significantly different from the average price of properties in the control group. Since
these statistics do not control for size, the reason for the extreme average value of
portfolio sales is, as their name suggests, that they simply involve larger transactions than
with single sales. Sale-leaseback transactions are associated with insignificantly higher
average prices than those of the control group.
Next I perform regressions based on estimating equation (18) for each of the 15
office markets
LNPRICF, = o + aEXREPL+ 2EXRELQ+a3RELQ_REPL + 4AGE+aA GE + 6SQFT
3
+ acSQFT2 + aLANDSQFT+ a9gLANDSQF72 + PARKING+ a FLOORS+ YI- CONDITION[
z=2
2005 P 43
+ a2BUYEROUT+a13SALELEASEBFCK+a14PORTSALE+ z,,YRn + Y6 \u\i)\ i' (18)
n=2000 s=2
Similarly to the previous chapter, each of the regressions is estimated using
stepwise regression, in which I allow the procedure to select which of the submarket
dummies to leave in the final model, based on their contribution to the fit of the model.
All other dependent variables are not subject to the stepwise procedure.
Table 15 reports coefficient estimates for each variable by market. The reported
results are based on regressions in which standard errors are adjusted to account for
potential heteroskedasticity. P-values are reported below the coefficient estimates. R-
squared varies by market from 84 percent (Las Vegas and Riverside / San Bernardino) to
92 percent (Washington, DC model). Although the R-squares suggest very good fits of
the models, they are driven primarily by the strong relationship between price and square
footage, and price and lot size. It is important to note that, in general, non-residential
commercial real estate is harder to value than residential commercial real estate. With
office, industrial and retail properties, lease structures can be very complicated and vary