Beta Lies in Drawdowns: How to Measure Real Downside Co-Movement

If you size positions using the beta on your broker's quote page, you are using an average that quietly assumes up-days and down-days are the same. They are not. The single most damaging property of a stock — its tendency to fall harder when the market falls — is exactly what a full-sample beta dilutes away.

This piece walks through why reported beta misleads in drawdowns, what "real" downside co-movement looks like, and three concrete calculations you can run in a spreadsheet this afternoon.

Why reported beta understates drawdown risk

Reported beta is the slope of a regression of a stock's returns against the market's returns, usually over 2-5 years of daily or weekly data. It mixes every market state — quiet uptrends, choppy ranges, vol spikes, crashes — into a single number.

Two problems follow from that.

First, beta is an average, and averages hide tail behavior. A stock that tracks the S&P loosely on normal days but moves 1-for-1 in crashes can carry a reported beta of 0.9. That number looks defensive. The behavior it describes is not.

Second, correlations rise in drawdowns. This is one of the most robust empirical findings in finance: when the market drops sharply, cross-asset correlations converge toward 1. Diversification benefits are largest when you need them least. Reported beta, computed across all days, cannot see this asymmetry.

The practical consequence: a portfolio engineered to a target beta of, say, 0.8 using trailing regressions can deliver an effective beta materially higher during severe drawdowns — the only market environment that actually matters for sequence-of-returns risk.

Downside beta and semi-correlation

The simplest fix is to compute beta conditionally — only on days when the market was down. This is usually called downside beta (sometimes "bear beta").

The recipe:

Pull daily total returns for the stock and a benchmark (SPY works) over 3-5 years.
Filter to days where the benchmark return is negative.
Run the same regression: stock return = alpha + beta × market return.

The slope coefficient is the stock's downside beta. Run the same calculation on positive-market days to get upside beta. The ratio (downside beta / upside beta) is your asymmetry factor.

A stock with an asymmetry factor above 1.2 captures more of the downside than the upside relative to the market. A factor below 0.9 is the profile you actually want — participates in rallies, lags drawdowns. Most high-beta growth names sit in the 1.1-1.4 range, which is why they feel worse to own than their reported beta suggests.

Semi-correlation is the same idea applied to correlation rather than beta. Compute the correlation of the stock and market only on down-market days. Compare it to the unconditional correlation. The gap tells you how much your reported diversification benefit evaporates when the tape turns.

Tail dependence and the 5% threshold

Downside beta still treats every down day equally. A market that's off 30 basis points is not the regime you're protecting against. What matters is the left tail.

A cleaner cut: conditional beta in the worst decile (or worst 5%) of market days. Same regression, but the sample is restricted to the bottom 10% or 5% of market returns by magnitude.

This is closer to what statisticians call lower tail dependence — the probability that a stock posts a tail-loss given the market posted one. You don't need a copula or a PhD to approximate it. Two simple proxies work:

Hit rate: of the worst 20 market days in your sample, on what fraction did the stock close in its own worst quintile?
Tail beta: the beta computed only on those worst 20-30 market days.

During periods of severe market stress, many "low-beta" staples and utilities posted tail betas materially above their reported numbers. Conversely, some single-stock stories with reported betas above 1.4 — think regulated utilities mid-litigation, or special-situations names — actually decoupled in crashes because their idiosyncratic risk dominated.

Practical workflow for prosumer investors

You don't need Bloomberg. You need a benchmark, daily total return data (Yahoo, Stooq, or your broker's export), and a spreadsheet. For each position you care about, compute and store four numbers:

Full-sample beta (the sanity check against what you see quoted).
Downside beta (market < 0 days).
Tail beta (worst 5-10% of market days).
Asymmetry factor (downside / upside).

Do this on a 3-year rolling window. Re-run quarterly. Patterns become obvious fast: certain "defensives" carry tail betas above their reported beta; some growth names carry tail betas below their reported beta because their drawdowns are idiosyncratic, not market-driven.

For portfolio-level work, weight each position's tail beta by its dollar weight. The portfolio's weighted tail beta is a far better input to position sizing than the weighted reported beta.

What to watch next

Run the four-beta table on your five largest positions this week. If your weighted tail beta is materially above your reported portfolio beta, you are taking more drawdown risk than you think.
Check the asymmetry factor on any position you hold for "defensive" reasons. A staple, utility, or healthcare name with an asymmetry above 1.1 is not doing the job you bought it for.
Re-run after every vol regime change. Tail betas computed in a calm 3-year window will understate risk; include periods of high volatility in your sample whenever possible.
Stop optimizing to reported beta in screeners. Use it as a starting filter, but never as the final risk input.