Mixed-frequency approaches to nowcasting GDP: An application to Japan

https://doi.org/10.1016/j.japwor.2021.101056Get rights and content

Highlights

  • This paper examines the performance of mixed frequency models for now-casting Japan’s GDP.

  • Some Mixed-Data Sampling (MIDAS) models and bridge equation models outperform a benchmark naïve constant growth model.

  • Forecasting performance of mixed frequency models improves with the addition of factors of soft indicators estimated with sparse principal component analysis.

  • Forecasting performance also improves when model forecasts are combined with professional survey forecasts.

Abstract

In this paper, we discuss the approaches to nowcasting Japan’s GDP quarterly growth rates, comparing a variety of mixed frequency approaches including a bridge equation approach, Mixed-Data Sampling (MIDAS) and factor-augmented version of these approaches. In doing so, we examine the usefulness of a novel sparse principal component analysis (SPCA) approach in extracting factors from the dataset. We also discuss the usefulness of forecast combination, considering various ways to combine forecasts from models and surveys. Our findings are summarized as follows. First, some of the mixed frequency models discussed in this paper record out-of-sample performance superior to a naïve constant growth model. Second, albeit small, the SPCA approach of extracting factors improves predictive power compared with traditional principal component approach. Furthermore, we find that there is a gain from combining model forecasts and professional survey forecasts.

Introduction

Understanding the current state of economy is crucial for policy makers. However, due to the inevitable publication delays of some key economic data, such as GDP, policy makers are forced to set policies without knowing the current state, and sometimes, even without knowing the past state, of the economy.

Nowcasting, the prediction of the current state of the economy, consequently has a growing body of literature around it. Although GDP is compiled mostly at a quarterly frequency and released with a lag, many business cycle indicators are timelier and more available at higher frequencies; e.g., monthly industrial production data, high-frequency financial data, or big-data obtained from internet/electronic transactions. Economists want to exploit such data in the most efficient way to monitor current state of economy in a timely manner. On the policy-making front, some central banks utilize nowcasting as a method of capturing current economic conditions in a timely fashion. For example, the Federal Reserve Bank of New York (NY Fed) and The Federal Reserve Bank of Atlanta (Atlanta Fed) release the nowcast of US GDP growth regularly.1 Bank of England (BOE) reports the nowcast of GDP in the Inflation Report.2 In addition, Norges Bank regularly releases short-term forecasts for GDP growth rates and inflation rates produced by the system for averaging models (Aastveit et al., 2011).

In this paper, we discuss the usefulness of mixed-frequency approaches to nowcasting quarterly GDP growth rates using Japanese data. To effectively utilize timely indicators which are available monthly, ranging from hard data (e.g., industrial production index) to soft data (business surveys), we employ mixed frequency approaches, Mixed-Data Sampling (MIDAS) and bridge equation approach. In doing so, we also examine factor models that utilize a novel sparse principal component approach, and also examine combination of models and professional forecasts.

Our findings are outlined as follows. First, in nowcasting the quarterly GDP growth rates, we find that out-of-sample forecasts produced by the employed models, namely MIDAS and bridge equation models, outperform those of an in-sample mean benchmark.3 In addition, we find that there is a gain from employing a sparse principal component approach in extracting factors. Furthermore, we find that some of the forecast combinations that combine model forecasts and survey forecasts improve predictive power.

This paper employs Japanese data instead of US data. The empirical success of mixed-frequency models and forecast combination schemes in nowcasting US GDP raise the question of whether such a procedure works well in other countries. In this regard, the research on nowcasting Japan’s GDP is still scarce, despite the fact that Japan is the third largest economy and the second largest advanced economy. As for the previous studies, Hara and Yamane (2013); Urasawa (2014), and Bragoli (2017) develop short-term forecasting models to conduct GDP forecasts for Japan and assess performance.4 These papers show that the forecasts generated by the proposed models are comparable with or outperform simple univariate models/professional forecasts. Our paper contributes to the literature through the following points. First, while their approaches are dynamic or static factor models, we employ a variety of mixed frequency models, including MIDAS models and Factor MIDAS models. Second, in extracting factors, we employ a novel sparse principal component analysis (SPCA) approach. Third, we consider a variety of forecast combination schemes and examine the usefulness of combining model forecasts and professional forecasts.

The paper is organized as follows. In Section 2, we discuss mixed-frequency approaches to nowcasting quarterly GDP growth rates. Section 3 discusses forecast combination schemes. Section 4 discusses data employed and empirical results. Section 5 concludes.

Section snippets

Nowcasting models

This paper focuses on forecasting the quarterly GDP growth of Japan, which is only available with a delay of two months. To estimate quarterly GDP growth, economists want to efficiently utilize data available at higher frequencies such as industrial production or survey data available at monthly. In this section, we compare several mixed frequency approaches: the bridge equation approach, Mixed-Data Sampling (MIDAS), and factor-augmented version of these approaches (see, for example, Bańbura et

Forecast combination

In Section 2, we discuss individual mixed-frequency forecasting models. In this section, we discuss the ways to combining multiple forecasts. There is a large body of literature that suggests that forecast combinations can provide more accurate forecasts by combining multiple models rather than relying on a specific model (see Hendry and Clements, 2004; and Timmermann, 2006).8 One justification for using forecast

Data

Typically, short-term indicators, such as business surveys or industrial production indexes, are released at a monthly frequency and are partially available before the release of quarterly GDP growth data. Thus, in this paper, we consider a set of monthly hard data and soft data for forecasting quarterly GDP growth (see Appendix B for data description).

For hard data, we employ the Index of Industrial Production (IIP), the Index of Tertiary Industry Activity (ITA), and the Current Survey of

Conclusion

In this paper, we discuss the mixed-frequency approaches to nowcasting Japan’s quarterly GDP growth rates.

First, we attempt to nowcast the quarterly GDP growth rates using monthly indicators, such as the Index of Industrial Production, the Index of Tertiary Industry Activity (ITA), and some soft indicators. In doing so, we employ a variety of mixed frequency approaches, bridge equation approach, MIDAS approach, and factor-augmented version of them, to utilize those data effectively. We find

References (43)

  • N. Hara et al.

    Real-time analysis on Japan’s labor productivity

    J. Int. Econ.

    (2011)
  • H.H. Kim et al.

    Mining big data using parsimonious factor, machine learning, variable selection and shrinkage methods

    Int. J. Forecast.

    (2018)
  • C. Schumacher

    A comparison of MIDAS and bridge equations

    Int. J. Forecast.

    (2016)
  • S. Urasawa

    Real-time GDP forecasting for Japan: a dynamics factor model approach

    J. Int. Econ.

    (2014)
  • K.A. Aastveit et al.

    Short-term forecasting of GDP and inflation in real-time: norges bank’s system for averaging models

    Norges Bank Staff Memo

    (2011)
  • M. Aiolfi et al.

    Forecast combination. Ch. 12

  • F. Altissimo et al.

    New eurocoin: tracking economic growth in real time

    Rev. Econ. Stat.

    (2010)
  • E. Andreou et al.

    Should macroeconomic forecasters use daily financial data and how?

    J. Bus. Econ. Stat.

    (2013)
  • N. Anesti et al.

    Peering into the present: the Bank’s approach to GDP nowcasting

    Quarterly Bulletin 2017 Q2

    (2017)
  • A. Baffigi et al.

    Bridge models to forecast the euro area GDP

    Int. J. Forecast.

    (2004)
  • M. Bańbura et al.

    Maximum likelihood estimation of factor models on datasets with arbitrary pattern of missing data

    J. Appl. Econom.

    (2014)
  • This paper was previously circulated under the title “Nowcasting Japanese GDPs.” We are grateful for helpful comments from Naoko Hara, Hibiki Ichiue, Naoya Kato, Tomonori Murakoshi, Ichiro Muto, Shinsuke Ohyama and Toshitaka Sekine and an anonymous referee. The views expressed herein are those of the authors alone and do not necessarily reflect those of the Bank of Japan or International Monetary Fund.

    View full text