LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
Fama-MacBeth regression asks whether a characteristic, such as market beta, lines up with next month’s returns across stocks, without pretending that one stock’s beta is a constant. Each month you regress the cross-section of future returns on the characteristic, store the slope, then average those slopes and test them. In the lecture’s small beta study the average relationship had the expected sign and was not statistically significant. A long-short book built from it showed an annualized return of 5.64 percent, volatility of 13.42 percent, a Sharpe ratio of 0.42, and a 20 percent maximum drawdown. The lecture would not trade that sample.
Why it matters
Pooled regressions throw every stock-month into one line. That mixes two clocks: the characteristic changes through time, and the return distribution changes through time. A phone unlock model fails in sunglasses because the data-generating distribution moved. Return distributions move for the same reason. Local clouds of points can slope up in one regime and down in another while a single pooled line pretends there was one relationship.
The idea in plain words
Each month, measure the characteristic with information you already had. For beta, regress the stock on a market proxy such as the S&P 500 ETF. Then regress the next month’s cross-section of returns on those betas. The slope is that month’s premium. Average the slopes and test the average. The universe might be a broad index, and stocks that die or merge have to stay in the accounting or the test flatters survivors. The lecture’s panel is a small industry sample. It claims the months are ordered without look-ahead, and it leaves survivorship mostly outside the talk.
In that run, higher beta lined up with a higher premium during a hot market. A top-quartile beta portfolio showed a 31.2 percent maximum drawdown and a Sharpe ratio of 1.15. That is leverage to the sample, not a new skill. Nvidia’s own beta over 2021 onward averaged about 2.14, touched about 2.53 on one day, and ranged from near 1 to about 4.95. Those are sample bounds, not a confidence interval.
The hedged construction is the point, and the slopes were not statistically significant. A long-short residual showed a 5.64 percent annualized return, 13.42 percent volatility, a Sharpe ratio of 0.42, and a 20 percent maximum drawdown. A CAPM fit on one portfolio in the talk had a beta of 1.39 and some alpha, which the insignificant slopes undercut. Combining the residual with another positive-Sharpe stream raised the illustrated Sharpe to 1.33, before costs. Signal quantiles were not a clean monotone ladder: the top bin had the highest forward return and the next bin the lowest.
How to use it
Pick the characteristic and the pricing model before you rank the Sharpe. Lag the characteristic so it is known before the return month. Average the monthly slopes. Only then build a high-minus-low portfolio, and only then hedge the factors the regression already said were in the return. If the average slope is noise, the pretty equity curve is a drawing.
What usually goes wrong
One regression for the whole panel, so time variation is averaged away. A beta estimated with future months in the window. Dropping dead stocks. Reading a hot-market, high-beta portfolio as alpha. And shipping a long-short book whose own Fama-MacBeth test did not clear a significance bar. The lecture’s 0.42 Sharpe is what that mistake looks like when you at least show the number.
Questions people ask
What question does one Fama-MacBeth slope answer?
For that month, across stocks, how much extra future return lined up with a one-unit difference in the characteristic. The research result is the average of those slopes, not any single month.
Were the lecture’s beta premiums significant?
No. The talk says the coefficients were statistically insignificant, even though the sign matched the CAPM story and a high-beta slice rode a strong market.
Why combine the long-short book with something else?
The hedged book was built to be orthogonal to the market. Pairing two positive-Sharpe, low-correlation streams raised the illustrated Sharpe from 0.42 on the long-short alone to 1.33 combined, before costs.