Three forecasts of the same thing, on one axis — what settled, what the day-ahead market itself said, and what our model says — each scored against what actually cleared.
The forecast peaks at 37 $/MWh on Mon 7 PM, against a low of 13 $/MWh — a swing of 24 $/MWh across the window. HB_NORTH is the heaviest zone at 41 $/MWh.
Plotted window Aug 2, 11:00 PM CDT → Aug 3, 10:55 PM CDT. Metiscast reissues every 15 minutes; ERCOT clears day-ahead once daily.
Every line above, and who produced it.
What settled — the realized values every forecast here is scored against.
The grid operator's own published forecast for these hours. Source: ERCOT (settled and day-ahead) and Metis (Metiscast).
The grid operator's own published forecast for these hours. Source: ERCOT (settled and day-ahead) and Metis (Metiscast).
The grid operator's own published forecast for these hours. Source: ERCOT (settled and day-ahead) and Metis (Metiscast).
Each day of the window on its own.
Congestion is why these price apart. Each hub’s settled price with every forecast published for it.
Each day of the window on its own.
Every hour of the window, shaded by load.
| 0 | 6 | 12 | 18 | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sun, 8/2 | ||||||||||||||||||||||||
| Mon, 8/3 |
Darkest cells are the tightest hours. The week peaks at 37 $/MWh on Mon 7 PM, and the daily band across the late afternoon and early evening is the solar ramp-down meeting air-conditioning load.
| Hub | Peak ($/MWh) | Peak hour | Low ($/MWh) | Average ($/MWh) |
|---|---|---|---|---|
| HB_NORTH | 41 | Mon 6 PM | 12 | 21 |
| HB_SOUTH | 37 | Mon 7 PM | 13 | 22 |
| HB_WEST | 44 | Mon 6 PM | 11 | 22 |
| HB_HOUSTON | 91 | Mon 9 AM | 15 | 28 |
| Forecast | Produced by | Method | Horizon |
|---|---|---|---|
| HB_SOUTH settled | ERCOT (settled and day-ahead) and Metis (Metiscast) | ERCOT SPP/DAM settlement point prices; Metiscast nodal price model | 24 hours ahead, 5-minute resolution for the model; hourly for day-ahead |
| HB_WEST settled | ERCOT (settled and day-ahead) and Metis (Metiscast) | ERCOT SPP/DAM settlement point prices; Metiscast nodal price model | 24 hours ahead, 5-minute resolution for the model; hourly for day-ahead |
| HB_HOUSTON settled | ERCOT (settled and day-ahead) and Metis (Metiscast) | ERCOT SPP/DAM settlement point prices; Metiscast nodal price model | 24 hours ahead, 5-minute resolution for the model; hourly for day-ahead |
Metiscast publishes p10/p50/p90 quantiles; the band on the chart is that range. ERCOT's day-ahead price is a single cleared number with no uncertainty attached. ERCOT prices are heavy-tailed — scarcity intervals move faster than these forecasts resolve.
Most price pages show you one number. This shows the settled real-time price at ERCOT's North Hub, the day-ahead price the market itself cleared for the same hours, and Metis's own forecast — plotted together so the disagreements are visible. Where day-ahead and real-time diverge is where the money is; where our model disagrees with both is where we think the market is wrong.
ERCOT's day-ahead market is an auction, not a model. Participants bid, the market clears, and the resulting price is the market's collective forecast of the next day — financially binding, which is what makes it a serious benchmark rather than an opinion. It closes around noon the day before delivery.
Metiscast is a fine-tuned time-series foundation model, not a market simulation. It reads seven days of five-minute real-time prices alongside day-ahead prices, load forecasts, and weather, and issues a probabilistic 24-hour path for each settlement point every 15 minutes. It outputs quantiles rather than a single number, so the band on the chart is the model's own stated uncertainty.
The day-ahead auction reflects what participants were willing to commit to financially, including risk premia and hedging positions that are not pure price expectations. A statistical model has no position to protect. Neither is automatically right, which is why both are shown against what actually settled.
curl "https://askthegrid.com/api/v1/grid/series?iso=ERCOT&metric=load_forecast&zone=SYSTEM&hours=168"
Connect your agent to mcp.askthegrid.com and this forecast is a tool call away — the same host Claude and ChatGPT connectors already use.
The day-ahead market is a financially binding auction that clears around noon for the following day. Its cleared prices are the market's own collective forecast of what power will cost — the benchmark any model has to beat.
It is a serious benchmark but a systematically imperfect one: day-ahead prices carry risk premia and hedging positions, so they diverge from real-time in predictable ways. The scorecard above measures that divergence over the last 24 hours at this hub.
Day-ahead clears on expectations; real-time clears on what actually happened. Unforecast outages, weather misses, and demand surprises all move real time away from day-ahead — and the spread between them is what virtual traders and batteries are positioned against.
Metiscast, a fine-tuned time-series foundation model that reads a week of five-minute prices along with day-ahead prices, load forecasts, and weather, and issues a probabilistic 24-hour path per settlement point every 15 minutes. It outputs quantiles, so its uncertainty is stated rather than implied.
The settled and day-ahead layers are public. The Metiscast layer requires an account with forecast access — the chart shows where it sits and what it covers either way.