A rules-based, multi-asset trend-following book across 23 futures-referenced markets. The research, data and execution stack is complete and running daily. What it does not yet have is a live track record, and this page is clear about what that means for the results.
26 Dec 2018 – 10 Aug 2026, 2,312 trading days. Costs are modelled at retail spread-bet rates (spread, slippage, financing): about 12 points a year, measured on the current book from April 2025 to September 2026. A closely related book earned about 19.4% a year before costs over a shorter period, so costs take roughly two-thirds of gross return. Cutting them is the largest single lever in the research roadmap below.
| Worst S&P months | S&P 500 | Book | Peer CTAs |
|---|
The six worst S&P 500 months in the backtest period. The book was positive in all six. Peer CTAs = equal-weight composite of six established trend managers.
Benchmark: the AQR Time Series Momentum (TSMOM) factor, the standard academic trend benchmark (Moskowitz, Ooi and Pedersen, 2012), monthly to May 2026. It is scaled by a constant to the book's 16.4% volatility so the two lines compare risk for risk. It is an excess return over cash, gross of fees and costs, so it understates a total-return comparison by the cash rate. Over January 2019 to May 2026 the book compounded at 8.4% a year against 3.8% for the scaled benchmark, with a monthly correlation of 0.41. Data: AQR Capital Management, LLC.
The profile is the classic one for trend-following: flat-to-negative years in calm, range-bound markets (2019, 2023) and its strongest years when other assets struggle (2022) or trends run (2020, 2024). 53% of months are positive; best month +13.8%, worst −8.6%.
The book runs at a 15% volatility target, which realised 16.4% in the backtest. Trend programmes are often run at higher risk budgets. This illustration scales the book's daily net returns by a constant 1.83× so realised volatility is 30%. The positions and their timing are unchanged; only the leverage is different.
| Year | At 16.4% vol | At 30% vol |
|---|
Leverage leaves the Sharpe ratio unchanged but makes compounding less forgiving. The 2022–2024 drawdown deepens from −24% to −42%, and losing years roughly double: 2023 goes from −13% to −24%.
It is a linear rescaling, not a re-run. The live drawdown controls, which cut exposure in stages as losses deepen, would have reduced exposure during the 2022–2024 drawdown and changed the path. Margin use scales with leverage: at 1.83× the demo account's 20% median would be about 36%, and its 32% peak about 58%, just under the 60% margin cap. Costs are scaled in proportion, so they are about right per unit of risk.
Scaling leverage leaves the Sharpe ratio where it is, so the useful ceiling comes from that ratio, not from the broker. Compound growth peaks when volatility equals the Sharpe ratio. Beyond that point, extra leverage lowers long-run return.
| Realised vol | Leverage vs book as run | CAGR | Max drawdown | Worst month | Worst year | Peak margin use |
|---|---|---|---|---|---|---|
| 15% | 0.9× | 6.3% | −22% | −8% | −12% | 29% |
| 16.4% (book as run, 15% target) | 1.0× | 6.7% | −24% | −9% | −13% | 32% |
| 20% | 1.2× | 7.8% | −29% | −10% | −16% | 39% |
| 25% | 1.5× | 9.0% | −35% | −13% | −20% | 48% |
| 30% | 1.8× | 9.9% | −42% | −15% | −24% | 58% |
| 35% | 2.1× | 10.4% | −47% | −18% | −29% | 68% |
| 41% (growth peak) | 2.5× | 10.7% | −54% | −21% | −34% | 80% |
| 50% | 3.1× | 10.1% | −63% | −25% | −41% | 97% |
Backtest Dec 2018 – Aug 2026. Each row scales the daily net returns of the book as run (15% target, 16.4% realised) by a constant, so the first column is realised volatility, not a target; CAGR in calendar time. Peak margin use scales the demo account's peak by the leverage column: the demo account runs the same book at the same 15% target, and its margin used a median 20% and a peak 32% of equity, taken over its 50 daily account records from 12 Aug to 30 Sep 2026 (intraday snapshots give the same median and a 32.3% peak). Figures in red exceed the 60% margin cap. The drawdown controls and equity-drop halt are not applied; at these risk levels they would cut exposure and change the path.
Compound growth peaks where volatility equals the backtest Sharpe of 0.41. That figure is an in-sample estimate with wide uncertainty; on a Sharpe in excess of cash (0.26) the peak falls to about 26%.
Growth-optimal sizing is sensitive to overestimating the Sharpe ratio, and this one is an in-sample estimate. The research book runs at its 15% target; the other rows of the table and the 30% section are historical rescalings for illustration, not recommended risk levels.
On the current retail venue, the demo account's 32% peak margin use would reach the 60% cap at about 1.9× the book as run, a realised volatility of about 31% in the backtest. Brokers also raise margin rates in stressed markets. Exchange-traded futures would lift this limit; the Sharpe ceiling would remain.
Mulvaney Global Markets has reported close to 40% volatility over 27 years, alongside a higher Sharpe ratio than this backtest. Scaled to Mulvaney's volatility over the same months, the backtest would have compounded at 14% a year, against Mulvaney's 40%.
| Series | CAGR | Vol | Sharpe | Max drawdown | Worst month |
|---|---|---|---|---|---|
| Mulvaney Global Markets, May 1999 – Apr 2026 | 19.3% | 37% | 0.65 | −61% | −41% |
| Mulvaney Global Markets, Jan 2019 – Apr 2026 | 40.4% | 49% | 0.94 | −61% | −41% |
| This book, Jan 2019 – Apr 2026 | 7.8% | 18% | 0.51 | −19% | −9% |
| This book scaled to 49% vol, same months | 14.4% | 49% | 0.51 | −47% | −24% |
Monthly returns; Sharpe without a cash adjustment. Mulvaney figures are computed from monthly returns as reported (net of fees, USD); this book's are hypothetical and before fees. The path to a higher risk budget runs through a higher Sharpe, starting with execution cost, about 12 points a year on the current venue.
A trend sleeve is usually assessed by what it does to the portfolio around it, not by its return alone. In this historical illustration, 20% of a classic 60/40 equity-bond portfolio is moved into the backtested book, rebalanced monthly. Return barely changes, but volatility and drawdown fall sharply, largely because the book gained in periods when stocks and bonds fell together, as in 2022.
| Jan 2019 – Jul 2026 | CAGR | Vol | Sharpe | Max drawdown | Worst month | 2022 |
|---|
60/40 = 60% SPY and 40% AGG (US aggregate bonds), total return with dividends reinvested, in USD; rebalanced monthly with no rebalancing costs. Book returns are the hypothetical backtest, net of modelled trading costs but before fees, and are mixed in without currency adjustment. 91 months; Sharpe without a cash adjustment. The book's monthly correlation to 60/40 is −0.31.
Monthly return correlation over the overlap period, January 2019 to April 2026 (88 months; shorter where a fund launched later). Moderate correlation to the peer group says the book captures the same broad trend premium. The part that does not correlate comes from its own market mix and construction.
Roughly half of the peer group's monthly moves show up in the book. It overlaps with established trend managers without simply duplicating them.
−0.30 to the S&P 500, −0.22 to 7–10y Treasuries, −0.15 to 20y+ Treasuries over the same months.
Book returns are hypothetical; fund returns are as reported, net of fees and mostly in USD. With 88 monthly observations, each correlation is uncertain by roughly ±0.2. Peer composite: Mulvaney, Transtrend, Aspect, Winton, Man AHL and Dunn, equal weight. The AQR factor is an academic index (gross, excess of cash), not a fund.
Metals, agriculturals, US Treasuries, FX and equity indices, across 23 markets. The market list was fixed on practical grounds such as liquidity, not by backtest contribution.
Risk is budgeted at market, sleeve and book level, so no single market or asset class dominates the book.
The book is scaled to a 15% annualised volatility target within leverage bounds, and trading is cost-aware: rebalances that would cost more than they add are avoided.
Signal, sizing and construction rules and their parameters are not published.
Most backtests fail because of what researchers do, not what markets do. The programme is set up to make self-deception expensive, and its research record shows more rejected ideas than accepted ones.
Major studies fix their constants, pass/fail bars and kill criteria before any result is seen. The forward incubation runs against a pass/kill rule written down on day one.
Each backtest is stamped with hashes of its specification, code, price data, roll schedule, calendar and cost model, so any figure can be traced to the exact inputs that made it.
Pipelines are tested on random-walk prices that should score nothing. One such control caught a look-ahead defect in mid-2026. It was fixed, and every result on this page post-dates the fix.
Energy as a trend sleeve, FX as a return source, a diversified second return premium and an intraday session strategy were each tested to a pre-set bar and rejected. FX stays in the book only for its diversification value.
About 280 configurations have been tried across the programme. Selection-adjusted statistics show that backtests alone cannot prove the edge, which is why the plan below leans on a forward record.
Costs are modelled per market from broker spread tables and checked against live fills: live demo fills averaged 1.8 bp per side over 214 fills, slightly above the 1.5 bp slippage assumption.
Exchange-sourced continuous futures from 2010; a deep-history set covering 39 markets back to 1969; 413 million one-minute bars of FX and CFD prices from 2003; plus forward capture of the broker's own live prices.
Vendor data lands unchanged and tagged with its source. Cleaning steps (outlier checks, gap handling, quarantine of removed bars) are logged operations that can be undone and replayed.
Continuous contracts are rebuilt from individual contracts with each roll ratio recorded. A replayed series matched the vendor's own to the last digit across 76,671 bars.
Every bar carries a single UTC timestamp, with each dataset's exchange and provider time zones declared. Missing data is recorded as a gap, never silently filled from a proxy.
Vendor feeds and live broker stream into a time-series database
Logged cleaning, quarantine, replayable continuous contracts
Backtest, walk-forward, cost model; freezes a versioned parameter snapshot
One library turns parameters plus prices into positions, for history and live alike
Pre-trade checks, broker orders, day ledger
Across the data, research, execution and risk services. Safety guards must also pass mutation tests, which prove the test fails when the guard is broken.
An append-only, hash-chained event log; updates and deletes are refused at the database.
Continuous integration on every service, review-gated merges, and a deploy gate that refuses uncommitted or regressing code.
| Control | Setting |
|---|---|
| Volatility target | 15% annualised, bounded leverage |
| Drawdown controls | Exposure reduced in stages as drawdown deepens, to flat at a hard limit |
| Equity-drop halt | Executor freezes on a sharp equity loss |
| Margin cap | 60% of equity (demo use 12 Aug – 30 Sep 2026: median 20%, peak 32%) |
| Pre-trade checks | 11 controls (size, price collars, staleness, exposure) |
| Kill switches | 4 independent, one fully outside our systems; drilled |
| Unattended operation | Halts when operator sign-off expires |
The book's backtest figures above are net of modelled retail spread-bet costs; the AQR factor is gross of costs, and the manager figures are net of their own fees. Moving the same positions to exchange-traded futures through an institutional broker removes the daily funding charge and narrows the trading spread. On a model of one-tick fills plus published commissions, the cost drag falls by roughly four-fifths.
| Cost per year, % of NAV | Spread bets (today) | Futures (modelled) |
|---|---|---|
| Trading: spread, or ticks plus commission | 6.8% | 2.1% |
| Daily funding on rolling positions | 4.9% | 0.0% |
| Contract roll | 0.4% | 0.3% |
| Total | 12.1% | 2.4% |
| Futures, if fills are 1.5 / 2 / 3 ticks wide | 3.3% / 4.2% / 6.0% |
It takes the gross return of a closely related book, about 19.4% a year over a shorter period than the main backtest, and subtracts 2.4% to 6.0% of modelled futures costs. The spread-bet costs it replaces (12.1%) are measured on the current book from April 2025 to September 2026. The futures costs are modelled, not measured, so the scenario mixes bases and is not a forecast.
A futures contract cannot be split, so a market can be held only when one contract fits its share of the risk budget. Micro contracts make currencies, metals and the S&P 500 accessible at $250k. Treasuries and agriculturals trade only in full-size contracts, and two soft commodities and one equity index need well over $1m to fit.
The futures figures apply to the full 23-market book. A $250k account would hold a 13-market subset, a different portfolio that needs its own pre-registered backtest. Comparisons are before tax; for a UK individual, spread-bet gains are tax-free while futures gains are subject to capital gains tax.
Orders have run daily on a broker demo account since July 2026; real money has not been committed. The model's forward incubation (since 12 Aug 2026) is +1.8% with a −5.8% maximum drawdown over seven weeks, far too short to judge.
A backtest Sharpe of 0.41 after costs is in line with the trend asset class's long-run record, not above it. The final walk-forward fold was negative, and 2023 and 2026 to date were losing periods.
Retail spread-bet execution costs about 12 points a year on the current book (April 2025 – September 2026), against about 19 points of gross return in a closely related book. Exchange-traded futures are modelled at 2.4 to 6 points, but have not been measured, and need at least $250k to hold a reduced 13-market book.
Demo positions track the model's book closely (0.95 correlation since early September), but accounting-level reconciliation between the simulator and the broker ledger is still being fixed.
A small personal account at the 15% vol target, on the same code that runs today.
A pass/kill rule written before launch: a minimum forward Sortino to continue, and a hard kill on a deep drawdown or a strongly negative forward Sortino.
Execution timing first, then exchange-traded futures: a 13-market book from about $250k with its own pre-registered backtest, 20 markets from about $500k, and the full 23-market book from about $1.8m. The open question is how much gross return survives execution: futures costs are modelled at 2.4 to 6 points a year against about 12 on spread bets, and have yet to be measured.
Live and simulated returns must agree to a pre-set correlation bar before the live results are relied on.
Measured futures costs and a market-by-market capacity study, replacing the modelled figures on this page.
This page describes a personal research project. It is not an offer, solicitation or invitation to invest, and it is not investment advice. No fund, managed account or other investment product based on this research exists, and none is offered.
Performance shown is hypothetical. It comes from a computer simulation applied with hindsight and has inherent limits: no actual trading took place, and results may not reflect the impact of liquidity, market conditions or decisions made under real risk. Simulated results are net of modelled transaction and financing costs. The futures cost scenario is a hypothetical calculation on mixed bases, not a backtest result or a forecast. Forward incubation figures are model-generated, not account results. Demo-account activity uses no real money. Past performance, simulated or actual, is not a reliable indicator of future results. Trading futures and leveraged derivatives involves substantial risk of loss.
Third-party fund returns are used only to calculate correlations and summary statistics; monthly return series are not reproduced. Fund names belong to their respective managers, who are not affiliated with and have not reviewed this material. Market data as of 10 Aug 2026 (backtest), 30 Sep 2026 (forward), 30 Apr 2026 (peer returns) and 31 May 2026 (AQR factor). The Time Series Momentum factor is published by AQR Capital Management, LLC in its data library and is used here with credit; AQR has not reviewed this material.