Forecasting and nowcasting emerging market GDP growth rates: The role of latent global economic policy uncertainty and macroeconomic data surprise factors

• Date: 01 January 2020
• Author: I. Ethem Guney,Oguzhan Cepni,Norman R. Swanson
• Published date: 01 January 2020
• DOI: http://doi.org/10.1002/for.2602

Received: 10 December 2018 Revised: 26 April 2019 Accepted: 4 May 2019

DOI: 10.1002/for.2602

RESEARCH ARTICLE

Forecasting and nowcasting emerging market GDP growth

rates: The role of latent global economic policy uncertainty

and macroeconomic data surprise factors

Oguzhan Cepni1I. Ethem Guney1Norman R. Swanson2

1Central Bank of the Republic of Turkey,

Ankara, Turkey

2Department of Economics, Rutgers

University, NewBrunswick, New Jersey

Correspondence

I. Ethem Guney, CentralBank of the

Republic of Turkey,Anafartalar Mah.

Istiklal Cad N

̇o:10 06050 Ulus, Altındag,

Ankara, Turkey.

Email: Ethem.Guney@tcmb.gov.tr

Abstract

In this paper, we assess the predictive content of latent economic policy

uncertainty and data surprise factors for forecasting and nowcasting gross

domestic product (GDP) using factor-type econometric models. Our analysis

focuses on five emerging market economies: Brazil, Indonesia, Mexico, South

Africa, and Turkey; and we carry out a forecasting horse race in which predic-

tions from various different models are compared. These models may (or may

not) contain latent uncertainty and surprise factors constructed using both local

and global economic datasets. The set of models that we examine in our exper-

iments includes both simple benchmark linear econometric models as well as

dynamic factor models that are estimated using a variety of frequentist and

Bayesian data shrinkage methods based on the least absolute shrinkage operator

(LASSO). We find that the inclusion of our new uncertainty and surprise fac-

tors leads to superior predictions of GDP growth, particularly when these latent

factors are constructed using Bayesian variants of the LASSO.Overall, our find-

ings point to the importance of spillover effects from global uncertainty and data

surprises, when predicting GDP growth in emerging market economies.

KEYWORDS

economic policy uncertainty, emerging markets, factor model, forecasting, lasso, shrinkage

1INTRODUCTION

In many countries, initial real gross domestic product

(GDP) estimates are released at least 3 weeks after the

calendar quarter to which the data pertain. For example,

in the euro area and the USA, GDP reporting lags are 3

and 4 weeks, respectively; and in Turkey, first-release GDP

data are available only after as many as 10–12 weeks. At

the same time, tracking economic activity in real time is

crucial to the decision-making process of macroeconomic

policymakers. Fortunately, there is now an abundance of

both real-time and big datasets available to researchers,

allowing for the construction of ever more accurate early

forecasts and nowcasts (i.e., signals) of the current state

of the economy. In this context, as pointed out by Gian-

none, Reichlin, and Small (2008), dynamic factor models

(DFMs) have become one of the workhorses for short-term

forecasting, and are now widely used in central banks and

research institutions for both forecasting and nowcasting.

For further discussion, see Artis, Banerjee, and Marcellino

(2005) for the UK; Schumacher (2010) for Germany; Liu,

Matheson, and Romeu (2012) for Latin America coun-

tries; Bessec (2013) for France; Luciani and Ricci (2014)

for Norway; Girardi, Gayer, and Reuter (2015) for the Euro

Area; Modugno, Soybilgen, and Yazgan(2016) for Turkey;

Bragoli (2017) for Japan; Kim and Swanson (2018a) for

Journal of Forecasting. 2020;39:18–36.wileyonlinelibrary.com/journal/for© 2019 John Wiley & Sons, Ltd.18

the USA; Kim and Swanson (2018b) for Korea; Luciani,

Pundit, Ramayandi, and Veronese (2018) for Indonesia;

Bragoli and Fosten(2018) for India; and Cepni, Güney, and

Swanson (2019) for emerging markets.

In this paper, we contribute to the literature on now-

casting and forecasting real GDP growth in emerging

economies by empirically assessing the importance of eco-

nomic policy uncertainty and data surprises in factor-type

econometric forecasting models for five emerging mar-

ket economies: Brazil, Indonesia, Mexico, South Africa,

and Turkey. Our analysis centers around the use of DFMs

for constructing GDP predictions, although we also evalu-

ate benchmark linear autoregressive (AR) models. Impor-

tantly, our DFMs are specified both with and without

uncertainty and data surprise factors constructed using

both local and global economic variables. Moreover, in

addition to standard econometric estimation methods, we

estimate factors using a variety of data shrinkage methods

including the standard least absolute shrinkage operator

(LASSO), the adaptive LASSO, the Bayesian LASSO, and

the Bayesian adaptive LASSO.

As mentioned above, we utilize both local and global

datasets when constructing the factors used in our predic-

tion models. Although there are many empirical papers

that focus on using only local macroeconomic data in the

context of GDP forecasting, the importance of uncertainty

regarding policymakers' decisions on international eco-

nomic policies has received increasing attention since the

beginning of 2018. For example, concerns over US–China

trade tensions, Brexit negotiations with the EU, Italy's fis-

cal planning, and how the Federal Reserve Board of the

USA will determine the timing and pace of policy nor-

malization all weigh heavily on the global economy.These

sorts of international spillovers are particularly impor-

tant for emerging markets, in particular those with high

foreign portfolio ownership and weak macroeconomic bal-

ance sheets. For this reason, it is crucial to assess the

relevance of uncertainty and data surprises in the con-

text of forecasting emerging market GDP. Needless to

say, the impact of uncertainty on economic activity has

received considerable attention in the economics litera-

ture in recent years (see, e.g., Bloom, 2014). For example,

Bontempi, Golinelli, and Squadrani (2016) propose uncer-

tainty indicators based on Google Trends. Their results

suggest that online search data can provide early signals of

uncertainty, and can be used in macroeconomic forecast-

ing. Baker, Bloom, and Davis (2016) construct an index of

economic policy uncertainty (EPU) based on newspaper

coverage frequency. They find that economic policy inno-

vations foreshadow declines in investment, output, and

employment, using a panel vector autoregressive (VAR)

model for 12 major economies. Thorsrud (2018) develops

a new coincident index of business cycle activity based on

quarterly GDP and textual information contained in a daily

business newspaper.

In addition to the above proxies for economic pol-

icy “uncertainty,” a growing strand of the literature uses

consensus forecasts to disentangle macroeconomic uncer-

tainty from more “general” uncertainty. In particular, it is

argued in this literature that professional forecasters (e.g.,

those forecasters contributing to the Survey of Professional

Forecasters in the USA) closely monitor macroeconomic

data and often base their predictions on sophisticated

econometric models. Thus departures of their (consensus)

predictions from actual realizations can be viewed as data

“surprises,” which are themselves measures of macroe-

conomic uncertainty. There are different proxies for this

sort of macroeconomic uncertainty that are proposed in

the empirical literature. For example, Rossi and Sekh-

posyan (2015) construct a macroeconomic uncertainty

index based on comparing the realized forecast error of

a variable of interest with the sample distribution of the

forecast errors of that variable. If the realization is in the

tail of the distribution, they conclude that the macroe-

conomic environment is more uncertain. Carriero, Clark,

and Marcellino (2016) develop a model to identify uncer-

tainty by modeling the common component of the volatil-

ity of the forecast errors of a large set of macroeconomic

and financial variables. Finally, Scotti (2016) proposes a

macroeconomic surprise index that exploits the difference

between actual releases of data and Bloomberg forecasts

to capture economic agents' expectations about the state of

the economy. In this paper, we construct global economic

policy uncertainty and surprise indices based on a vari-

ety of different local and global datasets. More specifically,

we incorporate uncertainty into our prediction models

in three different ways. First, as our benchmark we uti-

lize only local macroeconomic data. Using this approach,

we estimate both DFMs and simple AR models, but do

not explicitly include any uncertainty or surprise indexes.

Second, we augment our DFMs with “surprise” indices

constructed using professional forecasters' expectations.

Finally, we additionally augment our DFMs with factors

extracted from a wide variety of uncertainty indices of eco-

nomic policy,trade policy, monetary policy,and migration.

It should be noted that we focus on the prediction of

GDP growth in emerging markets (EM) for two main rea-

sons. First, official releases of EM GDP figures are sub-

ject to significant publication lags and data revision, as

discussed above. Second, again as discussed above, it is

likely that the effects of uncertainty on economic activ-

ity are particularly significant in environments character-

ized by large budget deficits, high current account/GDP

ratios, and high external funding needs. As a case in point,

note that Carriere-Swallow and Cespedes (2013) investi-

gate the effects of an uncertainty shock from the USA on

CEPNI ET AL. 19