The Capital Asset Pricing Model makes one prediction everybody remembers: riskier assets earn higher expected returns. Within the stock market, it has been consistently wrong in a specific way. The calmest stocks have historically delivered returns comparable to the most volatile ones, with a fraction of the volatility — which means dramatically better risk-adjusted returns.
This is the low-volatility anomaly, and it is uncomfortable precisely because it is not a small residual. It inverts the central relationship the model is built on.
The evidence
Robert Haugen and James Heins noticed it in 1972 and it has survived every attempt to explain it away since. The two most-cited modern treatments:
Blitz and van Vliet’s “The Volatility Effect” (2007) sorted global stocks into deciles by past volatility and found the lowest-volatility decile outperformed the highest by roughly 12 percentage points a year on a risk-adjusted basis, and that the effect held independently in the US, Europe and Japan.
Baker, Bradley and Wurgler’s “Benchmarks as Limits to Arbitrage” (2011) is the more interesting paper because it proposes a mechanism rather than just documenting the effect. Their argument runs like this.
Why it survives
If low-volatility stocks are underpriced, why has nobody arbitraged it away?
Because the natural arbitrage requires leverage, and most of the money that could do it is not allowed to use any.
Suppose you believe low-volatility stocks offer better returns per unit of risk. The way to exploit that is to buy them and lever the position up until its total risk matches your target. That is what the theory says you should do. But the large majority of institutional equity capital sits in mandates that prohibit or severely limit leverage — mutual funds, pension mandates, insurance portfolios. A manager in one of those who wants higher expected returns cannot lever a low-beta portfolio. Their only available lever is to buy higher-beta stocks.
That produces persistent, structural demand for volatile stocks from exactly the investors with the most capital, and no offsetting demand for the calm ones. The high-volatility end stays bid, the low-volatility end stays cheap, and the gap does not close because the people who see it cannot act on it at scale.
There is a second, behavioural leg. Volatile stocks are lottery tickets, and retail investors demonstrably overpay for lottery-like payoffs — Bali, Cakici and Whitelaw documented this as the MAX effect: stocks with the highest recent single-day returns subsequently underperform. Preference for skewness is not a mistake investors correct once it is pointed out; it is what makes lottery tickets sell.
Both explanations predict the same thing, and both predict it will persist, because neither depends on an information gap that can close.
Measuring volatility without computing it
NSE publishes a daily volatility figure for every stock it trades, as part of the risk-management framework that sets margin requirements. It is exponentially weighted, updated daily, and available going back a decade. That is a better input than a rolling standard deviation you calculate yourself, for a practical reason: it is the number the exchange itself uses to size margins, so it is the market’s own operational estimate of each stock’s risk rather than one of many defensible academic constructions.
One trap: the field is stored as a fraction, not a percentage. An annualised volatility of 0.6725 means 67.25%. Writing a filter as annualised_volatility < 30 intending “under 30%” will pass every stock in the market. Unit errors of this kind do not throw an error — they produce a screen that silently does nothing, which is worse.
Weighting by the inverse of volatility
Having selected the calmest thirty names, equal weighting them undoes part of the work. Within a low-volatility basket there is still a three- or four-fold spread between the calmest name and the most volatile, and equal rupee weights mean the most volatile holding contributes several times the risk of the least.
Inverse-volatility weighting assigns each position a weight proportional to 1 / volatility, so risk contribution is roughly equalised across the book. This is the construction NSE uses for its Nifty100 Low Volatility 30 index — select the 30 least volatile Nifty 100 constituents, weight by the inverse of volatility, rebalance quarterly.
What you are actually buying
Two honest caveats, because low volatility is the factor most often oversold.
It lags badly in recoveries. A low-volatility portfolio is structurally underweight high-beta cyclicals, which are exactly what leads off a market bottom. In the twelve months after March 2009, and again after March 2020, low-volatility strategies underperformed the index substantially. If you deploy this expecting it to keep up in every regime you will abandon it at the worst moment.
It becomes a sector bet. Low-volatility screens naturally concentrate in FMCG, pharma and utilities and hold almost no metals, realty or PSU banks. Over a five-year window the strategy’s return will owe as much to that sector tilt as to the volatility factor itself. If you want the factor without the sector concentration you need to score within sectors rather than across the whole universe.
The reason to hold it anyway is behavioural rather than statistical. The strategy most people can actually stay invested in is the one whose worst drawdown they can tolerate, and drawdown tolerance is the binding constraint on real portfolios far more often than expected return is.
Try it
The Low Volatility 30 template selects the 30 calmest Nifty 200 constituents on NSE’s published annualised volatility, weights them inversely to that volatility, and rebalances twice a year to keep turnover low.
The comparison worth running: put it next to the Momentum 30 template over the same window and look at the two drawdown charts side by side. They bottom at different times. That non-overlap is the entire case for blending factors instead of choosing one.
Further reading
- Why blending factors beats picking one on what low volatility contributes alongside momentum and quality, and when it drags
- Volatility targeting: sizing is a more reliable lever than timing uses the same predictability of volatility at the portfolio level rather than in selection
- The quality factor: screening for businesses, not charts for the factor whose holdings overlap with this one most
Glossary: volatility, beta, inverse volatility weighting, minimum variance, factor investing.