The Boring Stocks Won: Clarke, de Silva and Thorley on Minimum-Variance Portfolios

There is one portfolio on the Markowitz efficient frontier that is special: the leftmost point, the minimum-variance portfolio. It is the combination of assets with the lowest possible volatility. And it has a wonderful property that Clarke, de Silva and Thorley built an entire research program around.

Building it does not require you to forecast returns at all.

Every other point on the frontier needs expected returns as an input. The minimum-variance point needs only the covariance matrix. Since expected returns are the hardest thing in finance to forecast, and since errors in them are the single most destructive input error an optimizer can receive, that is a very attractive property. You get to skip the hardest step.

What Clarke, de Silva and Thorley did in 2006 was actually build the thing, carefully, across the US stock market over decades, and report what happened.

The problem: theory says low risk should mean low return

Here is what standard finance predicts. The CAPM says return is compensation for bearing market risk. Take less risk, earn less return. A portfolio deliberately engineered to have the lowest possible volatility should therefore deliver correspondingly low returns. It should be a defensive, boring, low-return holding, the equity equivalent of parking in a savings account.

That is the prediction. Now check it.

The key idea via analogy: the tortoise that also finished first

The authors constructed minimum-variance portfolios from the large-capitalization US stock universe, rebalanced over a multi-decade sample, using a covariance matrix estimated with care. Because a naive covariance estimate on hundreds of stocks is a noisy disaster, they paid attention to how they estimated it, using structured approaches like factor models and shrinkage rather than raw historical covariances. This is not a footnote: the whole exercise stands or falls on whether your risk model is any good.

The result: the minimum-variance portfolios delivered substantially lower volatility than the capitalization-weighted market, while earning average returns that were comparable or better.

Read that again, because it is the whole paper. Less risk. Same or more return. Under the CAPM, that combination is not supposed to exist. It is the tortoise beating the hare, not by being slow and steady over a long race, but by simply being faster and slower at the same time.

The portfolios that came out of the optimizer were also interesting in themselves. They tilted, unsurprisingly, toward low-beta, low-volatility stocks: utilities, consumer staples, stable large companies. They tilted away from the glamorous high-beta names. And they achieved much of their risk reduction not through picking calm stocks alone, but by exploiting the correlation structure, combining stocks that offset each other. That is the part a naive "just buy low-volatility stocks" screen misses, and it is why the optimizer earns its keep.

Why it mattered

• It is a clean statement of the low-volatility anomaly. Others had found that low-beta and low-volatility stocks earn higher risk-adjusted returns than they should. This paper made the point in the language allocators actually use: here is a portfolio you could hold, here is its volatility, here is its return, and the trade-off is not what the textbook says.
• It launched a product category. Minimum-volatility and low-volatility equity funds and ETFs, now a very large chunk of the smart beta industry, descend directly from this line of work. The authors were practitioners as well as researchers, and the strategy went from paper to portfolio quickly.
• It gave an unusually robust way to use optimization. Because minimum variance sidesteps expected returns entirely, it avoids the single worst failure mode of mean-variance optimization. In the DeMiguel, Garlappi and Uppal horse race, minimum-variance strategies were among the better performers relative to 1/N for exactly this reason.
• It pointed a finger at the CAPM. If deliberately taking less risk does not cost you return, then the central prediction of the CAPM is not describing the world. This paper is part of the evidence stack that eventually produced explanations like Frazzini and Pedersen's leverage-constraint story: investors who want high returns but cannot use leverage bid up high-beta stocks instead, which makes high-beta stocks expensive and low-beta stocks cheap.

The honest limitations

• It lives or dies on the covariance matrix. The whole method is one giant bet that you can estimate how hundreds of stocks co-move. Get that wrong and the optimizer will concentrate in stocks that merely appear calm, which is exactly the estimation-error-maximizing behavior Michaud warned about. The careful risk modeling in the paper is doing enormous work.
• The portfolios have real, unhedged tilts. Minimum-variance portfolios end up loaded on particular sectors, particular styles, and typically on interest-rate-sensitive defensive names. They can behave like a bond proxy. When rates rise sharply, "defensive" equity portfolios can be anything but.
• Low volatility is not low risk. A portfolio can have low day-to-day volatility and still be exposed to a nasty tail. Volatility does not capture crash risk, and a strategy that is long boring stocks can be quietly short a rare disaster.
• The anomaly may be getting crowded. Since these results were published, enormous amounts of money have flowed into low-volatility strategies. Any anomaly that becomes a mass-market product risks having its own premium bid away, and there have been long stretches where low-volatility investing has been painfully out of favor.
• The comparison is to cap-weighting, which is a low bar in some respects. Beating the cap-weighted index on risk-adjusted terms is a real achievement, but many alternative weighting schemes (equal weight, fundamental weight, risk parity) also do so, and it is not obvious minimum variance is the best of them.

The one-line takeaway

Clarke, de Silva and Thorley built the lowest-risk portfolio the US stock market allows, using no return forecasts whatsoever, and found that it delivered substantially less volatility with returns that were comparable or better, a result that is a direct embarrassment for the idea that return is simply payment for risk.