Case 084Regression and model reviewCore
Regressed on the market alone, Suryamandal's fund shows alpha of 0.8% a month (t = 2.9). Adding size and value factors gives alpha of 0.35% (t = 1.4), with loadings of 0.95 on the market, 0.6 on size and 0.4 on value. Where did the alpha go, and what does the fund really deliver?
1The situation
Suryamandal Asset Management markets an active equity fund on its record: over 60 months, a regression of its monthly excess return on the market's gives an alpha of 0.80% a month with a t-statistic of 2.9, close to 10% a year. A pension client's consultant reruns the regression with two more factors, a size factor (small companies minus large) and a value factor (cheap stocks on book value minus expensive).
The three-factor regression gives alpha 0.35% a month (t = 1.4) and loadings of 0.95 on the market, 0.6 on size and 0.4 on value. Over the same 60 months the market earned 0.60% a month over cash, the size factor 0.45% and the value factor 0.45%. The fund charges 1.8% a year.
2Your task
Explain where the alpha went with numbers, say whether what is left is evidence of skill, and say what the client is really buying.
Quick check
Why does alpha fall from 0.80% to 0.35% when size and value are added?
Worked solution
Try it on paper, then open one step at a time.
30-second answerThe answer to give first
Of the 0.80% monthly alpha, 0.45% was the fund's tilt to small and value stocks: 0.27% from size and 0.18% from value. What remains, 0.35% a month with a t-statistic of 1.4, is not significant; its 95% interval runs from -0.15% to 0.85%. The client is mainly buying a small-cap value tilt at a market beta of 0.95, which can be held far more cheaply than a 1.8% fee.
Step 1Where did the alpha go?
Imagine judging a cricket coach by how many runs the team scored, without noticing he only picked players from the strongest club in the state. Part of the result was the pool, not the coaching. Alpha is whatever return the model's factors cannot explain, so a model with fewer factors calls more of the return skill. The fund's average excess return is 1.37% a month. The market model explains 0.95 x 0.60 = 0.57% and leaves 0.80%. The three-factor model also credits 0.6 x 0.45 = 0.27% to size and 0.4 x 0.45 = 0.18% to value, and leaves 0.35%. The 0.45% that moved is the tilt's reward over these 60 months.
| r bar | fund's average monthly excess return, % |
| M bar | market's average excess return, 0.60% |
| SMB, HML | size and value factor returns, 0.45% each |
| beta | the fund's loading on each factor |
Step 2Is the remaining 0.35% evidence of skill?
Read the standard errors, not just the point estimates. A t of 2.9 on 0.80 means a standard error of about 0.28; a t of 1.4 on 0.35 means about 0.25. The three-factor alpha's 95% interval, roughly -0.15% to 0.85% a month, includes zero, so 60 months cannot tell this fund's skill apart from none. That is not proof of no skill either: an alpha of 0.35% a month would be worth having if real, and with this much noise it would take roughly four times as many months to confirm it. The fair statement is that the record does not show it.
Step 3What is the client really buying?
Read the loadings as a description. A market beta of 0.95, a size loading of 0.6 and a value loading of 0.4 describe a fund that holds smaller, cheaper companies than the index, which is a style, not a secret. Styles can be bought through low-cost index funds or rules-based products for a fraction of a 1.8% fee, so the question for the client is whether the 0.35% that might be skill is worth the fee gap. It also matters which way the style runs next: small and value premia vary for years at a time, and a fund that looked brilliant in a decade when small value led can look poor in one when it lags, with no change in the manager's skill.
State the limits. The factor returns used here are long-short portfolios before costs, which an investor cannot hold for free, so comparing the fund with them flatters the factors a little. Adding a momentum or quality factor could move the alpha again in either direction. The answer is conditional on the model, and the honest conclusion is that most of the market-model alpha was exposure the client could have bought more cheaply.
Where candidates lose it
The common loss is saying the alpha fell because the second regression is noisier, or that the extra factors 'stole' the alpha unfairly. The drop is exactly loadings times factor returns, and computing it is the point of the question.
The second is concluding the manager has no skill because t = 1.4. Insignificant is not zero. The record is too short to show skill, and that is a different, more honest statement.
What the interviewer asks next
- How would you test whether the fund's size loading is stable over time?
- What would a negative value loading with the same alpha tell you?
- Why might the market beta change when you add size and value factors, and what would that mean here?
- How many months of data would you need for an alpha of 0.35% to reach a t-statistic of 2?
Asked at State Street, Investment Banking, Cambridge, 2019 (Wall Street Oasis): some basic market knowledge, such as factor model (Fama French), portfolio optimization, risk analysis
Company names and figures are illustrative.
