Short-term forecasting and composite indicators construction with help of dynamic factor models handling mixed frequencies data with ragged edges
Short-term forecasting and composite indicators construction with help of dynamic factor models handling mixed frequencies data with ragged edges

Summary/Abstract: This paper aims at presenting practical applications of latent variable extraction method based on second generation dynamic factor models, which use modified Kalman Filter combined with Maximum Likehood Method and can be applied for time series with mixed frequencies (mainly monthly and quarterly) and unbalanced beginning and the end of the data sample (ragged edges). These applications embrace short-term forecasting of Polish GDP and construction of composite coincident indicator of economic activity in Poland. Presented approach adopts the idea of short-term forecasting used by Camacho and Perez-Quirioz in Banco de Espana and concept of Arouba, Diebold and Scotti index compiled in the FRB of Philadelphia. According to the author’s knowledge, it is the first such adaptation for Central and Eastern Europe country. Quality of the forecast obtained with these models is compared with standard methods used for short-term forecasting with series of statistical tests in the pseudo real-time forecasting exercise. Moreover described method is applied for construction of composite coincident indicator of economic activity in Polish economy. This newly-created coincident indicator is compared with first generation coincident indicator, based on standard dynamic factor model (Stock and Watson) approach, which has been computed by the author for Polish economy since 2006.

• Details
• Contents