RDP 2014-04: Home Price Beliefs in Australia Appendix E: Additional Robustness

Callan Windsor, Gianni La Cava and James Hansen

May 2014 – ISSN 1320-7229 (Print), ISSN 1448-5109 (Online)

E.1 Data Restrictions

The exclusion restrictions applied in this paper are detailed in Table E1. In this appendix, we examine the sensitivity of our main results to additional exclusion restrictions on sample sizes and outliers.

Table E1: Sample Selection
Dropped Remaining
Criteria for selection: sale prices – 1992–2012
Recorded private final sale price   3,877,815
$28,000 ≤ sale price ≤ $50,000,000 6,812 3,871,003
Non-reported bedrooms in dataset 2,352,132 1,518,871
≤ 7 bedrooms 1,117 1,517,754
≥ 2 sales per postcode-quarter 2,865 1,514,889
Criteria for selection: self-assessed home values – 2002–2011
Self-assessed home valuation   47,820
Respondent located in Sydney, Melbourne or Brisbane 27,679 20,141
$28,000 ≤ home valuation ≤ $50,000,000 3 20,138
Non-reported bedrooms in dataset 14 20,124
≤ 7 bedrooms 18 20,106
≥ 2 assessments per postcode-year 1,142 18,964

Sources: APM; HILDA Release 11.0; authors' calculations

In this robustness test, the number of sales per postcode per quarter is further restricted to at least 20 sales and the number of self-assessed home valuations per postcode per year is further restricted to be at least 6 valuations. We also trim the top and bottom 1 per cent of home valuation differences.

The original results from Table 2 are shown again in Table E2 based on the additional exclusion restrictions. The local unemployment rate result is robust to the additional restrictions. So too are the results regarding the effect of tenure on home valuation differences. While the sign on the age variable is consistent with the results presented in Table 2, the effect is now statistically insignificant.

Table E2: Explaining Home Valuation Differences across Postcodes
Age 0.007 0.013
Age2 −0.0000 −0.0001
Tenure −0.013** −0.008*
Tenure2 0.0002** 0.0002**
Log income 0.151*** 0.023
Unemployment −0.023*** −0.056***
Education 0.110*** 0.070*
Sale price Inline Equation −0.190*** −0.502***
Time fixed effects No Yes
Postcode fixed effects No Yes
R2 0.195 0.800
Observations 1,323 1,323

Notes: Robust standard errors clustered at the postcode level; ***, ** and * indicate significance at the 1, 5 and 10 per cent level, respectively

Sources: APM; HILDA Release 11.0; authors' calculations

The original results from Table 3 are shown again in Table E3 based on the additional exclusion restrictions. We find that home valuation differences have very similar effects on all the dimensions of household decisions considered in the main text.

Table E3: Homeowner Decisions and Home Valuation Differences
SPENDINGpt
Self-assessed home prices 0.221***  
Sale prices   0.225***
Valuation difference   0.174**
DEBTpt
Self-assessed home prices 0.231  
Sale prices   0.203
Valuation difference   0.486*
HDEBTpt
Self-assessed home prices 0.455***  
Sale prices   0.450***
Valuation difference   0.519***
FINSHAREpt
Self-assessed home prices 0.002  
Sale prices   −0.016
Valuation difference   0.164***
EQSHAREpt
Self-assessed home prices 0.040**  
Sale prices   0.035**
Valuation difference   0.092***
Time fixed effects Yes Yes

Notes: Bootstrapped robust standard errors clustered at the postcode level; ***, ** and * indicate significance at the 1, 5 and 10 per cent level, respectively; the dependent variables SPENDINGpt , DEBTpt and HDEBTpt are in log levels; the dependent variables FINSHAREpt and EQSHAREpt are measured as ratios; the SPENDINGpt regression is estimated over the period 2006 to 2011, for which there was comprehensive expenditure data; the DEBTpt , FINSHAREpt and EQSHAREpt regressions are estimated on the wealth module years of 2002, 2006 and 2010; the HDEBTpt regression is estimated over the period 2002 to 2011

Sources: APM; HILDA Release 11.0; authors' calculations

E.2 Weighted Least Squares

A further robustness check, which accounts for estimation uncertainty without restricting the sample, is to weight each estimated home valuation difference by the uncertainty around its estimate.

For example, to examine the robustness of our results regarding the determinants of home valuation differences, Equation (6) can be re-estimated after pre-multiplying both the left- and right-hand side variables by the inverse of the standard errors of the estimated home valuation differences (as estimated in Equation (5)).

The results from this weighted least squares (WLS) approach are shown in Table E4. The results are similar to those presented in Table 2. Reiterating, home valuation differences are positively associated with age; negatively associated with tenure (although this effect is now insignificant in the regression with fixed effects); and negatively associated with the regional unemployment rate.[16]

Table E4: Explaining Home Valuation Differences across Postcodes – Weighted Least Squares Estimates
Age 0.0136** 0.0232***
Age2 −0.0001 −0.0002***
Tenure −0.0091** −0.0043
Tenure2 0.0001 0.0001
Log income 0.0997*** 0.0440*
Unemployment −0.0245*** −0.0098*
Education 0.0634 −0.0129
Sale price Inline Equation −0.107*** −0.0710***
Time fixed effects No Yes
Postcode fixed effects No Yes
R2 0.104 0.736
Observations 2,376 2,376

Notes: Time and postcode dummies omitted from fixed effects column; robust standard errors clustered at the postcode level; ***, ** and * indicate significance at the 1, 5 and 10 per cent level, respectively; this table shows the results from re-estimating Equation (6) in the main text after pre-multiplying each variable by Inline Equation, where κpt denotes the estimated home valuation difference obtained from Equation (5) in the main text

Sources: APM; HILDA Release 11.0; authors' calculations

Footnote

The results regarding home valuation differences and household decision-making were also robust to a WLS approach, and are available upon request. [16]