Will Natural gas price on Nov 4, 2026 at 5:00 PM EST close above $2.890?
6 seats worked this question alone; the panel agreed within a few points, so no debate was run. Code ran every test and computed the number. How the desk works
Judged by Kalshi KXNATGASD-26NOV0417-T2.890 settlement at 2026-11-04: If the close price of the 1-minute candlestick for natural gas on November 04, 2026 at 5:00 PM EST is above 2.890 USD/MMBtu, then the market resolves to Yes.
The dispute that matteredAll six analysts agree directionally (53-57% YES) but disagree sharply on the magnitude of daily volatility feeding their models: A/C report sd≈2.7-2.8%, while E reports sd≈7.9% on what appears to be the same instrument/window. This is not a disagreement about the sign of the edge but about how much noise surrounds the 0.24% cushion, which should set the width of the credible interval. Resolved by treating E's 7.9% figure as likely a different return-window/annualization artifact (not reproduced by any other analyst) and anchoring volatility assumptions on the A/C/D cluster (~2.7-3%), while still widening the final CI beyond a pure vol-implied band to account for the unresolved discrepancy and for the acknowledged intraday-settlement-vs-daily-close model gap (flagged independently by Analyst B).
Synthesis
- Nov 3, 2026 close = $2.897, which is $0.007 (0.24%) above the $2.890 Kalshi strike.
- 21-day price series: sd=0.211, autocorr_1=0.906 (level series).
- Daily log-return distributions reported with sd ranging 2.69%-2.78% (A,C,D) vs. 7.9% (E) — an unreconciled discrepancy.
- ADF test on price series: p=0.069 (cannot reject unit root at 5%).
- Lag-return regression: beta=-0.054, p=0.673, with one high-leverage observation (Cook's D=1.23).
- Probability of closing above strike modeled via normal/empirical approximation to the daily log-return distribution, applied to a 0.24% downside threshold.
- Calibration-weighted log-odds pooling across six analyst probabilities (deterministic pool = 0.55).
- Historical-analog empirical quantile model (D) estimating P(return > -0.24%) from 345 past daily moves.
- The series behaves approximately as a mean-zero random walk at daily frequency — no reliable momentum or mean-reversion signal survives statistical scrutiny.
- The 0.24% cushion is small relative to typical daily volatility (roughly 0.08-0.09 standard deviations under the A/C/D vol estimates), implying the outcome is close to a coin flip with a modest statistical tilt toward YES from the starting-position effect alone.
- Weather/storage/HDD signals are directionally mildly bullish but statistically insignificant (p=0.084) and should not materially move the probability away from the random-walk base rate.
- Settlement close on 2026-11-04 at 5:00 PM EST will be above $2.890 with probability ≈0.55 (90% CI: 0.49-0.61), i.e. only marginally better than a coin flip.
- No single fundamental catalyst (weather, storage, momentum) is expected to produce a decisive move away from the random-walk base rate within this 1-day horizon.
- Which volatility estimate (2.7% vs 7.9%) correctly characterizes the relevant return window for Nov 3→Nov 4 — this materially affects how tight the true probability band should be.
- Whether the actual Kalshi settlement rule (reported as a specific intraday/1-minute price, per Analyst B's caveat) could diverge meaningfully from a simple daily-close random-walk model.
- Whether any market-implied probability (bid/ask on the Kalshi contract itself) is already priced near 0.55-0.57, which would determine whether this view has any tradeable edge at all — this was not supplied to the committee and is a key missing input.
| Causal Skeptic (F) | 0.22 | Best calibration in segment (brier 0.2596, lowest/best of panel) and broad cross-domain relevance; explicitly tests and rejects momentum/mean-reversion (ADF p=0.069, regression beta p=0.673) rather than assuming it, making this the most rigorous 'null hypothesis' check. Treated as near-independent of the shared d1 fact since its contribution is methodological (unit-root/regression tests) not just position-vs-strike. |
| Commodities Analyst (A) | 0.20 | Highest domain relevance (1.0 commodities) and solid calibration (brier 0.267). Adds a genuinely distinct data point (autocorr_1=0.906 on levels) not duplicated by others, though its implication is partly in tension with F's unit-root finding — flagged as model uncertainty, not resolved. |
| Weather Analyst (B) | 0.16 | Strongest calibration accuracy (0.587) in segment but lower confidence (0.45) and admits a key model gap (daily vol model vs. actual intraday/1-min settlement mechanic). Weather/storage signal itself is non-significant (p=0.084), so contributes mainly a caveat, not fresh directional evidence. |
| Quantitative Researcher (C) | 0.15 | Good calibration (0.9597 influence) but its core input (return mean -0.12%, sd 2.69%) duplicates the same return-distribution logic used by A, D, E (differing only in window/parameters), so down-weighted for non-independence. HDD signal is directionally consistent but weak/unquantified. |
| Historical Analog Analyst (D) | 0.14 | Smallest sample (n=26) and weakest calibration (brier 0.292) of the panel — lower trust. Empirical-quantile approach is methodologically distinct from parametric vol models (adds some independence) but ultimately re-derives the same ~53-57% band. |
| Market Positioning Analyst (E) | 0.13 | Lowest confidence (0.35) and reports an outlier volatility estimate (sd 7.9%, mean -0.38%) sharply inconsistent with A/C's 2.7-2.8% sd on what should be the same underlying series — this internal inconsistency is itself a data-quality flag and is the basis for down-weighting rather than treating as a fresh independent signal. |
- If the natural gas settlement reference price used by Kalshi for Nov 4, 2026 5:00 PM EST closes/settles at or below $2.890, this forecast (P(YES)=0.55) is falsified as directionally wrong, though a single observation cannot invalidate the probabilistic calibration.
- If realized daily volatility over Nov 3→Nov 4 is confirmed (via subsequent data) to be close to 7.9% rather than ~2.7-2.8%, the credible interval used here was too narrow and should be revised wider in future near-ATM settlement forecasts of this instrument.
The seats
Commodities Analyst57%
53% Unconditional probability a single day's close lands above a strike set ~0.2% below the prior close, under a random-walk/no-drift assumption using the realized daily-return distribution (sd≈2.78%).
Daily NG returns have a wide realized range (min -6.96%, max +8.68% in d2) that easily exceeds the thin 0.24% buffer; natural gas is one of the most volatile commodities intraday, and the 'last' price I'm anchoring to (2.897) is a prior-day settle, not the current live price — if price has already drifted down between 03:41 UTC and now, my edge could already be gone or reversed. Additionally, d6 (likely a storage/flow series) shows a widening negative trend (-255,738 most recent vs mean -214,200), which could signal either tightening (bullish) or an unusually large surplus/injection (bearish) depending on what the series actually measures — I cannot fully resolve its directional implication without a label, which weakens my fundamentals-based tilt.
- Last observed close 2026-11-03 = 2.897, already 0.007 (0.24%) above the 2.890 strike, so YES requires only that price not fall; starting position favors YES.
- 21-day price series sd=0.211, autocorr_1=0.906 — strong day-to-day persistence in level, meaning a close near 2.897 one day ahead of settlement tends to stay in that neighborhood rather than revert sharply.
- Daily log-return sd=0.0278, mean≈0, skew≈0.04 (near-symmetric) — implies roughly 53-55% unconditional probability of staying above a threshold only 0.24% below last close.
If the Kalshi 1-minute candle close for natural gas at 5:00 PM EST on Nov 4, 2026 is ≤ 2.890 USD/MMBtu, this forecast (P(YES)=0.57) is falsified on the downside; if settlement is comfortably above 2.890 (e.g., >2.950), it would support the bullish fundamentals tilt beyond pure random-walk expectation.
- d1 last value (2026-11-03=2.897) is the most recent settle/reference price for NG front-month, used as starting point for Nov 4 close.
- d2 is daily log-return series for the same contract; its unconditional sd (~2.78%) is a reasonable proxy for one-day forward volatility absent new information.
- d3/d4/d5 represent heating-degree-day or weather-demand forecasts (rising into early Nov) and d6/d7 represent storage/rig or supply-side series; directionally these matter but I cannot fully resolve their labels.
- No major scheduled data release (EIA storage report typically Thursday) falls between as_of (Nov 4 03:41 UTC) and settlement (Nov 4 22:00 UTC) that would be unusually large.
- Random-walk-with-zero-drift approximation is adequate for a 1-day horizon; autocorr_1 of returns is near zero (0.013), supporting no strong momentum in returns themselves.
debate skipped: panel unanimous within 0.04
Weather Analyst56%
Loading…
Quantitative Researcher54%
Loading…
Historical Analog Analyst55%
Loading…
Market Positioning Analyst56%
Loading…
Causal Skeptic53%
Loading…
Claims and evidence
The debate
| Commod | Weathe | Quanti | Histor | Market | Causal | |
|---|---|---|---|---|---|---|
| Commodities | ||||||
| Weather | ||||||
| Quantitative | ||||||
| Historical A | ||||||
| Market Posit | ||||||
| Causal Skept |
Nov 3 close (2.897) sits only 0.24% above the 2.890 Kalshi strike. With daily NG volatility near 2.78% (sd of log returns) and strong level-persistence (autocorr_1=0.906), this is close to a coin-flip tilted slightly toward YES by starting position, reinforced by mixed-but-net-mildly-bullish weather/storage indicators.
Base rate for a mean-zero random walk suggests ~53% probability of closing above a strike just 0.24% below the last observed price. Given the lack of strong directional signal from weather (d3 choppy, no trend), insignificant storage correlation (s5, p=0.084), and a failed attempt to compute a clean degree-day anomaly (d4), I see no strong catalyst to deviate much from the random-walk base rate. I nudge slightly above 50/50 to ~0.56 because the last close is already above the strike and recent 2-day price action was modestly upward (2.856→2.897), but I hold only moderate confidence given the proximity of price to strike and the intraday (1-minute) settlement mechanic that my daily-volatility model does not fully capture.
Price is currently just above the strike (2.897 vs 2.890, a 0.24% cushion) with one trading session remaining before settlement. Under a near-random-walk model of daily log returns (empirical mean≈-0.12%, sd≈2.69%, from s1/s2, n=83), the probability of staying above the 0.24% downside threshold is only marginally above 50%, since the required move is small relative to typical daily volatility. Weak evidence of rising heating-degree-day forecasts (d5/d6, trending up through early Nov) suggests modest upside bias to gas demand/price, partially offsetting the slightly negative unconditional drift.
Given the strike (2.890) sits only $0.007 (0.24%) below the last observed close (2.897), this is a near-ATM settlement question best resolved by the empirical daily-return distribution rather than any strong directional thesis. Base rate from quantile interpolation of 345 daily returns suggests ~55% probability of staying above the threshold, modestly supported by seasonal bullish fundamentals (declining storage, rising heating-degree-day forecasts) but tempered by well-documented negative-skew tail risk in this market's historical returns.
Price is sitting almost exactly on top of the strike ($2.897 vs $2.890, a 0.24% cushion) going into a settlement window where daily nat-gas volatility (sd≈7.9%) is roughly 30x larger than that cushion. This is a near-coin-flip event; any apparent edge from 'last close was above strike' is largely swamped by noise, and near-zero return autocorrelation means momentum offers little extra confidence. A persistent, separate bearish-trending series (d3) and a collapse in traded volume (d4) both argue for caution about over-trusting the current price level as the market's settled view. I lean marginally YES because the median historical daily return is positive and the starting price already clears the strike, but the edge is thin and this is a trade with poor risk/reward for conviction positioning — the 'fair value' probability is close to 50-56%, not a high-confidence call either direction.
No causal driver in the supplied evidence reliably predicts whether Nov 4 5PM EST settlement will be above 2.890; the question reduces to a near-random-walk price currently sitting only ~0.24% above the strike, with no exploitable momentum, regime, or fundamental signal that survives scrutiny.
debate skipped: panel unanimous within 0.04
debate skipped: panel unanimous within 0.04
debate skipped: panel unanimous within 0.04
debate skipped: panel unanimous within 0.04
debate skipped: panel unanimous within 0.04
debate skipped: panel unanimous within 0.04
SYNTHESIS: The panel's apparent six-way agreement is largely an artifact of duplicated evidence — every analyst's thesis rests on the same core fact (last close $2.897 is 0.24% above the $2.890 strike), so this counts as ONE independent line of evidence, not six. Stripping duplication, the committee effectively offers only two genuinely distinct additional inputs: (1) volatility/return-distribution estimates, which disagree with each other by nearly 3x (2.7-2.8% vs. 7.9% sd) without reconciliation — a data-quality red flag — and (2) weak, statistically insignificant weather/storage signals that do not survive scrutiny (p=0.084) and are correctly discounted by the Causal Skeptic. Given this, I weight the best-calibrated, most skeptical analyst (F, Causal Skeptic, brier 0.26) and the most domain-relevant analyst (A, Commodities) most heavily, and down-weight the Positioning Analyst (E) whose vol estimate is an unreconciled outlier. The pooled/deterministic estimate of 0.55 is reasonable and I adopt it as my consensus view, but I widen the CI to [0.49, 0.61] to reflect the unresolved volatility discrepancy and the flagged gap between daily-close models and the actual (likely intraday) settlement mechanic. BASE RATE: for a near-ATM, mean-zero random-walk commodity price one session before settlement with a cushion of ~0.24%, the base rate is ~50-53% (Φ(cushion/sd) for sd≈2.7-3%); the committee's 0.55 is a modest, defensible update above this base rate driven purely by the starting-position effect, not by any fundamental catalyst. TRADEABILITY: This is NOT a high-conviction trade. The edge over a pure coin flip (0.55 vs 0.50) is thin, sits well within the panel's own CI, and importantly no market-implied/Kalshi-quoted probability was supplied for comparison — without knowing if the contract is priced at, say, 0.50 or 0.58, I cannot determine if 0.55 represents real edge or is already priced in. If the Kalshi YES price is observably below ~0.52, a small long-YES position sized for a thin edge (not a conviction bet) is justifiable; if priced above ~0.57, no trade. Absent that market price, my recommendation is to treat this as informational only, not actionable, pending confirmation of market pricing. OBSERVED: last close 0.24% above strike; conflicting vol estimates across analysts; insignificant weather/storage stats; unit-root test inconclusive. INFERRED: near-random-walk dynamics with a thin positive tilt from starting position. MODELED: pooled log-odds probability of 0.55 from six return-distribution-based individual estimates. PREDICTED: ~55% probability of YES settlement, CI 0.49-0.61. UNCERTAIN: true intraday volatility magnitude, settlement-mechanic fidelity to daily-close models, and whether any edge survives against actual market pricing.
Assumptions
- holdingd1 last value (2026-11-03=2.897) is the most recent settle/reference price for NG front-month, used as starting point for Nov 4 close.
- holdingd2 is daily log-return series for the same contract; its unconditional sd (~2.78%) is a reasonable proxy for one-day forward volatility absent new information.
- holdingd3/d4/d5 represent heating-degree-day or weather-demand forecasts (rising into early Nov) and d6/d7 represent storage/rig or supply-side series; directionally these matter but I cannot fully resolve their labels.
- holdingNo major scheduled data release (EIA storage report typically Thursday) falls between as_of (Nov 4 03:41 UTC) and settlement (Nov 4 22:00 UTC) that would be unusually large.
- holdingRandom-walk-with-zero-drift approximation is adequate for a 1-day horizon; autocorr_1 of returns is near zero (0.013), supporting no strong momentum in returns themselves.
- holdingd2 and d6 represent daily (log or pct) returns of the front-month natural gas contract used for settlement.
- holdingd1 is the daily close price series for the same contract, with 2026-11-03 close = 2.897.
- holdingReturns are approximately mean-zero and serially uncorrelated at 1-day lag (autocorr_1 ≈ 0.013 per s2), consistent with a quasi-random-walk.
- holdingNo major scheduled EIA storage report or extreme weather event occurs between 2026-11-03 close and 2026-11-04 5PM EST settlement that would be a > 2-sigma shock.
- holdingThe 1-minute candle at 5:00 PM EST will reflect the prevailing daily price level, not a momentary illiquid print far from fair value.
- holdingDaily log returns are approximately normally distributed and independent (weak autocorrelation of 0.051 per s1 supports near-independence).
- holdingThe 1-minute candle close at 5PM EST is a reasonable proxy for the 'daily close' used in the historical d1 series.
- holdingNo major scheduled news (e.g., EIA storage report) between as_of time and settlement is assumed to be already priced in or of typical magnitude; a surprise could invalidate the normal-return assumption.
- holdingLast observed daily close (2026-11-03) of 2.897 is a reasonable proxy for the pre-settlement price level entering Nov 4.
- holdingThe daily return series (d2/s1) is representative of the kind of single-day move expected between 11-03 close and 11-04 5pm EST settlement.
- holdingNo major scheduled catalyst (EIA storage report, weather shock) between 11-03 close and 11-04 5pm EST is assumed absent evidence otherwise.
- holdingSettlement candle close behaves like a continuation of the daily closing price process, not a distinct intraday regime.
- holdingd1 is daily settlement/close price series for the nat gas front-month contract (last obs 2.897 on 2026-11-03, one day before resolution).
- holdingd2 is a daily return series (fractional) with sd≈7.9%, consistent with historical Henry Hub front-month volatility in shoulder/winter-onset season.
- holdingd5 is a higher-frequency (intraday) return series with smaller sd (~2.8%), used as a check on d2's dispersion at shorter horizon.
- holdingd3 is a slowly-trending structural series (e.g., term-structure/basis spread) not the spot price itself, used only as a weak directional signal.
- holdingd4 is volume/open-interest-like series; the sharp drop to 2.179e4 on 2026-11-03 signals reduced participation/liquidity right before settlement.
- holdingNo Kalshi order-book/implied-probability snapshot was provided, so market-implied probability must be inferred from price-distance-to-strike plus volatility, not observed directly.
- holdingd2's ~6.2% sd reflects full daily close-to-close returns; actual intraday variance from 03:41 UTC to 5PM EST close is a fraction of this, but I lack a tool-verified intraday vol estimate, so I retain the full-day sd as a conservative (upper-bound) measure of uncertainty.
- holdingNo new fundamental shock (weather, storage report, pipeline outage) occurs between now and settlement that the current data does not already reflect.
- holdingThe d1 series level and d2/d3 return series refer to the same underlying natural gas contract tracked by the Kalshi settlement source.
Simulated trade
The portfolio manager proposed no paper trade. No market-implied/Kalshi-quoted probability was supplied to compare against the panel's 0.55 model estimate. The edge (0.55 vs. 0.50 base rate) is thin, sits inside the panel's own CI [0.49,0.61], and rests on duplicated evidence (one real independent signal: last close 0.24% above strike) plus an unreconciled ~3x volatility disagreement across analysts (2.7-2.8% vs 7.9% sd) that directly determines whether 0.55 or something closer to 0.50/0.60 is correct. At realistic transaction costs for a 1-day near-ATM gas settlement wager (~150-300bps round-trip), a 5-point edge is not demonstrated to survive costs, let alone crowding (0.75 — a widely-watched level everyone is pricing off the same last-close fact). Recommendation: informational only, not actionable, unless a binary contract on this exact strike is observed trading at a YES price below ~0.52, in which case re-open as a small thin-edge long. Continuous proxies (HENRY_HUB futures, UNG) don't fix this: they just re-express the same already-priced-in uncertainty with added costs and convexity mismatch versus the binary payoff being asked about.
Forecasts
| Seat | Stage | P | Base | Resolves | Outcome | Brier |
|---|---|---|---|---|---|---|
| Committee | final | 55% | — | 2026-11-04 | Yes | 0.201 |
| Commodities Analyst | prior | 40% | 53% | 2026-11-04 | Yes | 0.360 |
| Commodities Analyst | independent | 57% | 53% | 2026-11-04 | Yes | 0.185 |
| Commodities Analyst | revised | 57% | 53% | 2026-11-04 | Yes | 0.185 |
| Weather Analyst | prior | 45% | 53% | 2026-11-04 | Yes | 0.303 |
| Weather Analyst | independent | 56% | 53% | 2026-11-04 | Yes | 0.194 |
| Weather Analyst | revised | 56% | 53% | 2026-11-04 | Yes | 0.194 |
| Quantitative Researcher | prior | 45% | 52% | 2026-11-04 | Yes | 0.303 |
| Quantitative Researcher | independent | 54% | 52% | 2026-11-04 | Yes | 0.212 |
| Quantitative Researcher | revised | 54% | 52% | 2026-11-04 | Yes | 0.212 |
| Historical Analog Analyst | prior | 40% | 55% | 2026-11-04 | Yes | 0.360 |
| Historical Analog Analyst | independent | 55% | 55% | 2026-11-04 | Yes | 0.202 |
| Historical Analog Analyst | revised | 55% | 55% | 2026-11-04 | Yes | 0.202 |
| Market Positioning Analyst | prior | 45% | 52% | 2026-11-04 | Yes | 0.303 |
| Market Positioning Analyst | independent | 56% | 52% | 2026-11-04 | Yes | 0.194 |
| Market Positioning Analyst | revised | 56% | 52% | 2026-11-04 | Yes | 0.194 |
| Causal Skeptic | prior | 42% | 52% | 2026-11-04 | Yes | 0.336 |
| Causal Skeptic | independent | 53% | 52% | 2026-11-04 | Yes | 0.221 |
| Causal Skeptic | revised | 53% | 52% | 2026-11-04 | Yes | 0.221 |