What Is Beta? The Slope That Measures Market Exposure

A slope, not a mystery

Beta is the slope of a best-fit line. Plot the market's return on the horizontal axis and a stock's return on the vertical axis, one dot per week or month, run a regression through the cloud of dots, and the tilt of that line is beta. Slope 1.0: the stock has historically moved one-for-one with the market. Slope 2.0: two percent, on average, for the market's one. Slope 0.5: half. The formula behind the slope is the covariance of the stock with the market divided by the variance of the market, but the picture is the whole idea.

A toy dataset makes it exact. Four months of returns: the market prints +2%, -3%, +1%, -4%, and the stock prints +4%, -6%, +2%, -8%. Every stock return is exactly double the market's, so the regression slope is exactly 2.0 and the fit is perfect. Real stocks never fit like that. The dots scatter around the line, beta is only the line's tilt, and the scatter is measured separately — by R-squared — and matters just as much as the slope.

Beta is always relative to a benchmark, almost always the S&P 500 for US stocks. A stock with a beta of 1.3 against the S&P can show a different beta against the Nasdaq 100 or the Russell 2000. The number carries no meaning without knowing what it was regressed against, over what period, at what frequency, which is why the same ticker shows different betas on different sites.

What 0.5, 1.0, and 2.0 feel like in a real drawdown

Apply the slope to a real-sized decline. The S&P 500 fell about 25% peak to trough in 2022. Market exposure alone implies a beta 0.5 name down about 12.5%, a beta 1.0 name down 25%, and a beta 2.0 name down 50%. The recoveries required are then 14.3%, 33.3%, and 100% respectively, so drawdown asymmetry doubles as the punishment schedule for high beta.

The word implies is doing real work in that sentence. Beta describes the market-driven component of a move; the company's own news adds or subtracts on top. In 2022 plenty of beta 1.0 stocks fell 60% because their earnings collapsed, and some high-beta names beat the model because their businesses improved. Beta sets the baseline exposure, not the outcome. The practical mistake is sizing a beta 2.0 holding like an index position: at the same dollar size it carries roughly twice the market risk, so the beta-consistent adjustment is half the dollars for the same market exposure.

High beta and volatile are not synonyms

Beta can be rewritten as correlation with the market multiplied by the ratio of the stock's volatility to the market's. That decomposition separates two things people habitually merge: how much a stock moves, and how much of that movement lines up with the market. A stock needs both high volatility and high correlation to earn a high beta.

Work the numbers on two very different stocks, with market volatility at 16% annualized. A clinical-stage biotech swings 80% a year but trades on trial results, with a market correlation of just 0.2: beta = 0.2 × (80 / 16) = 0.2 × 5 = 1.0. A large bank swings 24.5% a year with a 0.85 correlation: beta = 0.85 × (24.5 / 16) = 0.85 × 1.53 = 1.3. The biotech is more than three times as volatile yet carries the lower beta. Its risk is real; it just lives almost entirely outside the market's influence.

R-squared is the tell. It is the squared correlation: the fraction of a stock's variance that the market explains. The biotech's R-squared is 0.2 squared = 0.04, meaning the market accounts for 4% of what the stock does, and the beta of 1.0 sits on top of 96% noise. The bank's is 0.85 squared = 0.72. Two similar-looking betas deserving completely different levels of trust, and the difference is printed in a regression statistic most quote pages never show.

Why every site shows a different beta

Beta is an estimate with settings, and vendors choose different settings. Yahoo Finance uses five years of monthly returns; many brokers use two or three years of weekly returns; some quant screens use one year of daily. Each choice is defensible and each produces a different number for the same stock — a company that transformed itself three years ago still drags its old self through a five-year window.

Return frequency matters on its own. Daily betas pick up short-term noise and run low for thinly traded stocks whose prices react to the market a day late (the stale-price effect). Monthly betas smooth that out but leave only 60 observations in five years, which makes the estimate statistically loose. There is no correct setting, only a disclosed one: a beta quoted without its window and frequency is a number with no units attached.

Where beta breaks

Beta assumes the past relationship persists, and companies change. New leverage, a divested division, a business-model pivot: each rewrites the true sensitivity while the trailing regression keeps reporting the old one. Bank betas estimated through 2006 said little about 2008. Any beta computed across a structural change in the company is an average of two different animals.

Regimes break it from the market side. Correlations across stocks rise sharply in crashes — nearly everything falls together in a panic — so effective betas converge upward exactly when the low-beta label is being leaned on. A defensive name with a calm-market beta of 0.6 can trade like a 0.9 during a liquidation. Beta measured over a quiet period systematically understates panic exposure.

For low R-squared names, beta is close to meaningless. When the market explains 4% of a stock's variance, the regression slope is an artifact riding on noise; hedging that stock with index futures, or sizing it by beta, manages the 4% and ignores the 96%. Check R-squared before trusting any beta, and treat every beta as a rear-view estimate rather than a forward measurement.