Fundamental Market Modelling
We analyse assets, portfolios and markets through fundamental optimization models, using two complementary engines: KL.Grid — our proprietary market-simulation platform — and SDDP (developed by PSR). Every engagement is delivered as reproducible KL.Grid runs: deterministic scenario paths, stochastic price tubes (P5–P95) and backtests against actual market data — so you see how the model performs on history before relying on its forecasts.
Current validation: Panama, 2022–2026 – 39,936 consecutive hours across three ENSO regimes. Hourly mean absolute error $13/MWh (WAPE 13.5%), correlation r = 0.83, near-zero bias +$1.4/MWh; monthly correlation 0.97 — the model tracks the 2024 El Niño peak (~$224/MWh) and the sub-$50 October crash, including the structural break of Gatún’s commissioning, which cut average market prices roughly 37% year-on-year.
Scope covers demand assessment and projection; generation, reservoirs and fuel; endogenous value of water; hourly dispatch and system-marginal-price formation; Monte-Carlo price and generation distributions; and short-, mid- and long-term price and dispatch views for greenfield and brownfield evaluation, PPA pricing, portfolio and risk work.
Typical outputs from real runs
Three ready examples from actual Panama runs, each produced by the engine and viewed in the Grapher.
Lookback – reproducing history
A playback of January–July 2026 (4,872 hours) laid over the operator’s published CMS: the model tracks the calm ~$60–90/MWh regime and the dry-season scarcity peaks, near-unbiased.
Lookback: KL.Grid daily-mean price over the published CMS, Jan–Jul 2026, with per-month accuracy.
Forecast – the most-likely forward path
A forward unit-commitment run gives the hourly generation mix and the system marginal price, with reservoirs scheduled at their water value — here a representative week showing renewables and run-of-river filling the base while reservoir hydro and thermal set the margin.
Forecast: hourly generation by technology with the CMS, one representative week.
Stochastic — the price tube
1,000 Monte-Carlo runs of full-year 2026 with time-varying drivers: the median price with P25–P75 and P5–P95 bands that fan out with lead time — the honest forward view for pricing PPAs, sizing merchant upside and stress-testing debt.
Stochastic: the CMS price tube — median, P25–P75 and P5–P95 — fanning out with lead time.
The Grapher — interactive gallery
All are live, interactive panels; the analyst aggregates (hourly / daily / monthly), drags to zoom, toggles the legend and exports any chart.
Grapher — price (CMS) and demand, with the aggregation controls (Hourly / Daily / Weekly / Monthly) and drag-to-zoom.
Roll-up / roll-out: the same price shown hourly, daily-mean and weekly-mean at one click.
Per-hour economic stack: two hours compared, units stacked in merit order to demand, with an explicit curtailment (Vertido) wedge.
Price-duration curve (X-Y explorer): CMS sorted high-to-low — read the P90 / median / P10 straight off the curve.
Stochastic histogram: distribution of the run-average CMS across the Monte-Carlo scenarios.
Stochastic histogram: distribution of a focus unit's generation across the same scenarios.
Network tab (forecast): per-node prices (LMP) vs the system price, and inter-zone flows.
