In finance, the beta coefficient measures an asset’s sensitivity to movements in the market as a whole, estimated as the slope of the asset’s returns regressed on the returns of a market index. In statistics, a beta coefficient most often means a standardized regression coefficient, also called a beta weight: the expected change in the dependent variable in standard deviation units for a one standard deviation change in a predictor, holding the other predictors constant. They are different quantities sharing a name and a Greek letter, and only the field tells you which one is meant.
How a Finance Beta Is Estimated
The estimator is a ratio built from two return series, the asset’s and the index’s.
B = Cov(Ra, Rm) / Var(Rm)
B the asset's beta, the output, the slope of its returns on the index's returns
Ra the asset's returns over the estimation window
Rm the returns on the market index standing in for the market
Cov(Ra, Rm) the covariance of the two series, the degree to which they move together
Var(Rm) the variance of the index's returns, the size of the market's own movementThat ratio is the slope of a regression of the asset’s returns on the index’s returns, which is how a Federal Reserve Bank of New York study states it: the equity beta is calculated as the slope coefficient in the regression of the equity return on the market return (Green, Lopez and Wang, FRBNY Economic Policy Review, 2003).
The arithmetic on stated inputs:
Cov(Ra, Rm) 0.00260 stated input
Var(Rm) 0.00200 stated input
B = 0.00260 / 0.00200 = 1.30Three choices are made before that arithmetic: the index that stands in for the market, the interval the returns are measured over, and the length of the window. Regression betas are clearly affected by the time period used, the return interval and the index, and the beta will vary widely depending on how the regression is set up and run (Damodaran, Estimating Risk Parameters, Stern School of Business, New York University).
Reading the Number
The scale belongs to the model doing the reading, not to the asset. Inside a single-factor pricing model the number is read against 1: a beta of 1 tracks the index, between 0 and 1 dampens it, above 1 exaggerates it, 0 leaves the asset priced at the risk-free rate, and a negative beta moves the opposite way. Change the index and the same asset can cross those boundaries, because the boundary is drawn at the index’s own movement rather than at anything fixed.
Models carrying more than one risk term replace the single market beta with a loading on each factor, a separate slope measuring sensitivity to that factor alone, which is where arbitrage pricing theory starts.
Levered and Unlevered Beta
A beta regressed on a company’s own share price carries the risk of the business and the risk its borrowing adds on top. The valuation profession splits the two: a levered beta reflects a capital structure that includes debt, also called the equity beta, and an unlevered beta reflects one without debt, also called the asset beta (International Valuation Glossary, Business Valuation, 2022).
BL = Bu x (1 + (1 - t) x (D/E))
BL the levered beta, the beta of the company's equity
Bu the unlevered beta, the beta the company would have carrying no debt
t the corporate tax rate
D/E the ratio of debt to equityThat form assumes the debt itself bears no market risk, which overstates the true beta at higher debt ratios, and Damodaran gives an expanded version carrying a beta for the debt (Estimating Risk Parameters).
A private company has no traded return series to regress, so its beta is built from guideline public companies, meaning traded firms close enough in business and size to stand in for the subject. Unlever each comparable’s beta to strip out its own borrowing, average the unlevered betas, then relever the average at the subject’s capital structure. Unlevering with the average debt to equity ratio over the regression period and relevering with the current one keeps a recent change in leverage from being priced off an old balance sheet.
Where a Firm Meets a Beta Estimate
A beta estimate reaches a firm as somebody else’s input rather than as a finished answer. In a valuation engagement it feeds the cost of equity, one leg of the weighted average blend set out in cost of capital, and that rate then does its work inside the income approach, one of the three set out in valuation. The rate it feeds is worked through in CAPM equation.
In an issuer audit the beta arrives as one of the assumptions sitting behind such a rate. AS 2501 establishes requirements for auditing accounting estimates including fair value measurements, and it lets the auditor test an estimate by any of three approaches, individually or in combination: testing the company’s process, developing an independent expectation, or evaluating evidence from events after the measurement date (paragraph .07, PCAOB, AS 2501).
The assumption work sits inside the first of those. An auditor testing the company’s own process has to identify which of the assumptions used are significant to the estimate and evaluate whether those are reasonable, individually and in combination, while an auditor building an independent expectation picks that reasonableness duty back up only where the company’s own significant assumptions are reused inside it (paragraphs .15, .16 and .24, PCAOB, AS 2501).
What a Beta Is Not
Beta is not total volatility. Standard deviation measures how far an asset moves, while beta measures only how steeply it moves per unit of index movement, so beta says nothing about size on its own. A low-beta asset measured against a violently moving index can move further in absolute terms than a high-beta asset measured against a calm one. It is not R-squared, which the same regression also reports: R-squared says how much of the asset’s variation the index explains, beta says how far the asset is expected to move when the index moves. And it is not a return forecast: it describes how an asset has moved alongside an index over a chosen window, not what it will pay.
The statistics sense is not always the standardized one either. Some sources use beta for an unstandardized coefficient, which is expressed in the variables’ own units and changes when those units change, whereas a standardized beta is unitless because standardizing has already removed the units.
Common Questions
Two things, and the field decides which: in finance a property of an asset’s returns, their sensitivity to an index, and in statistics a property of a fitted model, a predictor’s effect expressed in standard deviations.
Beta is the slope of the asset’s returns on the index’s returns, so the number is a multiple: the asset is estimated to move that many times as far as the index, up and down, over the window the regression covered.
Index return 10.0% stated input
Beta 1.3 stated input
Expected asset return = 1.3 x 10.0% = 13.0%That is an estimate rather than a fact about the company: change the index, the interval or the window and the same asset can produce a different number.
No, and not worse either. A higher beta is more market sensitivity, which raises the return the model requires and widens the swing both ways. Whether that is wanted depends on the portfolio it sits in and the risk its holder will carry.
The prediction for the dependent variable moves by that many standard deviations when the predictor moves by one standard deviation and the others are held constant. Because the standardization uses the sample’s own standard deviations, a beta weight describes the spread inside that sample and does not carry cleanly to a sample with a different spread.
Accountably places trained offshore accountants and tax preparers inside US CPA and EA firms, and a firm can grade real work from its own files through a trial task before anything depends on it.
