Key words: Big Data analysis, climate econometrics, spatial modelling, time-varying parameters, unemployment estimation and nowcasting

"Multivariate State Space Methods for Official Statistics and Climate Modelling"

This thesis explores how state space models, which are a type of econometric models designed to analyse time series data, can be employed to achieve more accurate and realistic estimates of official statistics, and to model and forecast regional concentrations of air pollutants. Specifically, a novel approach is presented, which incorporates survey-based, claimant counts and Google Trends data in order to provide more timely and accurate estimates of Dutch unemployment, over only employing survey-based data. A new method is proposed to model the relationship between the latter and claimant counts data as time-varying, which allows us to promptly tackle changes in such relationship and therefore achieve more realistic real-time estimates of Dutch unemployment. Time-varying relationships can potentially be modelled with other, already existing, econometric techniques, than the one proposed in this thesis, and the reasons why they have not been considered further are here documented. Finally, a novel spatial type of state space model is employed in order to model regional concentrations of nitrogen dioxide (NO2) in the Netherlands. The (time-varying) effects on this air pollutant of meteorological conditions, traffic intensity and geographical location of the Dutch regions, are accounted for in the model. The latter is further used to forecast regional NO2 concentrations for different scenarios of traffic intensity, and can therefore be potentially employed for evaluation of pollution-reduction policies.

Click here for the full dissertation.

Click here for the live stream.