Loading…
—
- Gas, 5 days
- —
- VIX
- — close
- Last spike
- —
Natural gas and its spikes
Does a gas spike mean anything for stocks?
S&P 500 after each spike, against every day in the sample and against days with a similar VIX. The p-value is the share of placebo tests (the same spike pattern shifted in time) that show a gap at least as large; below 0.05 would be evidence of an effect.
| Horizon | Spikes | S&P after a spike | S&P, all days | Gap | Gap at same VIX | Fell ≥5%: after a spike | Fell ≥5%: all days |
|---|
Does the answer depend on the definition?
The same 3-month test with other thresholds, in sub-periods, and with the fixed-percentage rule this site used before.
| Variant | Spikes | S&P after a spike | S&P, all days | Gap | Gap at same VIX | Fell ≥5%: after / all |
|---|
Every spike
S&P 500 returns are measured from the close on the spike day. A dash means not yet known.
| Date | Gas price | Gas, 5 days | Size (z) | VIX | S&P +1m | S&P +3m | S&P +6m | Worst fall in 3m |
|---|
Methodology
How a spike is defined and tested
- Roll-adjusted gas returns. NG=F is the front-month Henry Hub future, unadjusted. On the second-to-last trading day of each month it switches to the next contract, and the price gap between the two contracts shows up as a move that never happened (about +6% on average in autumn, −6% in January). Those daily returns are dropped.
- Spike = 5-day gas return of at least 2.5 standard deviations, where the standard deviation is taken from the 252 trading days before the window. Gas volatility ranges from 30% to over 100% a year, so a fixed threshold such as +15% means different things in different years.
- One event per episode. After a spike, no new one can fire for 21 trading days. The date is the first day the threshold is crossed and never moves afterwards.
- Outcome = S&P 500 return 1, 3 and 6 months after the spike-day close, and whether it closed 5% or more below that level at any point in between.
- Baseline. The same outcome averaged over every day, and over days in the same VIX bucket (<15, 15–20, 20–30, ≥30). High VIX on its own goes with larger drawdowns, so without this control a "gas + VIX" signal mostly measures the VIX.
- p-values by circular shift. Spikes cluster and the forward windows overlap, so the events are not independent draws. The whole set of spike dates is shifted through the sample one day at a time, the gap is recomputed for every shift, and the p-value is the share of shifts with a gap at least as large (two-sided).
What changed from the first version
- No baseline. It reported that 36% of spikes were followed by a 5% S&P drawdown within 3 months. About a third of all days are.
- Roll artifacts counted as spikes. 43 of its 102 events had a contract switch inside the 5-day window.
- Filters fitted to the outcome. Four "fakeout" rules were chosen after seeing which spikes failed, then scored on the same spikes. That guarantees a good-looking result and says nothing about the next one.
- Look-ahead. An event was replaced when a larger spike arrived within 15 days, and the drawdown window included the spike day's own low.
- Hard-coded statistics in the page that did not update with the data, and trading advice built on them.
Limitations
- Around twenty spikes in 25 years is a small sample. The test can rule out a large, reliable effect; it cannot rule out a small one.
- The roll adjustment removes one day per month by calendar rule. It is an approximation of a properly back-adjusted series built from individual contracts.
- The VIX control uses full-sample bucket averages. That is fine for asking whether gas adds information, not for trading.
- Three horizons and several variants are shown; with that many tests, an occasional p-value under 0.05 is expected by chance.
- Henry Hub is a US domestic price. European gas (TTF), which drove the 2022 energy crisis, is not covered.