Skip to content
BiVelio joins the NVIDIA Inception Program NVIDIA Inception Program BiVelio is part of the ElevenLabs Grants Program ElevenLabs Grants

Article 7 · Leverage indicator

A leverage indicator for 2x and 3x ETFs: we built it before looking

An indicator preregistered before it was built and evaluated once on 2019–2026: it protected capital, but carried so little leverage that it grew little.

Read the full article on Substack (in Spanish)

An educational tool, not investment advice. Historical figures, computed with the Modeler's engine; past performance does not guarantee future results.

How much leverage did the indicator carry versus a fixed 3x?

0.68xaverage leverage with the indicator
8.0%a year with the indicator
33.3%a year with the fixed 3x
Index → average leverage with the indicator
Compound annual return in the test (2019–2026)

Each day it picks T-bills, 1x, 2x or 3x; on average it carried less than the unleveraged index, which is why it fell little and grew little.

How it's computed

Article 7 holdout test, Jan 2, 2019 to Sep 25, 2026: the frozen indicator, computed with the site's engine.

Main figure: matches the article From the article: The indicator's average exposure: 0.68 on the S&P 500 and 0.83 on the Nasdaq-100; the LRS's, 2.44. The site's engine, with the frozen indicator, gives the same figure (0.68x).

Open it in the Modeler

What happened to the leveraged long bond while the indicator sat in T-bills?

−20.5%a year with the fund
−93.7%its worst drawdown from a peak
2.8%a year with the indicator (in T-bills)
Long-bond fund → a year with the fund
Compound annual return in the test (2019–2026)

The long-bond indicator didn't hold bonds on a single day of the test: the drift it saw, cut by its standard error, was negative.

How it's computed

Article 7 holdout test, Jan 2, 2019 to Sep 25, 2026: the frozen indicator, computed with the site's engine.

Main figure: matches the article From the article: The indicator held no bonds during the whole test: it earned what T-bills did (2.8% a year) while TMF lost 20.5% a year and fell as much as 93.7%. The site's engine, with the frozen indicator, gives the same figure (−20.5%).

Open it in the Modeler

And the indicator's stock-and-bond portfolio versus the HFEA?

8.1%a year with the indicator's portfolio
−15.4%its worst drawdown from a peak
−70.7%worst drawdown with the 55/45 HFEA
Variance horizon → a year with the indicator's portfolio
Compound annual return in the test (2019–2026)

It applies the indicator to the S&P 500 and the long bond with UPRO and TMF; the HFEA always holds 55% UPRO and 45% TMF, rebalanced every quarter.

How it's computed

Article 7 holdout test, Jan 2, 2019 to Sep 25, 2026: the frozen indicator, computed with the site's engine.

Main figure: matches the article From the article: In the test it returned 8.1% a year with a maximum drawdown of −15.4%; the HFEA, 14.9% with −70.7%. The site's engine, with the frozen indicator, gives the same figure (8.1%).

Open it in the Modeler