Leverage For Retirement

I finally got around to properly playing with testfol.io’s Monte Carlo engine. With the Pro version, I was able to stretch the analysis out to a 40-year horizon and run 5,000 simulations. The underlying dataset only goes back 38 years, which isn’t ideal, but it’s enough to work with.

Here’s what I wanted to know: if you take a diversified portfolio and apply leverage to it, does it actually make retirees better off, or does it just add risk without adding much benefit?

Across every percentile I tested, the safe withdrawal rate (SWR) went up when leverage was applied, even in the worst-case scenario. At the 5th percentile (essentially the “things went about as badly as they could” case), the SWR improved by 71 basis points.

That might sound small. It isn’t. Think about how much ink has been spilled debating whether the classic 4% rule is even reliable anymore. A 71 basis point bump on top of that is roughly an 18% increase in your annual retirement budget. Ask yourself when you last got an 18% raise. That’s the scale we’re talking about here, and it shows up in the worst 5% of outcomes, not just the lucky ones.

And it’s not like this improvement came at the cost of portfolio quality. The Sharpe and Sortino ratios stayed essentially flat. In other words, you’re not paying for that extra withdrawal capacity with meaningfully worse risk-adjusted returns, at least not by these two measures.

Some of the tail outcomes do look alarming at first glance. But context matters a lot here. The single worst scenario for the extra-leveraged portfolio (“Portfolio 2”) had a max drawdown of 30% and a longest drawdown period of about 4.2 years. Both numbers are reasonable and manageable.

If we look at the “worst” numbers in each category, they all come from different paths. The path with the deepest drawdown (51%) must have had a relatively fast recovery. The path with the longest time underwater (18 years) must have gone sideways for a long time. It is only a hypothesis, testfol.io doesn’t allow me to look at each single generated path, but the combination of all the ‘ingredients’ cannot be worse…than the worst scenario.

That distinction matters more than it might seem. An 18-year real drawdown is a genuinely difficult thing to sit through, even if the portfolio never dropped that far below its peak. All of these figures are inflation-adjusted, by the way: in nominal terms, that 18-year drawdown was actually under 8 years.

I bring this up because the psychological side of this can’t be separated from the math. You’re trusting a process while you’re pulling money out of it, and every day you spend underwater is not a stress-free day. You can tell yourself the portfolio will recover eventually, but in the moment you’re never 100% sure. Maybe your trend-following sleeve (I used KMLM as the proxy) has stopped working. Maybe you’re living through another Global Financial Crisis and every voice around you is saying the system is broken for good. Those are legitimate questions to sit with when you’re the one living through the drawdown, not just backtesting it years later.

For what it’s worth, at a 5% inflation-adjusted withdrawal rate, the leveraged portfolio only ran out of money before year 40 in the single worst simulated scenario. The 1st-percentile cohort (i.e., still a bad outcome, just not the worst one) finished the 40-year period with roughly the same amount of money they started with.

IEF vs TLT

One limitation of the main simulation: it excludes the 1970s, because that’s as far back as the KMLM data goes. So I ran a separate, simpler comparison — just global stocks, bonds, and gold — swapping in IEF versus TLT as the bond sleeve, this time with a dataset that includes the ’70s.

The portfolio with IEF has a higher floor. What I did find useful is that swapping TLT for IEF in the original, leveraged portfolio only slightly hurt the results. Given that trade-off, IEF looks like the more responsible choice.

CRRY

Adding CRRY as an additional diversifier made the results look almost silly good. But I want to flag the obvious caveat: the dataset only runs from 2008 to 2026. There simply isn’t a long-history alternative strategy dataset available to check whether that holds up over a longer stretch, so take it with real skepticism.

Optimisation

I then run testfol.io optimisation engine using historical data (I lowered a bit CRRY return and increased its vol):

The goal was to find the portfolio with the highest CAGR at the 5th percentile. Unsurprisingly, I would say, the result was this one:

The candidate portfolio is an unlevered EW of the 5 ingredients:

Honestly, you don’t need to be a statistics genius to see why the optimizer landed there. If your portfolio components are uncorrelated and their returns are roughly normally distributed, leverage becomes a free lunch: mathematically, you come out ahead in almost every scenario. Of course, we all know how shaky those two assumptions get in the real world.

But… even though correlations tend to converge toward 1 during genuine stress events, this portfolio has what they call first and second responders built in. Looking back at past crises, correlations did spike, but there was consistently at least one component moving in the opposite direction while everything else fell apart. Adding a long volatility strategy on top of that should make the whole thing more resilient to exactly this failure mode.

Even though the backtest technically “survives” 3x leverage, I’d treat that more as weak reassurance that 2x is reasonable with some safety margin left over, not as a green light to run it at 3x for real.

If I had to boil this whole exercise down to one sentence: the original safe-withdrawal-rate research was pointed in the right direction, but it probably wasn’t bold enough.

A well-diversified portfolio gets you a long way on its own (say hi to Scott Cederburg, who in his own research treated a 99% drawdown as a “success”). But adding leverage on top of genuine diversification, along with a mix of inflation-reactive assets, appears to do meaningfully better than the standard all-stocks approach. And with newer instruments like MFEH, savers can now dial in the right amount of currency exposure too, rather than taking on FX risk as an unpriced side effect of certain instruments.

None of this is a reason to go max leverage tomorrow. It’s a reason to take the boring, diversified, moderately-levered version of this idea more seriously than it usually gets credit for.

What I am reading now:

Follow me on Bluesky @nprotasoni.bsky.social