The Hidden Look-Ahead: Point-in-Time Leakage in Electricity Market Data

Abdelkrim Alfalah

cs.LGOct 18, 2026

The Hidden Look-Ahead: Point-in-Time Leakage in Electricity Market Data

Authors: Abdelkrim Alfalah

Organizations: Goldsmiths, University of London

Abstract

Empirical studies of electricity prices routinely state that their models use only information available at the time of the trading decision. For weather-driven forecasts drawn from commercial archives, this condition is harder to meet than it appears, and whether it holds can change over time. We identify three timing pitfalls. A forecast run is labelled by the issue time of its weather model, but becomes available hours later and keeps changing after publication. Auction gates follow local clock time, while archives use UTC. Outturn series are revised after publication without earlier values being kept. Using the timestamps of 128,656 forecast runs from ten weather-model configurations, matched to the day-ahead and first intraday auction gates in Great Britain over 2022-2026, we find that selecting "the latest run issued before the gate" picks a run not yet published at the gate on more than 99% of delivery days for six of the ten configurations at one of the two gates. For GFS at the day-ahead gate, the share is 0.6% in winter and 99.8% in summer. The exposure also shifts as publication delays drift, and an action margin of a few minutes can cost a full six-hour forecast cycle. The wrongly selected inputs differ materially from those available at the gate, but their apparent accuracy gain ranges from -9% to +24% across variables, models, gates and seasons, so its effect cannot be assumed and must be measured. We argue that point-in-time correctness is a dynamic property of the data pipeline, and propose a protocol that audits it from timestamps before any model is fitted.

Figures & tables

Figure 1: Gate closures and the life cycle of forecast runs on the day before delivery, in UTC, for a winter day (top) and a summer day (bottom). Each bar is one forecast run. It starts at the run’s issue time (black tick), remains unpublished until its publication time (orange, hatched), and is then available but still changing until its finalisation time (blue); after finalisation no bar is drawn. Bar lengths are the 2022–2026 median delays for each configuration. The dashed lines mark the N2EX day-ahead gate (09:50 UK clock time) and the EPEX IDA1 gate (17:30 UK clock time), which fall one hour earlier in UTC in summer. A gate line crossing an orange segment marks a run that the naive rule would select although it was not yet published; a gate line crossing a blue segment marks a run that was published but not yet final.

Figure 2: Selection rules at one gate, illustrated for ICON at the summer day-ahead gate (09:50 BST, 08:50 UTC), using the configuration’s median publication and finalisation delays. The 06 UTC run is published four minutes before the gate and finalised fifty minutes after it. The naive rule and the publication rule without an action margin both select the 06 UTC run, whose archived (final) values were not those visible at the gate. With an action margin of 15 minutes (shaded), the publication rule falls back to the 00 UTC run, six hours older. The finalisation rule selects the 00 UTC run, whose values were final before the gate. All rules are evaluated at the true gate; the effect of stating the gate in UTC (Pitfall 2) is shown in Figure 1 and Table 4.

Configuration	Underlying system	Cycles (UTC)	Horizon	Runs per variable
GFS	NCEP Global Forecast System	00, 06, 12, 18	up to 16 days	6,957
GFS-ENS	NCEP global ensemble	00, 06, 12, 18	n/a	6,952
ICON	DWD ICON global, main runs	00, 06, 12, 18	120 h	6,955
ICON-SR	DWD ICON global, short-range intermediate runs	03, 09, 15, 21	30 h	6,945
ECMWF	ECMWF IFS high-resolution, main runs	00, 12	360 h	3,482
ECMWF-ENS	ECMWF IFS ensemble, main runs	00, 12	360 h	3,483

Table 1: Weather-model configurations in the archive.

Configuration	Publication: median	Publication: 95th pct.	Finalisation: median	Finalisation: 95th pct.	No finalisation time (%)
ICON-SR	2.75	3.00	3.20	3.30	3.3
ICON	2.76	3.09	3.67	3.83	0.4
GFS	3.55	3.64	5.22	5.37	0.1
GFS-ENS	3.86	3.94	6.52	6.75	0.3
AIFS (2026)	5.42	5.80	5.48	5.88	8.4
ECMWF-SR	5.77	5.78	6.21	6.22	6.9

Table 2: Delays after issue time (hours), wind generation, 2022–2026, runs with both times recorded.

Figure 3: Status at the true gate closure of the run selected by the naive rule, for every delivery day from January 2022 to September 2026 (columns) and each weather-model configuration (rows), at the day-ahead gate (top) and the IDA1 gate (bottom). Orange: the selected run had not yet been published. Blue: it had been published but its values were not yet final. Grey: it was final before the gate. Wind generation; solar and demand give the same picture. AIFS configurations, available only from 2026, are omitted. Summer stripes (GFS and ICON at the day-ahead gate, GFS at the IDA1 gate) reflect the daylight-saving shift of the gates. Changes of regime reflect drifting publication delays: ECMWF at the IDA1 gate in winter from late 2024, after its delays shortened, and ICON at the day-ahead gate in summer 2026, after its delay lengthened past the gate. The block in mid-2024 marks ECMWF runs without a recorded finalisation time (May to September 2024), which are counted as not final.

Configuration	DA, not published (BST / GMT)	DA, not final (BST / GMT)	IDA1, not published (BST / GMT)	IDA1, not final (BST / GMT)
GFS	99.8 / 0.6	99.8 / 100.0	0.9 / 0.1	99.7 / 5.0
GFS-ENS	99.8 / 93.2	100.0 / 100.0	2.3 / 0.0	99.8 / 99.6
ICON	23.5 / 1.0	99.8 / 4.0	0.5 / 0.6	1.3 / 1.4
ICON-SR	99.7 / 99.6	99.8 / 99.6	100.0 / 99.9	100.0 / 99.9
ECMWF	0.1 / 0.0	11.5 / 1.2	100.0 / 57.0	100.0 / 99.9
ECMWF-ENS	0.3 / 0.1	9.9 / 1.2	100.0 / 100.0	100.0 / 100.0

Table 3: Naive-rule leakage and exposure: share of delivery days (%) on which the selected run was not yet published (run-availability leakage), or not yet final (value-revision exposure), at the true gate closure. Wind generation, 2022–2026.

Gate	Configuration	Publication rule, UTC gate: not published	Finalisation rule, UTC gate: not final
DA	GFS	98.9	0.4
DA	GFS-ENS	11.4	0.5
DA	ICON	23.3	94.6
IDA1	ECMWF	35.6	0.0
IDA1	GFS	0.3	96.3

Table 4: Leakage caused by stating the gate in UTC alone: share of summer (BST) delivery days (%) on which the run selected with a fixed-UTC gate was not available at the true gate. Wind generation, 2022–2026,

\Delta

= 0. Configurations with no entry above 1% are omitted.

Figure 4: Median age at the gate of the run selected by the finalisation rule, as a function of the action margin

\Delta

, for the day-ahead gate (top) and the IDA1 gate (bottom), in winter (left) and summer (right). Coloured lines: ICON, GFS and ECMWF. Grey band: the freshest admissible run across all eight configurations with a full history. Wind generation, delivery days January 2022 to September 2026. Each step marks the margin at which the most recent run is lost and the selection falls back to the previous cycle.

Gate	Configuration	Rule	Season	Age at $\Delta$ = 0	Step at $\Delta$ (min)	Age after step
DA	ICON	publication	BST	2.8	5	8.8
DA	ICON	finalisation	GMT	3.8	11	9.8
DA	GFS	publication	GMT	3.8	19	9.8
DA	ICON	publication	GMT	3.8	64	9.8
DA	ECMWF-ENS	finalisation	BST	8.8	69	20.8
DA	ECMWF	finalisation	BST	8.8	114	20.8

Table 5: Every combination for which the selected run changes within

\Delta

\leq

120 minutes (median age of the selected run at the gate, hours; thresholds at one-minute resolution).

Variable	Configuration	Gate	Days with a different run (%)	Mean $D$ , MW (% of mean outturn)	$G$ , all days (%) [95% CI]	$G$ , summer (%)
Wind	ECMWF-SR	DA	100	611 (6.1)	11.4 [8.9, 14.0]	9.2
Wind	ECMWF	IDA1	100	550 (5.5)	10.1 [7.9, 12.2]	9.0
Wind	GFS	DA	100	486 (4.9)	5.4 [3.7, 7.0]	4.9
Wind	GFS	IDA1	55	240 (2.4)	4.2 [3.0, 5.6]	7.9
Wind	ICON	DA	56	305 (3.1)	0.0 [ $-$ 1.4, 1.2]	$-$ 0.2
Solar	ECMWF-SR	DA	100	126 (6.6)	2.7 [0.4, 5.1]	1.0

Table 6: Naive rule versus finalisation rule: difference between the selected inputs and apparent accuracy gain of the naive selection, delivery days 2024–2026.

Data type	When known	Revised after publication?	Auditable from the archive?	Safe use
Forecast runs	Publication time, 2.8–6.7 h after issue (Table 2)	Yes, until finalisation, up to 2.7 h later	Yes: issue, publication and finalisation times stored per run	Select by the finalisation rule with an explicit margin $\Delta$ (Section 7)
Auction prices	At results publication (DA 10:00, IDA1 18:00 UK time, D $-$ 1)	None observed (two retrievals)	Not tested	As a feature only after publication; never as a feature at its own gate
Outturns	Close to real time ( $\approx$ 0.3–1.3 h after the half-hour)	Yes; observed up to 32 days after delivery (median changes 0.2–4.4%)	No: only the latest values are retained	Lag conservatively (revisions seen up to 32 days), or record first-published values going forward; otherwise declare the revision exposure

Table 7: Point-in-time exposure by data type (GB, 2022–2026).

Temporal Data Leakage

Temporal data leakage occurs when models use information unavailable at prediction time, making performance estimates and forecasts misleading.

See all 35 papers

Aug 11, 2026

Market-Information-Aware Gated-LoRA of Foundation Models for Transferable Day-Ahead Electricity Price Forecasting

Electricity price forecasting is crucial for market participants but remains difficult because prices are volatile, market-specific, and closely tied to anticipated system conditions. Existing supervised methods depend largely on market-specific historical data, limiting their use in newly established or data-scarce markets. This paper proposes a market-information-aware adaptation framework that transfers the Chronos-2 time-series foundation model to day-ahead electricity price forecasting. It first constructs a multi-source market information (MSMI) interface aligning 7-day price context with pre-clearing supply--demand, reserve, maintenance, generator-capacity, and intertie variables, and then trains a source-domain gated low-rank adapter (LoRA), updating about

1\%

of model parameters without target-market labels. The gate scales the frozen source adapter according to reserve-tightness and operating-state signals. A leave-one-market-out protocol is adopted for evaluating cross-market transferability. Experiments on four Chinese provincial day-ahead spot markets show that the proposed framework reduces the average MAE/RMSE by

6.24\%/7.99\%

relative to market-information-aware zero-shot Chronos-2 and by

3.05\%/3.52\%

relative to vanilla Source-LoRA. Experiments show that the gain is not reproduced by a learned global scalar or by random gate initialization, while the additional improvement over Source-LoRA is limited. These results suggest that market-structured inputs and state-dependent gated LoRA can provide a practical transfer path for data-scarce electricity markets.

Hang Fan, Wei Wei, Shengwei Mei

Electricity Price Forecasting Time Series Forecasting Low-Rank Adaptation Transfer Learning Time-Series Foundation Models

Ask PDF

Sep 19, 2026

Stealing profits: Spread-based temporal hierarchy forecasting for day-ahead electricity markets

Day-ahead electricity price forecasts support trading and storage decisions, but for battery arbitrage predicting intraday price spreads is more relevant than predicting individual hourly prices. Here we show that a temporal hierarchy forecasting (THieF) framework that jointly reconciles forecasts of hourly electricity prices and all intraday price spreads consistently improves performance across two major European electricity markets and three different forecasting architectures. Using five years of out-of-sample data from Germany and Spain, we obtain accuracy improvements of up to 19.7% and profit gains of up to 10.4% relative to unreconciled hourly price forecasts. The gains persist even for a highly accurate pretrained TabPFN foundation model. Our results demonstrate that exploiting coherent relationships between economically relevant forecasting targets can improve both predictive accuracy and decision value, and that better statistical forecasts do not necessarily imply better economic decisions.

Arkadiusz Lipiecki, Nikolaos Kourentzes, Rafal Weron

Department of Computational Social Science, Wrocław University of Science and Technology, Poland · Indicio Technologies AB, Stockholm, Sweden · Department of Operations Research and Business Intelligence, Wrocław University of Science and Technology, Poland

Electricity Price Forecasting Time Series Forecasting

Ask PDF

Aug 31, 2026

Foundation models for electricity price forecasting and battery arbitrage: Can they replace market-specific forecasting models?

Foundation models promise accurate forecasts with little or no task-specific training, but whether they can replace models designed specifically for electricity price forecasting remains unclear. We compare nine variants from five foundation model families, evaluated in zero-shot mode, with two state-of-the-art electricity price forecasting benchmarks in Germany, Poland, and Spain over 2021-2025. Their performance is assessed in terms of point and probabilistic forecasting accuracy, as well as economic value in battery energy storage arbitrage. Only the TabPFN models consistently and significantly outperform the benchmarks across all three markets and all statistical measures. However, this statistical dominance does not translate directly into economic dominance: TabPFN performs best under unlimited bids and riskier quantile-based strategies, whereas the Distributional Deep Neural Network benchmark is more profitable when risk tolerance is lower. Thus, foundation models cannot universally replace market-specific models, and their value depends on both model architecture and the decision problem.

Arkadiusz Lipiecki, Rafał Weron

Department of Computational Social Science, Wrocław University of Science and Technology, Poland · Department of Operations Research and Business Intelligence, Wrocław University of Science and Technology, Poland · Center for Research in Energy (CoRE), Aarhus University, 8000 Aarhus C, Denmark

Tabular Foundation Models Electricity Price Forecasting Time Series Forecasting Probabilistic Forecasting Time-Series Foundation Models

Ask PDF

Abstract

Temporal Data Leakage

Explore similar work

Market-Information-Aware Gated-LoRA of Foundation Models for Transferable Day-Ahead Electricity Price Forecasting

Stealing profits: Spread-based temporal hierarchy forecasting for day-ahead electricity markets

Foundation models for electricity price forecasting and battery arbitrage: Can they replace market-specific forecasting models?