HOUSING DYNAMICS OVER THE BUSINESS CYCLE

• DOI: http://doi.org/10.1111/iere.12193
• Author: Peter Rupert,Roman Šustek,Finn E. Kydland
• Date: 01 November 2016
• Published date: 01 November 2016

INTERNATIONAL ECONOMIC REVIEW

Vol. 57, No. 4, November 2016

HOUSING DYNAMICS OVER THE BUSINESS CYCLE∗

BYFINN E. KYDLAND,PETER RUPERT,AND ROMAN ˇ

SUSTEK1

University of California–Santa Barbara and NBER, U.S.A.; University of California–Santa

Barbara, U.S.A.; Queen Mary, University of London and Centre for Macroeconomics, U.K.

Housing construction, measured by housing starts, leads GDP in a number of countries. Measured as residential

investment, the lead is observed only in the United States and Canada; elsewhere, residential investment is coincident.

Variants of existing theory, however, predict housing construction lagging GDP. In all countries in the sample, nominal

interest rates are low ahead of GDP peaks. Introducing long-term nominal mortgages, and an estimated process for

nominal interest rates, into a standard model aligns the theory with observations on starts, as mortgages transmit

nominal rates into real housing costs. Longer time to build makes residential investment cyclically coincident.

1. INTRODUCTION

Over the U.S. business cycle, fluctuations in residential investment (newly constructed homes)

are well known to systematically precede fluctuations in real GDP; see, e.g., Leamer (2007).

Perhaps, due to this leading indicator property, new housing construction attracts considerable

attention by professional economists. It has also been repeatedly documented that this obser-

vation is at odds with the properties of business cycle models once the aggregate capital stock

is disaggregated into two basic components: residential and nonresidential (e.g., Gomme et al.,

2001; Davis and Heathcote, 2005). The theory predicts that nonresidential investment should

lead output whereas residential investment should lag output.

Although the cyclical properties of residential (and nonresidential) investment have been

well established for the United States, little is known about the properties of these data in other

countries. Is the United States experience unique, and data from other countries support the

existing theory? Or do the data from other countries make the need for improving the theory

even more pressing? This article has two goals: first, to shed light on the cyclical dynamics of

the two types of investment beyond the United States and, second, to use these observations to

guide the development of the theory.

In a sample of developed economies, only Canada is found to exhibit the lead in residen-

tial investment observed in the United States. Nonetheless, international data do not support

existing models either. In other countries, residential investment is, more or less, coincident

with GDP, not lagging as the theory predicts. And in all countries, nonresidential invest-

ment is either lagging or coincident with GDP, not leading as in existing models. The case

against the theory is even stronger when international data on housing starts—the number

of housing units whose construction commenced in a given period—are taken into account:

∗Manuscript received September 2014; revised June 2015.

1We thank Tom Cooley, Carlos Garriga, Paul Gomme, Grant Hillier, Haroon Mumtaz, Peter Phillips, Don Schla-

genhauf, and Norman Sch ¨

urhoff for invaluable comments and suggestions. Special thanks go to Martin Gervais, Erik

Hembre, and Vincent Sterk for insightful conference discussions. We are also grateful for comments and suggestions to

two anonymous referees and the editor as well as to seminar participants at Birkbeck, Cal Poly, Cardiff, Cleveland Fed,

Concordia, Czech National Bank, Dallas Fed, Edinburgh, Essex, Exeter, Glasgow, Norges Bank, NYU Stern, Sogang,

USC, and Yonsei and to conference participants at the Bank of England, Nottingham, Regensburg, Sciences Po, SED

(Cyprus), HULM (St. Louis Fed), and Shanghai.

Please address correspondence to: Peter Rupert, Department of Economics, 2127 North Hall, Santa Barbara, CA

93106. Phone: (805) 893-2258. Fax: (805) 893-8830. E-mail: rupert@econ.ucsb.edu.

1149

C

(2016) by the Economics Department of the University of Pennsylvania and the Osaka University Institute of Social

and Economic Research Association

1150 KYDLAND,RUPERT,AND ˇ

SUSTEK

Nearly all countries in the sample exhibit housing starts strongly leading GDP. In other words,

residential construction picks up a few quarters before GDP.

Data on housing completions point to longer residential time to build in some countries than

in the United States. During the time to build period, national accounts record in each quarter

a construction project’s “value put in place” as a part of residential investment. Time to build

thus spreads recorded residential investment over a number of quarters, making it less of a

leading indicator in countries where time to build is longer.

An important aspect of housing markets in developed economies is reliance of homeowners

on mortgage finance to purchase a property. Mortgage finance takes the form of nominally

denominated loans that homeowners gradually repay, with interest, over many years.2Further-

more, the cyclical dynamics of nominal mortgage rates—and nominal interest rates, both long

and short, more generally—are strikingly similar across countries. Specifically, mortgage rates

are negatively correlated with future GDP and positively correlated with past GDP, suggesting

that mortgage finance is relatively cheap ahead of a peak in GDP.3

Motivated by these observations, we investigate (i) if the dynamics of nominal interest rates

observed in the data transmit into similar cyclical variations in the real cost of new mortgage

finance and if such variations are sufficient to overturn the standard predictions of the theory and

(ii) if time to build in residential investment can then account for the discrepancies between the

timing, in relation to output, of housing starts and residential investment. Various idiosyncrasies

of individual countries are abstracted from. To this end, long-term fully amortizing mortgages

and residential time to build are introduced into a business cycle model of Gomme et al. (2001).

Two main types of mortgages are considered: fixed-rate mortgages (FRMs) and adjustable-rate

mortgages (ARMs).4The exogenous input into the model is an estimated vector autoregression

(VAR) process for total factor productivity, the nominal mortgage interest rate, and the inflation

rate. In the absence of an off-the-shelf structural model for the observed lead–lag dynamics of

nominal interest rates described above, this guarantees that the cyclical pattern of the nominal

mortgage rate (and inflation) in the model is as in the data.

In a baseline case with one-period residential time to build and multiperiod nonresidential

time to build, the model exhibits lead–lag patterns of residential and nonresidential investment

similar to those in the United States and Canada while also being in line with standard busi-

ness cycle moments as much as other models in the literature. Introducing into the model a

multiperiod time to build in residential construction facilitates the distinction between housing

starts, completions, and residential investment. Although mortgage finance is crucial for pro-

ducing housing starts leading output, longer time to build pushes residential investment toward

being coincident with output. In both versions of the model, mortgage finance has also an indi-

rect effect on nonresidential investment—as households want to keep consumption relatively

smooth, when movements in residential investment of the magnitude observed in the data occur

ahead of an increase in GDP, nonresidential investment is delayed, making it lag output. The

relative price of newly constructed homes responds to housing demand and exhibits cyclical

volatility and positive comovement with output similar to those in the data.

A key to understanding the role of mortgages is in the form of an endogenous time-varying

wedge in the Euler equation for residential capital. The wedge, working like a tax/subsidy

2In contrast, nonresidential fixed investment in advanced economies is predominantly financed by retained earnings

and other forms of equity. Rajan and Zingales (1995) document that typically only about 20% of the value of long-term

assets in the nonfinancial corporate sector is financed through debt.

3In all countries in the sample, nominal mortgage rates have similar dynamics as yields on nominal government

bonds of comparable maturities. The “inverted” lead-lag property of U.S. government bond yields in relation to output

has been noted by, for instance, King and Watson (1996) and, more recently, Backus et al. (2010). The same pattern

is documented for other countries by Henriksen et al. (2013). Unfortunately, a theory that would successfully account

for this phenomenon is yet to be developed.

4Most countries can be broadly classified as having either FRM or ARM as their typical mortgage contract. Research

is still inconclusive on the causes of the cross-country heterogeneity in the use of FRM versus ARM, but likely reasons

seem to be government regulations, historical path dependence, and whether mortgage lenders raise funds through

capital markets or bank deposits (e.g., Miles, 2004; Campbell, 2012).