Counting significant monthly price moves in ERCOT from February to July 2026, with January's winter event excluded and the SEC's codified 10% test as the threshold. Then asking what actually caused the moves that remain.
Companion to ERCOT's $500K Months, which covers Jan–Jul against a fixed-dollar threshold.
Under the SEC's 10% test, measured against the February–July average, ERCOT day-ahead moved significantly in 4 of 6 months and real-time in 3 of 6. On a month-on-month basis it is 4 of 5 and 3 of 5. So removing January does not remove the moves — roughly two months in three still qualify.
In money, on a 100 MW book: $1.95 million of variance moved through six months of ordinary trading — 15.4% of spend — and settled at a net of $42,000. The average was a non-event; the path was not.
The cause is the finding. Over these six months Texas gas traded in an 18% band ($2.30–$2.72/MMBtu). Power traded in an 82% band ($20.39–$37.16/MWh). The implied heat rate — everything that is not fuel — moved 105%. April is the clean case: the month with the year's cheapest gas produced the period's highest power price.
Which means a buyer hedged on gas was, over this window, essentially unhedged.
January 2026 contained Winter Storm Fern. Day-ahead cleared at $95.47/MWh for the month against a Jan–Jun average of $39.50 — an observation roughly eleven standard deviations from the mean of the other six. Any statistic computed with it in is a statistic about that one month.
What the exclusion buys: a clean read on how much ERCOT moves under ordinary conditions, without a single event dominating every average. That is a real and useful question, and it is the question this study answers.
What it costs, stated plainly. January was the only month of 2026 that was an overrun — the only month where price finished above the period baseline by a material margin. Removing it removes the entire downside tail from a dataset about downside risk. This study therefore measures ordinary market drift, not risk. Both numbers are true; they answer different questions, and quoting the Feb–Jul count in a conversation about protecting a buyer from cost shocks would be misleading. Use this study for "how much does ERCOT move normally"; use the Jan–Jul study for "what happens when it doesn't".
SEC Regulation S-K, Item 305, Instruction 3.A: registrants "should use changes that are not less than 10 percent of end of period market rates or prices" absent economic justification for a smaller number. It is the only hard bright line in the materiality literature, and because it is a percentage of price rather than a dollar total it needs no volume assumption — every customer gets the same count.
"A price move" needs a reference point. Three defensible ones, all measured with the same 10% test.
| Month | Hours | Gas | DA $/MWh | Δ% | ≥10% | RT $/MWh | Δ% | ≥10% |
|---|
| Baseline | What it represents | Day-ahead | n | Real-time | n |
|---|
The three agree on the direction and disagree on the count, which is itself worth knowing. Against a fixed strike at the period average — the closest analogue to a hedge — it is 4 and 3. Against the previous month, which is what a buyer actually experiences as "the price moved", it is 4 of 5 and 3 of 5. Against February as a start-of-period anchor it is 5 of 5 on both, because February was the cheapest month of the period and everything after it rose.
Worth testing, since the whole count rests on one number. Day-ahead is robust — the answer stays at 4 of 6 anywhere between a 5% and a 10% threshold. Real-time is more sensitive, moving from 5 of 6 at a 5% threshold to 3 of 6 at 10%.
| Threshold | $/MWh (DA) | Day-ahead | $/MWh (RT) | Real-time |
|---|
A wholesale power price is, to a first approximation, the price of fuel multiplied by the efficiency of the marginal generator: P = heat rate × gas. So any move decomposes into a fuel part and a heat-rate part — and the heat-rate part is grid tightness, outages, renewable output and reserve scarcity, not commodity cost.
Indexed to February, the two lines below barely resemble each other.
Splitting each month's deviation into its two causes makes the point arithmetically. The fuel effect holds gas variable and the heat rate at its period average; the heat-rate effect does the reverse. In every one of the six months the two components point in opposite directions — gas pushing prices up while the grid pushed them down, or the reverse. Fuel was never the reason a month moved; in most months it was working against the move.
| Month | Implied heat rate | Total move | Fuel effect | Heat-rate effect | Fuel share |
|---|
April 2026 is the case to remember. Gas averaged $2.30/MMBtu — the cheapest month of the entire period. Day-ahead power averaged $37.16/MWh — the most expensive month of the entire period. The implied heat rate hit 16.16 against a period average of 11.58, a 40% spike. Fuel subtracted $2.53/MWh from the month while the grid added $11.52. A buyer indexed to gas would have seen their hedge move the wrong way while their bill rose.
The count above is volume-free — that is the point of the SEC test. The consequence of those moves is not. A move of Δ dollars per MWh applied to V MWh of delivered energy moves the bill by Δ × V, so the count stays the same for every customer while the dollars scale straight down the book.
One property is worth noticing before the table: because the threshold is a percentage, the movement expressed as a share of spend is identical at every size. A 10 MW customer and a 1 GW book both saw 15.4% of their six-month day-ahead spend move around. Only the dollar figure changes.
| Avg load | MWh / 6 mo | Feb | Mar | Apr | May | Jun | Jul | Gross movement | Net |
|---|
Read the shaded cells and the last two columns together. At 100 MW the four qualifying months alone moved $1.85M — and the two months that failed the 10% test contributed only $100k between them. The threshold is doing real work: it isolates 95% of the movement into four of six months.
Dollars on a book are abstract; margin multiples are not. Both figures below are $/MWh rates, so they compare directly.
| Month | Move $/MWh | × index adder ($3–8/MWh) | % of fixed-price margin ($26–50/MWh) |
|---|
For an index retailer this is not a margin question, it is a solvency question. The average qualifying month moved $6.51/MWh against an index adder of $3–8/MWh. That is 0.8 to 2.2 times the entire per-MWh margin on the book, four times in six months, in a half-year with no weather event in it. For a fixed-price retailer carrying $26–50/MWh the same move is 13–25% of margin — uncomfortable rather than existential, which is precisely why fixed-price books survive and thin index books do not.
The single most important number in the table above is the one that looks like nothing. Across the six months, a 100 MW day-ahead book ended $42,000 away from where a fixed strike at the period average would have put it — 0.3% of a $12.7M spend. On the average, the period was a non-event.
On the path, it was not. Follow the cumulative variance month by month.
By the end of March the index buyer was $957,000 ahead of where a fixed contract would have left them. By the end of July that entire advantage had gone — the position closed $42,000 behind. Same book, same six months, opposite conclusions depending on when you looked.
Anyone reporting this period as "prices came in at about the average" would be arithmetically correct and commercially useless. The average is the one number that contains none of the information. What a treasurer needs is the trough, the worst month, and the spread — and at 100 MW those were $957k, $590k, and $1.17M against a net of $42k.
Since ERCOT eliminated unsecured credit in October 2023, every dollar of exposure is collateralised, with calls due at the close of the second business day. A book that ends the period flat still has to fund the swing while it is happening. The relevant number for sizing a credit facility is the month-to-month spread, not the net.
A buyer who reviewed the fixed-versus-index decision in March would have concluded, on real data, that going index had saved them nearly a million dollars. By July that conclusion was wrong. This is the case for continuous evidence of what was actually paid rather than a quarterly look — a mid-period snapshot of this book supports the opposite decision to the one the full period supports.
At 1 GW — a large retail book or a hyperscale campus — the same six ordinary months produce a $19.6M gross swing, a $5.9M worst month, and a cumulative trough of −$9.6M, against a net of $417k. Nothing in that half-year made the news.
What these figures are and are not. They are the variance between settling at the actual monthly index and a hypothetical fixed strike struck at the six-month average — a clean way to size the movement, not a claim about any real contract. A real buyer holds shaped, blocked or partially-hedged positions, pays a retailer's adder on top, and settles at a zonal rather than an ERCOT-wide price. Every one of those makes the real number differ from this one, and zonal settlement in particular tends to make it larger, not smaller.
Running the identical test on February–July 2025 gives an uncomfortable and useful result: it does not produce a clean "volatility is rising" story.
| Series | Six-month avg | Peak-to-trough | Coefficient of variation | Significant months | n |
|---|
Day-ahead went from 0 of 6 in 2025 to 4 of 6 in 2026 — 2025 was remarkably flat, a 19% peak-to-trough band with a coefficient of variation of 8%. Real-time went the other way, from 4 of 6 down to 3 of 6. Dispersion on day-ahead roughly tripled; on real-time it rose more modestly.
The honest reading is that day-ahead behaved very differently in the two years while real-time did not. That is a change in the day-ahead/real-time relationship, and there is an obvious candidate cause that has nothing to do with volatility.
The confound. ERCOT's real-time market was redesigned on 5 December 2025 — "RTC+B", co-optimising energy and ancillary services, replacing the ORDC price adder with per-product demand curves and splitting the single system-wide offer cap into separate day-ahead and real-time caps. It is the largest change to the market since 2010, and it sits exactly between the two comparison windows. The 2025 and 2026 series are not the same market. Do not present this year-on-year comparison as evidence of rising volatility — it cannot carry that weight, and anyone who follows ERCOT closely will know why.
This was the thing worth testing. If the only material moves in ERCOT were winter events, a risk product would be selling insurance against a once-in-five-years headline — a hard sale. Instead, two months in three clear a regulator's own significance threshold in a completely ordinary half-year. That is a recurring condition, not a catastrophe narrative.
Over this window fuel moved 18% and the non-fuel component moved 105%. A gas-indexed or gas-hedged position would have tracked a small fraction of the exposure, and in April would have moved against the buyer. If a prospect tells you they are "hedged", the useful follow-up is hedged against what — because heat-rate exposure is the part that moved and it is the part most buyers neither price nor cover.
4 of 6, 4 of 5, or 5 of 5 — all true, all from the same data, all under the same threshold. A number quoted without its reference point is not a finding, it is a choice. Say "four of six months moved more than 10% from the period average" and the claim survives scrutiny; say "prices move most months" and it does not.
Six observations, with the largest one removed on purpose. This is an illustration of a mechanism, not a frequency estimate, and it cannot support a probability, an expected value, or an annualised rate. The mechanism — heat-rate dispersion far exceeding fuel dispersion — is the durable part and is worth testing against several more years of data before it goes in front of an investor.
ERCOT-wide average energy price, day-ahead and real-time, plus the monthly natural gas price used in the same reports. Read from the Independent Market Monitor's monthly reports and cross-checked across three report vintages. January is excluded by design; August is not yet published.
(a) the arithmetic mean of the six months — a fixed strike at the period average; (b) the previous month — the move a buyer actually experiences; (c) February, the first month of the period, held flat.
A month is significant when |Pm − B| ≥ 0.10 × B. No volume enters the calculation, so the count is identical for every customer regardless of size.
With implied heat rate HRm = Pm / Gm, and bars denoting period averages: fuel effect = (Gm − Ḡ) × H̄R; heat-rate effect = (HRm − H̄R) × Ḡ; interaction = (Gm − Ḡ)(HRm − H̄R). The interaction term never exceeds $1.00/MWh here and is reported but not charted.
The same six calendar months of 2025, each measured against its own six-month average, as a base rate — with the RTC+B market redesign flagged as a confound.
Solid: the source is ERCOT's own Independent Market Monitor; the series definition is unambiguous; and the values were read from three separate report vintages with agreement to the cent across all six months in this window.
To harden: the figures were read off charts in published PDFs rather than pulled from raw settlement data. Re-derive from ERCOT's Historical DAM and RTM Settlement Point Price files before external use. Doing so would also allow the two things this study cannot do: zone-level prices, since Houston, North, West and South diverge materially and an ERCOT-wide average hides that, and hourly decomposition, where the heat-rate story almost certainly gets more extreme rather than less. The gas series is the monthly index used in the IMM reports, not a basis-adjusted delivered price at any particular hub.