Portfolio and risk · Measuring return and risk
Alpha: the return that market exposure does not explain
How Jensen’s alpha is calculated, why it needs a t-statistic, and what ARKK, QQQ and Berkshire Hathaway show about alpha that appears and then disappears.
On this page
Alpha is the part of an investment's return that is left after subtracting what its market exposure would have earned. If a fund carries 1.5 times the market's risk and the market rises, the fund should rise more than the market, and that extra is not skill. Alpha is whatever remains once that is accounted for.
The word is used loosely to mean "beating the market". The precise version, introduced by Michael Jensen in 1968 to test whether mutual fund managers added anything, is stricter and more useful. It separates return earned by taking more market risk, which anyone can do with an index fund and a loan, from return earned some other way. This lesson calculates alpha on real funds and shows why most measured alphas cannot be told apart from zero.
The idea
Imagine two ways to earn 14% in a year when the market earned 10% and Treasury bills paid nothing.
The first investor bought an index fund with borrowed money, 1.4 times their capital. Their result is exactly what 1.4 units of market exposure delivers: 14%. No decision in that year added anything. Their alpha is zero.
The second investor held a portfolio that moves one-for-one with the market and still earned 14%. Four points came from something other than market exposure. Their alpha is 4%.
Both beat the index by four points. Only one did something an index fund and a loan cannot copy. Alpha is the measure that tells them apart, and to do that it needs the portfolio's beta.
The figure applies this to ARKK, an actively managed fund of fast-growing technology companies. One line is the fund. The other is what its market exposure alone would have earned: Treasury bills plus 1.76 times SPY's return above bills, because 1.76 is the fund's beta over the period. The gap between the lines, month by month, is alpha.
How it works
In practice alpha and beta are estimated together. Subtract the Treasury bill return from each month's portfolio return and from each month's market return, then fit a straight line through the pairs. The slope is beta. The intercept, where the line crosses the vertical axis, is the monthly alpha: the return the portfolio averaged in months when the market's excess return was zero. Multiplying by 12 gives the yearly figure.
A worked example on QQQ
The data: 262 monthly total returns, January 2005 to October 2026, with SPY as the market.
- QQQ's average return above Treasury bills: 13.80% a year.
- SPY's average return above Treasury bills: 9.56% a year.
- QQQ's beta to SPY from the regression: 1.13.
- Return explained by market exposure: 1.13 × 9.56% = 10.80%.
- Alpha: 13.80% − 10.80% = 3.0% a year.
QQQ beat SPY by about 4.2 points a year on average. Of that, 1.2 points came from simply carrying more market risk and 3.0 points from holding a different mix of companies, heavy in large technology firms, which did better than their beta predicts.
Is it different from zero?
An alpha is an average of noisy monthly leftovers, so it comes with a standard error. The usual test divides the alpha by its standard error to get a t-statistic. As a rule of thumb, a t-statistic beyond about 2 in either direction means the alpha is unlikely to be the product of chance alone; anything smaller could easily be luck.
QQQ's alpha of 3.0% has a t-statistic of 1.8, just short of the mark after almost 22 years. Starting the measurement in May 1999 instead, which takes in the collapse of 2000 to 2002, the alpha falls to 1.3% with a t-statistic of 0.6.
The arithmetic of the test explains why significance is so hard to reach. The standard error of a yearly alpha is roughly the volatility of the leftover returns divided by the square root of the number of years. For a fund with leftover volatility of 6% a year to show a true alpha of 2% a year at a t-statistic of 2 takes (2 × 6 ÷ 2) squared, or 36 years. Few track records are that long, and few strategies stay the same for that long.
Alpha across funds
| Fund | Period | Beta | Alpha a year | t-statistic |
|---|---|---|---|---|
| IVV (another S&P 500 fund) | July 2000 – 2026 | 1.00 | 0.0% | 0.2 |
| RSP (S&P 500, equal weight) | June 2003 – 2026 | 1.08 | −0.2% | −0.2 |
| VTV (large value stocks) | March 2004 – 2026 | 0.93 | −0.2% | −0.2 |
| USMV (low-volatility stocks) | Dec. 2011 – 2026 | 0.69 | +1.1% | 0.7 |
| QQQ (Nasdaq-100) | May 1999 – 2026 | 1.30 | +1.3% | 0.6 |
| BRK-B (Berkshire Hathaway) | July 1996 – 2026 | 0.62 | +5.0% | 1.7 |
| ARKK (active growth fund) | Dec. 2014 – 2026 | 1.76 | −3.8% | −0.5 |
Every return is measured above Treasury bills. BRK-B uses price changes, which is its total return because Berkshire pays no dividend. Costs inside each fund are already in its returns; taxes and trading costs for the holder are not.
None of the seven reaches a t-statistic of 2. The S&P 500 fund IVV shows the baseline: beta 1.00 and alpha zero against another fund tracking the same index. Low-volatility stocks returned less than SPY (11.7% a year against 14.8% since December 2011), yet their alpha is positive, because with a beta of 0.69 they were expected to return less still. Alpha and "beat the market" point in opposite directions there.
Using it: alpha that came and went
Split ARKK's history at its peak, and the single number becomes two.
| ARKK | Beta | Alpha a year | t-statistic | Fund's CAGR | SPY's CAGR |
|---|---|---|---|---|---|
| December 2014 – February 2021 | 1.60 | +15.4% | 2.1 | 35.9% | 12.4% |
| March 2021 – October 2026 | 1.92 | −24.7% | −2.0 | −7.8% | 14.4% |
| Whole period | 1.76 | −3.8% | −0.5 | 13.0% | 13.3% |
At the start of 2021 the fund had more than six years of history, an alpha above 15% a year and a t-statistic above 2. By the usual test that was significant. $1 invested at the start had become $7.19 by January 2021, when SPY had turned $1 into $2.02. Over the following years the fund lost 77% from its peak, and by October 2026 the $1 stood at $4.30 against $4.45 for SPY.
Two lessons sit in that table. A statistically significant alpha over six years was not evidence that it would continue; the fund's holdings shared an exposure, to expensive, fast-growing companies, that the single market beta did not capture and that reversed. And the whole-period alpha of −3.8% understates the damage to compound wealth. The beta-matched line in the figure ends at $9.55, more than twice the fund's $4.30, because alpha is an average of monthly gaps while money compounds, and the fund's leftover volatility of 25% a year is expensive to compound through. CAGR: compound annual growth rate, with the formula and its limits explains that drag.
Berkshire Hathaway shows a slower version. Its alpha was 6.8% a year from 1996 to 2008 and 2.8% a year from 2009 to 2026, and neither half is significant alone.
Where it fails or misleads
The benchmark decides the answer
AOR, a fund holding about 60% global stocks and 40% bonds, shows an alpha of −1.3% a year against SPY since 2009, with a t-statistic of −2.0. That is not a judgment of the fund. It holds foreign stocks and bonds, both of which lagged US stocks in that period, and SPY is the wrong yardstick. Alpha against an inappropriate benchmark measures the difference between two asset mixes.
One beta misses other exposures
The formula credits to alpha anything the market beta does not explain. Funds tilted toward small companies, cheap stocks, momentum or a single industry earn or lose returns for carrying those exposures. Multi-factor models add more benchmarks to the regression so that those returns are not mistaken for skill; Factor models: explaining a portfolio’s returns with several betas covers them. Alpha usually shrinks as benchmarks are added.
Luck among many funds
With thousands of funds, some will show high alphas by chance alone, and those are the ones that get advertised. Fama and French's 2010 study of US stock mutual funds found that, after costs, the distribution of fund alphas looked much like what luck alone would produce, with few funds showing skill sufficient to cover their fees.
Alpha is before the holder's costs and taxes
A fund's returns include its own expenses. They do not include the holder's trading costs, or the taxes on distributions and on switching funds, all of which come out of any alpha.
Compared with its neighbours
| Measure | What it compares | Adjusts for beta? | Unit |
|---|---|---|---|
| Excess return | Portfolio return minus benchmark return | No | Percent a year |
| Alpha | Portfolio return minus what its beta explains | Yes | Percent a year |
| Information ratio | Excess return per unit of tracking error | No | Ratio |
| Sharpe ratio | Return above the risk-free rate per unit of volatility | Not applicable | Ratio |
| Multi-factor alpha | Return left after several exposures are explained | Yes, several | Percent a year |
How Finabase does it
Finabase does not calculate alpha for a fund, a stock or a portfolio.
What it offers is the raw comparison. The comparison chart redraws a symbol and a benchmark as percent changes from a common starting bar, and the table beneath lists their price returns over periods up to two years together with each symbol's beta as supplied by the market-data provider. The gap between two lines on that chart is the excess return, not alpha: it makes no allowance for beta. A fund with a beta of 1.5 that is ahead of SPY by a few points over a rising two years may have a negative alpha.
A rough alpha can be worked out from those numbers: subtract the Treasury bill yield from both returns, multiply the benchmark's by the fund's beta, and subtract. With two years of data, and a beta whose measurement window is not stated, the result is a description, not a test.
Frequently asked questions
Is alpha the same as outperformance?
No. Outperformance is the fund's return minus the benchmark's. Alpha first subtracts the return that the fund's beta accounts for. The two are equal only when beta is exactly 1.
What is a good alpha?
Any positive alpha that is statistically reliable and larger than the cost of obtaining it would be valuable, and that combination is rare. Among the funds above, none has an alpha that can be distinguished from zero.
Can an index fund have alpha?
Measured against its own index, its alpha should be about zero, slightly negative by the amount of its fee. IVV's alpha against SPY is 0.0%.
Why did my fund have positive alpha one year and negative the next?
A single year contains twelve monthly observations, far too few to estimate anything. Yearly alphas are mostly noise around the fund's persistent exposures.
Does a negative alpha mean a fund is bad?
It means that, against that benchmark and over that period, the fund returned less than its market exposure explains. The benchmark may be wrong for it, the period may be unlucky, or the fund's costs may be the cause.
Summary
Alpha is return above the risk-free rate minus beta times the market's return above the risk-free rate: the intercept of a regression of one on the other. It distinguishes return from market exposure, which is free, from return from anything else. QQQ's 3.0% a year against SPY since 2005 and Berkshire's 5.0% since 1996 are large and still not statistically distinguishable from zero. ARKK's alpha was significant and positive for six years and then reversed completely. An alpha needs its benchmark, its period and its t-statistic stated, and even then it describes the past.
Check your understanding
5 questions on this lesson. Each answer comes with the reason, and nothing is scored.
The market returned 12%, Treasury bills 2%, and a fund with a beta of 1.5 returned 16%. What was the fund’s alpha?
- +4%
- −1%
- +1%
Show the answer
B. −1% Market exposure explains 1.5 × (12 − 2) = 15 points above bills. The fund earned 16 − 2 = 14 points above bills. Alpha = 14 − 15 = −1%. The fund beat the market by four points and still had negative alpha.
A fund shows an alpha of +2% a year over five years with a t-statistic of 0.8. What is the reasonable conclusion?
- The manager adds 2% a year
- The alpha cannot be distinguished from zero with this much data
- The fund is certain to underperform
Show the answer
B. The alpha cannot be distinguished from zero with this much data A t-statistic below about 2 means the estimate is within the range that chance alone produces. Five years is rarely enough to tell a 2% alpha from none.
A low-volatility fund with a beta of 0.7 returned less than the index over ten rising years. Can its alpha be positive?
- No, alpha requires beating the index
- Yes, if it returned more than 0.7 times the market’s excess return would explain
- Only if its fees were zero
Show the answer
B. Yes, if it returned more than 0.7 times the market’s excess return would explain Alpha compares the fund with what its own market exposure predicts. A beta of 0.7 predicts less than the market’s return in a rising market, so a fund can trail the index and still have positive alpha, as USMV did.
A balanced fund of global stocks and bonds shows a significantly negative alpha against the S&P 500. What is the most likely explanation?
- Poor stock selection
- The S&P 500 is the wrong benchmark for what the fund holds
- An error in the beta
Show the answer
B. The S&P 500 is the wrong benchmark for what the fund holds Alpha against a benchmark that does not match the fund’s holdings mostly measures the difference between two asset mixes. Foreign stocks and bonds lagged US stocks, which shows up as negative alpha against SPY.
ARKK’s alpha through February 2021 was +15.4% a year with a t-statistic above 2. What did its later history show?
- A significant alpha over several years was no assurance that it would continue
- That t-statistics above 2 guarantee future outperformance
- That the alpha was calculated incorrectly
Show the answer
A. A significant alpha over several years was no assurance that it would continue From March 2021 the fund’s alpha was −24.7% a year, and over the whole period it was negative and insignificant. The earlier alpha reflected an exposure that later reversed.
Sources and further reading
- Michael C. Jensen, “The Performance of Mutual Funds in the Period 1945–1964,” Journal of Finance, vol. 23, no. 2 (1968)
- Eugene F. Fama and Kenneth R. French, “Luck versus Skill in the Cross-Section of Mutual Fund Returns,” Journal of Finance, vol. 65, no. 5 (2010)
- Board of Governors of the Federal Reserve System — 3-month Treasury bill secondary market rate (TB3MS), via FRED
- Finabase chart data — monthly bars and dividends of ARKK, QQQ, BRK-B, USMV, RSP, VTV, IVV, AOR and SPY through October 2026; regressions computed by Finabase
