Forecasting Labour Market Indicators: Micro vs Macro

This work analyses the dynamics of a set of labour market indicators using micro-data instead of aggregate time series. The goal is to produce short-term (up to four quarters) forecasts of aggregate labour force participation rates, employment rates and unemployment rates and to compare the performance of the repeated cross-sections predictor with that of the traditional time series estimators proposed in the literature. Exploiting the ISTAT continuous labour force survey, an extensive analysis of the trends is carried out in order to understand if different socio-demographic groups follow different trends. As trend heterogeneity between groups emerges, a cross-section estimator which decomposes behavioural and compositional effects may add useful information for out-of-the-sample predictions. ARIMA, SARIMA and state space models are implemented and compared with the decomposed cross-section estimator which includes a logistic specification. Through exponential smoothing, traditional time series can model the trend in a very efficient way. Besides that, they are not able to catch compositional effects that are instead fundamental in the patterns of considered indicators. Average forecast errors of the cross section predictor, calculated on a rolling basis, show that the latter is competitive with the traditional models. It outperforms ARIMA and SARIMA. A crucial factor is that time series models improve their efficiency as the sample length increases. The proposed cross section estimator instead is built in order to work on a short sample length, like in the continuous labour force survey case.