Prognose und strukturelle Analyse mit kontemporär aggregierten Zeitreihendaten
Zusammenfassung der Projektergebnisse
Time series data in economics and social sciences are often constructed by aggregating over cross-sectional units such as individuals, sectors, states or countries. This type of aggregation is referred to as contemporaneous aggregation. Examples that highlight the practical importance include the sectoral aggregation in the construction of quarterly national accounts data, aggregation of different price index subcomponents, and the construction of area-wide time series e.g. for the Euro area. Contemporaneous aggregation may change the dynamic properties (e.g. the persistence) of the time series data in a substantial way, such that often more complex time series models are needed for the aggregate. In this project, we have explored methods for econometric forecasts and structural analysis of contemporaneous aggregates. To this end, we have analyzed the role of stochastic aggregation weights and have found that exploiting timevariation in the weights improves the forecasting accuracy of the aggregate Euro area time series. Moreover, we have developed a factor-model based alternative to standard aggregation methods that is useful to construct historical Euro area aggregate time series in situations where the time series information is missing for some cross-sectional units. Our empirical analysis has shown that some key Euro area variables, e.g. real GDP and inflation, can be forecasted more precisely with the factor-backdated data. We have also studied different methods of exploiting disaggregate information in forecasting aggregate Euro area macroeconomic time series. We have found that the inclusion of selected disaggregate components into the forecasting model for the aggregate improves the forecasting accuracy in a number of empirical applications. In addition, we have found that the selection of disaggregate components by a machine-learning technique is a useful alternative to including disaggregate information in the form of factor-time series extracted from the disaggregate components. The research on choosing weights for combining disaggregate forecasts to form the aggregate Euro area forecast has shown that using adaptive weights instead of standard time-invariant aggregation weights improves the forecasts for Euro area aggregates, in particular in cases where the time series is subject to structural change. Moreover, we have studied inference on structural impulse responses from vector autoregressive models under departure of standard assumptions on the innovations of the model. We have found that standard inference methods from the literature are not valid under more general assumptions. For these cases, we have suggested asymptotic and simulation based methods that lead to correct inference. The results from this project are useful for empirical economists and forecasting practitioners in research institutions and central banks.
Projektbezogene Publikationen (Auswahl)
- (2016). Combining Country-specific Forecasts when Forecasting Euro Area Macroeconomic Aggregates, Empirica. Journal of European Economics
Zeng. J.
(Siehe online unter https://doi.org/10.1007/s10663-016-9330-x) - (2013). Forecasting Contemporaneous Aggregates with Stochastic Aggregation Weights, International Journal of Forecasting, 29, pp. 60- 68
Brüggemann, R. & Lütkepohl, H.
- (2015). Forecasting Euro Area Macroeconomic Aggregate Variables, Dissertation. University of Konstanz
Zeng, J.
- (2015). Forecasting Euro-Area Macroeconomic Variables Using a Factor Model Approach for Backdating, Oxford Bulletin of Economics and Statistics, 77(1), pp. 22-39
Brüggemann, R. & Zeng, J.
(Siehe online unter https://doi.org/10.1111/obes.12053) - (2016) Inference in VARs with Conditional Heteroskedasticity of Unknown Form, Journal of Econometrics, 191, pp. 69-85
Brüggemann, R., Jentsch, C. & Trenkler, C.
(Siehe online unter https://doi.org/10.1016/j.jeconom.2015.10.004) - (2016). Forecasting Aggregates with Disaggregate Variables: Does Boosting Help to Select the Most Relevant Predictors? Journal of Forecasting
Zeng, J.
(Siehe online unter https://doi.org/10.1002/for.2415)
