Updated 3 Nov 2026, backtest 2012–2026

We keep score. Publicly.

80% ranges since 2012, scored against the next close, the misses kept in.

1 trading day

79.2–80.8%

inside the 80% range, 7 markets, 3,681–3,701 days each

Uncalibrated normal range 80.2–82.4%

5 trading days

78.4–80.6%

inside the 80% range, 7 markets, 3,677–3,697 days each

Uncalibrated normal range 79.1–81.2%

21 trading days

77.2–80.6%

inside the 80% range, 7 markets, 3,661–3,681 days each

Uncalibrated normal range 76.2–82.2%

Direction

0 of 18

tests where up or down beat a naive baseline, 6 markets, 3 horizons

So we don’t sell it

The live record.

Published every trading morning at 06:30 Dubai, scored against the close the next day.

  • Brent 1-day 80% range todayfirst range on the next trading morning
  • Yesterday against the closenot scored yet
  • Held since launchcounted from the first scored range

Announced against observed.

A dot on the dashed line holds exactly as often as it claims.

80% ranges, observed share of closes inside, 2012–2026
72%76%80%84%88%Brent, 1 day: 80.4% inside the 80% range [95% CI 79.1–81.7]; uncalibrated 81.7%Brent, 5 days: 80.2% inside the 80% range [95% CI 78.0–82.2]; uncalibrated 79.9%Brent, 21 days: 79.6% inside the 80% range [95% CI 74.1–84.2]; uncalibrated 79.0%BrentWTI, 1 day: 80.5% inside the 80% range [95% CI 79.2–81.8]; uncalibrated 81.4%WTI, 5 days: 80.2% inside the 80% range [95% CI 78.0–82.3]; uncalibrated 80.5%WTI, 21 days: 80.1% inside the 80% range [95% CI 75.5–84.1]; uncalibrated 78.9%WTIULSD NY Harbor, 1 day: 80.2% inside the 80% range [95% CI 78.8–81.6]; uncalibrated 80.2%ULSD NY Harbor, 5 days: 80.6% inside the 80% range [95% CI 78.6–82.7]; uncalibrated 79.1%ULSD NY Harbor, 21 days: 78.6% inside the 80% range [95% CI 73.9–82.9]; uncalibrated 76.2%ULSDRBOB gasoline, 1 day: 80.8% inside the 80% range [95% CI 79.5–82.2]; uncalibrated 81.5%RBOB gasoline, 5 days: 79.8% inside the 80% range [95% CI 77.7–81.9]; uncalibrated 80.2%RBOB gasoline, 21 days: 80.6% inside the 80% range [95% CI 76.4–84.8]; uncalibrated 78.0%RBOBDiesel crack (HO − Brent), 1 day: 79.3% inside the 80% range [95% CI 77.8–80.7]; uncalibrated 82.4%Diesel crack (HO − Brent), 5 days: 79.5% inside the 80% range [95% CI 77.4–81.7]; uncalibrated 81.2%Diesel crack (HO − Brent), 21 days: 77.2% inside the 80% range [95% CI 72.9–81.4]; uncalibrated 79.5%Diesel crackGasoline crack (RB − Brent), 1 day: 79.2% inside the 80% range [95% CI 77.7–80.7]; uncalibrated 81.9%Gasoline crack (RB − Brent), 5 days: 78.4% inside the 80% range [95% CI 76.2–80.7]; uncalibrated 81.2%Gasoline crack (RB − Brent), 21 days: 78.4% inside the 80% range [95% CI 74.6–82.2]; uncalibrated 82.2%Gasoline crackHenry Hub gas, 1 day: 79.9% inside the 80% range [95% CI 78.6–81.2]; uncalibrated 80.5%Henry Hub gas, 5 days: 80.4% inside the 80% range [95% CI 78.3–82.5]; uncalibrated 80.4%Henry Hub gas, 21 days: 79.7% inside the 80% range [95% CI 75.4–83.8]; uncalibrated 79.0%Henry Hubeach market at 1, 5 and 21 trading days, bars show the 95% interval, the dashed line is the 80% announced

Coverage by market and horizon.

Backtest 2012 to 2026 on public NYMEX contracts, two naive baselines next to every result.

  • 1 trading day, 80% announced79.2–80.8%
  • 5 trading days78.4–80.6%
  • 21 trading days77.2–80.6%
  • 95% range, uncalibrated then calibrated91.1–94.8% → 92.8–95.3%
  • Trading days per cell3,661–3,701
  • Live record, kept apart from the backtestfrom launch day
The full table
Out-of-sample coverage of the calibrated ranges, by market and horizon. Backtest 2012–2026, public data. Generated 2026-11-03 03:13.
MarketHorizonDays tested Real outcomes inside the rangeUncalibrated 80%normal range, same volatility Worst year80% rangeSize of move: model vs no changemean absolute error
50% announced80% announced[95% CI]95% announced
BrentNYMEX BZ1 d3,68348.9%80.4%[79.1–81.7]95.0%81.7%2018: 77.4%1.590 vs 1.592%ratio 0.998 · no gain
5 d3,67948.8%80.2%[78.0–82.2]94.6%79.9%2014: 73.3%3.685 vs 3.691%ratio 0.999 · no gain
21 d3,66351.0%79.6%[74.1–84.2]94.2%79.0%2014: 52.2%7.872 vs 7.839%ratio 1.004 · no gain
WTINYMEX CL1 d3,70049.8%80.5%[79.2–81.8]94.5%81.4%2026: 76.6%1.767 vs 1.766%ratio 1.001 · no gain
5 d3,69651.2%80.2%[78.0–82.3]94.9%80.5%2018: 71.8%3.986 vs 3.980%ratio 1.002 · no gain
21 d3,68050.1%80.1%[75.5–84.1]94.1%78.9%2014: 65.1%8.656 vs 8.537%ratio 1.014 · no gain
ULSD NY HarborNYMEX HO1 d3,70048.9%80.2%[78.8–81.6]95.0%80.2%2013: 76.6%1.595 vs 1.596%ratio 0.999 · no gain
5 d3,69650.2%80.6%[78.6–82.7]94.4%79.1%2014: 75.9%3.632 vs 3.626%ratio 1.002 · no gain
21 d3,68051.4%78.6%[73.9–82.9]93.8%76.2%2014: 63.1%8.252 vs 8.100%ratio 1.019 · no gain
RBOB gasolineNYMEX RB1 d3,69949.2%80.8%[79.5–82.2]95.0%81.5%2022: 78.0%1.670 vs 1.674%ratio 0.997 · no gain
5 d3,69548.8%79.8%[77.7–81.9]95.0%80.2%2014: 72.3%3.770 vs 3.788%ratio 0.995 · no gain
21 d3,67951.7%80.6%[76.4–84.8]94.3%78.0%2014: 67.5%7.928 vs 8.012%ratio 0.990 · no gain
Diesel crack (HO − Brent)HO − Brent1 d3,68247.1%79.3%[77.8–80.7]94.7%82.4%2019: 75.5%0.929 vs 0.930 $/bblratio 0.998 · no gain
5 d3,67849.2%79.5%[77.4–81.7]94.1%81.2%2017: 74.7%2.035 vs 2.035 $/bblratio 1.000 · no gain
21 d3,66248.9%77.2%[72.9–81.4]92.8%79.5%2021: 67.7%4.184 vs 4.205 $/bblratio 0.995 · no gain
Gasoline crack (RB − Brent)RB − Brent1 d3,68150.0%79.2%[77.7–80.7]95.3%81.9%2026: 73.6%0.826 vs 0.830 $/bblratio 0.996 · no gain
5 d3,67748.3%78.4%[76.2–80.7]95.3%81.2%2026: 69.6%1.823 vs 1.826 $/bblratio 0.998 · no gain
21 d3,66148.9%78.4%[74.6–82.2]94.7%82.2%2026: 62.8%3.618 vs 3.638 $/bblratio 0.994 · no gain
Henry Hub gasNYMEX NG1 d3,70149.7%79.9%[78.6–81.2]94.9%80.5%2024: 75.9%not tested
5 d3,69751.6%80.4%[78.3–82.5]94.2%80.4%2024: 73.5%not tested
21 d3,68151.4%79.7%[75.4–83.8]94.8%79.0%2022: 64.4%not tested

Coverage data in JSON Direction backtest in JSON

What does not work.

Up or down beat a naive baseline in 0 of 18 tests, so we price the range.

  • Direction, main model against no change, tests won0 of 18
  • Accuracy of the up-or-down calls45.2–55.9%
  • Size of the move, mean absolute error against no change0.990–1.019
  • Managed-money positioning, tests significant after Holm correction0 of 6
  • Direction when the 21-day model said 70% sure or more26.7% right on 176 cases
  • The same protocol on a planted signal, out of sample53–62%
Every test
Share of up-or-down calls that were right, main model against the best naive baseline for that market and horizon, with the 95% interval of the gap in points. Backtest 2012–2026, about 3,660–3,680 days per cell.
Market1 day5 days21 days
Brent52.5% vs 52.3%drift · gap [-1.3, +1.8]51.7% vs 52.3%drift · gap [-4.2, +2.9]51.6% vs 55.2%drift · gap [-10.7, +3.2]
WTI50.9% vs 51.2%drift · gap [-1.9, +1.4]49.9% vs 50.7%momentum · gap [-4.5, +3.0]47.7% vs 50.8%momentum · gap [-10.7, +4.5]
ULSD NY Harbor51.1% vs 51.6%drift · gap [-1.6, +0.7]49.4% vs 51.7%momentum · gap [-6.0, +1.4]45.2% vs 52.8%momentum · gap [-14.2, -1.0]
RBOB gasoline51.8% vs 52.9%drift · gap [-2.5, +0.2]51.6% vs 53.1%drift · gap [-3.7, +0.9]55.9% vs 57.4%drift · gap [-5.3, +2.2]
Diesel crack (HO − Brent)49.3% vs 51.2%momentum · gap [-4.2, +0.3]50.3% vs 49.9%mean reversion · gap [-3.3, +3.8]49.8% vs 52.2%mean reversion · gap [-7.9, +3.3]
Gasoline crack (RB − Brent)52.6% vs 52.4%drift · gap [-0.8, +1.2]51.3% vs 53.1%drift · gap [-4.2, +0.4]52.6% vs 55.4%drift · gap [-5.3, -0.6]
Managed-money positioning from the CFTC and ICE Commitments of Traders, used as a direction signal. Model against the drift baseline, 95% interval of the gap in points, Holm correction over the 6 tests.
MarketHorizonWeeksModelDriftGapSignificant
Brent ICE1 wk61653.1%53.4%-0.3 [-4.7, +4.2]no
Brent ICE2 wk61555.4%49.8%+5.7 [-0.3, +11.9]no
Brent ICE4 wk61356.8%55.1%+1.6 [-4.6, +8.0]no
WTI NYMEX1 wk86046.4%51.6%-5.2 [-10.5, +0.1]no
WTI NYMEX2 wk85950.5%52.4%-1.9 [-8.5, +4.7]no
WTI NYMEX4 wk85745.5%53.0%-7.5 [-14.6, -0.7]no

A backtest is not a track record.

Only the live record counts, and it is published with its misses.

  • Worst cell in fifteen years52.2% for Brent at 21 days in 2014
  • Public NYMEX proxies until a Gulf feed is connectedlimit
  • Ranges centred and symmetric, beyond 21 days untestedlimit
  • On your own book, ours next to yours in dollarsstatement on day 14

Start on your own book.

One trading week on your own book, with no card and no call.