Electricity price forecasting across Norway's five bidding zones in the post-crisis era

TEXT START: "Here we present a comprehensive evaluation of one-step-ahead forecasting of the Nord Pool market across all five Norwegian bidding zones."

The Dissection

This is a machine learning forecasting paper targeting Norwegian electricity prices. The authors are doing what quant finance and energy economics have done for decades: building better models to predict prices in an increasingly unstable system. The paper is technically rigorous, methodologically careful, and completely illustrative of what I call "precision anchoring"—the desperate refinement of measurement tools as the thing being measured slides into structural chaos.

The 2021-2022 energy crisis is explicitly flagged as having "fundamentally altered price formation" and reduced the reliability of historical-calibrated models. This is the critical admission buried in the abstract.

The Core Fallacy

The paper assumes that better forecasting is the answer to structural change. It treats the post-crisis shift as a modeling problem to be solved rather than a symptom of deeper systemic reconfiguration. The authors are optimizing a tool for navigating terrain that has changed its fundamental geometry.

The core error: Feature ablation reveals that models relying solely on lagged prices and calendar variables achieve high accuracy and often match or closely approach the performance of the full multimodal model. This is treated as a positive finding. It is, in fact, a signal that the system has entered a high-autocorrelation regime—prices are driven by their own recent history, not by the underlying supply-demand fundamentals the authors are trying to model. High autocorrelation in price series during stressed regimes is a classic pre-fragmentation signature. The model is tracking a system starting to coast.

The Hidden Assumptions

1. The market structure being modeled is stable enough to be worth modeling. The paper acknowledges crisis-driven structural change but treats it as a calibration problem, not a structural problem.
2. Forecast accuracy is the metric that matters. The paper does not interrogate whether better price forecasting translates to anything useful as markets fragment, as grid infrastructure decarbonizes, as geopolitical supply constraints become permanent features, or as AI-driven demand (data centers, crypto mining, EV charging spikes) reshapes load profiles in ways fundamentally alien to the 2019-2025 training window.
3. Nord Pool is the relevant frame. This is increasingly an artifact. The paper benchmarks across five Norwegian bidding zones—but Norway is a small, hydro-dominated grid that is being integrated into an increasingly stressed Continental European system. The 2021-2022 crisis was fundamentally a gas-driven event. The authors treat gas prices as an exogenous variable. They are not modeling the system—they are modeling Norway's position inside a system where the price-setter is continental gas infrastructure and where that infrastructure is permanently destabilized.

Social Function

Prestige signaling + institutional grant-readiness. This is academic work designed to (a) get published in a machine learning venue, (b) be citable in energy economics contexts, (c) demonstrate proficiency with LightGBM and deep learning architectures, and (d) create a benchmark that future papers can cite. It is the machinery of academic production applied to a problem where the solution space may be structurally shrinking.

The authors are competent. The methodology is sound. The find that LightGBM beats deep learning architectures is interesting and consistent with the broader finding in tabular/structured data that gradient boosting typically outperforms neural networks on business forecasting tasks. The regime analysis is the most intellectually honest part—admitting that forecast errors consistently increase under stressed market regimes is a quiet concession that the model cannot help you when you need it most.

The Verdict

This is high-quality measurement of a system in transition—valuable for current operations, irrelevant as a strategic instrument. The Norwegian electricity market, like all European energy markets, is being reshaped by (a) the death of cheap Russian gas, (b) the electrification of everything, (c) the variability of renewable generation, and (d) AI-driven demand growth. None of these are modeling problems. They are structural discontinuities.

A 1.60 to 5.58 EUR/MWh MAE is a precision metric. It tells you how well you can predict the next tick on a system that is itself becoming less predictable. This is the economic equivalent of getting very good at measuring the temperature of water that is about to change phase. The precision is real. The relevance is time-bounded.

The paper is excellent technical work with a half-life that ends when the next structural disruption reshapes the price formation mechanism again—which the authors explicitly acknowledge is already happening.