Early Estimates of the Industrial Turnover Index using Statistical Learning Algorithms

Doc: 03/2023

We use statistical learning algorithms to improve timeliness of the Spanish Industrial Turnover Index. The main idea is to use a gradient boosting algorithm to make a prediction for every single industrial turnover value not yet collected during the data collection, data editing and estimation phases. Regressors are constructed from the historical unit-level time series, current aggregated turnover moments and quantiles, and aggregated values of related industrial surveys. Accuracy indicators are also computed so that a quantitative trade-off between accuracy and timeliness can be appraised. This mass imputation exercise provides us with a nowcasting proposal which can be readily extended to many similar design-based surveys.

Palabras clave / Key words: Machine Learning, Statistical Learning, Industrial Turnover Index, timeliness improvement, missing data imputation

Early Estimates of the Industrial Turnover Index using Statistical Learning Algorithms (Pdf 5594 KB)
S. Barragán, L. Barreñada, J.F. Calatrava, J.C. Gálvez Sáenz de Cueto,J.M. Martín del Moral, E. Rosa-Pérez† and D. Salgado