Do leading indicators forecast U.S. recessions? A nonlinear re‐evaluation using historical data

DOIhttp://doi.org/10.1111/infi.12111
Date01 December 2017
Published date01 December 2017
DOI: 10.1111/infi.12111
ORIGINAL MANUSCRIPT
Do leading indicators forecast U.S. recessions?
A nonlinear re-evaluation using historical data
Vasilios Plakandaras
1
|
Juncal Cunado
2
|
Rangan Gupta
3
|
Mark E. Wohar
4
1
Department of Economics, Democritus
University of Thrace, Greece
2
Department of Economics, University of
Navarra, Spain
3
Department of Economics, University of
Pretoria, South Africa
4
College of Business Administration,
University of Nebraska at Omaha USA,
and School of Business and Economics,
Loughborough University, UK
Correspondence
Mark Wohar, College of Business
Administration, University of Nebraska at
Omaha, Mammel Hall 300, 6708 Pine
Street, Omaha, NE 68182.
Email: mwohar@mail.unomaha.edu
Abstract
This paper analyses to what extent a selection of leading
indicators is able to forecast U.S. recessions, by means of
both dynamic probit models and Support Vector Machine
(SVM) models, using monthly data from January 1871 to
June 2016. The results suggest that the probit models predict
U.S. recession periods more accurately than SVM models
up to six months ahead, while the SVM models are more
accurate over longer horizons. Furthermore, SVM models
appear to distinguish between recessions and tranquil
periods better than probit models do. Finally, the most
accurate forecasting models are those that include oil, stock
returns and the term spread as leading indicators.
1
|
INTRODUCTION
In the wake of the Great Recession, considered by the International Monetary Fund the worst global
recession since World War II (IMF, 2009), the usefulness of economic models in forecasting recessions
has been questioned (Gadea & Perez-Quiros, 2015). By way of background to the recent increase in
interest in this topic, a large body of literature has tried to find leading indicators of U.S. economic
activity since the late-1980s (Berge, 2015; Berge & Jordà, 2011; Estrella & Hardouvelis, 1991; Estrella
& Mishkin, 1998; Giacomini & Rossi, 2006; Hamilton & Kim, 2002; Harvey, 1988, 1989; Levanon,
Manini, Ozyildirim, Schaitkin, & Tanchua, 2015; Liu & Moench, 2016; Stock & Watson, 1989).
Despite the great volume of papers on this topic, accurately predicting business-cycle turning points is
still a pertinent research topic, and increasingly so since the largely unpredicted Great Recession.
In this context, the objective of this paper is to determine to what extent a selection of leading
indicators is able to forecast U.S. recessions. The contributions of the paper are threefold. The first
International Finance. 2017;20:289316. wileyonlinelibrary.com/journal/infi © 2017 John Wiley & Sons Ltd
|
289
is the use of a long period of data, J anuary 1871June 2016, which includes very distinc t episodes in
the U.S. economy. Although many economic variables (e.g. the yield curve) provide useful
information about future states of the economy, the relationship between these indicators and the
state of the economy has declined since the 1980s (Gertler & Lown, 1999; Mody & Taylor, 2003;
Rossi & Sekhposyan, 2011). Moreover, Chauvet and Potter (2005) and Stock and Watson (2003)
find evidence of structural br eaks in the relationship between the yield curve and eco nomic activity.
The existence of these changes justifies the use of a long time period of data to analyse the
forecasting ability of different methods.
Second, thepaper uses an ample selection of leadingindicators and analyses the forecastingability of
each of them.The academic literature has proposeda wide variety of variables to predict U.S.recessions.
The slope of the yield curvethat is, the (log) difference between long- and short-term interest rates
has been found to be one of the most informative leadingindicators for predicting U.S. recessions (Ang,
Piazzesi,& Wei, 2006; Bernanke & Blinder, 1992;Dueker, 2005; Harvey, 1988; Hamilton& Kim, 2002;
Liu & Moench,2016; Rudebusch & Williams, 2009;Stock & Watson, 2003). A flat curve indicatesweak
growthand, conversely, a steep curve will be followedby stronger growth. Other variables thathave been
consideredinformative are stock prices(Estrella & Mishkin, 1998; Hamilton,2011; Killian & Vigfusson,
2013), the index of leading economic indicators (Berge & Jordà, 2011; Stock & Watson, 1989), credit
market activity (Levanon et al., 2015), and financial intermediary leverage indicators (Liu & Moench,
2016). Engemann,Kliesen, and Owyang (2010), Hamilton (2011) and Killianand Vigfusson (2013) also
find that oil prices have considerable predictive power for U.S. recessions. In order to account for
monetary policy, the literature has also included the short-term interest rate (Estrella & Hardouvelis,
1991) and various monetary aggregates (Hamilton & Kim, 2002) as explanatory variables.
The third contribution of this paper is the use of both linear dynamic probit and nonlinear Support
Vector Machines (SVM) models to predict U.S. recessions. As in most previous studies (Berge, 2015;
Dueker, 2005; Estrella & Mishkin, 1998; Kauppi & Saikkonen, 2008; Liu & Moench, 2016), we use
probit models based on the business cycle chronology proposed by the National Bureau of Economic
Research (NBER) to define recessions. Although SVM models have seldom been used to predict
recessions, among the few papers using this methodology, Gogas, Papadimitriou, and Chrysanthidou
(2015)applied the SVM model to analyse the abilityof the yield curve to forecast U.S. outputfluctuations
around its long-runtrend, using quarterly data for the period 1976:Q32011:Q4. Their results show that
the SVM methodology outperformed classic econometric models (probit models) on overall forecast
accuracy. Inorder to evaluate the accuracy of the predictions,the paper analyses both in-sampleand out-
of-sample quadratic probability scores (QPSs) (Diebold & Rudebusch, 1989) for each of the models.
The remainder of the paper is structured as follows. Section 2 describes the data and discusses the
methodology used in the paper. Section 3 shows the empirical analysis. Section 4 summarizes the main
findings.
2
|
DATA AND METHODOLOGY
2.1
|
The data
We compile a data set of monthly observations spanning the period January 1871 to June 2016,
which covers most of the historical information available on U.S. recessions.
1
The data set consists
of the S&P500 index, zero-coupon Treasury bills with maturities of 3 months and 10 years, and
West Texas Intermediate (WTI) oil prices. The stock price and the long-term interest rate data are
obtained from the website of Professor Robert J. Shiller.
2
The CPI data, also obtained from
Professor Robert J. Shiller' sw ebsite were used to generate real prices from the nominal stock and oil
290
|
PLAKANDARAS ET AL.

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT