How to Predict Tomorrow’s Electricity Prices?

Témavezető:	Sebestyén Géza
email:	sebestyen.geza@mcc.hu

Projekt leírás

Description of the Research Topic

Electricity prices exhibit daily and weekly seasonality, sudden price spikes, and complex relationships with weather, electricity demand, generation capacity, and cross-border flows.

The project aims to compare different forecasting methods based on fundamentally different modelling approaches for predicting electricity prices over the next several hours and days.

The methodological spectrum ranges from a very simple seasonal naïve benchmark through a statistical SARIMAX model and the tree-based XGBoost algorithm to the more sophisticated N-BEATSx and Temporal Fusion Transformer deep-learning architectures.

Models will be evaluated using the same information set and a rolling-window backtesting procedure, allowing the student to assess whether greater model complexity produces a genuine improvement in point and quantile forecasts.

The analysis will examine different forecast horizons, extreme prices, changing market regimes, computational requirements, and model interpretability. The final output may include reproducible Python code, a systematic comparison of the models, and an interactive dashboard presenting forecasts, uncertainty intervals, forecast errors, and the most influential explanatory variables.

By choosing this topic, the student will acquire highly transferable and marketable skills in time-series forecasting, machine learning, deep learning, probabilistic modelling, data engineering, model validation, interElectricity prices exhibit daily and weekly seasonality, sudden price spikes, and complex relationships with weather, electricity demand, generation capacity, and cross-border flows. The project aims to compare different forecasting methods based on fundamentally different modelling approaches for predicting electricity prices over the next several hours and days. The methodological spectrum ranges from a very simple seasonal naïve benchmark through a statistical SARIMAX model and the tree-based XGBoost algorithm to the more sophisticated N-BEATSx and Temporal Fusion Transformer deep-learning architectures. Models will be evaluated using the same information set and a rolling-window backtesting procedure, allowing the student to assess whether greater model complexity produces a genuine improvement in point and quantile forecasts. The analysis will examine different forecast horizons, extreme prices, changing market regimes, computational requirements, and model interpretability. The final output may include reproducible Python code, a systematic comparison of the models, and an interactive dashboard presenting forecasts, uncertainty intervals, forecast errors, and the most influential explanatory variables. By choosing this topic, the student will acquire highly transferable and marketable skills in time-series forecasting, machine learning, deep learning, probabilistic modelling, data engineering, model validation, interpretability, and decision-support system development, opening up career opportunities not only in energy trading and utilities but also in banking, insurance, asset management, fintech, consulting, retail, logistics, manufacturing, telecommunications, virtually any organization that relies on forecasts of prices, demand, sales, risks, capacity needs, or market developments.pretability, and decision-support system development, opening up career opportunities not only in energy trading and utilities but also in banking, insurance, asset management, fintech, consulting, retail, logistics, manufacturing, telecommunications, virtually any organization that relies on forecasts of prices, demand, sales, risks, capacity needs, or market developments.

Skills and Knowledge to Be Acquired:

The project will provide the student with an opportunity to progressively learn and apply:

• the mathematical foundations of time-series analysis and multi-horizon forecasting;
• the different principles underlying autoregressive, machine-learning, and deep-learning models;
• Python-based data analysis and modelling, particularly using pandas, statsmodels, scikit-learn, XGBoost, PyTorch, and PyTorch Forecasting;
• the collection, cleaning, and temporal alignment of ENTSO-E, HUPX, and meteorological data;
• the construction of lagged variables, rolling statistics, calendar features, and known future inputs;
• rolling-window backtesting and methods for preventing temporal data leakage;
• the evaluation of point, quantile, and interval forecasts;
• statistical tests for determining whether differences in forecast performance are significant;
• model-interpretability techniques, including SHAP values, TFT variable-selection weights, and the careful interpretation of attention weights;
• the development of a simple forecasting dashboard, for example using Streamlit;
• the interpretation of English-language methodological literature;
• the presentation of technical findings to non-technical business decision-makers.

The project can be completed at several levels of difficulty. Its core version would consist of a rigorous comparison of point forecasts. An extended version could also include probabilistic forecasting, model interpretability, forecast combinations, regime-specific evaluation, and an analysis of the forecasts’ economic value.

Hivatkozások

• Lago, J. et al. (2021): “Forecasting Day-Ahead Electricity Prices: A Review of State-of-the-Art Algorithms, Best Practices and an Open-Access Benchmark.” Applied Energy, 293, 116983. link
• Wang, D. et al. (2022): “Electricity Price Instability over Time: Time Series Analysis and Forecasting.” Sustainability, 14(15), 9081. link
• Yılan, Y. and Beykent, A. (2026): “Day-Ahead Electricity Price Forecasting Using the XGBoost Algorithm: An Application to the Turkish Electricity Market.” Computers, Materials & Continua,
• Olivares, K. G. et al. (2021): “Neural Basis Expansion Analysis with Exogenous Variables: Forecasting Electricity Prices with NBEATSx.” International Journal of Forecasting. link
• Jiang, H. et al. (2024): “Probabilistic Electricity Price Forecasting Based on Penalized Temporal Fusion Transformer.” Journal of Forecasting, 43(5), 1465–1491. link