Short-Term Forecasting of the Japanese Economy Using Factor Models

Abstract

While the usefulness of factor models has been acknowledged over recent years, little attention has been devoted to the forecasting power of these models for the Japanese economy. In this paper, we aim at assessing the relative performance of factor models over different samples, including the recent financial crisis. To do so, we construct factor models to forecast Japanese GDP and its subcomponents, using 38 data series (including daily, monthly and quarterly variables) over the period 1991 to 2010. Overall, we find that factor models perform well at tracking GDP movements and anticipating turning points. For most of the components, we report that factor models yield lower forecasting errors than a simple AR process or an indicator model based on Purchasing Managers' Indicators (PMIs). In line with previous studies, we conclude that the largest improvements in terms of forecasting accuracy are found for more volatile periods, such as the recent financial crisis. However, unlike previous studies, we do not find evident links between the volatility of the components and the relative advantage of using factor models. Finally, we show that adding the PMI index as an independent explanatory variable improves the forecasting properties of the factor models.Keywords: Japan; Forecasting; Nowcasting; Factor models; Mixed frequency Bank topics: Econometric and statistical methods; International topics JEL codes: C50, C53, E37, E47 1 Non-technical summaryOver recent years, factor models have proven useful forecasting tools for dealing with large datasets. Forecasting Japanese GDP and its components is challenging, as it also involves dealing with very volatile data. Resorting to large databases could in principle help to single out common patterns (the 'factors') from multiple data series. To this extent, it is surprising to notice that while the usefulness of factor models in obtaining short-term forecasts for many advanced and developing economies has been documented extensively, little attention has been devoted to the Japanese economy. To the best of our knowledge, no mixedfrequency factor model to forecast the Japanese GDP and its components has appeared in the literature. We see this as a shortcoming, given that policymakers have a clear interest in providing forecasts not only for monthly series (e.g. inflation or industrial production), but also for (quarterly) GDP and its subcomponents. Factor models provide a simple and convenient way to accomplish this task.In this paper we explore the usefulness of factor models for forecasting real activity in Japan by resorting to different specifications. More specifically, we construct forecasts of past-, current-and next-quarter GDP, as well as its subcomponents, by using information available on the first, the second and the third month of the quarter. We also assess the performance of factor models over a simple AR specification, as well as a tougher benchmark, an indicator model based on Purchasing Managers index (PMIs).Overall, we find that factor models perf...

Nowcasting Czech GDP in real time

“…Dynamic factor models can be useful in nowcasting other policy-relevant quarterly variables, for example, the expenditure components of GDP. Indeed, several papers have employed dynamic factor models to successfully nowcast the components of GDP (Angelini et al, 2010;Godbout and Lombardi, 2012).…”

Factor Models and Time‐varying Parameter Framework for Forecasting Exchange Rates and Inflation: A Survey

“…However, in the literature we find econometric models that yield accurate short‐ and long‐term forecasts. On the one hand, Godbout and Lombardi (), Jakaitiene and Dées (), Barhoumi et al . () and Angelini et al .…”

“…Thus, basing the forecast model on data summaries that include the unobservable components of a large number of predictors incorporates abundant information into the forecasts (Stock and Watson, ). Godbout and Lombardi () suggest that extracting the best ‘signal’ out of a wide set of indicators has become a popular way to judge the path that an economy may take. Nonetheless, Stock and Watson () note that despite ‘the extraordinary complexity and regional and sectoral variation of large modern economies, it would seem surprising a‐priori that such a simple idea would have much empirical support.…”

Short Term Forecasting Performances of Classical VAR and Sims-Zha Bayesian VAR Models for Time Series with Collinear Variables and Correlated Error Terms

“…[5] considers short term load forecasting methods based on European data, while [6] [7] and [8] use different statistical models to forecast GDP in their respective countries. [9] uses factor models for short term forecasting of the Japanese economy; [10] considers factors affecting short term load forecasting and, [11] in their review considers short term weather forecasting from the physic and mathematics point of view. For other examples on short term forecasting see [12] [13] and [14].…”