Iron ore futures finished −15.2% over the period. The Permutable strategy returned +27.4% after an assumed 25 bps bid/ask spread on every trade. It matched iron ore in the rebounds and sold the declines.
A paper evaluation of Permutable Copilot signals, not a marketed product return. Cumulative, unit-notional returns on roll-adjusted COMEX iron ore futures, after an assumed bid/ask spread of 25 bps per trade (half on entry, half on exit); before fees, commission, roll costs and financing.
Iron ore slipped through January and February, recovered to a high on 11 May, then fell through the summer and autumn to a low on 15 October, and finished the period −15.2%. A holder of iron ore futures finished −15.2%. The Permutable strategy, which reads the direction of iron ore news every hour and can be long, short or flat, earned 27.4% after trading costs: it sold the declines and joined the rebounds, and finished 8 months out of 10 in profit against 4 for iron ore. Iron ore is not in the live book, so this page does not quote a book contribution. The window ends on the last iron ore price, 23 October 2026.
This page sets out what the strategy does and how it decides, the decisions that delivered most of the return, where the outperformance came from, and the basis on which the figures are measured.
The lines separate in the three biggest declines. In February the strategy was up 6.0% and iron ore was down 6.1%; in June the strategy was up 7.9% and iron ore was down 8.0%; and in October the strategy was up 5.8% and iron ore was down 5.4%. All figures are after costs.
A decision is a confirmed change in the strategy's view of iron ore. Three Sell decisions delivered about 23 of the strategy's 27 points of return after costs, and each beat iron ore by 11 to 21 points over its own days. The Buy trades matched iron ore, as any long trade does; the edge came from selling the declines. Each trade is recorded in the trade log with its entry and exit dates.
Sell from 20 Jan. Trade 20 Jan – 26 Feb.
The signal's explanations cited rising supply from major miners and new sources, falling steel output in China and recent price declines. 97% of hourly readings during the trade were Sell.
Iron ore moved −7.0 points over the same days, so the short earned 6.8%. February closed with the strategy up 6.0% and iron ore down 6.1%.
Sell from 21 May. Trade 21 May – 13 Jul.
The explanations cited China's stricter steel capacity controls and regulatory changes reducing demand, high stock levels in China, and added supply. 98% of hourly readings during the trade were Sell.
Iron ore moved −10.7 points over the same days, and the short earned 10.2%, the largest single trade of the period. June closed with the strategy up 7.9% and iron ore down 8.0%.
Sell from 29 Sep. Trade 29 Sep – 20 Oct.
The explanations cited recent price declines, weakening demand from China and supply pressures. 99% of hourly readings during the trade were Sell.
Iron ore moved −5.5 points over the same days, to its low of −15.5% on 15 October, and the short earned 5.8%. October closed with the strategy up 5.8% and iron ore down 5.4%.
| Trade | Direction | Dates | Strategy | Iron ore, same days | Versus iron ore |
|---|---|---|---|---|---|
| Summer slide | Sell | 21 May – 13 Jul | +10.2% | −10.7% | +20.9 |
| Winter sell-off | Sell | 20 Jan – 26 Feb | +6.8% | −7.0% | +13.8 |
| Autumn slide | Sell | 29 Sep – 20 Oct | +5.8% | −5.5% | +11.3 |
| Early-August drop | Sell | 28 Jul – 3 Aug | +4.5% | −4.3% | +8.8 |
| Spring rebound | Buy | 13 Apr – 18 May | +3.0% | +3.3% | −0.2 |
Iron ore's move is the change in cumulative long-only return between each trade's open and close; “versus iron ore” is the strategy's return minus that move, in points. A long trade can at best match the asset over its own days, so the long trades sit near zero. Trade returns are unit-notional and after the assumed 25 bps spread; a long trade nets slightly below the asset's move for that reason.
The 7 long trades matched iron ore, as any long trade does: they earned 6.6% after costs (8.4% before) against 8.1% for iron ore over the same days. Almost all of the outperformance came from the 8 short trades, which earned 20.8% after costs (22.8% before) while iron ore moved −22.5 points over those days, a gain of 45.2 points on holding iron ore before costs. On the days the strategy had no position, iron ore moved −0.8 points. Trading costs took 3.8 points: 15 round trips at 25 bps each.
With iron ore −15.2% over the period, a strategy that sells the declines is rewarded when the market falls. In a rising market the same strategy would earn from its long trades, which can only match the asset. Iron ore is not a sleeve of the live book, so there is no book-weighted P&L to set beside this result.
Points versus holding iron ore, 1 Jan – 23 Oct 2026, unit notional. The first three rows are before costs; the total is after the assumed 25 bps spread. Components are unrounded, so they may differ from the headline figures by 0.1. Across 212 weekdays the strategy was short on about 58%, long on about 27% and flat on about 15%.
The strategy is rules-based. Every step below is fixed in advance and applied the same way throughout the period.
The strategy's settings were fixed on data up to 31 December 2025. Iron ore signals begin on 1 October 2025, so that window is short, about three months and five trades. The period on this page, 1 January to 23 October 2026, falls after that date and was not used to choose them. It ends on the last iron ore futures price, which can be a few sessions behind the book's last mark.
| Nature | A paper evaluation of Copilot signals. Permutable does not trade this strategy and the figures are not a marketed product return. |
| Period | 1 January to 23 October 2026. 15 trades, 2 Jan – 20 Oct (7 long, 8 short); 10 were profitable. Iron ore is not in the live book, so this page does not deduct a book weight or quote a book contribution. |
| Out-of-sample | Strategy settings were fixed on data to 31 December 2025, a window of about three months. 2026 is out-of-sample. |
| Comparison | Long-only is a buy-and-hold position in the same roll-adjusted iron ore futures, with no financing, carry or trading costs. Charging it one 25 bps round trip would lower it by 0.25 points. |
| Risk metrics | Sharpe is a simple daily Sharpe with a zero risk-free rate, and volatility is the standard deviation of daily marks, both annualised by √252. They use unit-notional daily P&L over about ten months, are not GIPS-compliant, and are shown for comparison only. |
| Return basis | Unit notional: each trade is % of entry price. Returns are after an assumed bid/ask spread of 25 bps per round trip (3.75 points over 15 trades); they exclude fees, commission, roll costs and financing. Before costs the strategy returned 31.1%. |
| Cost sensitivity | The result depends on the assumed spread. At 12.5 bps per round trip the strategy would have returned 29.3%; at 50 bps, 23.6%. |
| Scope | One asset, one period, in which iron ore finished −15.2%. A strategy that can sell short is favoured when the market falls, and this page is not a forecast. It is the iron ore strategy only, not the return of a fund or portfolio, and iron ore is not in the live book. |
| Concentration | The three largest trades contributed about 23 of the 27 points after costs. The other 12 trades contributed about 5 points in total, and 5 of the 15 trades lost money or broke even. |
| Explanations | The drivers quoted for each decision are the model's own written explanations. They describe what the signal cited, not a proven cause of the price move. |
Past performance is not a guide to future results. This page is published to show how Permutable's dataset and models behave in a defined evaluation. It is for information only and is not investment advice or a recommendation to buy, sell or allocate capital.
Discuss the dataset or API. Email enquiries@permutable.ai or book a demo. See the live commodity book on the Performance page.
Get in touch