7.4Lesson 7.4 · Module 7 — The CAPM and APT

Estimating Beta: From Return Data to Market Exposure

CAPM can compute a required return once beta is known — but beta is not directly observable. It must be estimated from return data, and the estimate is uncertain.

β̂ = slope of the stock-vs-market regression
Excess returns on both axes
Beta ≠ R²: slope vs fit
Residuals are the vertical gaps
Beta is an estimate, with standard error

Start

Central question

CAPM needs beta, but where does a company's beta actually come from?

7.4.1Section 1 · Beta is not directly observable

A required input that nobody can see

From Lesson 7.3, the CAPM required return is:

E [R_{i}] = R_{f} + β_{i} (E [R_{M}] - R_{f})

The question

We can observe market prices and historical returns, but where do we obtain beta?

Required statement

Beta is not printed on a security and is not known with certainty. It is estimated from how the security moved relative to a selected market benchmark.

Definition · The hat notation

An estimate is written with a hat:

\hat{β}

. The hat means “estimated from data.” Throughout this lesson,

\hat{β}

is an estimate of the unknown true beta

β

7.4.2Section 2 · Pair stock returns with market returns

One point per period

To estimate beta, pair each period's market excess return with the stock's excess return over the same period.

Month	Market excess return	Stock excess return
1	+3%	+5%
2	-2%	-4%
3	+1%	+0%
4	+4%	+7%
5	-3%	-2%

(x, y) = (R_{M} - R_{f}, R_{i} - R_{f})

Five months of data, one point each. No fitted line yet.

Definition · What each point asks

Each point asks: when the market produced this excess return, what excess return did the stock produce during the same period?

7.4.3Section 3 · Fit the market relationship

A line through the cloud

The market model expresses each period's stock excess return as a market-linked component, an average intercept, and a period-specific residual.

Market-model regression

R_{i, t} - R_{f, t} = α_{i} + β_{i} (R_{M, t} - R_{f, t}) + ε_{i, t}

The stock's excess return in each period is decomposed into a market-linked piece, an average intercept, and a residual.

$β_{i} (R_{M, t} - R_{f, t})$: Market-linked component: how the stock tends to respond to market movements.
$α_{i}$: Average fitted return not explained by market exposure (the intercept).
$ε_{i, t}$: Period-specific difference between actual return and the fitted line (the residual).

Full performance interpretation of alpha belongs in Lesson 7.5. Here it is only the fitted intercept.

Note

Alpha appears here as a regression intercept. It is not yet treated as proof of skill — that interpretation is deferred to Lesson 7.5.

7.4.4Section 4 · Beta is the slope

The slope of the fitted line

The central result of beta estimation: the fitted slope is the beta estimate.

\hat{β}_{i} = slope of the stock-versus-market regression line

A fitted example

R_{i} - R_{f} = 0.10% + 1.25 (R_{M} - R_{f}) + ε_{i}

$\hat{β}$ = 1.25

Over the sample period, when the market excess return changed by 1 percentage point, the stock's excess return tended to change by approximately 1.25 percentage points in the same direction, on average.

Three words that matter

tended toapproximatelyon average

Beta is not an exact next-period prediction. The stock will not always move exactly 1.25 times the market.

7.4.5Interaction · Build the regression line

Classify, estimate, then reveal

Read the scatter, pick a slope, then reveal the fitted line and its residuals.

Each point pairs one period's market excess return with the stock's excess return.

Step 1 · Classify the cloud

Is the relationship positive or negative?

Is the sensitivity low or high?

Is the fit tight or loose?

7.4.6Section 5 · Slope and scatter are different

Beta asks what; R² asks how well

Beta and R² answer two different questions about the same regression.

Beta asks

How strongly does the asset tend to respond to the market?

R² asks

How much of the asset's historical variation did the market regression explain?

R^{2} = 32%

Approximately 32% of the stock's return variation in the sample was associated with the fitted market relationship.

The remaining 68% is variation not explained by this particular regression. It is not necessarily pure firm-specific risk — it may include omitted factors and noise.

Same beta, different R²

Asset	Beta	R²
A	1.2	70%
B	1.2	18%

Same estimated sensitivity; different tightness around the line.

Different beta, same R²

Asset	Beta	R²
C	0.6	45%
D	1.5	45%

Same proportion explained; different response magnitude.

Required conclusion

Beta measures slope. R² measures fit.

7.4.7Section 6 · Residuals explain period-by-period deviations

Why points do not sit on the line

Beta describes the fitted market component; residuals explain the gap between each actual point and the line.

$\hat{β} = 1.25$
Market excess return: $4%$
Predicted market-linked stock return: $1.25 \times 4% = 5%$
Actual stock excess return: $2%$
Approximate residual: $ε = 2% - 5% = - 3%$

Beta describes the fitted market component.
Residuals explain why actual points do not lie exactly on the line.
Company events and omitted influences may create residuals.

Required statement

Beta is an average relationship, not a mechanical return multiplier.

7.4.8Section 7 · Connect regression to the covariance formula

The same relationship, two forms

The regression slope and the covariance-over-variance formula describe the same stock-versus-market relationship.

Beta from covariance and variance

\hat{β}_{i} = \frac{Cov ( R _{i} , R _{M} )}{Var ( R _{M} )}

\hat{β} = \frac{0.024}{0.020}

= 1.20

The covariance formula and the regression slope describe the same simple stock-versus-market relationship. You are not asked to compute covariance from a long raw dataset here.

7.4.9Section 8 · Beta is estimated with uncertainty

An estimate, not a fact

Because beta is computed from a sample of returns, the slope estimate comes with a standard error.

Estimate

\hat{β} = 1.25

Standard error

S E (\hat{β}) = 0.18

Definition · Standard error, intuitively

The standard error describes how imprecisely the slope has been estimated from the sample.

A rough uncertainty range

\hat{β} \pm 2 S E = 1.25 \pm 0.36

\approx 0.89 to 1.61

An approximate uncertainty range — not a guarantee or a permanent interval.

Required conclusion

The data do not prove that the asset's beta is exactly 1.25.

7.4.10Interaction · Slope versus scatter

Separate sensitivity from noise

Two controls: market sensitivity (steepens the line, raises beta) and company-specific noise (spreads points, lowers R²). Build the three target combinations.

Estimated β

1.25

R²

73%

Market sensitivity

True β = 1.2

Company-specific noise

σ(ε) = 0.07

Try these targets

Raising market sensitivity

steepens the fitted line;
increases the estimated beta;
does not by itself improve the fit.

Raising company-specific noise

spreads points farther from the line;
generally lowers R²;
does not automatically change beta.

Note

$β$ measures slope. $R^{2}$ measures fit. Two assets can share the same beta with very different $R^{2}$ , or share the same $R^{2}$ with very different betas.

7.4.11Section 9 · Why estimated beta changes

Five sources of instability

The same stock can produce different beta estimates from different reasonable estimation choices.

1Time window

A two-year estimate may differ from a ten-year estimate.

2Return frequency

Daily, weekly, and monthly observations can produce different estimates.

3Market proxy

Beta relative to a domestic index may differ from beta relative to a global index.

4Unusual periods

Crises, sharp rallies, regulatory events, or structural shocks may influence the result.

5Business and capital-structure changes

Acquisitions, divestitures, operating leverage, and financial leverage may change equity beta.

Required statement

Beta is always relative to a benchmark, sample period, and estimation method.

7.4.12Section 10 · Historical beta versus future beta

CAPM needs the beta relevant to future returns

Regression provides an estimate of how the asset behaved in the past — not a guarantee about the future.

historical estimated beta \neq = guaranteed future beta

Business example

A historically defensive company sells its stable division and enters a highly cyclical industry.

Should its old historical beta still be used automatically?

No. A statistically correct historical estimate may no longer represent the company's current business risk.

Required conclusion

A useful beta should make sense statistically and economically.

7.4.13Interaction · Beta across different windows

Same company, different estimates

Helix Industries moved through three phases. Choose a window and watch both the beta estimate and its uncertainty change.

Company · Helix Industries (fictional)

Over ten years Helix moved through three phases: stable operations, acquisition-driven expansion, and highly leveraged cyclical operations. The window you choose changes both the beta estimate and its uncertainty.

Stable operations

Mature, low-leverage consumer-staples business.

Acquisition-driven expansion

Rapidly integrating cyclical targets; rising operating leverage.

Highly leveraged cyclical operations

Debt-financed capital-intensive cycle business.

Years 1–10

3 phases covered

Estimated beta0.95

Approximate uncertainty range

\hat{β} \pm 2 S E

0.67 – 1.23

An approximate range from the sample, not a guarantee or a permanent interval.

SE(β)

0.14

R²

41%

Phases

3 of 3

Read this window

Blends three very different business phases into a single average. Statistically long, but economically mixed.

Note

Which estimate is most relevant for a forward-looking CAPM analysis? There is no single automatic answer. Recent data may reflect the current business better, but short samples are noisier. The estimate should reflect the company's current business and capital structure, not just the longest available history.

7.4.14Section 11 · Read a regression output

Nova — interpret every line

A single regression printout contains beta, the standard error, R², and alpha. Interpret each one carefully.

Nova — fitted regression

R_{Nova, t} - R_{f, t} = 0.12% + 1.30 (R_{M, t} - R_{f, t}) + ε_{t}

Statistic	Estimate
Alpha	0.12% monthly
Beta	1.30
Beta standard error	0.20
R²	27%
Sample	60 monthly observations

Beta

Nova historically had greater market exposure than the benchmark.

R²

The market regression explained approximately 27% of Nova's sample return variation.

Standard error

The beta estimate is uncertain and should not be treated as exactly 1.30.

Alpha

The fitted intercept was positive, but this alone does not establish skill or future outperformance.

7.4.15Exercise · Read the regression

Read off beta, then interpret

Given the fitted regression $R_{i} - R_{f} = - 0.05% + 0.85 (R_{M} - R_{f}) + ε_{i}$ , with $R^{2} = 40%$ and $S E (\hat{β}) = 0.12$ .

Estimated beta

Beta is the coefficient on (R_M − R_f).

Estimated β̂

7.4.16Explicit ending · The takeaway

Beta is an evidence-based estimate, not a permanent fact

This conclusion must be visible before the completion gate.

\hat{β} R^{2} ε S E (\hat{β}) : estimated market sensitivity : fraction of sample variation explained : period-specific unexplained return : uncertainty in the beta estimate

Regression provides an evidence-based beta estimate, not a permanent or perfectly precise fact.
A useful beta should be evaluated both statistically and economically.
Historical beta is an input to judgment, not a substitute for judgment.

7.4.17Final check · The core conclusions

Confirm what beta estimation does and does not mean

Five questions on the regression, R², residuals, and uncertainty.

1. What is beta in the stock-versus-market regression?

2. What does R² measure?

3. Are residuals only firm-specific risk?

4. What does the standard error of beta describe?

5. Is historical beta the same as guaranteed future beta?

7.4.18Transition · Toward alpha and performance

If returns repeatedly exceed what beta justifies

The regression intercept reappears in the next lesson as a measure of performance.

Beta gives CAPM its benchmark required return. But the regression also produced an intercept — alpha:

α_{i}

If an investment repeatedly earns more than its estimated beta appears to justify, does that prove skill — or did CAPM fail to capture another risk?

That question is the heart of Lesson 7.5 — Alpha, Performance Evaluation, and the Limits of CAPM.

Try itMastery check

Pass with 4 of 6 correct

Answer all questions, then check your work. You can retry any time — mastery is based on correctness, not speed.

01
Where does an asset's beta come from?
02
In the stock-versus-market scatterplot, what is beta?
03
If two assets share the same beta but one has much lower R², what differs?
04
What does the beta standard error describe?
05
Should a historical beta be used permanently without revision?
06
Does a beta below one mean the stock is safe?

Lesson summary

1Beta is not printed on a security; it is estimated from how the asset moved relative to a market benchmark.
2Each period pairs the market excess return with the asset's excess return: one point per period.
3Beta is the slope of the fitted stock-versus-market regression line (β̂).
4Over the sample, a beta of 1.25 means the asset tended to move about 1.25 points per 1-point market move, on average.
5R² measures fit (fraction of variation explained); beta measures slope (sensitivity). They answer different questions.
6Residuals are the period-specific vertical gaps between points and the line — not only firm-specific risk.
7The covariance formula Cov( $R_{i}$ , $R_{M}$ ) / Var( $R_{M}$ ) and the regression slope describe the same relationship.
8Standard error describes how imprecisely the slope was estimated from the sample.
9Beta changes with the time window, return frequency, market proxy, unusual periods, and business or capital-structure changes.
10Historical beta is an input to judgment, not a substitute for it — a useful beta must make sense both statistically and economically.

Replay this lesson Continue to Alpha, Performance, and the Limits of CAPM