In OLS Model II, I add controls for absolute location by using a third order
expansion of the property's latitude and longitude coordinates. This expansion yields nine
additional explanatory variables.1 The resulting model specification has the following
form:
LNPRICEm = ao +a EXREPL+a2EXRELQ+a RELQ _REPL+ A 4GE+a 5AGE2 + 6SQFT
+ aiSQFT2 + acLANDSQFT+ c9LANDSQF72 + PARKING+ 1 FLOORS+ a,2UNITS
+ ZfCONDJITIOY+a 3BUYEROUT+ a4SENIOR+ axSUBSIDIZE+ al6CONDO
2005 P_43 r r
+cO7CONDOCONV+a1,PORTSALE+ X,,YR, + '6 -'/i( 'I + 2 YZ,1(Xk ,Y) (19)
n=2000 s=2 k=0l=0
Table 11 presents the regression results for Model II. Controlling for absolute
location, in addition to relative location, improves slightly (by about 0.01) the fit of the
models. Adjusted R-squared for the market regressions range from 0.81 to 0.92. The
regression results are virtually unchanged from the results reported for Model 1.
Therefore, I will only discuss results related to the key variables of interest.
The coefficient on the variable indicating a replacement exchange, EXREPL, is
positive and significant in 12 of the 15 regressions. This reconfirms the previous result
that buyers of replacement properties are paying statistically significant price premiums
in the majority of the sample markets. The coefficient on the variable indicating a
relinquished exchange, EXRELQ, remains generally positive and significant in 7 out of
the 15 regressions. The size of coefficient estimates remains largely unchanged. The
coefficient on RELQ REPL is again positive and significant in 13 out of the 15 markets
and its magnitude remains larger than the coefficient on replacement exchanges.
1 I also experiment with a model where I control for differences in the effects of key structural
characteristics across submarkets by interacting age and the square footage of improvements with the sub
market dummies. The Third Model yields a slightly better fit R-squares are increased by 1-5 percent
across markets. However, it sacrifices even more degrees of freedom. In addition, coefficients on age and
square footage, due to the presence of interaction variables, are mostly insignificant and hard to interpret.
Therefore, I do not report the results from the third model in the dissertation.