The Arbitrage Value of Snowy 2.0
A backtesting simulation of what storage arbitrage could be worth to a Snowy 2.0-sized pumped-hydro plant, on four years of real NSW electricity spot prices.
Abstract. This project estimates what energy arbitrage (buy cheap electricity, store it, sell it back when it is expensive) could be worth to a Snowy 2.0-sized pumped-hydro plant, using four years of real half-hourly New South Wales spot prices (2022 to 2025). Three dispatch schemes frame the answer: a perfect-foresight linear program that sets the theoretical ceiling of $0.71 to $1.43 billion a year, a deployable rule of thumb whose two trigger prices are frozen on the training years, and a forecaster that learns the market’s statistical pattern by stochastic dynamic programming but never sees the future. The forecaster captures 84 to 93 per cent of the ceiling in the three years after the 2022 fuel crisis, including 93 per cent on the held-out test year it was never trained on. A financial appendix converts the gross margins into NPV and IRR under the plant’s disclosed cost envelope.
Keywords: energy arbitrage · pumped hydro · linear programming · stochastic dynamic programming · Australian NEM
Data: AEMO dispatch prices for the NSW1 region, 2022 to 2025 (public NEMWeb archive), averaged to half-hourly steps. Committed snapshots make every number reproducible offline.
Tools: Python (cvxpy with the HiGHS solver, pandas, NumPy), Quarto. The report is written for non-technical readers; all code is folded away by default.
Read the report · Interactive demo · Open in Colab · Source on GitHub
How it was made: built with Claude (via Claude Code) in July 2026, working from my own model design. Every number is recomputed on each render and guarded by in-document assertions, and the optimisation results were cross-checked by independent re-solves during review. Independent clean-room work: open-literature methods applied to public data and public plant specifications.