Joint work with Jennifer L. Castle and J. Isaac Miller
Extreme weather events occurring around the world are a daily reminder that our climate is rapidly changing due to human activity. The long-term pattern of weather, or climate, is determined by human behaviour interacting with the physical properties of Earth’s climate system. To understand how the climate is changing and prepare for the future,
both in terms of mitigation and adaptation, we need accurate forecasts of the climate. However, human behaviour is non-stationary from both stochastic trends and location shifts. Consequently, the time-series data generated by the interaction of humanity and the climate are non-stationary from distributional and trend shifts, resulting in forecasts that are uncertain and prone to failure howsoever they are generated. The success of forecasts made by either climate scientists or econometricians hinges on the ability to handle unanticipated shifts as climate change is characterised by `the change in the change’. While unanticipated changes cannot be avoided, they later become in-sample, so empirical modelling must take account of those to avoid distortions in parameter estimates and the resulting forecasts. It is important to identify and model location shifts as doing so improves the verisimilitude of the model and its forecasts. Using indicator saturation estimators to capture in-sample shifts, improved econometric models and their forecasts can be achieved, as demonstrated within a system of four of the key climate variables, atmospheric CO2, global mean surface temperature deviations, ocean heat content deviations and global sea-level rise.