A trip to Monte Carlo

I’ve been messing around with the Testfol.io Monte Carlo Simulator. The free tier is pretty crippled — 500 simulations, 15-year cap — but Testfol.io is literally the only place where we have a decent Trend Following proxy that actually goes back decades. Thirty-eight years of data isn’t exactly blowing my hair back, but it beats the hell out of the 26-ish years you’ll find everywhere else, and when you’re trying to stress-test a portfolio, you take what you can get.

Testfol.io lets you run 5-year blocks, which matters more than most people realize. It tries to preserve the mean-reverting nature of equities, basically keeping the simulation honest. That’s not me talking, by the way, that’s something I picked up from Cederburg.

Should I pay 15 bucks to run better tests? Maybe.

This was my starting configuration:

The idea – which I think I borrowed from Roger Nusbaum, but I am not 100% sure – is to have half stocks, half diversifiers, equally weighted. This is also close-ish to my portfolio, where I have more stocks than each diversifier.

The real equal-weight version, 25/25/25/25, performs better than this. But I guess we are all humans and I cannot fathom at this moment to go really there with my portfolio. To be fair, my stock allocation is 36% (before leverage), so I am closer to 25 than to 50? Anyway, all of this will become more relevant later in this post.

The results, on the whole, appear fairly robust. The downside scenarios are more contained than one might expect, and the key metrics show a notable degree of consistency across simulations. The ranges remain compressed rather than sprawling in ways that would undermine confidence in the analysis.

That said, intellectual honesty requires acknowledging the limitations here. This is not a deep-dive analysis, and I cannot rule out the existence of structural issues that simply didn’t surface in this particular exercise. Nothing alarming emerged, but absence of evidence is not evidence of absence. My hope is that the 50% equity allocation does some meaningful work in reducing mental tracking error, the psychological distance between what you’re holding and what “the market” is doing, while still delivering a portfolio with genuine resilience across scenarios.

One area where Testfol.io falls short is asset selection. The platform does an admirable job extending datasets further back in time than most competing tools, which is no small feat. But the universe of available assets, particularly when it comes to diversifiers, is narrow. For a tool this useful in some respects, that limitation is a real constraint…and a somewhat frustrating one.

A few, partial, observations:

• TLT seems to add value but the simulation is limited to 1988 due to KMLM. What if we add the ’70s? I cannot recall the authors, but there is a research paper that demonstrates (?) how the stock outperformance is…actually…just a recent outlier (I guess if you saw the study, you know what I mean). Even taking into account the grim outlook for duration – increasing debt levels, inflation, Dredge’s Hunger Games – I do not think it is correct to give up on it. Also, yields are responding to the outlook: the US 30-year is at 5%. I cannot say if 5% is enough to compensate for the risk, but at least it is not the 1% we had 6 years ago.
• TIPS do not outperform TLT in this analysis. The natural question, again, is whether the time horizon being examined is simply the wrong lens through which to evaluate this. That’s possible. But the result is also consistent with what theory would suggest. TIPS, by design, offer a form of insurance against inflation-driven sequence risk. And as with most insurance, you pay a price for that protection in the form of lower expected long-run returns. You are not supposed to come out ahead with TIPS over a long horizon. That’s not what they’re for. But the test is important because we do not care about the asset alone, we care about what the asset does in the portfolio.
• MTUM provides more upside but also more downside: not really worth it. But if I do a 25% MTUM 25% VRB (and I compared it to a 50% SPY allocation) the improvement is material. The issue for me, in pursuing this strategy, is that I would not be able to use any portable alpha instrument. But if you are scared about leverage, pushing into factors’ combinations seems the way to go (I mean, they won a Nobel prize for this, not really an original insight).
• ZROZ is the same as MTUM, works better than TLT when things are better.

Then by chance, I discovered this post on the LETFs subreddit. The user stress-tested the following portfolios:

to the following results:

The B4 portfolio is not that far from what I was testing (lil more bonds, lil less stocks) plus leverage.

What makes the post genuinely valuable is the rigor applied to testing the various allocations. The decision to use stacking ETFs like NTSX rather than leveraged products like SSO, for instance, meaningfully improves robustness across different rebalancing frequencies — a detail that matters more in practice than it might appear on paper. The walk-forward optimization further demonstrates that equal weighting (and I use that term loosely, given that the underlying exposures are anything but equal) holds up as well as any more sophisticated weighting scheme. The author also deserves credit for including an honest caveats section, which is rarer than it should be in this kind of analysis.

A word of caution, however. These are simulations and backtests, not guarantees. I would not stake anything meaningful on reliably achieving a 0.7 Sharpe ratio running these strategies in live markets. But the structural advantages are difficult to dismiss. The B4 portfolio, with four ETFs and no changes otherwise triggered only by a 10% weight drift, is genuinely low-maintenance once the underlying logic is internalized.

More importantly, this is a framework, not a prescription. The diversifiers, weights, rebalancing windows, and degree of leverage are all adjustable parameters. The architecture is robust enough to accommodate meaningful discretion and still outperform conventional portfolios including, notably, a 100% equity allocation, across a wide range of specifications.

What I am reading now:

Follow me on Bluesky @nprotasoni.bsky.social