Forecasting the E³CI and food inflation under evidence of non-linear interactions between extreme weather events and prices

This thesis investigates the non-linear relationship between extreme climate events, measured by the European Extreme Events Climate Index (E³CI), and unprocessed food prices, captured by the Harmonised Index of Consumer Prices (HICP). The motivation for this research consists in two main considerations: climate change is intensifying extreme events, and fresh food prices are highly exposed due to their reliance on agricultural supply. The analysis proceeds in three main steps. First, a Smooth Transition Autoregressive model is applied to Slovenia, replicating the Bank of Slovenia’s study, and then extended to Italy to test robustness across countries. Second, the forecastability of the E³CI is investigated with a variety of machine learning and deep learning models in order to evaluate the predictive potential of climate indicators. Third, given the challenges in accurately forecasting the index itself, the observed E³CI is incorporated into inflation models together with farm-gate prices and macroeconomic drivers, using both econometric and machine-learning approaches. Overall, the forecasting analysis revealed significant limitations. The E³CI tracks extreme and rare events, whose intrinsic unpredictability makes accurate prediction extremely challenging, underscoring the persistent difficulty of forecasting extremes. At the same time, food price inflation reflects multiple supply and demand shocks, rigidities, and external drivers, adding further complexity. Within this framework, forecasting serves to test the central non-linear hypothesis: to what extent climate-induced price dynamics — state-dependent, varying across regimes, and potentially disproportionate — can be captured by econometric and machine-learning models. The findings from both forecasting exercises indicate that even advanced models capture these dynamics only partially, and sometimes fail to capture them at all.