Goldeneye

Signal Lab

Model ensemble · directional signal
View
Vol estimator
Ensemble signal · 1-dayBullish
Expected move+5.19%range 0.16% – 9.90%
Agreement
3 of 4
elevated regime
Model agreement3 0 1 of 4input diversity: high

No proven directional edge at this 1-day horizon (walk-forward testing). Read this as a view to reason about, not a probability — the calibrated edge is the volatility range below.

· 3 of 4 models agree on bullish direction.

· Mixed price + fundamental signals (high input diversity).

· Votes weighted by measured calibration (Brier) — up: holt_trend; down: moving_average_directional.

moving average directional1d
Bullish
+13.05%
closes
SMA-20/50 cross
SMA-20 (3.046) is above SMA-50 (2.860) by 6.52%, suggesting bullish momentum.
RSI reading
RSI at 65 reads as moderately overbought; historically precedes short-term consolidation.
holt trend1d
Bullish
+0.69%
closes
Holt linear trend
Smoothed trend of +0.347%/day (alpha=0.9, beta=0.05) projects bullish over 1d.
Smoothing projection, not a guarantee
One-step residual scale is 14.12% of price; the projection extrapolates the recent trend and ignores regime shifts.
factor composite1d
Bullish
+0.50%
closesstoragecot
EIA storage delta vs 5-year seasonal norm (consensus-free proxy)
EIA storage delta vs 5-year seasonal norm (consensus-free proxy): -9.4 Bcf (smaller build / larger draw → bullish).
Not a trained model
Rules-based composite of storage, positioning, and momentum signals — every weight is hand-set, not learned. A trained model replaces it in a later phase.
logreg directional1d
Neutral
+0.22%
closes
trend gap vs SMA-20
raises the modelled P(up).
Trained per-call on a rolling window
In-sample fit accuracy 59% (not out-of-sample); coefficients re-estimate each call.
Backtest Performance · 1d horizon
Loading backtest summary…
Honest backtest · look-ahead-safe · 50% line = coin-flip baseline · how we validate →

Model Calibration

Loading…

Reliability of each model's stated confidence vs. its realized hit-rate over backtested history. The tick is what the model claimed; the bar is what happened. Brier scores the gap (lower is better; 0.25 ≈ coin flip). Descriptive research diagnostics, not advice.

Explanation

NG (logreg_directional) — Signal Explanation The ensemble reads as bullish on NG, with 3 of 4 models aligned in that direction and the highest-calibration model (holt_trend) among the bulls. The strongest supporting factor appears to be the EIA storage delta versus the 5-year seasonal norm (approximately -9.4 Bcf), which suggests the market is running below seasonal inventory levels — a structurally supportive fundamental backdrop that factor_composite treats as a consensus-free signal. The most significant contradicting factor is the managed-money net week-on-week delta of -18,027 contracts, which suggests speculative positioning is being trimmed and could indicate that large traders are fading the directional move; the logreg_directional model returning neutral, weighted against the bullish cluster, reinforces this caution. Confidence band: moderate-to-high, reflecting strong model agreement but tempered by elevated volatility regime and the mixed price-momentum picture (RSI and 10-day momentum both flagged as contradicting factors across models). The primary invalidation risk is a sharp reversal in the storage deficit — whether through demand destruction, supply surprise, or a revision to the EIA print — which would undercut the core fundamental premise driving the ensemble.

History
Horizon EndModelDirConfExp %Real %ΔOut
Supporting News · Live
Loading news…
Sources: EIA · Yahoo Finance · filtered for NG-relevance · auto-refresh 5m