Does the hedge actually work?
An out-of-sample test of the pure-play thesis against two documented 2026 market events, on real prices.
The question a peer reviewer asked
If these portfolios really group companies by shared risk, then when that shared risk materializes in the real world, a short on the portfolio should pay off — hedging a private holding exposed to the same risk. So we tested it: pick the most data-available portfolios, estimate how to weight a short from one window of real prices, then measure what that short would have returned through a later, independently-documented adverse event.
How the test avoids fooling itself
Regional Bank Net-Interest-Margin Compression
View this portfolio →The real event
Regional-bank credit-fraud & inflation selloff · event window 2026-02-20 → 2026-03-02
Zions Bancorporation disclosed a ~$50M charge-off and ~$60M credit hit from borrower 'misrepresentations and contractual defaults'; Western Alliance disclosed a fraud suit against a borrower; a hotter-than-expected January PPI (wholesale inflation +0.5% vs +0.3% expected) revived higher-for-longer rate fears. Regional-bank shares fell sharply across the board.
Shared risk factors this event materialized:
Private-side analogue: A private investor holding an interest in a private commercial lender or a loan fund exposed to the same borrower-credit and rate risk would have taken a simultaneous markdown.
Step 1 — Do these companies actually move together?
Using daily returns over a stable estimation window (2026-01-05 → 2026-02-13), before the event, the portfolio's companies show an average pairwise correlation of 0.77. High correlation is the precondition for a pure play: if they co-move, shorting the basket hedges a private holding exposed to the same risk.
Step 2 — Allocate the short from the covariance
From that same pre-event window, we compute minimum-variance weights from the covariance matrix — how much of a $100 basket to place in each name. These weights are fixed before the event (out-of-sample); the event-window column shows what each name actually did.
| Company | Weight of $100 | Return over event |
|---|---|---|
| Cullen/Frost Bankers (CFR) | 19% | -3.2% |
| Southside Bancshares (SBSI) | 16% | -2.2% |
| UMB Financial (UMBF) | 15% | -6.7% |
| Old National Bancorp (ONB) | 14% | -5.4% |
| Associated Banc-Corp (ASB) | 13% | -6.2% |
| Fulton Financial (FULT) | 12% | -5.1% |
| Renasant (RNST) | 10% | -3.1% |
Step 3 — Short the basket through the event
Holding those fixed weights as a short on $100 through the event window, versus shorting the market (SPY) over the exact same days:
(gain on $100)
(gain on $100)
the market
Basket fell -4.5% over the window while SPY moved +0.3%. The short captured the portfolio's shared risk firing — not market beta.
Mortgage-Rate Sensitive Homebuilders
View this portfolio →The real event
Mortgage-rate spike & housing-demand warning · event window 2026-04-28 → 2026-05-08
30-year mortgage rates pushed toward and past 7% into the spring selling season; builders including Lennar and KB Home flagged deteriorating demand, and housing data weakened. Homebuilder shares sold off.
Shared risk factors this event materialized:
Private-side analogue: A private investor in a local homebuilder, land developer, or building-products supplier faces the same rate-driven demand collapse.
Step 1 — Do these companies actually move together?
Using daily returns over a stable estimation window (2026-03-02 → 2026-04-10), before the event, the portfolio's companies show an average pairwise correlation of 0.85. High correlation is the precondition for a pure play: if they co-move, shorting the basket hedges a private holding exposed to the same risk.
Step 2 — Allocate the short from the covariance
From that same pre-event window, we compute minimum-variance weights from the covariance matrix — how much of a $100 basket to place in each name. These weights are fixed before the event (out-of-sample); the event-window column shows what each name actually did.
| Company | Weight of $100 | Return over event |
|---|---|---|
| D.R. Horton (DHI) | 17% | -7.3% |
| M/I Homes (MHO) | 16% | -1.7% |
| PulteGroup (PHM) | 16% | -8.4% |
| Lennar (LEN) | 13% | -5.2% |
| Toll Brothers (TOL) | 13% | -6.4% |
| KB Home (KBH) | 13% | -10.7% |
| Meritage Homes (MTH) | 12% | -6.6% |
Step 3 — Short the basket through the event
Holding those fixed weights as a short on $100 through the event window, versus shorting the market (SPY) over the exact same days:
(gain on $100)
(gain on $100)
the market
Basket fell -6.5% over the window while SPY moved +3.1%. The short captured the portfolio's shared risk firing — not market beta.
What this shows — and what it doesn't
In both tested cases, a documented real-world event that struck a portfolio's shared risk factors produced a profitable short on the pure-play basket, out-of-sample, while shorting the broad market over the same days did not. That is real, encouraging evidence that the portfolios capture genuine co-moving risk — exactly the property a pure-play hedge needs.
Caveats, stated plainly: this is two events on two portfolios, not a large-sample study; it uses only companies that still trade in 2026, so it cannot speak to delisted names; weighting and window choices affect magnitude; and past relationships need not persist. This page is a transparent experiment and a demonstration of method, not a performance guarantee or investment advice.
Reproduce it
The experiment is three Python steps in the repo's _build/: fetch real daily prices (06_backtest_fetch.py), compute correlation, covariance weights, and out-of-sample short P&L (07_backtest_analyze.py), and render the charts (08_backtest_charts.py). Event dates and triggers are sourced from contemporaneous financial-news reporting.