For business
Know where Colombian electricity demand and spot price are heading, with the margin of error of every forecast and a public record of how each one did.
- Mean error of the demand forecast
- 1.9 %
- Mean error of the price, 1 to 3 days
- 10.7 %
- Past dates used to test it
- 50
Measured on days the model had not seen. Lower is better.
Who it is for
- Electricity retailers (comercializadoras) that decide each week how much to buy in the spot market and how much to cover with contracts.
- Large consumers on the unregulated market who plan consumption and contract renewals.
- Anyone who wants a second, transparent opinion on where the spot price and the system's demand are heading, with the evidence in the open.
What you get
Demand, up to 30 days ahead
Daily national demand with a P10 to P90 range, tested against what really happened.
Spot price, up to 14 days ahead
The daily mean price with a range calibrated so that, on average, about 80 % of outcomes fall inside it, and the chance that it rises above a threshold.
Hydrology
Reservoir level and inflows by hydrological region and by reservoir, which are what moves the price in the medium term.
A record you can check
Every forecast is stored as first issued, before XM's public data show the outcome. The pipeline never rewrites a stored forecast, and they are scored in public afterwards.
Measured accuracy
Mean percentage error (MAPE) over 50 past dates the model had not seen, against two references. Lower is better. Days ahead are counted from the last day with a complete price in XM's public data, which is about three days before a forecast is published, so the first days are the nearest ones.
| Forecast | Volt-AI model | Seasonal naive | ETS |
|---|---|---|---|
| Demand, 1 to 30 days | 1.9 % | 3.4 % | 3.4 % |
| Spot price, 1–3 days | 10.7 % | 29.8 % | 15.3 % |
| Spot price, 4–7 days | 21.2 % | 33.1 % | 29.2 % |
| Spot price, 8–14 days | 40.0 % | 50.0 % | 45.3 % |
In the first days the price model beats both references; beyond a week it stays level with ETS, a strong statistical method. The method and the detailed comparison are on the Models page.
Check it yourself
Each day Volt-AI stores its forecast before the outcome is published. Once XM publishes the day, the Models page compares them with what happened, hits and misses included.
See the live recordScope
- Sharpest in the first week for the price and across the whole month for demand; further out, read the range more than the central value.
- Built on XM's public data, which carry the complete daily price about three days late, so every forecast projects forward from that point.
- Regulatory decisions, rationing and events with no precedent in the history are outside what a statistical model can see.
- It supports decisions; it is not advice to buy, sell or hedge energy.
See a real report
A short summary like the one that could be sent each week: forecast with ranges, price spike probabilities, hydrology and the record of hits. It uses real data: forecast issued on 26 October 2026, with prices complete up to 23 October.
Download the sample report (PDF, in Spanish)Let's talk
Tell me what you decide each week and I will show you how this forecast would fit your case. A trial month, a weekly report or a forecast adapted to your portfolio are all possible.
Write to memiguelangel.14130@gmail.com · I reply in Spanish or English.