Every night, four independent models read every name we cover. Before the US market open, we publish a Long or Flat call on each name and three Systematic Portfolios built only from the names the models agree on. This page explains what you receive, how it is built, why it has worked, and why it is hard to copy.
What you receive, and which service covers what
Two services, three portfolios.
1) Systematic Macro & Megacaps (MM)
The theme covers 91 names: 19 macro assets (index, sector, rate, credit, dollar, and Bitcoin ETFs), the Magnificent-7, 60 further megacaps and AI leaders, and 5 crypto-related names.
Subscribers to MM, or MM + Research, receive the MM and MM Crypto portfolios, plus Multi-model Signals for all 91 names. The Research tier adds the macro note and deep dives into portfolio holdings.
2) Systematic Commodities
The theme covers 35 names: 12 commodity ETFs across metals, energy, agriculture, natural gas and uranium, and 23 producers and miners.
Subscribers receive the Commodities (CO) portfolio and the signals for all 35 names.
Every trading morning, each service publishes the target book for each portfolio (every holding, its weight, the cash level), a Changes today block with one line per trade to make at the open, and the Multi-model Signals for every covered name (position, consensus, each model's call, 63 sessions of history, the next confirmed earnings date), by email, Google Sheet and website post. On Mondays, the Track Record page is updated with the live P&L.
The three Systematic Portfolios
MM is the core book: the whole US map, from index and rate ETFs to the megacaps and Bitcoin, in one long-only portfolio of at most ten names.
MM Crypto is MM plus a crypto sleeve. On top of MM's own rows, each of ETHUSD, SOLUSD, XRPUSD, MSTR and COIN holds 5% while its signal is long and broadly agreed, with the sleeve capped at 15% in total: when more than three qualify, the slices are scaled down pro-rata (four names at 3.75%, five at 3%).
The sleeve is funded first from cash, then from the weakest-ranked MM holding, and kept separate because crypto's volatility would otherwise dominate a risk-sized book.
Commodities (CO) is a separate book of at most ten commodity names. It tends to perform when equities don’t, making it the natural 25% partner to either equity book.
All three share one rule set.
A name is eligible only when its signal is Long and at least three of the four models agree.
Eligible names are ranked by how many models agree, momentum breaking ties.
Holdings are sized by risk, so each contributes a similar share of the book's risk, capped at 25% a name and by asset class.
The book is scaled to its volatility target: when recent volatility runs hot, every weight is cut and the difference sits in cash, so cash is a position and gross exposure never exceeds 100%.
The book is re-optimised once a week, with at most two names swapped in, and
a holding is sold the same day its signal turns Flat or agreement falls below three of four.
The four models
The four models read different evidence and excel at different things: one at context, one at speed, one at staying with a trend, and one at reading the season. Each also fails in its own way, so no single model decides a position; three of the four must agree.
M, the machine-learning model, reads the name's own price history together with the cross-asset picture: bonds, the dollar, credit, oil, gold and the volatility market. Its strength is context: it notices when a stock looks fine while credit and the dollar say otherwise, and is usually first to turn cautious before a broad sell-off. Its weakness is novelty: when a move runs beyond anything in its history, it stays cautious too long.
N, the neural-network model, reads the same evidence but learns it a different way, and it is the fastest of the four. Its strength is speed: it is usually the first to notice a holding deteriorating, and it triggers more of the book's same-day exits than any other model. Its weakness is restlessness: on its own it changes its mind too often, and many of its flips reverse within days, so it is never allowed to run a book alone; it counts as one vote.
T, the trend model, reads price only. Its strength is that it has no view on why, so it is never argued out of a trend; it keeps the book in a bull market after the others have grown nervous. Its weakness is turns and ranges: late at tops and bottoms, whipsawed when a market goes sideways.
R, the market-regime model, reads the state of the market as a whole, not the name. Its strength is the big picture: whether this is a season for holding risk at all, which pulls exposure down in 2020- and 2022-style markets. Its weakness is the individual name: right about the season, wrong about a specific stock, so it never acts alone.
Beside every call we print how many of the four agree: Broad when three or four do, Mixed when fewer. A Mixed long stays on the signals page and out of the books.
The process, in four steps
Run the models overnight for every covered name on the last end-of-day close.
Run the portfolio process: rank eligible names by agreement, size by risk, scale to the volatility target.
Build the book on re-optimisation day (the first session of each week); during the week, monitor for weakening signals and sell any holding that no longer meets the criteria.
Disclose and lock the record: the day's calls and books go to subscribers and are frozen, so the P&L is always trackable against a verified record.
Backtest results
The backtest runs from 21 June 2016 to 31 July 2026, the last session before the live record began, net of 5 basis points a side in costs, using only the data available on each day. The blends hold 75% in the equity book and 25% in Commodities, rebalanced quarterly. Sharpe is the annualised return divided by the annualised volatility.
Series | Annual return | Volatility | Sharpe | Worst drawdown | 2022 |
|---|---|---|---|---|---|
MM | 32.2% | 14.5% | 2.00 | -15.8% | -11.5% |
MM Crypto | 39.2% | 15.9% | 2.16 | -17.7% | -12.8% |
Commodities | 16.6% | 14.0% | 1.17 | -16.0% | +17.5% |
MM 75% + Commodities 25% | 28.6% | 12.6% | 2.06 | -10.3% | -4.6% |
MM Crypto 75% + Commodities 25% | 33.8% | 13.7% | 2.20 | -12.2% | -5.5% |
SPY (the S&P 500 ETF) | 15.3% | 17.1% | 0.91 | -32.0% | -18.7% |
Every book ran at lower volatility than the index with roughly half its worst fall. Commodities alone earns the least, but at 25% beside either equity book it cuts the worst drawdown by a third for a few points of return, because the two universes rarely have a bad month together; 2022 is the clearest case. MM Crypto adds about 7 points of annual return to MM for about 2 points of drawdown. A backtest cannot promise the next ten years; the live record is kept separately and never merged with it.
The record
Every call is locked the day it is made: positions, weights and the day's post freeze when they go out, a correction is a new dated entry beside the original, and red weeks stay on the page. The live P&L starts at zero on 3 August 2026 (MM Crypto on 5 October 2026) and is built only from the books that were published, priced at the open they were published for.
The Track Record page shows each portfolio's live return, volatility, worst drawdown and Sharpe beside the backtest, a monthly grid, and a blotter of every trade against SPY. Nothing is overruled by hand: a rule change is tested on every book, written up and announced before it takes effect.
Why it works, and why you cannot rebuild it at home
The edge is not a secret indicator. It is four models that disagree in useful ways, a consensus rule that waits for three of them, risk-based sizing, a volatility target that makes cash a position, and a weekly cadence with same-day exits.
Each part is simple, but together they have compounded at roughly twice the index with half its worst fall, with no human in the way.
An AI assistant can explain all of this; it cannot run it. The machine-learning and neural-network models are trained on a decade of cross-asset history and refit on a rolling schedule; the regime and trend models are calibrated per name; the construction rules were chosen by testing thousands of variants on frozen data.
Reproducing it requires clean daily data for 126 names, the trained models, a pipeline that runs and self-checks every night, and a locked record proving the calls came before the prices moved. A prompt gives you an opinion. This gives you a tested process, a verified record and a book you can act on at the open.
What this is not
Not a managed account or personalised advice. The portfolios are the models' own tracked positions: concentrated, long only, unlevered, holding cash on purpose, not tax-efficient. If you want intraday trades, a story for every move, or a 10x in a month, this service isn't for you.
Glossary
Term | Meaning |
|---|---|
Published position | The day's call for a name, Long or Flat, locked when published |
Consensus, Broad / Mixed | How many of the four models agree: three or four is Broad, fewer is Mixed |
Eligible | Long in the published record and Broad; the only names a portfolio can hold |
2x tier | A higher-risk sizing on the signals page; not used by the portfolios |
Re-optimisation | The weekly rebuild of a book on the first session of the week |
Carry-forward day | Any other day: yesterday's book, with same-day exits applied |
No-trade band | A held name within 5 percentage points of its target is not traded |
Volatility target | The level of risk each book is scaled to; when recent volatility runs above it, weights are cut and the difference sits in cash |
Gross | The share of the book invested; the rest is cash; never above 100% |
Sleeve | MM Crypto's capped crypto allocation on top of MM's rows |
Publish lock | The freeze on the day's positions, weights and post; corrections are new dated entries |
Backtest | The locked construction run over history on data available on each day, net of costs; never a track record |
CAGR | Compound annual growth rate |
Sharpe | Annualised return divided by annualised volatility |
Worst drawdown | The largest fall from a running peak |
DISCLAIMER: This article is intended for educational purposes only. Any information or analysis in this article does not constitute an offer to sell or a solicitation of an offer to buy any securities. Nothing in this article is intended to be investment advice, nor should it be relied upon to make investment decisions. Any opinions, analyses, or probabilities expressed in this article are those of the author as of the article's date of publication and are subject to change without notice.