Linear Mixed-Effect State-Space Forecasting: Completing the algorithm

Statistical downscaling techniques are used to translate global climate scenarios into local impact forecasts. In this study, I propose a new algorithm for downscaling by using a linear mixed-effect state-space model (LMESS). The rationale to use this model in a climate data context is that it allows for both time-varying and fixed relations between dependent and explanatory variables. My findings show the importance of identifying the correct random and fixed effects. I develop a new method for selection based on the state-space formulation with fixed parameters by Chow (1984). I apply the proposed methods to climate data at five different weather stations in the Netherlands. My findings show that the LMESS model is not able to consistently outperform a multivariate linear regression forecast method.