Yield Curve Point Triplets in Recession Forecasting

• Author: Theophilos Papadimitriou,Periklis Gogas,Efthymia Chrysanthidou
• Published date: 01 June 2015
• DOI: http://doi.org/10.1111/infi.12067
• Date: 01 June 2015

Yield Curve Point Triplets in

Recession Forecasting

Periklis Gogas, Theophilos Papadimitriou

and Efthymia Chrysanthidou

Department of Economics, Democritus University of Thrace, Greece.

Abstract

Several studies h ave highlighted the yield cur ve’sabilitytoforecast

economic activity. These studies use the information provided by the

slope of the yield curve—i.e., pairs of short- and long-term interest

rates.Inthispaper,weconstructthreemodelsforforecastingthe

positive and negative deviations of real US GDP from its long-run

trend over the period from 1976Q3 to 2011Q4: one that uses only

pairs of interest rates and two that draw on more than two points

from the yield curve. We employ two alternative forecasting methodol-

ogies:theprobitmodel,whichiscommonlyusedinthislineof

literature, and the support vector machines (SVM) approach from the

area of machine learning. Our results show that we can achieve a 100%

out-of-sample forecasting accuracy for negative output gaps (reces-

sions) with both methodologies and an overall accuracy (both infla-

tionary and unemployment gaps) of 80% in the case of the best SVM

This research was co-financed by the Europe an Union (European Social Fund, ESF) an d Greek

national funds through th e ‘Education and Lifelong Learning’operation al programme of the National

Strategic Reference Framework (NSRF ). Research funding also c ame from the program ‘THALES,

Investing in a knowledge socie ty through the European Social Fund ’(MIS 380292).

International Finance 18:2, 2015: pp. 207–226

DOI: 10.1111/infi.12067

© 2015 John Wiley & Sons Ltd

model. The forecasting performance of our model strengthens the

existing evidence that the yield curve can be a useful tool for gauging

future economic activity.

I. Introduction

The yield curve is a graphical representation of the relationship between the

maturity and the y ield to maturity of bon ds issued by a singl e debtor at a specific

point in time. In the empirical l iterature, the shape of the yi eld curve is often

associated with phases of bu siness cycles, whi ch are measured as fluctuati ons of real

GDP from its long-run trend (Stock and Watson 1989). Although the term ‘cycle’

implies repeated p atterns, the a ctual behavi our of these fluc tuations is usua lly

unpredictable. However, there is signi ficant evidence that the yield curve can be an

efficient indicator of future economic ac tivity.

1

Obviously, the interest in and importance of forecas ting future economic ac tivity

is not merely academic . The ability to forecast the phase of the business c ycle within

a reasonable marg in of error is of g reat interest to policy makers and investors.

Litterman (1986) applied a Bayesian vector autoregressive model (BVAR) using US

data from 1948Q1–79Q3 to forecast future econom ic activit y. The model included

variables such as real GDP, the GDP deflator, industrial investment, quarterly yield s

of government bonds, unemploym ent and the money supply. After comparing the

results of his metho dology wi th forecasts from leadin g models—the Data Resources

Inc. model, the Whar ton model and the Chas e Econometrics mode l—Litterman

concluded that his BVAR model outperformed t he others in most c ases. However,

Giacomini and Rossi (2005) provided evidence that the yield curve’s ability to

forecast growth has weakened since the 1980s, such that its forecast ing power now

remains strong only for recessions.

Whether the yield sprea d can provide signals about f uture recessions has been the

subject of the work of numerous researche rs, including Stock an d Watson (1989),

Estrella and Hardouvelis (1991) and Chen (1991). Estrella and Mishkin (1998)

investigated the perform ance of a probit model to forecas t US recessions up to eig ht

quarters in advan ce. They showed that when th e interest rate spread was fed into a

probit model, it yield ed more accurate results than models fed with othe r leading

macroeconomic indicators . Estrella and Trubin (2006) showed that the y ield spread

between the 3-month Treasury bill and the 10- year government bond can be an

excellent indicator of future economic act ivity. Galv~

ao (2006) supported this view by

suggesting a mod el for recession forecastin g in which the spread between sh ort- and

long-term rates is consid ered a leading indic ator of future output. She compared

four different model s—a VAR, a threshold VAR, a structural-break VAR and a

1

For example, an inverted yield cur ve is considered a sign of either a n upcoming recession or an

output gap.

208 Periklis Gogas et al.

© 2015 John Wiley & Sons Ltd