TAILFIFTY · RESEARCH

Will WTI oil on Sep 21, 2026 close above $95.99?

Oil Price (WTI) on Sep 21, 2026? — Above $95.99

commodities2dasked 2026-09-19 · resolves on WTI.close level>= 95.99

Committee55%
Market54%
OutcomeNOBrier 0.303 · market 0.286

Committee

Commodities Analyst55%was 57%
Market Positioning Analyst52%was 52%
Quantitative Researcher54%was 50%
Public Markets Analyst58%was 59%
Weather Analyst53%was 56%
Causal Skeptic55%was 52%

Top disagreement

The committee-wide fault line (raised most sharply by Analyst A against every other analyst, methodological weight 0.75) is: does the -8.6% three-session WTI crash (101.9 -> 96.73) represent a mean-reverting sentiment/positioning overreaction (favors YES, price stabilizes above the strike) or the first leg of a fundamentally justified re-rating driven by a real supply/demand shift -- OPEC+ output surprise, a large Cushing/EIA inventory build, or a ceasefire/geopolitical-premium unwind (favors NO, momentum continues below the strike)? This is unresolved because no analyst -- across six independent workstreams -- obtained futures curve structure (backwardation vs. contango), EIA/Cushing stock data, or OPEC+ signals. All six instead debated the shape of the unconditional historical return distribution (sd, skew, kurtosis, autocorrelation) computed on what is functionally one shared underlying series, which cannot distinguish these two regimes by construction: a random-walk/return-moments framework is regime-agnostic and silent on causal driver. RESOLUTION: I do not attempt to adjudicate this with data nobody has. Instead I (1) treat the fat-tail/negative-skew evidence (independently corroborated by B, D, E on the same series) as the correct correction to the naive symmetric random-walk estimate, pulling fair value down from ~0.57-0.60 toward 0.52-0.58 -- consistent with where the calibration-and-relevance-weighted panel actually converges; and (2) widen the credible interval substantially beyond the analysts' tight 0.52-0.58 point-estimate band to [0.47, 0.63] to price in the unmodeled tail risk that this is in fact a fundamental repricing, which the current evidence base is structurally unable to rule out either way.

Synthesis

SYNTHESIS: WTI close >= $95.99 on 2026-09-21 (2-day horizon from as_of 2026-09-19 00:54 UTC) CONSENSUS: P(YES) = 0.55, 90%-ish credible interval [0.47, 0.63]. This is a calibration-and-relevance-weighted synthesis, not an average: weighted point estimate using the weights above is 0.546, rounded to 0.55 -- essentially the modal cluster of D (0.58), F (0.55), A (0.55), C (0.54), with B (0.52) and E (0.53) pulling slightly lower. I widen the CI well beyond the analysts' tight 0.52-0.58 stated range because the panel's entire evidence base is a single, undifferentiated random-walk/return-distribution framework applied to what is functionally one shared price series -- it cannot see the one variable (futures curve structure) that would actually tell us whether the crash is transitory or structural. EXPECTED RETURN: 2-day expected log return of WTI is approximately 0.0% to +0.3% (point estimate +0.1%), consistent with near-zero measured autocorrelation across every correctly-identified WTI return series (A: 0.0205, B: 0.0149, D: 0.0118) and no validated drift signal from any analyst. TOP DISAGREEMENT AND RESOLUTION: The real fight is mean-reversion (overreaction, bullish) vs. fundamental re-rating (structural, bearish) -- and it is unresolved because nobody in the six-analyst panel obtained curve, inventory, or OPEC+ data (flagged repeatedly by Analyst A, the domain specialist, against every other analyst). I resolve this not by picking a side but by (a) using the fat-tail-corrected return-distribution estimates as the best available proxy, since they at least move off the naive symmetric-normal assumption that ignored the possibility of a fatter left tail, and (b) pricing my uncertainty about the unresolved regime question directly into a wider credible interval rather than a false-precision point estimate. DATA QUALITY (0.5): The core spot-price fact is solid and cross-verified (multiple analysts independently reconstruct 96.73 from the same log-return check), but the return-moment inputs (sd 2.61%-3.86%) show troubling window-sensitivity for what should be one number, and one analyst's entire evidence set (C) was built on a mislabeled instrument. No physical/fundamental data exists anywhere in the evidence base. STATISTICAL QUALITY (0.55): Standard toolkit (ADF, bootstrap CI, autocorrelation, skew/kurtosis) applied competently; F's regime-split and placebo event-study tests are the most rigorous falsification attempts in the panel and meaningfully raise the floor. No walk-forward or true out-of-sample backtest was run by anyone (explicitly conceded by C). NOVELTY (0.2): Low. This is a standard near-the-money random-walk problem; no analyst produced an insight that isn't a variant of "measure the return distribution and compare to the buffer." CROWDING (0.85): Very high. Three facts (spot level 96.73, the shared return-moment series, and the Kalshi 0.54 mid) are each cited by four to six of the six analysts as if independently confirmatory. Collapsed to their true information content, the panel really only contributes: (1) the spot-level fact, (2) one return-distribution-shape estimate with real window uncertainty, (3) C's retraction eliminating the lone momentum data point, (4) F's independent regime/placebo tests, and (5) B's volume observation. That is five lines of evidence dressed up as twenty. TRADEABILITY: NOT tradeable at meaningful size against the current 0.54 market mid. My synthesized edge (0.55 vs 0.54, ~1 point) is well inside both the panel's own estimation noise (sd estimates alone span a range that would move fair value by +/-5-8 points) and any realistic transaction cost/spread on this contract. Every analyst who attempted a rigorous out-of-sample or regime-stability test (C's full concession, F's t6/t8) found no validated edge beyond the market price. If forced to express a view, the only defensible trade is a small position sized to the genuine residual uncertainty (e.g., a few percent of capital long YES, reflecting the marginal 1-point tilt from the fat-tail-adjusted consensus), explicitly framed as a low-conviction convergence bet rather than an edge trade -- and I would not recommend it given fees/spread will likely exceed the modeled edge. The more useful action is non-tradeable: flag that resolving the curve-structure question (backwardation vs. contango over 9/16-9/18) would be worth more to this forecast than any further return-distribution analysis.

What is what

Observed

  • WTI closed at $96.73 on 2026-09-18, down -5.08% from $101.90 the prior session, and down ~8.6% over the preceding 3 sessions.
  • The $96.73 close is 0.77% (=$0.74) above the $95.99 strike.
  • Kalshi's mid-price for this contract at the as_of timestamp (2026-09-19 00:54 UTC) was 0.54.
  • Daily WTI log-return autocorrelation-1 is near zero across every correctly-scaled measurement (A: 0.0205 on n=249; D: 0.0118; B: 0.0149) -- no momentum or mean-reversion signal in the tape.
  • Trading volume declined monotonically through the selloff, hitting its lowest 6-session print (2.165e5) on the day of the largest price drop (Analyst B, d5).
  • Analyst C's original evidence series (values 3.2-3.5) is off by ~30x from any plausible WTI price band and was retracted by C as a likely Henry Hub natural gas series, not WTI.

Inferred

  • The retraction of C's series removes the only positive-autocorrelation (momentum-continuation) data point in the panel, which mechanically firms up the 'no exploitable momentum' read shared by A, B, D.
  • Estimates of daily return sd on the same underlying WTI series range from 2.61% to 3.86% across five analysts -- this spread is itself informative: it signals meaningful window/estimation sensitivity, not five independent confirmations of one number, and argues for wider uncertainty rather than a tighter point estimate.
  • Declining volume into the crash is consistent with either low-conviction/thin-crowding (bullish for mean-reversion) or holiday/weekend-adjacent thin liquidity mechanically amplifying a smaller net order flow (a market-microstructure artifact, directionally neutral) -- the data cannot distinguish these, so this evidence is weaker than Analyst B's framing implies.
  • The magnitude of the selloff (-8.6% in 3 sessions with no identified fundamental catalyst in any analyst's evidence) is somewhat atypical for physical crude absent a real supply/demand event, which is weak, non-quantified grounds for leaning slightly toward mean-reversion (YES) -- but this cannot be verified without curve or inventory data.

Modeled

  • A symmetric normal random-walk model scaling daily return sd to a 2-day horizon and centering on the +0.77% buffer, which produces raw fair-value estimates in the 0.57-0.60 range (A, B, D, E each ran a version of this).
  • A skew/kurtosis-adjusted correction to the above (negative skew approx -0.42 to -0.45, excess kurtosis approx 4.7-5.2) applied by B and implicitly by D and E, pulling the fair-value estimate down to approximately 0.50-0.53.
  • Analyst F's regime-split correlation test (full-sample corr approx -0.0056 with a sign flip across regimes, same-sign share 0.857) and event-study CAR-vs-placebo comparison (mean 0.0062, 95% CI [-0.021, 0.033], p=0.669), both of which fail to find a stable, causal, cross-series signal.
  • An ADF unit-root test on the WTI level series (non-stationary, consistent with normal commodity price behavior; this test cannot distinguish a transitory headline move from a structural balance shift, per Analyst A vs D disagreement).

Predicted

  • Consensus P(WTI close >= $95.99 on 2026-09-21) = 0.55, credible interval [0.47, 0.63].
  • Expected 2-day log return of WTI over the horizon is approximately 0.0-0.3%, statistically indistinguishable from zero given near-zero autocorrelation and no confirmed drift signal (point estimate used: +0.1%).
  • Falsification: WTI front-month settlement below $95.99 on 2026-09-21 falsifies the YES case; at or above $95.99 confirms it. A secondary falsification for the underlying regime debate: if the futures curve is later observed to have shifted from backwardation toward contango coincident with the 9/16-9/18 selloff, that would retroactively validate the 'fundamental repricing' (bearish/momentum) hypothesis that this synthesis could not test; persistent backwardation would validate the 'overreaction/mean-reversion' (bullish) hypothesis this synthesis leans toward.

Uncertain

  • Whether the -8.6% 3-day crash is a sentiment/positioning overreaction (mean-reverting, bullish for YES) or the start of a fundamentally justified re-rating (persistent, bearish for NO) -- unresolved by any analyst; no futures curve, EIA/Cushing inventory, SPR, or OPEC+ data was examined by anyone in the six-analyst panel.
  • The true daily-return volatility regime: estimates range 2.61%-3.86%/day depending on window, a ~50% relative spread that materially changes the random-walk breach probability (roughly 0.40-0.60 fair value depending on which sd is 'correct').
  • Whether declining volume through the selloff reflects genuine low conviction/low crowding (supports mean-reversion) or mechanical thin-liquidity amplification (regime-neutral) -- Analyst A's rebuttal of Analyst B's interpretation is not resolved by available data.
  • Whether Analyst D's calibration statistic (n=1, brier=0.2025, acc=1.0) is representative of true forecasting skill in this segment, given the extremely small sample.

Falsification

← all research

Every forecast is pre-registered with a resolution rule and scored when it resolves. Numbers come from deterministic tests over point-in-time data. Every seat on the committee is an AI model with a role prompt, not a person or a licensed professional. TailFifty is an educational research experiment run by an individual. Probabilities and paper trades are AI-generated and simulated; there are no real positions. Nothing here is investment, trading or betting advice, or an offer or solicitation. Not a registered investment adviser, broker or CTA. Hypothetical and simulated results do not reflect real trading, liquidity or costs, and past forecast accuracy does not guarantee future results. Data: NOAA, NHC, CFTC, Open-Meteo (CC BY 4.0), CoinGecko, Kalshi and others; derived figures only. © 2026 Harrison Johnson. About and disclosures.