| In this dissertation, we use various models under different scenarios with different assumptions. The models stem from the panel data framework and include spatial approaches to both the serial and contemporaneous correlation in the data. There are three main contributions. The first is an examination of the statistical performance of five test statistics for detecting unobserved group or cluster effects. Of these, two were recently proposed (C and SSRw), and the other three are traditionally used in panel data to find evidence of cluster effects (BP, SLM and F). The second is to implement the cost-distance model to develop flexible weights between locations. It gives us a method to model the high variation across space that is common in housing data. The third main contribution is to incorporate both the space and time correlation of north Dallas housing market into a modified VAR model. By considering the spatial interaction and contemporary relationship between the neighborhoods, the S_VAR model has better forecast power for short term forecast. |
