Alpha: Benchmark-Adjusted Return, Jensen Alpha, and Model Dependence

What is Alpha?

Alpha is a benchmark- or model-relative measure of investment performance.

In everyday investing language, people often use alpha to mean "return above a benchmark."

In a more formal asset-pricing context, Jensen's alpha is the portion of return not explained by the investment's exposure to the market under the Capital Asset Pricing Model, or CAPM.

A common CAPM form is:

1Alpha
2=
3Portfolio Return
4- [Risk-Free Rate + Beta × (Market Return - Risk-Free Rate)]

That makes alpha different from simple benchmark excess return.

If a portfolio earns 12% while its benchmark earns 10%, the simple active return is 2 percentage points.

Whether the portfolio also has 2% of alpha depends on the model and its estimated risk exposures.

A simple Jensen alpha example

Suppose a portfolio produced a 12% return during a period when:

1Risk-free rate:  3%
2Market return:  10%
3Portfolio beta: 0.8

Under CAPM, the model-implied return is:

1Expected model return
2= 3% + 0.8 × (10% - 3%)
3= 3% + 5.6%
4= 8.6%

The realized Jensen alpha for the period would be:

1Alpha = 12.0% - 8.6% = 3.4%

The portfolio's simple outperformance versus the market was only 2 percentage points.

The model-based alpha is larger because the portfolio carried less estimated market exposure than the benchmark.

This example shows why alpha and raw outperformance are not synonyms.

Alpha is model-dependent

There is no model-free alpha.

An alpha estimate depends on what the model says should explain return.

Under a one-factor CAPM regression:

1Portfolio Return
2=
3Alpha
4+ Beta × Market Return
5+ Residual

alpha is the fitted intercept after accounting for market sensitivity.

A multifactor model can instead control for several systematic exposures such as size, value, momentum, rates, credit, or other selected factors.

A strategy that appears to have positive alpha under a simple market model may show much less alpha after additional factor exposures are recognized.

Therefore:

1Alpha is return unexplained by the chosen model,
2not return proven to come from unique skill.

Benchmark choice can change alpha

A global equity manager should not automatically be evaluated against a narrow domestic index.

A small-cap portfolio should not be compared casually with a mega-cap benchmark.

A long-duration bond manager should not be judged using an equity index.

If the benchmark does not represent the manager's opportunity set and investment process, the resulting alpha can be misleading.

Benchmark quality matters because poor benchmark specification can attribute ordinary exposure differences to apparent manager skill.

CFA Institute's performance-evaluation framework emphasizes that benchmark misspecification can invalidate attribution and appraisal conclusions.

Alpha is not the same as active return

Active return is usually the portfolio's return minus the benchmark's return.

1Active Return = Portfolio Return - Benchmark Return

Alpha normally adjusts that performance through a selected risk model.

A portfolio can have positive active return but negative model-based alpha if it took enough systematic risk to make its performance disappointing relative to the model.

It can also have negative active return but positive alpha over a period if the model says its lower risk exposure should have produced an even lower return.

This distinction matters when evaluating active management.

Alpha is connected to beta

Beta estimates sensitivity to a selected market or benchmark factor.

In the CAPM version of alpha, beta determines the return that the model associates with market risk.

That means an inaccurate or unstable beta estimate can affect the alpha estimate.

The two metrics are therefore linked:

1Beta  -> modeled systematic exposure
2Alpha -> fitted return not explained by that exposure

Neither should be interpreted without knowing the benchmark and estimation method.

Historical alpha is not proof of skill

Positive alpha can arise from skill, but a historical estimate alone does not prove skill.

Observed alpha can also reflect:

• random variation;
• omitted risk factors;
• an inappropriate benchmark;
• leverage or nonlinear exposures;
• stale or biased data;
• survivorship or selection bias;
• transaction-cost assumptions; and
• a favorable market regime.

A manager with no true forecasting edge can produce a period of positive realized alpha by chance.

Likewise, a skilled process can experience negative realized alpha over a limited sample.

Evaluating manager skill therefore requires more than checking whether alpha is above zero.

Statistical uncertainty matters

An estimated alpha is normally subject to sampling error.

For example, suppose a monthly regression estimates:

1Monthly alpha: 0.30%

That number may look attractive, but it could be statistically weak if the residual variation is large and the sample is short.

A proper analysis may consider the standard error or t-statistic of the alpha estimate, the length and quality of the sample, and whether the relationship remains stable across subperiods.

A point estimate without uncertainty can create false precision.

Fees can materially change alpha

Gross-of-fee alpha and net-of-fee alpha answer different questions.

A strategy can add value before management fees, trading costs, taxes, market impact, and financing expenses yet deliver little or no alpha to the end investor.

When comparing strategies, investors should verify whether returns are:

• gross or net of management fees;
• before or after transaction costs;
• before or after financing costs; and
• measured consistently across products.

Economic alpha to an investor is ultimately affected by implementation costs.

Alpha does not measure consistency

Two managers can report the same average alpha with very different return paths.

One may generate modest positive relative returns frequently.

Another may suffer long periods of underperformance followed by a few large wins.

Alpha alone does not reveal that difference.

Tracking Error describes the variability of active returns, while the Information Ratio relates average active return to that benchmark-relative variability.

Maximum Drawdown can add another path-dependent risk perspective.

A single alpha number is therefore incomplete performance analysis.

Alpha can disappear when exposures are measured better

Suppose a manager consistently owns smaller, cheaper companies.

Against a broad market benchmark, the strategy may appear to deliver alpha.

If a multifactor model attributes part of the return to systematic size and value exposures, estimated alpha may shrink.

That does not automatically mean the strategy is bad.

It means the explanation for return changed from "unexplained residual performance" to "compensation associated with measured factor exposures."

This distinction is essential for understanding what an investor is paying active fees to receive.

Alpha can be negative even for a profitable investment

An investment can make money and still have negative alpha.

Suppose a high-beta portfolio earns 15% in a very strong market, while the selected model implies that its risk exposure should have produced 20%.

The investor gained money in absolute terms, but the portfolio underperformed the model-adjusted expectation.

Conversely, a portfolio can lose money in a market crash and still have positive alpha if it loses substantially less than its modeled risk exposure would imply.

Alpha is relative performance, not absolute profit or loss.

Alpha and the information ratio answer different questions

The Information Ratio is commonly:

1Mean Active Return / Tracking Error

It measures benchmark-relative return per unit of benchmark-relative variability.

Alpha can be model-adjusted and may incorporate beta or multiple factors.

Information ratio usually works directly with active return against a benchmark.

A portfolio can have attractive alpha but an unattractive information ratio if active results are highly inconsistent.

It can also have a high information ratio from small but steady benchmark-relative gains even if a richer factor model attributes those gains to systematic exposures rather than unique alpha.

How investors should use alpha

Before interpreting an alpha estimate, ask:

1. Which benchmark or factor model defines expected return?
2. Is this simple excess return or model-based alpha?
3. Are beta and other factor exposures estimated consistently?
4. Is the sample long enough to judge statistical uncertainty?
5. Are returns gross or net of fees and trading costs?
6. Has the manager's process or portfolio changed?
7. Could omitted factors explain the apparent alpha?
8. What do tracking error, information ratio, volatility, and drawdown show alongside it?
9. Is the benchmark actually appropriate for the strategy?

Grizzly Bulls' Models research can be evaluated with these questions without treating this encyclopedia page as live performance authority. The Cyclically Adjusted Risk Premium supplies separate market-valuation context, not a canonical alpha calculation.

Sources and further reading

• CFA Institute, 2026, Portfolio Risk and Return: Part II: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-risk-return-part-2
• CFA Institute, 2026, Analysis of Active Portfolio Management: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/analysis-active-portfolio-management
• CFA Institute, 2026, Portfolio Performance Evaluation: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-performance-evaluation
• CFA Institute, 2026, Using Multifactor Models: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/using-multifactor-models

Explore more topics in the Financial Research Encyclopedia.