Introducing Multimodel Signals (June 25, 2026)

Hi YXI friends,

For the last year, every Risk On/Risk Off signal has come from a single machine-learning model. It served us reasonably well (although sometimes also with frustration).

But one model, however good, sees the market through one lens.

Today, I want to introduce its replacement: Multimodel Signals (“MS”).

I plan to launch MS next week (w/e June 29, 2026). MS brings four independent models to every name. For each name, it deploys the combination that has earned its place there. Sometimes that is one model. More often, it is a blend of two or three. On a few names, it is all four.

This is the biggest upgrade to the signal book since launch. Here is how it works, in plain terms.

DISCLAIMER: This newsletter is intended for educational purposes only. Any information or analysis in this note does not constitute an offer to sell or a solicitation of an offer to buy any securities. Nothing in this note is intended to be investment advice, nor should it be relied upon to make investment decisions. Any opinions, analyses, or probabilities expressed in this note are those of the author as of the note's date of publication and are subject to change without notice.

The original signal came from a machine-learning model. Call it the pattern reader. It digests dozens of inputs at once and learns what tends to come before a move. Where it works, it works well.

But one model has one point of view. When a trend runs further than history says it should, it can lean the wrong way, and nothing checks it. We saw this in steady megacap uptrends. The model would turn cautious into a rally that simply kept climbing.

The fix was not a cleverer single model. It was a team of models.

Each model thinks in a genuinely different way. That difference is the point.

This is the evolution of our original engine. We have made upgrades to its inputs and parameters since Q1, but the core model logic remains.

It reads the name's own price behaviour, such as momentum and volatility. It also reads signals from other markets, such as bonds, the dollar, and credit. It spots links across many markets at once, the kind a human would miss.

It breaks a price series into its underlying trend and its repeating rhythms, then projects them forward. Different mathematics to the ML model above, but a genuinely independent read.

When a real move is in place, it stays with it. When the price is choppy, it steps aside. No cleverness about tops and bottoms, just staying on the right side of a trend.

It reads the market's hidden state, risk-on, risk-off, or neutral, like telling you the season.

None of the four is the whole truth. Together, they get a lot closer.

Think of the four models as candidates. For every name, we test each one alone and in combination. We judge them on data they were not trained on. That is the honest test of a real edge, not a model that has simply memorised the past.

The Machine-Learning model still does much of the heavy lifting. It is in the winning line-up for most names. The Neural-Network model plays a supporting role, stepping in where it has proven more useful than the others. The principle is simple: the right tool for each name, chosen by evidence.

Whichever line-up drives a name, you always see how all four models are positioned that day, shown as four dots on the page, one per model. They are a transparency layer. You see the full read behind every signal, including where the models disagree.

The pattern of agreement can be telling in its own right. Markets are made of players running different styles. When the models all swing the same way, it often means a name is already stretched and positioning has become one-sided. That is especially true of an all-out risk-off. It is exactly the setup from which sharp reversals come. We have seen an all-out bearish consensus snap back hard more than once. Traders call that a short squeeze. The daily commentary is there to flag those moments.

This is the idea worth taking away. Because the names we cover across a range of asset classes behave differently at different times, their off-days rarely line up. Weakness in one corner is often offset by strength in another. You do not need every name to work. You need the book to work.

We run the system across our full book, meaning we cover all the names at once. We are also expanding our coverage over time, with names spanning equities, crypto, commodities, rates, and FX.

Shorting is simply too difficult and not worth our time most of the time.

Having examined our live performance using leverage and short strategies and performed extensive testing, I have decided to keep the leverage strategy but remove the shorting strategy.

1-Year walk-forward backtest

Here is a one-year walk-forward backtest of MS (aggregated across all tickers), from June 2025 to June 2026. Walk-forward means each model only used data available at the time. The signals never saw the future.

Treat it as illustrative, not a forecast. Next year's market will look nothing like this one. What it shows is how the diversified approach behaves across the whole book.

Over that year, the MS signal against simply buying and holding:

There is also a 2× leverage option for subscribers who want to lean in harder. The leverage is not applied across the board. It only steps in when all four engines agree, and volatility is calm. Selective by design.

Over the same year, with selective leverage:

The point is not just the higher return. It is the quality of it. Nearly double the return for the risk taken, and roughly half the drawdown. That is what diversification across models and names is meant to deliver. A steadier ride, not a lucky streak.

One honest note. This is a backtest, and we label it as such. The live MS track starts fresh next week. That is the number that ultimately counts, and you will watch it build in real time.

Position. Our stance on the name. LONG means we are in the market at a normal size. FLAT means we are out. To be clear, FLAT is not a short. It simply means no position. In plain terms, this is the Risk On or Risk Off call.

A name can also show 2×, our selective leverage. It only appears when all four engines agree and volatility is calm. When the engines agree but volatility is not calm, leverage stays off, shown as "no lev." That is deliberate risk discipline.

Multi-model consensus. How broadly the four engines agree with today's call, whether long or flat. "Broad" means three or four of them back it. "Mixed" means a narrower split. The colour shows the direction: green for long, grey for flat. This is the context for the call, not the call itself.

Engines (M N T R). The four models behind the call, shown as four dots.

A green dot means that the engine backs a long position. A grey dot means it backs flat. A hollow dot means it dissents from today's call. This is full transparency. You always see where the models agree and where they split.

Signal history (3M). A bar showing the last three months of the position. Oldest on the left, newest on the right. Green is long, grey is flat. A dashed line marks the cutover on 2026-06-29. Everything to its left is a backtest. Everything to its right is the live track. Over time, you will watch the live side grow.

Today's action. What changed, if anything. "Maintain Long" or "Stay Out" means the stance is unchanged. "New Entry" or "Exit Position" means we flipped today. "Leverage → 2×" or "Deleverage → 1×" means a change in size. A small note such as "M flipped long" tells you when one engine changed overnight.

Below the table, each name gets a short written commentary. A few plain sentences on what is driving the call: the model read, the trend, the regime, and how firm the setup looks. Use it for the why behind a position, not just the what.

A quick word on expectations.

MS is systematic. It reacts to evidence, not to a forecast. So it will not catch every move, and it is not built to call exact tops and bottoms.

The hardest case is a sharp V-shaped reversal. As a market falls, the system often exits to flat to limit the damage. It steps back to long once the turn is confirmed. That means it sits out part of the drop, but it also gives up the very bottom and the first leg of the snap-back.

This is a deliberate trade-off, not a flaw to fix. We accept missing some of the rebound in exchange for sidestepping more of the fall. Over many cycles, that is the better deal. In any single V, it will look imperfect, and no model can capture every tick.

What it does do is keep you on the right side of the big moves more often than not, with less pain along the way.

In Q3, I will focus on bringing in a number of new equity names across the AI value chain and in commodities. I see both themes going hand in hand over the next 3-5 years, and I want to broaden the choices for our subscribers, especially those who cannot trade ETFs easily.

We will conclude the chapter on our existing ML model approach next week, with an overview of the performance, strengths, and weaknesses. It will be publicly available to everyone.