“A trend portfolio is not paid at the moment a trend is identified. It is paid for the distance travelled while it was already positioned, and it hands back the same way.”
About This Report
The ATS Classic Trend Benchmarking Portfolio is a systematic trend-following reference model. It runs a ten-strategy ensemble across sixty-eight futures markets spanning eight sectors, sized by a fixed-amount rule on prior-bar volatility, and serves as a clean benchmark for classic trend following. The figures in this report are produced by the ATS Classic Trend engine, a standalone Python implementation run on out-of-sample data using simple trend-following models. Results are computed as dollar profit and loss on a stated twenty-five-million-dollar capital base, with returns stated additively: the year-to-date figure is the sum of the monthly net returns, and drawdowns are measured on the same additive basis. The model carries no discretionary overlay; it is the rules, the markets, and the price data.
This is a high-volatility portfolio, and the figures should be read with that in mind. It is built to capture large, sustained trends rather than to smooth the path to them, so the return profile pairs strong accumulation through trending years with sharp monthly declines and deep drawdowns when trends reverse. June’s -19.77%, August’s +12.70% and September’s +2.66% are three consecutive expressions of the same architecture rather than departures from it, as is the -42.82 point peak-to-trough decline of 2024 to 2025.
The report is published in the first trading days after each month-end as a fast, factual read on the prior month for classic trend followers. It documents the model’s performance, decomposes the result by sector and by market, and sets the numbers against the month’s market backdrop. The aim is a clear account of what the trend-following environment delivered, and of what the model is carrying into the month ahead.
Commentary
September returned +2.66%, following August’s +12.70% and July’s +10.76%, and the modest headline conceals the widest internal spread of the year. Energy contributed +6.16 percentage points on the trading basis and Bonds +3.21, while Equities cost -2.53, Metals -2.18 and Grains -1.33. Three of the eight sectors contributed, against six in August. The two that led produced more than three times what the month returned, and the five that detracted gave back most of it. On the fixed capital base the portfolio closed September at a cumulative +204.9%, a new high for the track record, with the year to date at +30.1%.
The month’s path matters more than its size. Decomposed from the daily per-system curves, the portfolio returned +2.03%, +2.02%, +0.52% and -0.59% through the four reporting weeks, then lost 1.25 points across the final three sessions of 28 to 30 September. Those weekly figures are stated on the engine trading basis, before the portfolio-level cost and interest adjustments, and sum to the +2.73% trading result rather than to the +2.66% fully-loaded headline. The book held +3.98% at the close of 25 September and finished the month with +2.73% of it. About a third of what four weeks earned was returned in three days, and the reversal came from the same place the gain did.
The Month, Week by Week
The TTU Trend Barometer ran 55%, 59%, 45% and 52% across the four weeks of September, a path that rose to the top of its range, fell fourteen points in a week and recovered half of that.
The 59% reading on 11 September was September’s only move into the favourable band, and it arrived in a week when just 14 of 49 contracts finished higher.
A week later 18 contracts rose and the reading fell to 45%. Counting markets that are moving and counting markets that are trending are different exercises, and September separated them more clearly than any month this year.
Performance Snapshot
September net return of +2.66% against August’s +12.70%. The month is the smallest positive of 2026 and the sixth gain in nine months. It is also the month that carried the cumulative track to a new high, at +204.9% since the January 2020 inception.
How to Read the Track
The panel below steps through every month since the portfolio’s January 2020 inception, one at a time.
It carries two lines. The solid line shows the portfolio marked to market, which is how the performance tables report it. The dimmer line shows profit realised on closed trades. The distance between them represents profit still sitting in open positions. Watching that gap open and close provides one of the clearest pictures of what a long-horizon trend follower is doing between entry and exit.
At the September 2026 close, the marked line stands at +215.8% and the closed-trade line at +137.8%. That leaves 78.0 percentage points of the result in positions still running.
Both lines are presented on the engine trading basis, net of the embedded commission and before portfolio-level cost adjustments and the collateral interest credit. This is deliberate. It places both lines on the same basis, so the distance between them reflects open profit and nothing else. The published fully loaded track stands at +204.9%, approximately 10.9 points below the engine trading result.
Six declines deeper than eight points are shaded in red, from peak to trough. The three stretches in which the portfolio established new highs by fifteen points or more are shaded in green.
The deepest decline ran from March 2024 to May 2025, taking the portfolio 43.08 points below its previous high. That high was finally regained in February 2026, twenty-three months after the peak. The 2022 episode was shallower at 29.39 points, but still required twenty-two months to recover.
Against those declines, the advance from February to April 2022 added 45.52 points, while the two months to March 2024 added 29.06 points.
Neither the declines nor the advances represent a departure from the model. They are the same behaviour seen from opposite sides: a strategy that holds positions until a trend ends rather than leaving simply because the journey has become uncomfortable. The March 2024 peak makes the point especially clearly. It completed one of the strongest advances on the track and then became the high from which the deepest drawdown began.
Months carrying an explanation are numbered on the chart. The chronology beneath the panel fills in as the walkthrough reaches each one, and any numbered entry can be selected to return directly to that month.
Each explanation is sourced from the contemporaneous This Week in Trend record where one exists, from published sources where the account depends on broader market history, and from our own attribution where the claim concerns this portfolio. Every month, annotated or not, displays its largest contributors and largest costs, calculated from the portfolio’s own records rather than reconstructed from memory.
The chronology is being completed month by month. Months that do not yet carry a written explanation still show their underlying performance drivers.
The track, month by month
Every month since the January 2020 inception, stepped one at a time. The solid line is the portfolio marked to market; the dimmer line is profit actually realised on closed trades. The gap between them is money still in open positions. Red bands are declines of more than eight points, green bands are stretches making new highs. Where a month carries an explanation it appears on the right and is added to the chronology below it.
Largest contributors
Largest costs
Chronology of major events
Click the panel, then space plays and pauses and the arrow keys step one month. Click any entry in the chronology to jump to that month. Figures are on the engine trading basis, net of the embedded commission and before the portfolio level cost adjustments and the collateral interest credit, so both lines sit on one basis and the gap between them is open profit alone. The published fully-loaded track is about ten points lower over the period. Some months are not yet annotated; those show their drivers only while the research is verified.
Cumulative Performance and Drawdown
Since the January 2020 inception, the portfolio has generated cumulative profit equivalent to +204.9% of the initial capital base across 81 months. September closed at the peak of that fully loaded series, clearing the April high. Its maximum peak-to-trough decline was 42.82 points, reached in May 2025. The interactive chronology shows the same episode at 43.08 points on the engine trading basis, before portfolio-level adjustments. These are two measurements of the same decline on the two bases used throughout this report.
Monthly Performance Since Inception
Key Statistics Since Inception
Computed from the monthly series since the January 2020 inception. CAGR annualises the cumulative return over the capital base; the MAR ratio is CAGR divided by the maximum drawdown; win rate, profit factor and skew are measured on monthly returns.
The profile is characteristic of classic trend following: a high annualised return carried alongside a deep maximum drawdown, with the monthly distribution positively skewed. Annualised volatility of 34.1% and a MAR of 0.42 describe a model built to stay with large moves rather than to limit the variance of the path. The best month in the record is +28.9% and the worst -19.8%, and both sit inside the last two years.
Trade Statistics Since Inception
This section reports the portfolio on a realised, closed-trade basis since inception, the lens by which classic trend followers assess a programme. It complements the equity-curve statistics above and is stated on the trading basis, net of commission. It does not reconcile to the additive equity curve, because open positions and mark-to-market timing differ; it is a distinct, realised view rather than a restatement.
Direction: long and short participation
Of 3,680 closed trades, 72.8% were long and 27.2% short. Long positions contributed 115.5% of net trade P&L and short positions -15.5%. Thirty-two trades closed in September, and the month’s result sits largely in positions still open at month-end, which is why this realised view does not reconcile to the equity curve. The bond and petroleum positions that produced the month were both still running at the close.
R-multiple framework
Every trade is sized to a fixed initial risk, one R, so outcomes are directly comparable across all 68 markets. Average trade expectancy is +0.19R. The average winner returned +1.59R against an average loser of -0.64R, a win-to-loss ratio of 2.49. The largest winning trade reached +35.2R while the largest realised loss was -3.0R. That asymmetry, many small losses against occasional very large gains, is the historical shape of the approach. The stop discipline is intended to contain losses, and the realised distribution shows it doing so over this period, but no observed figure sets a limit on what a future loss can be.
Distribution and skew
The distribution of outcomes in R has a skew of +6.8: a majority of small losses (median trade -0.33R, win rate 36.9%) against a small number of very large gains, with 8.2% of trades closing beyond -1R and the right tail reaching +35.2R. The shape is the mechanism rather than a by-product of it. A model that cuts quickly and holds without a profit target produces this distribution by construction, and the same asymmetry is visible in the monthly series and in the per-system breakdown, at three different time scales.
Holding period
The average trade is held 145 days and the median 86 days, consistent with a low-frequency, high-conviction profile. Long positions are held 150 days on average against 132 for shorts. The gap between the mean and the median is itself informative: a minority of positions run far longer than the typical trade, and those are the positions the record is built on.
System Analytics
The portfolio is built from seven distinct trend systems, described here as System 1 through System 7. It does not trade one of each. On every market it runs a 10-model ensemble drawn from the seven systems, with each model’s parameters tuned to that market, so a system can appear more than once on a market, with different parameters, and need not appear on all of them. Across the 68 markets that is 680 models in total. This section groups those models by the system they come from and reports the seven resulting streams: how each contributes to the return, how each behaves in isolation, and how each moves in relation to the whole. The construction and parameters of the individual systems are proprietary and are not disclosed; what follows is the performance anatomy of the ensemble, not its recipe.
A note on basis: gross versus net
The cumulative figures in this section are stated on the engine trading basis. That basis is already net of the embedded USD 5 round-turn commission and is described here as gross only in relation to the portfolio-level overlays that follow it: the per-side commission charge that replaces the embedded one, slippage, roll execution and the collateral interest credit. It is not gross of every cost. On this basis the seven systems sum to +215.8% gross. This gross figure should not be confused with the headline net return. After the full cost stack, the portfolio returned +204.9% net, as reported in Key Statistics Since Inception. The difference of 10.9 points is a portfolio-level overlay and is not attributed to any single system, which is why the per-system curves reconcile to +215.8% gross and not to the +204.9% net headline.
How the ensemble works
Each of the seven systems expresses a different rule for defining and following a trend, and each is deployed several times over, with market-specific parameters, inside the per-market ensemble. Because their dollar results are additive on the fixed $25M base, the models, and the seven systems they roll up into, sum day by day to the portfolio equity curve: nothing is left unexplained, and no system is weighted by discretion. The benefit of combining them is not that any one is exceptional, but that they are imperfectly aligned. Across the period the average pairwise co-movement between systems was 0.57, and each system’s correlation to the whole ranged from 0.68 to 0.89. These are positive throughout, so the systems are distinct rather than independent. No single system is the portfolio.
That imperfect alignment is where the diversification benefit sits. The sum of the seven systems’ standalone monthly volatilities is 12.2% a month. The portfolio’s own monthly volatility was 9.9%, roughly four fifths of that sum. The comparison measures dispersion between the systems, not total portfolio risk, and no inference about risk-adjusted performance is drawn from it here. Each system also captures a slightly different slice of every trend: some enter earlier, some hold longer, some carry more of the short side, so the ensemble stays exposed across more of each move, and through more kinds of regime, than any single rule could.
The convex, positively skewed payoff that defines classic trend following is visible at more than one level of the book. The portfolio distribution and the seven per-system distributions reported below all show the same broad shape: many small losses funding a few large winners. The per-market distributions are not shown here and no claim is made about them. Diversifying across seven expressions of that signature does not dilute it; it stabilises the path by which the result accumulates, and it keeps the programme in contact with trends that any single system would miss.
Per-system cumulative contribution since January 2020, additive on the $25M base, stated on the engine trading basis, which is net of embedded commission and before the portfolio-level overlays. The seven curves sum to the portfolio’s +215.8% trading-basis result.
Per-system return characteristics
The table below reads each system on the equity-curve basis. Contribution is additive and sums to the portfolio; the CAGR, MAR ratio, maximum drawdown and correlation are properties of each system’s own curve.
Per-system R-multiple analysis
Every trade is sized to a fixed initial risk of one R, so outcomes are comparable across all systems and markets. The table applies the R-multiple framework of Trade Statistics Since Inception to each system as a separate column, with a total that reconciles to the portfolio. The Sum of R row is the reconciling line: the seven systems’ R contributions add to 682.1R, and the Total column reproduces the portfolio’s realised figures on the frozen book of 3680 closed trades.
None of these per-system differences is visible in the blended portfolio line, and that is precisely why the blend accumulates more steadily than its parts.
System concentration this month
The since-inception figures above describe the ensemble over six and three quarter years. Within a single month the seven streams can diverge sharply, and September 2026 did. System 2 contributed +4.15 percentage points on the trading basis while five of the seven lost money, and the month’s +2.73% trading result is smaller than that single contribution. A positive month therefore rested on diversification across the ensemble as much as diversification across sectors. Losses from five system families were more than offset by gains from the other two, without any discretionary change in their weighting.
This section is regenerated each month from the same frozen book of record and the append-only monthly loader. The per-system curves are decomposed from the engine’s daily mark-to-market and continue to sum to the portfolio, so the mark-to-market variance measured against the locked baseline each month applies to the systems and to the whole on one consistent basis. The System 1 to System 7 mapping is fixed and carried forward unchanged from month to month.
Sector Attribution
Three of the eight sectors contributed in September, against six in August. Energy led with +6.16% of the $25M base on the trading basis, Bonds added +3.21% and Meat +0.37%. The five detractors were Equities at -2.53%, Metals at -2.18%, Grains at -1.33%, Softs at -0.54% and Currencies at -0.44%. The spread from best to worst sector is 8.69 points against a portfolio result of +2.73%, the widest internal dispersion of any month this year.
This is the inverse of August in shape. August produced +12.70% from six contributing sectors with no single sector dominating, a broad month in which the aggregate was larger than any of its parts. September produced +2.73% from two large contributors and five offsetting detractors, a month in which the aggregate was less than half of what Energy alone delivered. The same ensemble produced both results eight weeks apart, which is the point of running it across eight sectors rather than one.
Energy remains the year’s dominant sector contributor at +32.41% year-to-date after September’s +6.16%, and the gap to the rest of the board has widened rather than closed. Bonds sit second at +1.67% for the year, having spent eight months contributing almost nothing before September supplied nearly twice their entire prior annual contribution in a single month. Softs remain the year’s largest detractor at -3.13%.
Attribution Highlights
September’s contributors were concentrated in two complexes and its costs were spread across five. Heating Oil led at +2.40%, followed by RBOB Gasoline at +1.42%, ICE Brent Crude at +1.36% and Light Sweet Crude at +0.86%. The entire US Treasury curve paid, the 5 Year Note contributing +0.75%, the 2 Year +0.44%, the 30 Year Bond +0.43% and the 10 Year +0.41%. On the other side the Australian Dollar was the month’s largest single cost at -1.11%, ahead of the Russell 2000 at -0.88%, the Dow at -0.79%, Cotton at -0.77% and Gold at -0.72%. Thirty-five of the 68 markets finished positive and 33 negative, the narrowest split of the year.
Across 2026 the petroleum complex still leads and has extended its lead. Heating Oil is the year’s largest contributor at +15.24%, joined by RBOB Gasoline at +6.95%, Light Sweet Crude at +5.95% and ICE Brent Crude at +4.36%. Those four markets account for more than the portfolio’s entire year-to-date result. Gold is the year’s largest single detractor at -2.70%, with Palladium at -2.07% and Platinum at -2.01% behind it, so the precious metals have cost the book almost seven points between them across nine months.
Market-Contribution Review
The since-inception contribution paths show the structure the portfolio is built to exploit: a small number of markets carrying the record over long horizons, a long tail contributing little in either direction, and the identity of the leaders changing slowly. The same shape repeats at the month scale. In September four petroleum contracts and four points of the US Treasury curve produced more than twice what the portfolio returned, and 60 other markets netted against them. That self-similarity across time scales is a feature of the return structure rather than an artefact of this particular month.
Trend Spotlights
Heating Oil is the month’s largest single contributor for the third month running, at +2.40%, and it is also the clearest illustration of the month’s shape. The contract gained 6.85% in the first week and 9.23% in the second, the latter one of the three largest advances on the board, carrying the petroleum complex above the range that had contained it through July and August. In the fourth week it fell 7.72%, the largest single move on the board in either direction, from the upper end of its chart, before recovering part of that on the final session of the month. The position was comfortably profitable throughout and the trailing stops gave back part of what the advance had marked.
RBOB Gasoline contributed +1.42% and traced the same arc one step removed from the crude pair. The contract rose 5.39% in the first week, extended for a third consecutive week with a 6.66% gain in the week to 18 September when the crude pair stalled, then fell with the rest of the complex in the fourth. Leadership rotating within a sector while the sector keeps advancing is a common shape in energy, and the ensemble captured it because it holds the complex through several models rather than expressing a single view on crude.
The Australian Dollar was September’s largest single cost at -1.11%, and it cost through a change of regime in the currency basket rather than through a trend. The basket scattered in the first week, four higher and four lower, after a fortnight of moving as a single bloc on the dollar. It then found the dollar factor again for the last two weeks of the month, with seven of eight crosses lower against a rising US Dollar Index. A position taken into the scattered middle of the month paid for the reorganisation at the end of it. The US Dollar Index itself contributed +0.47% and the Canadian Dollar +0.47%, so the sector’s -0.44% was a cost of internal reshuffling rather than of direction.
How to Read the Trade Playback
The panel below plays September’s five largest contributing markets bar by bar: Heating Oil at +2.40%, NYMEX RBOB Gasoline at +1.42%, ICE Brent Crude at +1.36%, Light Sweet Crude Oil at +0.86% and the 5 Year US Treasury Note at +0.75%.
Each chart begins shortly before the earliest position still open at the September close. The viewing window therefore opens on the underlying move rather than at the beginning of the month. The earliest of the five begins on 6 March 2025.
Every market is traded by a ten-model ensemble. Each model that takes a position draws a straight line from its entry to where that position stands now. A sustained trend therefore appears as a fan of trajectories opening from the separate points at which the ensemble entered, rather than as a single line.
Green represents long positions and red represents short positions. Entries are shown as triangles and exits as crosses. Each model’s stop appears as a dashed red line. The stepped path records where that stop has been, while the line extending across the chart shows where it sits now.
Across the five windows, there are 59 entries and 25 exits, most originating in earlier months. At the September close, 42 of the 50 model positions remained open.
The stop is worth watching rather than assuming. Its trailing component can only ratchet in the direction of the trade, but the effective stop sits at the wider of that trailing level and an ATR-based distance from entry. It can therefore widen when volatility expands. That occurred on 12% of open bars across these five markets.
Models are labelled Strategy 1 to Strategy 10 according to their position within each market. The numbering does not carry from one market to another, and the construction and parameters of the individual systems are not disclosed.
Two conventions apply to the figures shown. First, they are presented on the engine trading basis, before portfolio-level cost adjustments and the collateral interest credit. They will therefore not reconcile directly to the published +2.66% monthly return.
Second, playback ends on 30 September, with every open position marked to the final valid close for the reporting month. This now places the interactive playback on the same month-end valuation basis as the report.
Trade playback
The five largest contributors in September 2026, played bar by bar from the start of the move. Each market runs a ten-model ensemble, and each model carries its own stop. Each trade draws a straight line from its entry to where it stands now, so a long trend shows as a fan of trajectories opening from the points the ensemble stepped in. Stops are the dashed red lines. The trailing component only ratchets, but the stop sits at the wider of that and an ATR distance from entry, so it can widen when volatility expands. That happened on 12% of open bars here.
Click the panel, then space plays and pauses and the arrow keys step one bar. Click a strategy to isolate it, click again to show all ten. Green marks longs and red marks shorts; entries are triangles and trajectories are solid lines. Exits are red crosses, whichever way the trade was facing. Stops are dashed red throughout: the stepped path is where a stop has been, the rule across the chart is where it sits now. Playback ends 30 September, with each open position marked to the final valid close for the reporting month.
Sector Rundown
The Trend Environment
The barometer ran 55%, 59%, 45% and 52% over the four weeks of September. It opened at the threshold between Neutral and favourable, crossed into the favourable band for the first time on 11 September, fell fourteen points the following week, and recovered seven in the last. The SG Trend Index, an external industry index and not a measure of this book, advanced every week, reading +0.59%, +2.79%, +3.23% and +4.47% month-to-date at the Friday closes of 4, 11, 18 and 25 September. Source: ATS This Week in Trend, 4, 11, 18 and 25 September 2026. Both series are weekly and both stop on 25 September, so neither describes the final three sessions of the month. The reading that fell hardest, in the week to 18 September, came in a week when more markets rose than the week before. What dropped was not the number of markets moving but the number moving with the structures already beneath them, and that is the distinction the barometer is built to make.
The Lesson of the Month
September is a reminder that the environment reading and the portfolio result answer different questions, and that the gap between them can close very quickly. For four weeks this book and the trend environment broadly agreed. The portfolio earned +3.98% on the trading basis through 25 September, the barometer spent the month inside or just below the favourable band, and the external index advanced every week. Then three sessions returned about a third of it. No further weekly environment observation was available for those sessions, so nothing can be said here about what the board was doing. What changed in the portfolio was the direction of travel in the complex to which it was most heavily exposed. A favourable environment describes the persistence available on the board. It makes no commitment about which end of a position a given portfolio is holding when that persistence stops.
Monthly Wrap
September returned +2.66%, the smallest positive month of 2026 and the forty-sixth gain of the 81 months since inception, and it carried the cumulative track to a new high at +204.9%. The result was built on Energy at +6.16% and Bonds at +3.21% on the trading basis and reduced by five detracting sectors, with Equities the largest at -2.53%. The year to date stands at +30.1%. Costs took 0.26 points from the trading result, roll execution accounting for 0.19 of that, and interest on idle capital at a 3.94% T-bill rate added back 0.18 points.
Looking Ahead
The barometer at 52%, three points below the favourable threshold with its rate of change reading neutral, describes a board on which most of the structures that built the middle of the month remain in place and one, the petroleum complex, has been interrupted. The US Treasury curve has moved to new lows across all four maturities, the precious metals have resumed the declines they have run since February, and four grain contracts sit at the highest levels on their charts. The book carries positions in all three into October, along with the petroleum positions the final week reduced. Whether the Energy reversal was the end of that advance or an interruption inside it is the question October answers, and it is not one this report forecasts.
About these figures
All figures are generated by the ATS Classic Trend engine from back-adjusted continuous futures data across the sixty-eight-market universe. Returns are stated on a fully-loaded, pre-fee basis: net of estimated commission, a conservative slippage allowance and roll-execution costs on every trade and contract roll, and inclusive of collateral interest earned on the un-margined capital at the US 90-day Treasury bill rate. They are dollar profit and loss on a twenty-five-million-dollar capital base, additively. Results do not include performance or management fees, which would reduce the returns shown. Contribution figures are additive and sum to the portfolio result at the sector and market level. This report is a benchmark study and is not investment advice or an offer of any product.
Richard Brennan writes on systematic trading, complex adaptive markets, and the philosophical foundations of trend following at atstradingsolutions.com. His books include The Fractals of Finance, Complex Adaptive Markets, Carved by Impossibility and The Aussie Turtles Trend Following Guide.
Want to explore why structure exists at all?
Carved by Impossibility: What Remains When Everything Else Is Eliminated
The book explores the architecture of constraint, emergence, and reality itself, and what it means for how we understand markets, life, and the universe.
Available now on Amazon in paperback, hardcover, and Kindle.
Want the theoretical foundation for why trend following works?
The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets bridges complexity science with practical trading implementation. With a foreword by Jerry Parker, original Turtle Trader.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.