TL;DR
Contrary to the CAPM, low-risk stocks (low beta, low volatility, low idiosyncratic vol) have earned returns as high as — or higher than — high-risk stocks, on a far better risk-adjusted basis. The "low-risk anomaly" is really a family of closely related effects (beta, volatility, lottery), and it is one of the most robust and most-exploited anomalies in practice.
A 30-year arc
Black (1972); Haugen & Heins (1975) — early evidence that the security market line is too flat.
Ang, Hodrick, Xing & Zhang (2006) — the idiosyncratic-volatility puzzle: high-idio-vol stocks earn "abysmally low" returns (5-1 spread ≈ −1.06%/mo), which diversifiable risk shouldn't command.
Baker, Bradley & Wurgler (2011) — Benchmarks as Limits to Arbitrage: an agency/benchmarking story for why the anomaly persists.
Bali, Cakici & Whitelaw (2011) — the MAX effect: stocks with extreme recent daily returns (lottery tickets) are overpriced and underperform.
Frazzini & Pedersen (2014) — Betting Against Beta: leverage-constrained investors bid up high-beta assets; a beta-neutral long-low/short-high book earns alpha.
Sub-threads (the low-risk family)
Low beta (BAB) · total volatility · idiosyncratic volatility · the MAX / lottery effect. They are tightly correlated — a single "low-risk" complex viewed through different lenses.
Why it works
Leverage constraints (Frazzini-Pedersen) — investors who can't lever overweight high-beta stocks, flattening the SML.
Lottery preferences (Bali et al.) — some investors overpay for high-skew "lottery" stocks, depressing their returns.
Limits to arbitrage / benchmarking (Baker-Bradley-Wurgler) — fixed-benchmark managers won't underweight high-vol names, and shorting them is costly.
The dark side
Specification sensitivity — results depend on the vol measure, weighting, and microcap screens (Bali-Cakici critique).
Short-leg costs — profits lean on shorting hard-to-borrow, high-vol names.
Crowding & rate sensitivity — "defensive equity" became a popular product; low-vol stocks behave like bond proxies and can suffer when rates rise.
Does it survive out of sample?
The low-risk anomaly is robust but compressed by popularity; BAB and idio-vol replicate, though the standalone idio-vol effect is sensitive to construction. Our replications recompute Betting-Against-Beta, idiosyncratic volatility, and the MAX effect, scored on the holdout.
Run it yourself
Replications — Frazzini-Pedersen BAB, Ang et al. idiosyncratic volatility, Bali-Cakici-Whitelaw MAX.
Curriculum — the synthetic market ships a planted vol_1m factor (and the realistic-exchange missions cover how volatility shows up in microstructure).
Playground / Competitions — build a low-risk signal and score it OOS.