Dynamics of European Energy Prices: Forecasting and Transmission Mechanisms Between TTF Natural Gas and the Italian PUN

This thesis analyses the out-of-sample forecasting performance of time-series models across different market regimes and the long-run price transmission mechanism between the Dutch Title Transfer Facility (TTF) natural gas futures and the Italian Prezzo Unico Nazionale (PUN) electricity price. Predictive accuracy is evaluated using ARIMA, ARIMA-EGARCH and ARIMA-ANN models through a rolling-origin procedure. Market regimes are identified using Bai-Perron structural-break tests, while price transmission is examined through Johansen cointegration tests, Vector Error Correction Models (VECM) and Toda-Yamamoto Granger-causality tests. The TTF front-month contract behaves as a random walk, whereas the PUN displays predictability based on a weekly cycle and mean-reverting behaviour. Constant-variance models fail to measure forecast uncertainty during market shocks, while the ARIMA-EGARCH model maintains well-calibrated prediction intervals in most regimes. Predictive causality runs from the gas market to the electricity market, with the PUN adjusting to TTF shocks within two to three trading days. This relationship is regime-dependent. During the energy-crisis period following 2021, the estimated long-run elasticity of the PUN with respect to the TTF was approximately twice the point estimate for the preceding decade, while the share of the PUN forecast-error variance attributed to TTF shocks increased from about 2% to between 55% and 64%, depending on the Cholesky ordering.