Transmission Congestion in ERCOT: A $484/MWh Price Split

These are SCED locational marginal prices at two hubs from the same dispatch timestamp, not fifteen-minute settlement point prices. The source timestamps are UTC; this article uses Central Daylight Time, UTC−5.

At North Hub, $499.60 − $328.14307 leaves $171.46/MWh of congestion. At West Hub, $15.60 − $328.14307 leaves −$312.54. Subtract the two and the $484 split reappears. System energy cancels, and ERCOT publishes no marginal-loss component in its LMP.

The surrounding trace shows this was an episode, not a standing discount. At 5:05 p.m. the hubs were $23.76 apart. At 6:15 they were $484 apart. By 7:30 the gap was $37.10.

ERCOT’s SCED shadow-price report lists the constraint and contingency pairs that were active. Eight had positive shadow prices at this timestamp, and the four largest are below with ERCOT’s own identifiers.

A shadow price is the marginal dispatch cost of one more megawatt of capability on that constraint. It is not a hub’s price, a charge paid by customers, or a line’s loading percentage.

The contingency names the condition the limit is tested against, not an event that happened. And the reported flow and limit are fields in a dispatch record, not live telemetry from a physical line.

A constraint moves a location’s price by that location’s sensitivity to it. For a North-minus-West basis, the contribution is the shadow price times West’s sensitivity minus North’s, using ERCOT’s sign convention.

Ask the Grid’s completed September 18 sensitivity run put North’s factor for 35100__A/DWCSRAM5 at −0.03107 and West’s at +0.08007 against the system reference. Their difference times $4,458.48 gives about +$495.49/MWh, a bit above the net $484 because other constraints pull the other way.

Those individual factors have bootstrap standard errors of 0.00156 and 0.00207 and rest on 143 binding intervals over four distinct days. Each is more than twice its error, though that does not make a confidence interval for the difference.

The factors are estimated from historical price and shadow-price identities from July 20 to September 18, not published by ERCOT as a PTDF matrix and not a re-solved electrical case. The estimate is a strong lead, not proof of the initiating outage, weather or dispatch action.

The same run marks B4_V4_1/BASE CASE as thinly supported: 18 binding intervals on two days, and no published factor. A blank estimate is not zero exposure, and an attribution that quietly drops it would look more complete than it is.

The conclusion is layered. The price records establish the $484 congestion split. ERCOT’s constraint records establish which constraints bound. The dated model points to 35100__A/DWCSRAM5 as the dominant estimated contributor. None of it establishes why that capability became scarce.

Highlighting a line on a map would add location and still leave that causal gap open. Naming a weather event or an outage would take contemporaneous evidence tying it to the monitored constraint.

A generator exposed to West Hub and hedged at North Hub would see the opposite sign from the spread above, subject to its actual settlement and contract. The dollars follow the direction and the megawatt-hours present during the event.

A monthly average can hide an episode like this, which is why the production-weighted basis study looks at operating hours. A battery can move energy between intervals, but its power and state-of-charge limits decide how much of a brief price opportunity it can capture.

The three-hour trace covers 5–8 p.m. Central and has 37 matched SCED timestamps per hub. SCED executions are not evenly spaced, so the chart places each point at its actual timestamp rather than inventing a regular grid.

The sensitivity estimate comes from a completed run on September 18, 2026 using the ols-blockboot-v2 method. It is retrospective, built with information available after the event, and is neither a forecast nor a reconstruction of what a trader knew beforehand.