Betting Against Beta

Overview

The Capital Asset Pricing Model makes a clean prediction: riskier stocks, meaning higher-beta stocks, should earn higher returns. Take more market risk, get paid more.

The data says something more awkward. Higher-beta stocks do earn slightly higher raw returns, but nowhere near enough to justify the extra risk. On a risk-adjusted basis, the relationship inverts: low-beta stocks have historically delivered better Sharpe ratios than high-beta stocks. The line that should slope up is close to flat, and sometimes tilts the wrong way.

Betting Against Beta, from Frazzini and Pedersen, turns this into a trade. Buy low-beta stocks, short high-beta stocks, and then, crucially, lever the low-beta side up so that both legs have the same market exposure. Without that leverage step you have simply bought a defensive portfolio. With it, you have isolated the anomaly itself.

Thesis (why the edge exists)

The explanation is elegant and, unusually for a factor, mechanically plausible.

Imagine you want an expected return higher than the market's. The textbook answer is: buy the market and borrow to lever it up. That is the efficient way to do it.

Now imagine you are not allowed to borrow. Mutual funds face leverage limits. Pension funds have mandates. Many institutions simply cannot use margin. Retail investors mostly do not.

You still want the higher return. What is your only remaining option? Buy the riskiest stocks you can find. Skip the leverage and get your risk from the securities themselves.

That is exactly what happens, at scale, across the whole market. The result is chronic excess demand for high-beta stocks and chronic under-demand for low-beta stocks. High-beta gets overpriced, low-beta gets underpriced, and the relationship between beta and return flattens out.

The people who can borrow, hedge funds and other levered investors, are supposed to arbitrage this away. They cannot fully, because their own funding is limited and, more importantly, because their funding tends to dry up exactly when the trade goes against them.

That last point is the whole risk of the strategy, and it is not a footnote.

Strategy logic

• Estimate beta. Regress each stock's returns against the market. The refinement that matters: estimate volatility from a short window and correlation from a long window, because correlation is far more stable than volatility. Then shrink the result towards 1 to tame the noise.
• Rank and split. Sort all stocks by beta. Everything below the median goes in the long candidate pool, everything above goes in the short pool.
• Weight by extremity. Within each leg, weight by how far the stock is from the median beta. The lowest-beta names get the biggest long weights.
• Equalise the betas. This is the step people skip and it is the step that defines the strategy. Compute each leg's portfolio beta. Then scale the long leg up (with leverage) until its beta is 1, and scale the short leg down until its beta is also 1. Now the two legs cancel each other's market exposure exactly, and what remains is a pure bet on the flatness of the beta-return line.
• Rebalance monthly and re-equalise, because betas drift.

Parameters (knobs)

• Beta estimation window: 1 year of daily data for volatility, 5 years for correlation, is the paper's recipe. Shorter windows adapt faster and are much noisier.
• Shrinkage. Shrink estimated beta towards 1, typically with weights around 0.6 on the estimate and 0.4 on the prior. This meaningfully improves out-of-sample stability.
• Leverage cap. Uncapped, the long leg can require 1.5x to 2x gross. In practice you cap it, and capping it reduces both the return and the chance of ruin.
• Universe. Large cap only (safer, lower return) or full universe including small caps (higher return, worse borrow and liquidity).
• Neutrality. Beta-neutral only, or additionally sector-neutral and dollar-neutral. Sector neutrality removes a large accidental bet.

Portfolio construction

The construction is where the strategy lives, and it is unusual: your leg sizes are determined by your beta estimates, not by conviction. If the low-beta leg has a portfolio beta of 0.65, you must run it at roughly 1.54 times capital to get its beta to 1.

That leverage is the entire point, and it is also the entire danger. You are being paid for doing something that other people cannot do, which means you are being paid for bearing the risk of being forced to stop doing it.

Practical guardrails: cap gross leverage well below what the pure recipe demands, hold a cash buffer sized for a margin call in a stress scenario, and monitor the borrow situation on the short leg continuously.

Costs, capacity and turnover

Three costs, and papers routinely underweight all three.

Financing. You are borrowing to lever the long leg. That has a real rate attached, and the rate is not the risk-free rate, it is your broker's rate. In a rate environment where borrowing costs 5 percent, a 0.5x leverage overlay costs you 2.5 percent a year before you have earned anything.

Borrow. The high-beta short leg contains volatile, speculative, heavily shorted names. Borrow fees there can be brutal, and availability disappears exactly when you most want the short.

Turnover. Betas drift, and monthly re-equalisation means constant resizing. Expect turnover well over 100 percent per year.

Capacity is decent in large caps and constrained in the small-cap short leg.

Backtest design checklist

• Beta estimation must be out of sample. Estimating beta with data from the holding period is a look-ahead that will make the strategy look far better than it is.
• Financing cost. Model it explicitly as a drag on the levered long leg. A backtest without financing cost is not a backtest of this strategy.
• Borrow cost and availability. At minimum, exclude names that were hard to borrow. Ideally, charge a realistic fee.
• Margin call simulation. Build a scenario where the market drops 20 percent in a month and your leverage must be cut. See what forced deleveraging does to the return path.
• August 2007 and March 2020. Two events where levered, crowded quant books deleveraged simultaneously. Look at those windows specifically.
• Junk rallies. The high-beta short leg gets destroyed in the months after a market bottom. Isolate mid-2009 and the second half of 2020.
• Compare to unlevered low-beta. Run a simple long-only low-beta portfolio alongside. If the levered long-short does not clearly beat it after all costs, the complexity is not paying for itself.

Common failure modes

• Forced deleveraging. The signature failure. Your funding tightens exactly when the trade is losing, you are forced to cut, and you crystallise the loss at the worst possible price. The strategy is short liquidity by construction.
• Junk rallies. Coming off a bottom, high-beta trash rips. Your short leg bleeds and your defensive long leg lags.
• Beta estimation error. If your betas are wrong, your legs are not actually neutral and you are carrying an unintended market bet.
• Crowding. BAB is well known and widely run. Crowded levered trades unwind together.
• Costs eating the edge. Financing plus borrow plus turnover can consume most of the theoretical premium. This is not a hypothetical, it is the usual outcome for anyone not operating at institutional cost levels.

Variants

• Unlevered low-beta long-only. Drop the leverage and the short leg. Far more robust, far less exciting, deployable by anyone.
• Sector-neutral BAB. Rank betas within sector. Removes the implicit bet on defensive sectors.
• BAB plus quality. Low-beta junk exists and it is dangerous. Screening the long leg for profitability materially improves the trade.
• Idiosyncratic-volatility version. Rank on residual volatility rather than beta. Related but not identical, and it does not require the leverage machinery.
• Capped-leverage version. Cap gross at 1.3x and accept the resulting imperfect neutrality. This is what most real implementations do.

Our notes and suggestions

This is the most intellectually satisfying anomaly in equities because the explanation is not "investors are dumb", it is "investors are constrained", and constraints do not go away when they are published.

But be very clear about what you are signing up for. You are earning a premium for supplying leverage to a market that wants it and cannot get it. The price of that premium is that you will be forced to stop supplying it at the exact moment supplying it is most valuable. That is not a bug in the strategy, that is the strategy.

If you cannot survive a forced deleveraging, run the unlevered long-only low-beta version instead. It is a large fraction of the benefit with none of the ruin risk, and the honest answer for most people building this is that the unlevered version is the one they should trade.