Pillar
Edge is a compensated exposure discovered by ranking a cross-section, not a forecast of one series. You sort every name by a characteristic, go long the top and short the bottom, and ask whether the spread's return is a real risk premium, a behavioral mispricing, or a ghost that dies out of sample.
Factor investing runs on one equation: expected return equals alpha plus beta times lambda. Beta is exposure, lambda is the premium, alpha is the leftover. The market is factor #1, and CAPM fails on its own.
You never see factor exposure. So you rank stocks on a proxy, go long the top decile and short the bottom, and test the spread. Deciles, a t-stat, and a monotonicity check, from scratch.
Fama-MacBeth gets you the factor premium in two passes, even for untradable factors. But the second pass uses estimated betas, so skipping the Shanken correction inflates every t-stat.
Alpha is the gap between what a portfolio earned and what your model predicts. The GRS test grades all those gaps at once, and the usual verdict is blunt: you are missing a factor.
ML promises to triple your Sharpe. Bryan Kelly, who builds the models, says expect 20%. Headline numbers die under fair tests, 166% turnover, and alpha trapped in tiny illiquid stocks.
Test 316 factors at a t of 2 and about 16 are pure luck. The factor zoo is a multiple-testing failure: raise the bar to t above 3, expect a 36% out-of-sample haircut, but don't prune to five.
Regional models beat global, said 20 years of linear studies. Redone with neural nets across 24 markets, it flips: complex models want global data. The global NN hits 0.74%/mo, t=5.73.
The |t|>3 rule is right for testing one factor, wrong for building a portfolio. A book of 18,000 signals, 80% noise, beats the strict filter because diversification pays where selection does not.
Every factor model, from Fama-French to a neural net, is one equation: the SDF. Same skeleton, but the choice of characteristics and weighting function swings out-of-sample Sharpe from 0.45 to 3.4.
Value's 55% drawdown looked like death. But the structural premium stayed positive; the loss was the value-growth spread hitting the 100th percentile. That is a repricing, and repricings revert.
Value, momentum, volatility, and sentiment timing all lost to a plain equal-weight factor basket in China. Theory says timing is huge; estimation error eats it. Trust the plateau, not the peak.
Alternative data buys a short-dated edge, and Dessaint shows the tax: short-horizon accuracy rises while long-horizon accuracy falls. It is a lease, not a purchase, and the crowd rents it too.
Prospect theory turns broken psychology into two factors: a TK score of how attractive a stock looks and capital-gains overhang. The prettiest stocks pay least, and the spread runs 1.24% a month.
Two stocks end the month flat, one by recovering, one by fading. The shape between the endpoints predicts next month: low-convexity stocks beat high-convexity by 0.84%/mo, and no factor explains it.
Trade an asset off the momentum of everything it's linked to, not its own. A learned cross-asset graph delivers a 1.51 Sharpe, and the alpha lives in the links between asset classes.
The return you predict matters more than the 147 features you feed the model. Demeaning, standardizing, or ranking the target moves monthly alpha ~0.9pp; feature transforms move it ~0.4pp.
In low-signal economics and finance, the do-nothing zero forecast is Bayes-optimal. Ridge can beat it, Lasso cannot for any penalty, and signal weakness, not sparsity, is why.
Value, momentum, volatility, and sentiment timing all lost to a plain equal-weight factor basket in China. Theory says timing is huge; estimation error eats it. Trust the plateau, not the peak.
Two stocks end the month flat, one by recovering, one by fading. The shape between the endpoints predicts next month: low-convexity stocks beat high-convexity by 0.84%/mo, and no factor explains it.