RDP 9202: Some Tests of Competition in the Australian Housing Loan Market 4. Results

Jerome Fahrer and Thomas Rohling

February 1992

The parameter estimates for three models are presented in Tables 1(a) and 1(b). In one sense, the results are rather unsatisfactory, with only about one-third of the estimated parameters signifcantly different from zero. However, since we do not wish to place any emphasis on the reactions of any particular banks, the individual parameter estimates are of limited interest; rather it is the aggregate variations which are of interest. The estimates in Table 1(b) are disappointing in that most of the estimates on the interest rates are wrongly signed or insignificant. Our guess is that there is a lot of noise in the monthly data that is not captured by our specification. Possibly, a more elaborate supply function needs to be estimated to capture the variation in monthly lending.

Table 1(a): Conjectural Variation Parameters
(Standard Errors in Parentheses)
  βi1 βi2 βi3 βi4 βi5 βi6 αi
q1   −0.10
(0.03)		 0.11
(0.03)		 0.34
(0.12)		 −0.10
(0.09)		 0.02
(0.01)		 −0.88
 
q2 −0.90
(0.52)		   0.13
(0.11)		 −0.71
(0.55)		 0.67
(0.38)		 0.05
(0.05)		 −0.47
 
q3 −1.56
(0.73)		 −0.11
(0.18)		   0.41
(0.75)		 −0.03
(0.51)		 0.003
(0.04)		 0.24
 
q4 −0.01
(0.12)		 −0.08
(0.04)		 0.04
(0.03)		   0.02
(0.10)		 0.01
(0.01)		 −0.48
 
q5 −0.18
(0.17)		 0.08
(0.05)		 0.002
(0.04)		 0.48
(0.19)		   −0.01
(0.02)		 −0.52
 
q6 1.77
(1.78)		 −0.26
(0.44)		 −0.04
(0.27)		 −1.00
(1.89)		 −1.08
(1.26)		   0.07
 

1=CBA, 2=WBC, 3=ANZ, 4=NAB, 5=SBV, 6=RES
Table 1(b): Other Parameters
(Standard Errors in Parentheses)
  γi0 γi1 γi2 R2 DW
q1 0.57
(0.92)		 0.02
(0.08)		 −0.05
(0.03)		 0.67
 		 1.67
 
q2 16.52
(2.90)		 −1.23
(0.23)		 0.17
(0.10)		 0.52
 		 1.31
 
q3 −4.15
(4.89)		 0.56
(0.39)		 −0.07
(0.15)		 0.38
 		 1.91
 
q4 2.79
(0.99)		 −0.25
(0.07)		 0.07
(0.03)		 0.43
 		 2.08
 
q5 −4.63
(1.20)		 0.32
(0.10)		 −0.01
(0.04)		 0.65
 		 1.85
 
q6 14.94
(11.99)		 −2.00
(0.90)		 1.00
(0.36)		 0.31
 		 1.63
 

1=CBA, 2=WBC, 3=ANZ, 4=NAB, 5=SBV, 6=RES
Coefficients on dummy variables not reported.

The aggregate variations are the αi reported in the last column of Table 1(a). These are found by summing the βi in the relevant column. Thus, the αi corresponding to the CBA is equal to −0.88, found by summing the coefficients in the column headed βi1. In every case the point estimate of αi is much closer to −1 (corresponding to perfect competition) than it is to −1+1/si (perfect collusion). Nevertheless, in every case (except for the CBA), the departure from perfect competition appears to be significant.

This hypothesis is confirmed by the results of the formal tests for competition, collusion and Cournot oligopoly are reported in Table 2. The test statistics are calculated as

Table 2: Tests for Competition, Collusion and Cournot Oligopoly
Test Test Statistic Probability Value*
Competition:
αi=−1, i=1..6		 47.81
 		 0.00
 
Collusion:#
αi=−1+1/Si, i=1..6		 108.23
 		 0.00
 
Cournot Oligopoly:
αi=0, i=1..6		 3.65
 		 0.72
 

* significance levels at which null hypothesis is rejected.
# 1=CBA, 2=WBC, 3=ANZ, 4=NAB, 5=SBV, 6=RES
s1=0.23, s2=0.18, s3=0.11, s4=0.12, s5=0.13, s6=0.23

where T is the number of observations, while |Σr| and |Σu| are, respectively, the determinants of the covariance matrices of the errors in the restricted and unrestricted systems. The test statistics are distributed as Chi-squared, with degrees of freedom equal to six, the number of restrictions. The results of the tests are such that we can decisively reject the hypothesis of perfect competition, and even more decisively, perfect collusion in the market for housing loans. However, we are unable to reject the hypothesis that this market can be characterised as a Cournot oligopoly.

In interpreting these results, we note that the standard errors on the conjectural variation parameters are relatively large. Thus, our inability to reject the hypothesis of a Cournot oligopoly might be due to the imprecision of our parameter estimates. However, this imprecision strengthens our rejections of perfect competition and perfect collusion, since large standard errors reduce the likelihood that any hypothesis will be rejected.

Finally, we report calculations of both unadjusted and adjusted Herfindahl indices of concentration, where the market is defined as in the model (the five largest banks plus one residual bank). The unadjusted index is 0.18, while the adjusted index is 0.11. Thus the housing loan market can be characterised as containing about nine banks of equal size, and so appears to be reasonably competitive, but not perfectly so.