choice about how much of this uncompensated risk you are volunteering for.

There is a second silence worth naming. Beta is symmetric. A regression does not know that the upside surprises were the good ones, so a stock that ripped 22% on an earnings beat contributes to a high beta exactly the way a 22% collapse would. The vendors know this is unsatisfying, which is why LSEG publishes a Beta Up and a Beta Down alongside the normal beta, computed only on days when the benchmark's return was positive or negative respectively.^[1] When a data provider ships three versions of a statistic, it is quietly telling you one was never enough.

Where beta is legitimately useful, and where the professionals moved on

None of this means beta is junk. It has one real job and it does it adequately: estimating a cost of equity. Plug a beta into the CAPM, get a required return, discount your cash flows with it. Every discounted cash flow model built in a bank does this, and if you have ever wondered why a valuation swings so much on inputs nobody can observe, the beta is one of the suspects. That is also why beta belongs in the same mental drawer as the multiple you pay: see how a price-to-earnings ratio encodes an expectation rather than a fact.

The academic verdict on the model beta sits inside is not gentle. Eugene Fama and Kenneth French, reviewing forty years of CAPM tests in the Journal of Economic Perspectives in 2004, describe its empirical record as “poor enough to invalidate the way it is used in applications.”^[5] In a later paper on French's own site, working with US data back to 1926, they put the problem in one line. Variation in beta that is unrelated to firm size and value-growth goes unrewarded across the whole 1926 to 2004 period.^[6] Beta as a standalone explanation of returns did not survive contact with the data.

What replaced it is not a better beta but more factors. The Fama and French three-factor model added size (small minus big) and value (high minus low book-to-market) to the market factor; the 2015 five-factor version added profitability (robust minus weak) and investment (conservative minus aggressive). French publishes the whole series monthly from July 1963 forward, free, on the Dartmouth data library.^[7] Market beta is still in there. It is just one of five, and the other four exist because it wasn't enough.

What to actually do with the number

Three rules, and they take about ten seconds each.

1. Never read a beta without its R-squared. If the R-squared is low, the beta is describing a relationship that barely exists. Delete the number from your thinking rather than rounding it.
2. Ask what window produced it. Two-year weekly and five-year monthly are different measurements of different things, and the gap between them is not noise, it is methodology.
3. Use beta for portfolio construction, not stock selection. Knowing that a sleeve of holdings carries an aggregate beta of 1.3 is genuinely useful for sizing your exposure to a drawdown. Knowing that one company has a beta of 1.3 tells you almost nothing about that company, which is what the other ways of evaluating a stock for the long term are for.

The thing I find genuinely annoying is that none of this is hidden. It is in Morningstar's own methodology PDF, in LSEG's own developer docs, in a Nobel laureate's own summary of his life's work. The caveat has always shipped with the number. It just never made it onto the page where the number gets displayed, and a single decimal on a quote screen is a very convincing thing.

Next time you look one up, find the R-squared first. If the site doesn't publish one, that tells you what the site thinks you are doing with the beta.