submarkets are defined. In the smallest of the 15 apartment markets, Sacramento, there

are 21 submarkets.

 In the office sample there are 488 distinct submarkets and that these submarkets

largely overlap with the submarkets defined in the apartment sample. There are 51

submarkets present in the Los Angeles office sample, 35 submarkets distinguished in the

Phoenix area and only 6 submarkets identified in Tucson, the smallest of the 15 studied

markets.

 The retail sample contains 491 different submarkets. Los Angeles is once again the

largest market in this sample with 60 submarkets, Chicago, the second largest market in

the sample, has 37 submarkets, and San Francisco, which is the smallest market in the

sample with only 339 observations, has 33 submarkets defined.

 The estimated models differ for residential and non-residential (office and retail)

properties. For each property type I run the specified model by market. In order to avoid

the effect of outliers in the data, I winsorize all continuous dependent variables in the

regressions at the top and bottom one percent of the distribution. The winsorising

procedure takes the non-missing values of a continuous variable sorted in ascending order

and replaces its one percent highest and lowest values by the next value counting inwards

from the extremes. The only exceptions are FLOORS, PARKING and UNITS, where I

winsorize at the top and bottom 0.5 percent of the distribution, to account for the narrow

distribution. Longitude and latitude coordinates are not winsorized.

 The next chapter presents summary statistics of apartment transaction data by

market and the regression results from the models specified.