If you go to business school, you spend a depressing amount of time on a model called CAPM. It is simple, elegant, and almost completely wrong in practice. CAPM is also, in a way, the founding myth of modern finance. Understanding why pros stopped trusting it is the easiest way to understand where risk modeling actually went.
Stage 1: One number explains everything
CAPM, born in the 1960s out of work by Sharpe, Lintner, Treynor, and Mossin, said an asset's expected return is a function of one thing: its beta to the market. A stock with beta 1.2 should return 1.2 times the market premium above the risk-free rate. Done. Print the textbook.
It is a beautiful theory and Sharpe got a Nobel for it. It is also wrong. Low-beta stocks have historically outperformed what CAPM predicts. High-beta stocks underperform. Small-caps and value stocks crush their CAPM-implied returns. The model leaves the building.
Stage 2: Three factors. No, five. No, a hundred.
Fama and French rolled in with a 3-factor model in 1992. Market, size, value. Way better fit. They added profitability and investment in 2015 to get the 5-factor model. The academic world kept finding new factors. By the early 2010s, John Cochrane famously talked about a 'factor zoo.' Hundreds of published factors. Most of them noise.
The deeper problem was not just the noise. It was that factor premiums themselves are not stable. Value worked for decades, then quietly stopped working for a decade. Momentum works until a momentum crash. Low-vol works until the regime where it doesn't. A static factor model says expected returns are constant. The actual world says they are not.
Stage 3: Time-varying everything
Quants started noticing that expected returns, volatilities, and correlations all change over time. So the next wave of models let those parameters move. GARCH for volatility. DCC for correlations. Conditional CAPM for beta. The models got bigger, the math got harder, and the explanatory power got better.
But they shared a flaw. They treated parameters as if they drifted smoothly. Real markets do not drift. They jump. One day SPX vol is 12. A week later it is 35. A smooth model cannot get there in time. By the time the model catches up, the regime has changed again.
Stage 4: Regime switching
The current frontier is regime models. Instead of one model that tries to fit every market state, you build a small set of regimes (calm bull, choppy chop, crisis, recovery, whatever) and you switch between them. Each regime has its own expected returns, vols, correlations, behaviors.
Hidden Markov models, Bayesian regime switching, online clustering on realized features. The math varies. The intuition is the same: stop assuming the world is constant. Stop assuming returns come from one distribution. Build the model to know it does not know which world it is in right now, and let the data tell it.
For options-driven markets, the regimes that matter most are vol regimes. Low vol with stable correlation is a very different environment from high vol with shifting correlation. The same options strategy is a license to print money in one and a slow-motion blowup in the other.
What this means for an individual trader
You do not need a PhD to apply the lesson. The takeaway is short:
- No single strategy works everywhere. The phrase 'this always works' is the calling card of someone about to blow up.
- Your job is to identify the regime first, then pick a strategy that fits it. In positive-gamma, mean-reverting tape, sell premium and fade extremes. In negative-gamma, trending tape, buy gamma and ride.
- Position sizing should be regime-aware too. The same trade in a calm vol regime is a different bet than in a stressed one.
The dirty secret
The institutional world has been doing regime-conditioning for two decades. It is one of the reasons multi-strategy hedge funds (Citadel, Millennium, Point72, Balyasny) have such resilient returns. They are not betting on one strategy. They are running many, each sized to its current regime, and the firm-level edge is the regime selection more than the strategies themselves.
Retail can do a lighter version of the same thing. Pick two or three strategies, define the regimes they fit, and size them up or down based on which regime you are actually in. That is not magic. It is just refusing to assume the world is static. Which, if the last 60 years of finance research has taught anything, it never is.