Forecasting US GNP growth: The role of uncertainty

• Published date: 01 August 2018
• DOI: http://doi.org/10.1002/for.2517
• Author: Mawuli Segnon,Stelios Bekiros,Rangan Gupta,Mark E. Wohar
• Date: 01 August 2018

Received: 15 September 2016 Revised: 16 October 2017 Accepted: 23 January 2018

DOI: 10.1002/for.2517

RESEARCH ARTICLE

Forecasting US GNP growth: The role of uncertainty

Mawuli Segnon1,2 Rangan Gupta3Stelios Bekiros4,5 Mark E. Wohar6,7

1Center for Quantitative Economics,

University of Münster, Münster,Germany

2Mark E AG, Hagen, Germany

3Department of Economics, University of

Pretoria, Pretoria, South Africa

4Department of Economics, European

University Institute, Florence, Italy

5IPAGBusiness School, Paris, France

6Department of Economics, University of

Nebraska Omaha, Omaha, NE, USA

7School of Business and Economics,

Loughborough University,Loughborough,

UK

Correspondence

Mawuli Segnon, Center for Quantitative

Economics, Am Stadtgraben 9, 48143

Münster, Germany.

Email: segnon@uni-muenster.de

Abstract

A large number of models have been developed in the literature to analyze and

forecast changes in output dynamics. The objective of this paper was to compare

the predictive ability of univariate and bivariate models, in terms of forecasting

US gross national product (GNP) growth at different forecasting horizons, with

the bivariate models containing information on a measure of economic uncer-

tainty. Based on point and density forecast accuracy measures, as well as on

equal predictive ability (EPA)and superior predictive ability (SPA) tests, we eval-

uate the relative forecasting performance of different model specifications over

the quarterly period of 1919:Q2 until 2014:Q4. We find that the economic policy

uncertainty (EPU) index should improve the accuracy of US GNP growth fore-

casts in bivariate models. We also find that the EPU exhibits similar forecasting

ability to the term spread and outperforms other uncertainty measures such as

the volatility index and geopolitical risk in predicting US recessions. While the

Markov switching time-varying parameter vector autoregressive model yields

the lowest values for the root mean squared error in most cases, we observe rel-

atively low values for the log predictive density score, when using the Bayesian

vector regression model with stochastic volatility.More importantly, our results

highlight the importance of uncertainty in forecasting US GNP growth rates.

KEYWORDS

economic policy uncertainty, forecast comparison, US GNP, vector autoregressive models

1INTRODUCTION

Theoretical papers by Bloom (2009), Mumtaz and Zanetti

(2013), and Carriero, Clark, and Marcellino (2015), fol-

lowing the early works of Bernanke (1983) and Dixit

and Pindyck (1994), confirm that, besides productivity

and/or policy shocks, various forms of policy-generated

uncertainty lead to business cycle fluctuations. While the

(negative) influence of uncertainty on economic activ-

ity is well established theoretically, in the wake of the

“Great Recession” the focus has also been on quantifying

the impact of uncertainty. Understandably, this requires

a measure of uncertainty—an otherwise latent variable.

In this regard, there are two approaches to measuring

uncertainty: (i) the news-based approach of Brogaard and

Detzel (2015) and Baker, Bloom, and Davis (2016),

whereby the authors perform month-by-month searches

of newspapers for terms related to economic and policy

uncertainty to construct their measure of economic policy

uncertainty (EPU); (ii) alternatively, Mumtaz and Zanetti

(2013), Mumtaz and Surico (2013), Alessandri and Mum-

taz (2014), Mumtaz and Theodoridis (2015, 2017), Bali,

Brown, and Tang (2015), Carriero et al. (2015), Chuliá,

Guillén, and Uribe (2015), Jurado, Ludvigson, and Ng

(2015), Ludvigson, Ma, and Ng (2015), Rossi and Sekh-

posyan (2015), Rossi, Sekhposyan, and Soupre(2016), Shin

and Zhong (2016), and Creal and Wu (2017) recover mea-

sures of uncertainty from the estimation of various types of

Journal of Forecasting. 2018;37:541–559. wileyonlinelibrary.com/journal/for Copyright © 2018 John Wiley & Sons, Ltd. 541

542 SEGNON ET AL.

small- and large-scale structural models related to macroe-

conomics and finance. Irrespective of which approach

(news or model based) is pursued, these studies, along

with others that have used such indices (e.g., Bachmann

& Bayer, 2011; Bachmann, Elstner, & Sims, 2013; Balcilar,

Gupta, & Segnon, 2016; Benati, 2013; Caggiano, Casteln-

uovo, & Groshenny, 2014; Castelnuovo, Caggiano, & Pel-

legrino, 2015; Cheng, Hankins, & Chiu, 2016; Colombo,

2013; Jones & Olson, 2013; Kang, Lee, & Ratti, 2014; Karni-

zova & Li, 2014; Knotek & Khan, 2011) confirm the signif-

icant role of uncertainty in affecting economic activity.

However, apart from Karnizova and Li (2014) and Bal-

cilar et al. (2016), all the above-mentioned studies trying to

link uncertainty with economic activity (e.g., measures of

output and/or unemployment, investment) have entailed

in-sample analysis. Karnizova and Li (2014) depict the role

that the news-based EPU of Baker et al. (2016) can play in

forecasting US recessions based on probit models. By con-

trast, Balcilar et al. (2016) emphasize that forecasting gains

for US recessions can be obtained using mixed-frequency

Markov switching models. Against this backdrop of lim-

ited evidence on out-of-sample forecasting of a measure

of economic activity, and given the widely held view that

importance of variables and models require out-of-sample

validation (Campbell, 2008), the objective of our paper

is to use a wide array of univariate and multivariate lin-

ear and nonlinear models in analyzing the role played

by the news-based measure of the EPU of Baker et al.

(2016) in forecasting US gross national product (GNP)

growth rate. In our study, we analyze the forecasting per-

formances of the various models considered over the his-

torical quarterly period of 1919:Q2 to 2014:Q4, using an

in-sample of 1900:Q1 to 1919:Q1. The decision to use the

news-based EPU, rather than model-based uncertainty,

simply emanates from the availability of a measure of

uncertainty for forecasting GNP growth over the longest

possible sample period, covering various phases of US

economic history.

Forecasts of output growth represent an important indi-

cator for both policymakers and financial investors. Such

figures reveal information about the current state of the

economy and play a key role in formulating appropriate

monetary and fiscal policies. As indicators of future growth

potential of the economy, they help financial investors

in their investment decision-making process. Hence the

need for accurately forecasting the growth rate of the econ-

omy cannot be overstated. Given the importance of fore-

casting economic growth, different powerful univariate

and multivariate econometric models have been devel-

oped in the literature to provide accurate GDP growth

forecasts, especially of the vector autoregressive (VAR)

variety (for an overview of different models for forecast-

ing output growth, see Chauvet & Potter, 2013; Eickmeier,

Lemke, & Marcellino, 2011; Giannone, Lenza, & Prim-

iceri, 2015; Rossi & Sekhposyan, 2010, 2014; Schorfheide

& Song, 2015; Schumacher, 2011). This paper consid-

ers the baseline autoregressive moving average (ARMA)

model, the Bayesian VAR (BVAR), the threshold VAR

(TVAR), the smooth transition VAR (ST-VAR), two-types

of time-varying parameter VARs (TVP-VARs), the Markov

switching VAR (MSVAR), the unobserved component

stochastic volatility (UCSV), the Bayesian VAR with CSV

and a mixed-frequency VAR (MF-VAR) to produce both

point and density forecasts of US GNP growth. As dis-

cussed in Rossi and Sekhposyan (2010), Bekiros and

Paccagnini (2013) and D'Agostino, Gambetti, and Gian-

none (2013), it is important to model nonlinearities when

forecasting US output, due to issues of structural insta-

bility, and also when relating movements of output with

uncertainty (Caggiano, Castelnuovo, & Nodari, 2014).

Hence we look at both linear and nonlinear models. In

addition, Herbst and Schorfheide (2012), Barnett, Mum-

taz, and Theodoridis (2014), and Rossi and Sekhposyan

(2014) argue that it is becoming more and more impor-

tant to assess the uncertainty associated with the forecasts

of models. This is specifically why central bankers wish

to evaluate how well models perform in forecasting a

range (uncertainty) of future values of relevant macroe-

conomic variables, rather than just the point forecasts of

these variables. In other words, the forecaster needs to

look not only at point forecasts but also analyze den-

sity forecasts (Bekiros & Paccagnini, 2015a). Note that

in this paper we basically take an atheoretical approach,

although there are of course theoretical models of fore-

casting output based on large-scale Keynesian-type models

and microfounded dynamic stochastic general equilibrium

models (for detailed reviews in this regard, see Bekiros &

Paccagnini, 2013, 2014, 2015b; Del Negro, Hasegawa, &

Schorfheide, 2016; Del Negro & Schorfheide, 2012). It must

be emphasized that our objective in this paper is not nec-

essarily to contribute to the model sets used in forecasting

output growth. Rather, the objective is primarily that of

forecasting US GNP growth, based on existing models, but

for the first time incorporating the role of EPU.

The remainder of the paper is organized as follows.

Section 2 presents the data used in our analysis. In Section

3 we describe the different forecasting models used in

this study. Section 4 provides the forecasting evaluation

methodologies. The empirical results are presented in

Section 5, and Section 6 concludes.

2DATA ANALYSIS

This study uses time series of different frequency that stem

from different sources. The quarterly data on real US GNP