Forecasting monthly industrial production in real-time: from single equations to factor-based models

Abstract

Short-term forecasting, Industrial production index, Realtime data-set, Aggregate and disaggregate forecasts, Conditional predictive ability, Forecast combinations, C22, C53, E01, E37,

Forecasting Capacity of ARIMA Models; A Study on Croatian Industrial Production and its Sub-sectors

“…However, some empirical literature (for example see Bulligan, Golinelli and Parigi (2010)) suggests that usage of alternative forecasting methods could assure statistically signifi cant accuracy gains, without relevant differences between static and dynamic approaches. This brings us to two major shortcomings of the paper that we as authors would like to emphasize.…”

“…Their fi ndings suggests that both seasonal and trend effects were present in India's industrial production so that the future forecasts could be made by this approach after adjusting the effects of short-term and long-term variations. Next, Bulligan, Golinelli and Parigi (2010) analyzed the performance of alternative forecasting methods to predict the index of industrial production of Italy as they used 12 different models, from simple (univariate) ARIMA to dynamic factors models exploiting the timely information of up to 110 short-term indicators, both qualitative and quantitative. They conclude, that though most of the factor based models outperform ARIMA model (by the sheer fact that short-run indicator signal always dominates the noise component), still this model can be used as a benchmark since it provides rather robust results in term of forecasting performance which do not deviate too much from the results of other methods.…”

“…Forecasts of industrial production can also be used as an additional input in larger models, which are often criticized for their (in)ability in tracking business cycle turning points. As Bulligan, Golinelli and Parigi (2010) rightly conclude, the index of industrial is probably the most important and widely analyzed high-frequency indicator, given the relevance of the manufacturing activity as a driver of a whole business cycle which is experienced by the extensive comments and reactions of the business analysts as soon as the index is published. Therefore, the index of industrial production is a crucial variable in the forecasting process of the short-term evolution of national output in most countries.…”

The Role of Indicator Selection in Nowcasting Euro Area GDP in Pseudo Real Time

“…Additional references and descriptions can be found in Bulligan et al (2010). and coincident indexes are used as indicators (Heij et al, 2011, reassessed this finding). However, if the aim of the pseudo real time exercise is to compare the relative forecasting ability of alternative approaches (rather than to measure absolute forecasting ability), then their ranking should not be greatly affected by neglecting data revisions, as shown for both BM and FM contexts in Bernanke and Boivin (2003), Golinelli and Parigi (2008), Schumacher and Breitung (2008), Bulligan et al (2010).…”

“…BM may appear excessively ad hoc because of the "incredible" exclusion restrictions underlying the list of the pre-selected indicators; FM may be biased by unbalanced sources of information (see Boivin 2 These two routes were first compared in their ability to forecast the US economy in the short run by Klein and Ozmucur (2008). Then, other studies extended the FM-BM comparison to other countries/areas: Bulligan et al, (2010) for Italy, Antipa et al (2012) for Germany, and Brunhes- for France and Ng, 2006). In fact, the main requirement emerging from the asymptotic properties of the FM approach (such as factor estimators, structure and convergence to optimal forecasts) is that the sources of common dynamics remain limited as the number of cross-sections increases to infinity.…”