Low Volatility and Minimum Variance: Construction, Interest-Rate Sensitivity and the Lottery Explanation

What you will learn

• The difference between a low-volatility ranking and a minimum-variance optimisation.
• How to calculate the two-asset minimum-variance weight and the resulting portfolio volatility.
• Why low-volatility portfolios tend to be sensitive to interest rates.
• The lottery-preference, leverage-constraint and benchmark-constraint explanations.

Explanation

The low-volatility anomaly is the observation that shares with low past volatility or low beta have earned returns similar to or higher than high-volatility shares, although theory says higher risk should be paid with higher expected return. Fischer Black noted the flat relation between beta and return in 1972; Baker, Bradley and Wurgler (2011) and Frazzini and Pedersen (2014) documented it across markets and decades. It is an anomaly because the risk-premium explanation does not fit: the "safe" stocks are the ones that did well.

Two construction methods are used, and they produce different portfolios:

• Low-volatility ranking: measure each share's realised volatility (standard deviation of returns over one to three years) or beta, and hold the lowest 20 to 30 percent, often weighted by inverse volatility. Simple and transparent, but it ignores correlations and tends to concentrate in a few defensive sectors.
• Minimum variance: an optimiser with a full covariance matrix finds the combination with the lowest expected portfolio volatility, subject to constraints such as maximum stock weight, sector and country limits and a turnover cap. Here a moderately volatile stock with low correlation to the rest can receive more weight than a low-volatility stock that moves with everything else. Widely used minimum-volatility indices work this way; as of 2026, check the current methodology for the exact constraints.

Both methods end up overweight in utilities, consumer staples, health care, insurers and listed real estate, and underweight in technology, energy and cyclicals. That profile brings interest-rate sensitivity: companies with stable cash flows are valued like long-duration bonds, so when yields rise their relative valuation tends to fall. For a low-volatility equity portfolio you can estimate an empirical duration by regressing its return relative to the market on changes in long-term government bond yields; published estimates vary with the period, so treat any single number as rough. The practical point: in a year of sharply rising rates, a low-volatility fund can lag the market substantially.

Why would a low-risk portfolio earn a fair or better return? Three explanations have research support. The lottery-preference explanation (Bali, Cakici and Whitelaw, 2011): many investors overpay for shares with a small chance of a huge gain, high volatility, high recent maximum daily return, low price, which pushes those shares' expected returns down and leaves calm shares relatively cheap. The leverage-constraint explanation (Frazzini and Pedersen, "Betting Against Beta"): investors who want higher returns but cannot or will not borrow buy high-beta shares instead, bidding them up; those who can use leverage are paid for holding low-beta shares with leverage. The benchmark-constraint explanation (Baker, Bradley and Wurgler): managers judged against a market-cap index avoid low-volatility positions because they add tracking error without adding expected return in their mandate, leaving the anomaly in place. All three suggest the effect could persist because the causes are structural; none guarantees it.

Worked example

Hypothetical example: two assets. Asset 1 has volatility σ₁ = 15% per year, asset 2 has σ₂ = 25%, and their correlation ρ = 0.4. Find the minimum-variance weights and the portfolio volatility. Fees and taxes are excluded.

The two-asset minimum-variance weight for asset 1 is w₁ = (σ₂² − ρσ₁σ₂) ÷ (σ₁² + σ₂² − 2ρσ₁σ₂).

• σ₁² = 0.0225; σ₂² = 0.0625; ρσ₁σ₂ = 0.4 × 0.15 × 0.25 = 0.015.
• Numerator: 0.0625 − 0.015 = 0.0475. Denominator: 0.0225 + 0.0625 − 0.030 = 0.055.
• w₁ = 0.0475 ÷ 0.055 = 0.864; w₂ = 0.136.

Portfolio variance = 0.864² × 0.0225 + 0.136² × 0.0625 + 2 × 0.864 × 0.136 × 0.015 = 0.01680 + 0.00116 + 0.00353 = 0.02148, so volatility = about 14.7%, below the 15% of the calmer asset alone because the second asset is imperfectly correlated. A 50/50 mix would have about 17.0%. Real indices do this with hundreds of assets and constraints, but the logic is the same: correlations, not just individual volatilities, decide the weights.

Common mistake

Practice activity

A hypothetical low-volatility fund has a market beta of 0.7 and, for this exercise, zero alpha. (a) Estimate its return if the market falls 30%. (b) Estimate its return if the market rises 20%. (c) Assume an empirical duration of 4, meaning it loses about 4 percentage points relative to the market for each 1 point rise in long-term yields. If the market rises 20% while yields rise 1 point, estimate the fund's return. State one limitation.

Key takeaways

• Low-volatility rankings ignore correlations; minimum-variance optimisation uses them, with constraints to keep the portfolio diversified.
• In two assets, the minimum-variance weight depends on both volatilities and their correlation, and the result can be calmer than the calmer asset alone.
• Defensive sector weights give low-volatility portfolios bond-like interest-rate sensitivity.
• Lottery preferences, leverage constraints and benchmark constraints are structural explanations; they support persistence but do not promise it.

Next step

The next lesson moves from theory to products: how MSCI and S&P build factor indices, screening versus tilting, and how to measure the exposure you really get.