sample. Therefore, I reject the null hypothesis that the difference between the mean price
in the comparison group and the mean price observed with replacement exchanges, out-
of-state buyers, condo conversions and portfolio sales correspondingly are equal to zero. I
was not able to reject the null hypothesis for sales involving relinquished exchanges and
two different exchanges. This result is consistent with our expectations.
Next, I present the results from estimating equation (17) for each of the 15 markets in
Table 10.
I perform each of the regressions using a stepwise estimation procedure to decide
which of the submarket dummies to leave in the final model. All other dependent
variables are not subject to the procedure. I use a robust estimation method to account for
potential heteroskedasticity; therefore all reported p-values are adjusted values.
The reported results in Table 10 show that the estimated coefficients on the
structural attributes are of the predicted sign and statistically significant in most of the
market regressions. PARKING tends to be positive and insignificant. However, in two
markets, San Diego and New York City, it is negative and significant. It also has a
negative sign in the Chicago and Los Angeles regressions, with p-values of 0.14 and 0.16
correspondingly.
This finding is not surprising and is specific to higher density areas in which more
parking is usually associated with apartment communities that are distant from the
centers of the city and hence tend to be cheaper. The coefficient on UNITS is positive in
all regressions but one, and is significant in 6 of the 15 models. In Phoenix, UNITS
reverses sign and becomes negative and significant at less than one percent level. Phoenix
is the market in which the largest average transactions are observed. The average