Title: Econometrics for climate modelling
Authors: David Hendry - University of Oxford (United Kingdom) [presenting]
Jennifer Castle - Oxford University (United Kingdom)
Abstract: Climate time series are non-stationary from both stochastic trends, tackled by cointegration, and distributional shifts-caused by everything from volcanic eruptions to policy interventions-tackled by indicator saturation. The tools include [a] model selection retaining theory information while selecting over other candidate variables; [b] software that can handle more variables, $N$, than observations, $T$, by expanding and contracting multi-path block searches; and [c] saturation estimation, including in that candidate set indicators for a range of potential contaminations, including outliers (impulse-indicator saturation, IIS), location shifts (step-indicator saturation, SIS), both (IIS+SIS, super saturation), changes in model parameters (multiple-indicator saturation, MIS), and designed indicators (e.g.) for impacts of volcanic eruptions on temperatures (DIS). Despite $N>T$ (often several fold), the costs of selection are small relative to mis-specification problems that might otherwise occur. Areas of application include modelling UK CO2 emissions from 1860 onwards.