Skip to main content

Demography and housing demand—what can we learn from residential construction data?

Abstract

There are obvious reasons why residential construction should depend on the population’s age structure. We estimate this relation on Swedish time series data and Organization for Economic Cooperation and Development panel data. Large groups of young adults are associated with higher rates of residential construction, but there is also a significant negative effect from those above 75. Age effects on residential investment are robust and forecast well out-of-sample in contrast to the corresponding house price results. This may explain why the debate around house prices and demography has been rather inconclusive. Rapidly aging populations in the industrialized world makes the future look bleak for the construction industry.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3

Notes

  1. The instrument set includes the third and fourth lag of the house price index and the second and third lag of inflation and real interest rates.

  2. This conjecture is based on the observation that the negative effect is much less when we do not restrict the coefficient on the first and second lag of prices to be equal in magnitude. Then, the positive first lag effect is greater. The theoretical restriction to a differenced term, thus, tends to underestimate the total price effect on investment and some of the negative relative price effects of the portfolio shift are caught by the age group instead.

  3. Instruments are age variables and lagged GDP deflators and the second and third lag of the relative price itself.

  4. Both of the instrumented estimates in Table 4 have an instrument list that includes—besides age variables—the second and third lag of real interest rates, real house prices, and inflation.

References

  • Andersson B (2001) Portfolio allocation over the life cycle: evidence from Swedish household data. Working paper 2001:4, Department of Economics, Uppsala University

  • Banks J, Blundell R, Tanner S (1998) Is there a retirement-savings puzzle? Am Econ Rev 88(4):769–788

    Google Scholar 

  • Chen N-K (2001) Bank net worth, asset prices, and economic activities. J Monet Econ 48(2):415–436

    Article  Google Scholar 

  • Chiuri MC, Jappelli T (2003) Financial market imperfections and home ownership: a comparative study. Eur Econ Rev 47(5):857–875

    Article  Google Scholar 

  • Davis MA, Heathcote J (2005) Housing and the business cycle. Int Econ Rev 46(3):751–784

    Article  Google Scholar 

  • Engelhardt GV, Poterba JM (1991) House prices and demographic change: Canadian evidence. Reg Sci Urban Econ 21(4):539–546

    Article  Google Scholar 

  • Fair RC, Dominguez KM (1991) Effects of the changing U.S. age distribution on macroeconomic equations. Am Econ Rev 81(5):1276–1294

    Google Scholar 

  • Green R, Hendershott PH (1996) Age, housing demand, and real house prices. Reg Sci Urban Econ 26(5):465–480

    Article  Google Scholar 

  • Hamilton BW (1991) The baby boom, the baby bust, and the housing market: a second look. Reg Sci Urban Econ 21(4):547–552

    Article  Google Scholar 

  • Harter-Dreiman M (2004) Drawing inferences about housing supply elasticity from house price responses to income shocks. J Urban Econ 55(2):316–337

    Article  Google Scholar 

  • Heiborn M (1998) Essays on demographic factors and housing markets. Ph.D. thesis, Deparment of Economics, Uppsala University

  • Higgins M (1998) Demography, national savings, and international capital flows. Int Econ Rev 39(2):343–369

    Article  Google Scholar 

  • Holland AS (1991) The baby boom and the housing market: another look at the evidence. Reg Sci Urban Econ 21(4):565–571

    Article  Google Scholar 

  • Leung CKY (2004) Macroeconomics and housing: a review of the literature. J Hous Econ 13(4):249–267

    Article  Google Scholar 

  • Malpezzi S (1999) A simple error correction model of house prices. J Hous Econ 8(1):27–62

    Article  Google Scholar 

  • Mankiw NG, Weil DN (1989) The baby boom, the baby bust, and the housing market. Reg Sci Urban Econ 19(2):235–258

    Article  Google Scholar 

  • Ortalo-Magne F, Rady S (2004) Housing transactions and macroeconomic fluctuations: a case study of England. J Hous Econ 13(4):287–303

    Article  Google Scholar 

  • Pettersson L (1990) Svenskt bostadsbyggande 1862–2005: en cyklisk historia (Swedish residential construction 1862–2005: a cyclical history). R58:1990, Statens råd för Byggnadsforskning

  • Pitkin JR, Myers D (1994) The specification of demographic effects on housing demand: avoiding the age-cohort fallacy. J Hous Econ 3(3):240–250

    Article  Google Scholar 

  • Poterba JM (1984) Tax subsidies to owner-occupied housing: an asset market approach. Q J Econ 99(4):729–752

    Article  Google Scholar 

  • Skinner J (1989) Housing wealth and aggregate saving. Reg Sci Urban Econ 19(2):305–324

    Article  Google Scholar 

  • Swan C (1995) Demography and the demand for housing—a reinterpretation of the Mankiw–Weil demand variable. Reg Sci Urban Econ 25(1):41–58

    Article  Google Scholar 

  • Tsay W-J, Chung C-F (2000) The spurious regression of fractionally integrated processes. J Econom 96(1):155–182

    Article  Google Scholar 

Download references

Acknowledgements

We greatly appreciate helpful comments from anonymous referees as well as from discussants and participants at the Research Seminar of the Institute for Housing Research (14–16 April 1999) in Gävle, at the Workshop for Age Effects on the Macroeconomy (2–4 June 1999) in Stockholm, at the European Network for Housing Research Conference 2000 (26–30 June) 2000 in Gävle, at the Macroeconomic Workshop at Uppsala University, and at the Population Association of America 2002 Annual Meeting (9–11 May) in Atlanta. We are grateful to Lars Fälting and Lennart Berg for help with the Swedish data. The initial research for this paper was funded by the Institute for Housing Research.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Thomas Lindh.

Additional information

Responsible editor: Junsen Zhang

Appendices

Data Appendix

We have used several sources for our data, and the Swedish time series are not publicly available, but we will provide the relevant data on request for anyone interested in replicating our results.

Our Swedish residential investment variable was kindly provided by Byggentreprenörerna, an organization for Swedish construction firms. Data from this source are available in a consistent series 1950–1996 while National Accounts data have to be linked due to changes in definition. The residential investment (expressed in constant 1991 Swedish currency, SEK) was divided by total GDP (also in constant 1991 SEK) to get the rate used in the regressions.

The house price measure is a real estate index constructed on the basis of real estate taxation values. Taxation values are based on local market transactions weighted by the distribution of local real estate over a number of standard categories. It, therefore, also reflects the upward trend in housing standards. In its current form, this index is available from 1975 and onwards. Lennart Berg at Uppsala University has interpolated the series backwards to 1952 by trends in local market transaction prices (so-called köpeskillingskoefficienter). We also owe him for the compilation of mortgage interest rates that we have used as controls. Demographic data are from Statistics Sweden, estimates and projections downloaded in 1997.

In the panel estimation, we use disaggregated investment data from OECD (1997) Economic Outlook. Our basic OECD sample contains 18 of the OECD countries: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany (BRD), Ireland, Italy, Japan, New Zealand, Norway, Spain, Sweden, Switzerland, the UK, and the USA. Missing data for several countries in the beginning of the 1960s made it simpler to start from 1964. We have a full set of annual observations from 1964–1995, except for West Germany, for which the economic series end in 1994. We then have at most 32 complete observations per country (31 for West Germany). As we lose one observation in calculating growth rates, our estimation sample adds up to a total of 557 observations. Annual demographic data are from United Nations (1994) starting in 1950. West Germany is again an exception; we obtained age structure data for this country from Statistisches Bundesamt.

The investment rates were, in this case, taken from current price ratios. The comparability over countries of price deflators is dubious, so we considered the current price ratios to be safer. We had difficulties in finding comparable interest and house price data for the whole sample and finally substituted the relative price of investment to get some proxy for these variables. These data were obtained from OECD (1997) National Accounts.

ECM equation

Standard tests cannot reject the null hypothesis that the residential investment share series is integrated of order 1. In the sample period, this will be the case for several of the age share series, too. Thus, there is hypothetically a risk that the high \(t\)-values and \(R^{2}\)s are spurious regression results. A fractionally integrated series—“long memory” or persistent series—may also give rise to spurious regression in the sense that \(t\)-values and \(R^{2}\) converge to spurious values (Tsay and Chung 2000).

However, if the age shares form a linear combination with the residential investment series and that linear combination is stationary, our results should still be consistent. In fact, they are superconsistent with substantially faster convergence to the true parameter values than an ordinary least-square estimate. To check whether spurious regression is a problem, we estimate an error correction model based on the specification we used above. Estimating a cointegrating vector by, e.g., the Johansen procedure cannot be implemented in this case. Johansen’s method is based on a vector autoregression (VAR) system where it is assumed that the cointegrating variables are endogenously determined. With so many variables with long memory, a VAR system with many lags runs into serious identification problems on annual data, so this is not a feasible option here.

The basic ECM builds on the generic model

$$ y_{t}=\beta _{0}+\beta _{1}x_{t}+\beta _{2}x_{t-1}+\beta _{3}y_{t-1}+\varepsilon _{t}.$$

Under the assumption that \(\beta _{1}+\beta _{2}+\beta _{3}=1\), an ECM representation, can be derived by assuming a stationary long-run relation \(y-x \)

$$ \Delta y_{t}=\beta _{0}+\beta _{1}\Delta x_{t}+(\beta _{3}-1)(y_{t-1}-x_{t-1})+\varepsilon _{t}.$$

With only around 45 usable observations—depending on the number of lags used as instruments—and six age shares plus another five explanatory variables, we necessarily have to restrict this generic model. As age shares are highly serially correlated, identification problems arise if we include both current and lagged values of the age shares. We use current values to achieve as little collinearity as possible with the other lagged explanatory variables. This does not invalidate the basic idea. Instead of the generic model above, we have

$$ y_{t}=\beta _{0}+\beta _{1}x_{t}+\beta _{3}y_{t-1}+\beta _{4}z_{t}+u_{t}$$

where \(z\) are other independent variables. Assuming now that \(\beta _{1}+\beta _{3}=1\), it is easily seen that we can replace \(\beta _{1}=1-\beta _{3}\) and subtract \(y_{t-1}\) from both sides to get the equivalent form

$$ \Delta y_{t}=\beta _{0}+\left( 1-\beta _{3}\right) \left( x_{t}-y_{t-1}\right) +\beta _{4}z_{t}+u_{t}$$

The basic level equation we have used previously is Eq. 5 and we only need to include a lag of residential investment in this to get an ECM representation of this form.

$$ \Delta I=\beta _{0}+\left( 1-\beta _{3}\right) \left( \sum\limits_{k=1}^{n}\beta _{k}a_{k}-I_{-1}\right) +\beta r_{-1}+\gamma \Delta Q_{-1}+\mathbf{\delta }^{\prime }\mathbf{X}+u $$
(9)

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Lindh, T., Malmberg, B. Demography and housing demand—what can we learn from residential construction data?. J Popul Econ 21, 521–539 (2008). https://doi.org/10.1007/s00148-006-0064-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00148-006-0064-0

Keywords

  • Residential construction
  • Age distribution
  • Housing demand

JEL Classification

  • R21
  • R23
  • R31