Identifying US business cycle regimes using dynamic factors and neural network models

• Author: Barış Soybilgen
• Published date: 01 August 2020
• Date: 01 August 2020
• DOI: http://doi.org/10.1002/for.2658

Received: 5 July 2018 Revised: 10 March 2019 Accepted: 3 January 2020

DOI: 10.1002/for.2658

RESEARCH ARTICLE

Identifying US business cycle regimes using dynamic

factors and neural network models

Bar𝚤¸s Soybilgen

Center for Financial Studies, Istanbul

Bilgi University, Istanbul, Turkey

Correspondence

Bar𝚤¸s Soybilgen, Center for Financial

Studies, Istanbul Bilgi University,

Santralistanbul, 34060, Istanbul, Turkey.

Email: baris.soybilgen@bilgi.edu.tr

Abstract

We use dynamic factors and neural networkmodels to identify current and past

states (instead of future) of the US business cycle. In the first step, we reduce

noise in data by using a moving average filter.Dynamic factors are then extracted

from a large-scale data set consisted of more than 100 variables. In the last step,

these dynamic factors are fed into the neural network model for predicting busi-

ness cycle regimes. Weshow that our proposed method follows US business cycle

regimes quite accurately in-sample and out-of-sample without taking account

of the historical data availability. Our results also indicate that noise reduction

is an important step for business cycle prediction. Furthermore, using pseudo

real time and vintage data, we show that our neural network model identifies

turning points quite accurately and very quickly in real time.

KEYWORDS

business cycle, dynamic factor model, neural network, recession

1INTRODUCTION

Whether the USA is in a recession or an expansion at any

given time is crucial information for all economic agents

in the USA and around the globe. In particular, identify-

ing the start of a recession as early as possible may help

policymakers to take necessary precautions for the econ-

omy. However, the business cycle dating committee of the

National Bureau of Economic Research (NBER), which

currently maintains the chronology of the US business

cycle, has historically announced business cycle turning

points with a significant delay. Therefore, over the years,

many business cycle dating methodologies have been

proposed in the literature.1In this study, we use dynamic

factor models (DFM) and neural network (NN) models to

determine business cycle states of current and past periods

in real time. Wepredict recessions and expansions in three

steps. First, we filter noise in the data set using a moving

average filter. Second, we use the DFM proposed by

1See Hamilton (2011) for a survey of models that aim to identify turning

points in real time.

Giannone, Reichlin, and Small (2008) to extract a handful

of dynamic factors from a large number of data series.

Finally, we feed these dynamic factors into NNs to deter-

mine recession and expansion periods in real time.

Predicting economic variables by factors extracted from

large/medium-scale data sets is a widespread approach in

the literature,2but this is not still common for predict-

ing business cycle regimes for future or current periods.

In one of the notable studies, Fossati (2016) uses probit

and Markov switching models with factors to determine

current business conditions. Forpredicting future business

cycle regimes, Bellégo and Ferrara(2009) extract static fac-

tors from 13 variables and feed them into a probit model to

forecast euro area recessions. Chen, Iqbal, and Lai (2011)

also follow a similar approach by extracting factors from

a data set including 131 variables and inserting them into

probit models to predict recessions in the US economy.

2See Eickmeier and Ziegler (2008) for a meta-analysis of factor fore-

cast applications for output and inflation and see Ba ´

nbura, Giannone,

Modugno, and Reichlin (2013) for factor nowcasting applications for

output.

Journal of Forecasting. 2020;39:827–840. wileyonlinelibrary.com/journal/for © 2020 John Wiley & Sons, Ltd. 827