SLS Blog
Perspectives on storage bidding, probabilistic forecasting, and trading in US wholesale electricity markets.
June 29, 2026
Battery revenue in ERCOT concentrates in a handful of hours a year, and dispatch during delivery is entirely mechanical. That combination changes what the strategy object is: not a price forecast, not a dispatch schedule, but the marginal value of stored energy, expressed as an offer curve.
May 18, 2026
The RL literature on electricity market bidding is large and growing, yet production bidding still runs on mathematical optimization. Here is why, and where RL credibly fits.
March 9, 2026
I trace a single grid battery trade through ERCOT, from the 10 AM day-ahead deadline through five-minute dispatch and two-settlement. The surprise is how much of the outcome is decided before the operating day begins, and how little anyone decides during it.
January 12, 2026
Why rolling-horizon LP/MILP over coherent scenarios, with CVaR and hard market-rule constraints inside the optimizer, remains the most defensible way to bid flexible assets, and why pure expected value fails in fat-tailed power markets.
November 3, 2025
Power-market backtests fail in ways equity backtests rarely do: revised forecasts, shifting grid topology, uplift charges, and fills that never would have happened. Point-in-time discipline and honest baselines are the actual deliverable of a quant research effort.
August 25, 2025
Point forecasts are insufficient for most high-value bidding decisions. Calibrated distributions and coherent scenarios, with tails and dependence intact, are what optimization actually consumes.
June 16, 2025
In US wholesale power markets, prices come out of a security-constrained optimization, not a learned model. That changes what 'AI for energy trading' should mean: forecast the optimizer's outputs instead of trying to rebuild the optimizer.
