LETFs: Optimally Harvesting Decay

[WITH CODE] An in-sample historical test of paired LETF decay strategies, assumptions, and limits.

Reviewed and updated 28 July 2026

Hello!

Welcome back to (likely) the last post in the LETF / volatility decay series. This one caps off the series with a historical simulation that produced a source-style risk-adjusted ratio of 2.64 (and a “dirty” ratio of 6.24) with annualized volatility of 1.28%. Using a conventional daily excess-return Sharpe calculation, the same result is about 2.44. The sample runs from July 23, 2024 through June 30, 2025.

Thank you all for the support over the last few posts. This topic may have been the most interesting one yet to investigate, and it makes me glad to see that it interests you all as well.

A research companion with the code, setup, results, and limitations is available to paid members below.

Let’s get into it.

---

Finding Optimal Pairs

Last post, we analyzed whether “Dual Shorting” (shorting both bull and bear LETFs) is more effective than shorting a bull LETF against its underlying index.

In the earlier simulation, shorting bear LETFs was less attractive under the assumptions used there, and the major-index pairs produced weaker results. Those observations motivated looking for other pairs; they did not isolate the cause or establish that the relationship persists.

By switching focus to the biotech sector, the earlier post reported higher source-style and “dirty” source ratios. Those are historical labels tied to the formulas used in the series, not conventional Sharpe estimates without further qualification.

However, I speculated that there might be a stronger pair in the available sample. The pairs below produced stronger in-sample results under the same modeling assumptions.

The exploratory notebook downloaded 19 tickers, so this was a broader search rather than a pre-registered three-pair test.

---

Harvesting Additional Rate Decay

Borrow cost, short availability, and liquidity are practical selection concerns. This simulation fixes annual borrow at 4% and does not test a causal relationship between AUM, borrow cost, and tracking error.

I will go over the strategy details below, but first I want to outline the economic hypothesis and why the simulated volatility might be low. The backtest tests the net ETF return gap rather than decomposing the mechanism.

The hypothesis focuses on volatility decay and financing-related return differences. The simulation does not decompose P&L into those components; it measures the net historical return gap among the ETF legs under the stated assumptions.

We have discussed volatility decay in great detail. The hypothesis also considers financing and implementation costs in leveraged exposure. That background motivates the test below, but the simulation does not measure those components separately.

Leveraged funds can obtain exposure through derivatives whose financing and implementation costs vary by product and over time. This backtest includes no product-level holdings or individual swap-financing schedules.

Futures basis and contract rolls can influence a fund’s return, but their direction and magnitude depend on the curve, holdings, and implementation. The simulation below contains no futures curve or holdings series, so it does not measure that contribution.

A high positive basis motivated testing the crypto pairs. It is not measured in this package, and the observed ETF return gap is not evidence that the strategy harvested a specific speculative premium.

In the next section, I test high-volatility crypto pairs using allocations that are first-order beta-matched at each rebalance. The historical simulation compares their net ETF return gaps; it does not establish exact delta neutrality or isolate why the result differs from earlier pairs.

---

Optimal LETF Pair for Harvesting Decay

Now, let’s go into the details of the strategy, and how I came across it. Once again, this section of the first post explains the strategy details more thoroughly.