Holding 3x Nasdaq only when the market allows it

The problem with holding TQQQ

A 3x fund resets its leverage every day. In a calm market that works for you. In a volatile one it loses value even when the index goes nowhere. The drag is roughly three times the index's variance each year, so at 40% volatility the fund bleeds about 48% a year whichever way the market moves.

Held through 2000–02, a TQQQ-like fund fell 99.98% and never came back. Over 41 years it ended about where unleveraged QQQ did, having taken three times the risk to get there. The question with leverage is when to hold it, and the conditions that wreck a leveraged fund are easier to see while they are happening than to forecast.

What the rule does

After each close it checks the Nasdaq-100 for three things: how volatile the market is, whether volatility is rising sharply, and how far the index has fallen from its high. When all three are within their limits it holds TQQQ. When any one is not, it holds 20-year Treasuries. A change has to hold before the position switches.

The exact measures and limits aren't published. They were chosen by a search that re-ran every January using only data available at the time, and it has picked the same settings every year since 2009.

What the money sits in while the gates are shut

The rule is out of TQQQ about 40% of the time, so the defensive asset matters almost as much as the gates. Everything below uses identical signals and dates. Only the asset held while out changes.

Gold was tested in the same slot and rejected. From 2000 it looked better than Treasuries, but that window starts at the bottom of gold's twenty-year bear market. Extended back to 1986, gold added only 1.3 points a year while pushing the worst drawdown from −46.8% to −56.3%, and from 1986 to 2000 it cost 5.2 points a year. Original study: T-bills while out of the market, switching at the close the signal is read on.

How it was tested

About 6,000 variants were searched to find this rule, which makes any single backtest suspect. These are the tests that searching can't fake.

Where it fails

The worst drawdown is still about half the account, twice in 41 years. That beats losing everything, but it isn't safe, and the slowest recovery took well over a year.

Before 2010, TQQQ is reconstructed. It tracks the real fund at 0.9989 daily correlation since 2010, but it assumes the fund could always borrow at the T-bill rate plus 0.45%. A real funding crisis would break that, and pre-2010 is 60% of the sample.

It beat holding TQQQ in 18 of 42 calendar years and lost money in 8. All the outperformance comes from the few years it avoids a crash. In an ordinary bull year it lags, so you pay for the insurance in most years.

Tax cost the base rule about 10.7 points a year in the original study, because nearly every gain is short-term. Buy & hold gave up 0.7. It belongs in a sheltered account.

It reads closing prices, so an overnight collapse is taken at full exposure. It reduces the number of crashes you sit through; it doesn't make any single one survivable. Unemployment and credit-spread indicators went untested. And 41 years holds perhaps five independent bear markets, under one long regime of monetary policy.

A calmer variant

The same gates can size the position down as volatility rises instead of switching all or nothing. It gives up about half the return for a much calmer ride, and beats unleveraged QQQ with a third of its drawdown.

Taxable accounts

Buy & hold never realises a gain, so it compounds untaxed and settles once. The switching rule realises almost everything short-term. Slowing the rule down to trade less made things worse: after-tax CAGR fell from 24.6% to 17.5%, and checking monthly instead of daily cut it to 16.9% with a deeper drawdown. Missing crashes costs more than the tax it saves.

What works is keeping the gates and changing which exits you take. When the gates say exit and the position is sitting on a short-term gain, the taxable variant holds until the gain turns long-term, unless the drawdown gate has fired, which always overrides. Losses are never deferred.

Data

Nasdaq-100 index levels from 1985 with an estimated 0.3% dividend, then QQQ total return from March 1999. TQQQ is three times the daily index return less financing at the T-bill rate plus 0.45% and a 0.95% fee, replaced by the real fund from February 2010. The Treasury leg is a 20-year bond rebuilt from the 10-year yield, less a 0.15% fee. Re-run tables use the site's execution: signal at the 4pm close; sell the old position after hours at about the close; buy the new one at the next open. 5bp per switch.