Alpha
The part of a return that its factor exposures don't explain: what's left after accounting for the risks taken.
4 min readLesson 13 of 57Interactive
In short
A fund that made 12% when its market and other factor exposures account for 10% has an alpha of about 2% a year. Alpha is what investors pay managers for, but most measured alpha is noise: check its t-statistic before believing it.
Why it matters
A raw return mixes skill with risk. A fund that holds riskier stocks than the market will beat it in a rising year, and that is not cleverness: you could have done the same by buying more stocks. To judge a manager, a strategy or your own picks, you need to subtract what the risks taken would have earned anyway.
Alpha is that leftover. It is what professional investors are paid for, and it is also humbling: measured carefully, most of it is small, and much of what looks like alpha turns out to be noise.
How it works
Alpha is the constant in a factor regression: the part of the daily return that the factor exposures don't explain. A daily alpha is tiny, such as 0.02% a day, so it's annualized. MarketLab multiplies the daily figure by 252 trading days, which turns 0.02% a day into about 5% a year.
Alpha depends on which factors you use. Measured against the market alone, it's called Jensen's alpha. Adding size, value, momentum and low volatility usually shrinks it, because much of what looks like stock-picking skill is a tilt toward small, cheap or trending stocks.
Every alpha comes with a t-statistic: the alpha divided by its standard error, the typical size of the error in the estimate. Roughly, a t-statistic above 2 means the alpha would rarely look this big by luck alone, if it were the only thing you tested.
The trouble is noise. The standard error of a yearly alpha is about the residual volatility (the swings the factors don't explain) divided by the square root of the number of years. A stock with 20% residual volatility and one year of data has a standard error of about 20%, so even an alpha of +10% has a t-statistic of only 0.5.
Alpha (a year) = daily intercept × 252; t-statistic = alpha ÷ standard error
Annualize the daily leftover, then ask how many standard errors it is from zero.
A worked example
A fund returned +14% in a year. Its market beta is 1.1 and its size beta 0.5; its other betas are about 0. The market returned +10% and the size factor +4%. The swings the factors don't explain, its residual volatility, come to 8% a year.
- Market part: 1.1 × 10% = 11%.
- Size part: 0.5 × 4% = 2%.
- Explained by factors: 11% + 2% = 13%.
- Alpha: 14% − 13% = 1%.
- With one year of data, the standard error is about 8% ÷ √1 = 8%, so the t-statistic is 1% ÷ 8% = 0.125.
- To reach a t-statistic of 2 with the same alpha and noise, you'd need (2 × 8 ÷ 1)² = 256 years of data.
Result: The fund beat the market by 4 points, but 3 of them came from extra market and size exposure. The 1% of alpha that's left is far too small, against that much noise, to tell apart from luck.
Try it
Where did the return come from?
- Market+24.0%
- Size+1.6%
- Value+4.8%
- Momentum+1.8%
- Low volatility+2.0%
- Alpha+0.0%
The stock's return
+34.2%
Explained by factors
+34.2%
From the market alone
+24.0%
Beta 1.2 × +20.0%
Made-up factor returns. Adding the parts is a simplification: real returns compound day by day.
How to read it
- Near zero, t-statistic between −2 and 2
- The usual result. The returns are explained by the exposures, and any leftover could be luck.
- Positive, t-statistic above 2
- Interesting, but ask how many stocks, funds or strategies were checked to find it. Among 100, about 2 or 3 will clear this bar by chance.
- Negative
- It did worse than its exposures alone would suggest. For funds, fees and trading costs are a common reason.
- Huge, from a short window
- A big alpha over one year is almost always mostly noise, as the example shows.
In the real world
- S&P's SPIVA reports find that most actively managed US stock funds trail their index over long periods: around 9 in 10 large-company funds over 15 years. On average, alpha after fees is negative.
- Hedge funds have long charged high fees for alpha. As cheap factor funds spread, much of what was sold as alpha was repackaged as cheap beta, and managers face harder questions about what's left.
- Researchers who studied Warren Buffett's record found that much of Berkshire Hathaway's return can be explained by exposure to cheap, safe, high-quality stocks, held with modest borrowing. Even legendary alpha partly turns into factors when examined closely.
Common mistakes
- Calling any return above the market alpha. Higher risk earns more in rising markets without any skill.
- Trusting a one-year alpha. With typical noise, it takes decades of data to prove a modest alpha is real.
- Searching many stocks or strategies and reporting the one with the highest alpha. That's Overfitting, and some winners always appear by chance.
- Forgetting costs. An alpha before trading costs and fees can disappear once they're paid.
See it in MarketLab
What drives a stock →
On a stock's Risk & technicals tab, the What drives it card shows annualized alpha with its t-statistic, next to the factor betas.
What drives your portfolio →
On the portfolio's Risk tab, check whether your portfolio's alpha is distinguishable from zero, or just the leftover of its exposures.
Test yourself
Optional: 4 quick questions on the words, using the idea, and how it plays out in real markets.
Loading your answers…