Please use this identifier to cite or link to this item: http://hdl.handle.net/10419/53764
Authors: 
Maier, Philipp
Year of Publication: 
2011
Series/Report no.: 
Bank of Canada Working Paper 2011,11
Abstract: 
We evaluate different approaches for using monthly indicators to predict Chinese GDP for the current and the next quarter ('nowcasts' and 'forecasts' respectively). We use three types of mixed-frequency models, one based on an economic activity indicator (Liu et al., 2007), one based on averaging over indicator models (Stock and Watson, 2004), and a static factor model (Stock and Watson, 2002). Evaluating all models' out-of-sample projections, we find that all the approaches can yield considerable improvements over naive AR benchmarks. We also analyze pooling across forecasting methodologies. We find that the most accurate nowcast is given by a combination of a factor model and an indicator model. The most accurate forecast is given by a factor model. Overall, we conclude that these models, or combinations of these models, can yield improvements in terms of RMSE's of up to 60 per cent over simple AR benchmarks.
Subjects: 
Econometric and statistical methods
International topics
JEL: 
C50
C53
E37
E47
Document Type: 
Working Paper

Files in This Item:
File
Size
466.86 kB





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.