Behavioral heterogeneity in the Australian housing market

ABSTRACT

We propose a heterogeneous agent model (HAM) of four groups of investors with Markov chain regime-dependent beliefs for the housing market. Within the Markov switching framework, we take into account how heterogeneous investors shift their trading behaviour in response to changes in housing market conditions. The model is estimated and compared with the benchmark rational expectation models using the Australian housing market data from 1982Q1 to 2013Q2. We find evidence of within- and between-group heterogeneity in the Australian housing market. We show that HAM with Markov switching beliefs provides a better in-sample estimation efficiency and outperforms the conventional rational expectation models in terms of out-of-sample prediction.

KEYWORDS:

• Regime switching
• behavioural heterogeneity
• housing market
• heterogeneous agent model

JEL CLASSIFICATION:

• C22
• G19
• R31

Acknowledgments

We thank the anonymous reviewer for very insightful and constructive suggestions. All remaining errors are our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ The capital cities include Sydney, Melbourne, Adelaide, Perth, Canberra, Brisbane, Hobart and Darwin.

² Alternatively, seasonal dummies could be introduced to account for seasonal effects, however, because of the limited length of the available time series data, this would comprise the degree of freedom in the estimations, especially for the benchmark model with structural breaks and the Markov switching model.

³ Similar to the notations for the directly estimated parameters, we use $m_{\mathrm{\ell }}{\sigma }_{\mathrm{\ell }}(0)$, $m_c{d}_c(0)$, $m_v{\beta }_v(0)$ and $m_s{\beta }_s(0)$ to denote the respective parameters in state 0, and $m_{\mathrm{\ell }}{\sigma }_{\mathrm{\ell }}(1)$, $m_c{d}_c(1)$, $m_v{\beta }_v(1)$ and $m_s{\beta }_s(1)$ to denote the respective parameters in state 1.

⁴ In comparing the in-sample fit, we use the AIC in addition to the log-likelihood. The advantage of AIC is that it can deal with the trade-off between the goodness of fit of the model and the complexity of the model. AIC not only rewards goodness of fit, but also includes a penalty that is an increasing function of the number of estimated parameters. This penalty discourages over-fitting.

⁵ Since the breakpoint tests suggest obvious breaks in the sample, we use rational expectation model with breakpoints to perform the predication comparisons with the Markov switching model and leave out the rational expectation model without any breakpoint.

Additional information

Funding

This work was supported by the Nanyang Technological University Tier 1 Research Project [RG180/14] and the Fundamental Research Funds for the Central Universities [grant number 20720161073].