Quant Analytics: from scratch to pro · Factors, volatility, regimes and lead-lag · Intermediate
Factor regressions and the evidence on momentum
Regressing a return series on factor returns gives loadings and an intercept called alpha. Whether that alpha is real depends on its t-statistic, which is often small. Momentum is the best-documented factor and also the one with the sharpest reversals.
In this lesson
- Read a factor regression: loadings, alpha, residual
- Compute the t-statistic of an alpha and the sample length it needs
- Distinguish cross-sectional from time-series momentum
- State what the momentum evidence does and does not show
Read first: Factors: what market, size, value and momentum mean, How sure is an average? Standard error and confidence intervals
A fund beats the market. Is that skill, a bet on a known factor, or luck? A factor regression splits the return into parts and shows what is left.
The regression
For month t, regress the excess return of the portfolio (return minus risk-free rate) on the factor returns:
r_p - r_f = alpha + b_mkt x MKT + b_smb x SMB + b_hml x HML + b_mom x MOM + error
The b terms are loadings: how much the portfolio moves with each factor. Alpha is the intercept, the average return not explained by the factors. The error is what is specific to the month. Fama and French (1993) used the first three factors; Carhart (1997) added momentum. Factor series are free from the French data library. See factor. See factors explained.
Worked example
A fund returned 10% in a year when the risk-free rate was 2% and the market returned 8%. Its market beta is 1.1.
- CAPM expected return:
2% + 1.1 x 6%= 8.6%. Alpha is 10% - 8.6% = 1.4%. - Add a momentum loading of 0.4 with a premium of 3% (illustrative). Momentum explains 1.2%.
- The alpha after the momentum factor is 0.2%.
Most of the apparent outperformance was a momentum exposure, available cheaply elsewhere. The more factors fit, the smaller the alpha usually becomes.
Is the alpha distinguishable from zero?
Suppose a regression over 120 months gives an alpha of 0.10% a month (1.2% a year) with residual volatility of 2% a month. The standard error is 2% / sqrt(120), which is 0.183%, so the t-statistic is 0.10 / 0.183, which is 0.55. A t-statistic should be near 2 to be taken seriously, and it grows only with the square root of the sample. Reaching 2 at this alpha needs (2 x 2% / 0.10%)^2, which is 1,600 months, or 133 years. Small alphas cannot be confirmed with the history available. See standard error.
What the momentum evidence says
Jegadeesh and Titman (1993) reported that buying past winners and selling past losers earned significant profits over holding periods of 3 to 12 months. Carhart (1997) used momentum to explain why last year's winning mutual funds tended to keep winning for a while: his sample of diversified equity funds ran from January 1962 to December 1993, and common factors and expenses accounted for nearly all the persistence.
Moskowitz, Ooi and Pedersen (2012) extended the idea across assets. Time-series momentum looks at each asset's own past return and not at its rank among others. They found significant time series momentum in equity index, currency, commodity and bond futures, in each of the 58 liquid instruments they examined.
The record is not smooth. Daniel and Moskowitz (2016) show that momentum has had prolonged losses, concentrated in panic states: after market declines, when volatility is high, and as the market rebounds. A premium that is real on average can still be painful to hold.
Common mistakes
- Reading alpha as skill without a t-statistic and without asking which factors are missing.
- Fitting many factors until alpha disappears or appears. Every extra choice is a test.
- Regressing on a short window: loadings drift, so rolling estimates vary.
- Using prices instead of returns. See spurious regression.
- Ignoring costs. Momentum trades often, and a gross premium shrinks net.
In practice
Report loadings, alpha and its standard error together, in a stated window, and look at how they move through time. Treat factor premiums as long-run averages with long droughts. A pattern is a claim to be measured with its uncertainty, not assumed. Analytics on FinUties applies measures like these to the markets it covers, so they are at hand when you read a chart.
t-statistic of a monthly alpha of 0.10% against sample length
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| CAPM expected return | 0.02 + 1.1 * (0.08 - 0.02) | 8.6% |
| CAPM alpha | 0.10 - exp_ret | 1.4% |
| momentum contribution | 0.4 * 0.03 | 1.2% |
| alpha after the momentum factor | capm_alpha - mom_part | 0.2% |
| standard error of a monthly alpha | 0.02 / sqrt(120) | 0.183% |
| t-statistic of alpha 0.10% a month | 0.001 / se_alpha | 0.55 |
| months for t = 2 | (2 * 0.02 / 0.001)**2 | 1,600 |
| years for t = 2 | months_needed / 12 | 133 |
Check yourself
A fund returned 10%, the risk-free rate is 2%, the market returned 8%, and the fund's beta is 1.1. What is its CAPM alpha?
Expected return = 2% + 1.1 x 6% = 8.6%. Alpha = 10% - 8.6% = 1.4%.
After adding a momentum factor with loading 0.4 and premium 3%, the alpha of the same fund is what?
Momentum explains 0.4 x 3% = 1.2%. The alpha falls to 1.4% - 1.2% = 0.2%. Much of the apparent skill was a momentum exposure.
A monthly alpha of 0.10% has residual volatility of 2% a month over 120 months. What is its t-statistic?
Standard error = 2% / sqrt(120) = 0.183%. t = 0.10 / 0.183 = 0.55, far below 2.
What separates time-series from cross-sectional momentum?
Cross-sectional momentum ranks assets against each other and buys the best. Time-series momentum looks at each asset's own past return and goes long if it is positive, short if negative.
Does a factor regression prove skill when alpha is positive and significant?
No. It shows the return was not explained by the included factors. An omitted factor, costs, or chance (many funds tested) can still explain it.
Key terms
Checked facts
Every figure below carries how it was checked: Primary checked (read in the cited primary, official, academic or data source), Two sources agree, or Approximate (an order of magnitude only).
- Primary checked
Jegadeesh and Titman (1993) reported that strategies buying past winners and selling past losers earn significant profits over holding periods of 3 to 12 months.
- Primary checked
Moskowitz, Ooi and Pedersen (2012) found significant time series momentum in equity index, currency, commodity and bond futures, in each of the 58 liquid instruments they examined.
- Primary checked
Carhart (1997) used a survivor-bias-free sample of diversified equity funds from January 1962 to December 1993 and found that common factors and expenses explain nearly all persistence in fund returns.
Sources
- Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33(1), 3-56 (Eugene F. Fama, Kenneth R. French), 1993. Accessed 9 Nov 2026
- On Persistence in Mutual Fund Performance. The Journal of Finance 52(1), 57-82 (Mark M. Carhart), 1997. Accessed 9 Nov 2026
- Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. The Journal of Finance 48(1), 65-91 (Narasimhan Jegadeesh, Sheridan Titman), 1993. Accessed 9 Nov 2026
- Time series momentum. Journal of Financial Economics 104(2), 228-250 (Tobias J. Moskowitz, Yao Hua Ooi, Lasse Heje Pedersen), 2012. Accessed 9 Nov 2026
- Momentum crashes. Journal of Financial Economics 122(2), 221-247 (Kent Daniel, Tobias J. Moskowitz), 2016. Accessed 9 Nov 2026
- Kenneth R. French Data Library. Tuck School of Business at Dartmouth (Kenneth R. French). Accessed 9 Nov 2026