What happened to our July picks

If you opened Numu this week, you saw a lot of red. We are not going to bury that in a careful paragraph three — here is what happened, plainly.

In the first five trading days of July, semiconductor and AI-hardware stocks sold off violently while the rest of the market barely moved. The S&P 500 was flat on the week. Several of our July picks — AMAT, KLAC, MRVL, LRCX, TER among them — fell 22% to 29%, much of it in a single overnight gap on July 1st. Because momentum strategies concentrate in whatever has been working, and what had been working was AI hardware, most of our July list was in the blast radius.

In backtests going back to 2010 — through COVID, through the 2022 bear market — we have never seen our picks fall this hard while the broader market stayed flat. This was, by that measure, the most extreme week in sixteen years of data.

Before anything else: we follow the same picks with our own money, every month, including this one. July hurt us the same way it hurt you. What follows is not damage control — it is what we found when we spent the week taking the event apart.

Could we have seen it coming?

We went back and re-ran every indicator we track — and a few classic ones we do not use in selection but that any experienced investor would reach for — against what the data actually showed on June 30th, the day the picks were selected:

Eight of the ten picks were at or within days of their 52-week high on selection day itself.
Every pick was trading above both its 10-day EMA and 20-day EMA — most by 10–24%, a sign of a strong, intact uptrend, not a weakening one.
SROC, our own momentum indicator, was still accelerating into month-end on every single name — not flattening, not diverging from price.
RSI(14) told a more mixed story, and it is worth being honest about it. Six of the ten picks were sitting above 70 — the textbook "overbought" line. If RSI-overbought were a reliable sell signal, this should have been a warning. But look closer: names like AMAT and KLAC had spent roughly a third of the prior three months above 70, flickering in and out, all the way up. Trading out every time RSI crossed 70 would have meant repeatedly selling a stock that kept climbing for weeks afterward — the well-documented reason professional momentum strategies do not use RSI-overbought as a standalone exit. It is a real, honest data point. It is also not, on its own, a warning anyone could have safely acted on.

Put together: no divergence, no fading momentum, no technical crack in the indicators that actually work for this style of investing. The crash launched from all-time highs, overnight, on sector-specific news. We also checked whether the moving averages would have caught it faster once the selloff started — they did not. Both EMA10 and EMA20 broke a full day after the −5% stop-loss had already triggered, at roughly double the loss. That is the uncomfortable truth about gap crashes: no indicator we tested, and none we know of, gives advance warning when a sector falls out of bed from its highs. Anyone who claims otherwise is selling something.

What actually protected people

Two things in the July list did their job.

The −5% stop-loss discipline. This is exactly the scenario it exists for. A buy-and-hold of the full July list is down roughly 17% as of this writing. Followers who honored the stop-loss exited the fallers in the first days of the month — in a gap-down you rarely get out at exactly −5%, but realistic exits kept the overall damage in the single digits instead of above twenty. If you have ever wondered whether the stop-loss rule is worth the discipline it demands, this month is the answer.

The picks that weren't in the cluster. The three healthcare names on the list — MRNA, MOH, CRL — were flat to up double digits through the same week. Diversification worked exactly as far as it existed. Which brings us to the real lesson.

The real lesson: it wasn't ten picks, it was one bet

The honest post-mortem is not "the market surprised us." Markets do that. The honest post-mortem is that on June 30th, seven of the ten picks came from a single, tightly correlated group. The portfolio looked diversified on the surface — ten tickers — but it was structurally one position: the AI-hardware trade continues. When that one trade cracked, seven picks fell together.

That part — unlike the timing — was visible in advance, and it is fixable.

What we are changing

We have spent the week testing a sector concentration cap: a rule that limits how many picks can come from any single sector, filling the remaining slots with the next-strongest stocks from elsewhere. Early results, tested across sixteen years of survivorship-bias-free data:

In months like this one, it roughly halves the damage before the stop-loss even kicks in.
Over the full sixteen years, it costs essentially nothing in long-term returns — in most configurations it actually improves risk-adjusted performance.
Combined with the existing stop-loss, the two protections cover different failure modes: the cap limits how many picks can fall together, the stop-loss limits how far any of them can drag you.

We are deliberately not rushing it. Changing how picks are selected is not a hotfix — it changes the strategy you have chosen to follow, and it changes the backtest history we publish. When the validation is complete, we will roll it out the way we do everything else: announced clearly, with the methodology explained and the history clearly versioned. Nothing will change silently.

Zooming out

One month — even a brutal one — does not change the arithmetic of the strategy. Momentum investing has never worked by avoiding bad months; it works by compounding good ones over years. Since 2020, through this crash, the strategy's backtested returns remain roughly double the S&P 500's on an annualized basis. The price of that has always been weeks like this one, and the strategy has always priced it in through discipline: mechanical monthly selection, the stop-loss rule, and no emotional overrides.

That discipline is also what we would gently point to now. The worst version of this month is the one where the losses are locked in by abandoning the process at the bottom — history's most reliable way to turn a drawdown into a permanent loss.

We built Numu to be the app we wished existed: transparent about method, honest about results, invested alongside you. That includes weeks where honesty means writing a post like this one. If you have questions about anything in here, reach out — and thank you for the trust you place in the process, especially this week.

— The Numu Team