Forecasting regional labour markets with GVAR models and indicators

The development of employment and unemployment in regional labour markets is known to spatially interdependent. Global Vector-Autoregressive (GVAR) models generate a link between the local and the surrounding labour markets and thus might be useful when analysing and forecasting employment and unemployment even if they are non-stationary or co-trending. Furthermore, GVARs have the advantage to allow for both strong cross-sectional dependence on ``leader regions' and weak cross-sectional, spatial dependence. For the recent and further development of labour markets the economic situation (described e.g. by business-cycle indicators), politics and environmental impacts (e.g. climate) may be relevant. Information on these impacts can be integrated in addition to the joint development of employment and unemployment and the spatial link in a way that allows on the one hand to carry out economic plausibility checks easily and on the other hand to directly receive measures regarding the statistical properties and the precision of the forecasts. Then, the forecasting accuracy is demonstrated for German regional labour-market data in simulated forecasts at different horizons and for several periods. Business-cycle indicators seem to have no information regarding labour-market prediction, climate indicators little. In contrast, including information about labour-market policies and vacancies, and accounting for the lagged and contemporaneous spatial dependence can improve the forecasts relative to a simple bivariate model.