Price competition in the spatial real estate market: allies or rivals?

ABSTRACT

This paper examines whether real estate firms can avoid price competition when properties in the vicinity are priced by allies. An oligopoly model with differentiated products generally suggests that real estate firms engage in price competition with their spatially closest rivals. Yet, they can raise property prices when the market share of their allies increases. To test this prediction, a spatial autoregressive model with spatial autoregressive disturbances, including a share of allies in the vicinity, is estimated using data on the prices of residential condos in central Tokyo, Japan. The model prediction is supported by the empirical results. In the data set, the magnitude of the market share on property prices increases with the expansion of the size of the spatial market.

KEYWORDS:

• real estate competition
• spatial econometrics
• strategic pricing
• Tokyo

JEL:

• C31
• D43
• L85
• R31

ACKNOWLEDGMENTS

The authors thank two anonymous referees, Arnab Bhattacharjee, Seung-Young Jeong, Koji Karato, Mark Lijesen, Takanori Nakade, Mitsuru Ota, seminar participants at Kanagawa University, Keio University, University of Tokyo, and conference attendees at the Applied Regional Science Conference in Toyama, the Asian Real Estate Society in Jeju, the European Network for Housing Research Housing Economics Workshop in Vienna, and the North American Regional Science Council in Miami for their valuable comments. The authors are grateful to Marketing Research Center Co., Ltd for access to its micro-data.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

ORCID

Shinichiro Iwata https://orcid.org/0000-0002-3161-7180

Kazuto Sumita https://orcid.org/0000-0001-7791-356X

Mieko Fujisawa https://orcid.org/0000-0002-5618-8538

Notes

1 The 2014 National Condominium Market Report. Retrieved June 23, 2015, from https://www.fudousankeizai.co.jp/.

2 Major 7 is a website for new condos run by the following major real estate firms: Sumitomo Realty & Development Co., Ltd, Daikyo Inc., Tokyu Land Corporation, Tokyo Tatemono Co., Ltd, Nomura Real Estate Development Co., Ltd, Mitsui Fudosan Residential Co., Ltd, and Mitsubishi Jisho Residence Co., Ltd. For details, see https://www.major7.net/ (accessed on June 23, 2015).

3 The SARAR model is also known as a spatial autoregressive combined (SAC) model (LeSage & Pace, 2009).

4 Because ${\theta }_i$ takes a small value, which is shown in the empirical section below, we assume that ${\theta }_i$ and ${\theta }_k$ ($k=1,\cdots ,N$, and $i\ne k$) are independent.

5 More precisely, we should specify a vector ${\mathrm{\Theta }}_i=\left({\theta }_{i1},\cdots ,{\theta }_{iN}\right)$ representing the state of cooperation, with elements defined by ${\theta }_{ik}=1$ if the properties $i$ and $k$ are allies, and 0 otherwise (Bresnahan, 1987). In our empirical context, however, ${\theta }_i$ in equation (1(1) $p_i=\rho {\sum }_j{w}_{ij}{p}_j+\gamma {\theta }_i+{\mathit{X}}_i\mathit{\beta }+{ϵ}_i,$(1) ) reduces the number of explanatory variables substantially compared with ${\mathrm{\Theta }}_i$.

6 Chasco, Le Gallo, and López (2018) proposed the scan test to check the spatial group-wise heteroskedasticity of the spatial models. Although they checked the heteroskedasticity and tried several specifications in their example, they could not specify the heteroskedasticity. To cope with the problem, they estimated the model using the method proposed by Kelejian and Prucha (2010), which is robust to spatial heteroskedasticity.

7 For details about the MRC, see http://www.mrc1969.com/ (accessed on October 5, 2017).

8 We also considered the five- and 10-nearest-neighbour weight matrices. Nevertheless, these alternative specifications did not change the main empirical results. See the discussion in the last subsection of this section.

9 In addition to the revenue market share, we also tested two alternative formulations for the share of the ally: unit market share and relative market share. None of these alternative specifications fundamentally changed the qualitative results. See the discussion in the last subsection.

10 The samples appear to show that the average share is distributed randomly in space. We cannot directly observe the spatial correlation between condo prices and the share of the ally at the geographical level.

11 Stakhovych and Bijmolt (2009) recommended their approach based on Monte Carlo experiments of SAR and SEM models under the assumption of normality of the error terms; however, we estimate the SARAR model under non-normality of the disturbances $\mathit{u}$. These results must be taken with caution because the model and the non-normality of the disturbances were not examined in the prior study. The selection of the weight matrices for the SARAR models under non-normality of the disturbances remains for future research.

12 This corresponds to the average total impact suggested by LeSage and Pace (2009).

13 We examine a polar case where all units are assumed to obtain the spatial weight matrix (full radius case). The estimated coefficients of the spatial lag and the share of ally have significant and positive signs. The semi-elasticity is 1.8, which is the largest of the three cases.

14 The empirical results based on ${\mathit{W}}_4$, ${\mathit{W}}_5$ and ${\mathit{W}}_6$ are not shown but are qualitatively the same pattern as the benchmark cases in Table 6.

15 Different from Western countries, markets of existing condos are quite thin in Japan. Therefore, both potential buyers and sellers may not consider used condos as substitutes.

16 Beck et al. (2012) estimated a fixed effects model for 25 Louisville neighbourhoods for the period 2000–08 to examine the effect of changes in the HHI of residential real estate brokerage on property prices. The empirical results suggested that prices are insignificantly correlated with HHI.

17 We incorporate the spatial lag of Distance and the spatial lag of Zoning into the benchmark model and estimate the GNS model. The empirical results suggest that semi-elasticity is 0.3 in the 2-km radius, while it is 1.1 in the 4-km radius. These results are similar to those of the SARAR model in Table 6.

Additional information

Funding

This research was supported by the Association of Real Estate Agents of Japan (Fudosan Ryutsu Keiei Kyokai).