comp tool alpha v1
Many people - appraisers, real estate agents, taxpayers, etc. - do not think of residential value in terms of models. They think about comps. We get many, many questions about 'what comps did you use to value my home?' While we do want to develop a really good comparison tool, we also need a framework by which to judge one algorithm against the other.
Deliverables
-
Script(s) to produce comparables data. -
Final data set with five fields: subject_pin, comparable_pin, comparable_deed_no, comparable_sale_price, comparable_score -
Power-Point presentation of work. -
Data: Define the universe of properties that you want to find comparables for. Define the bounds (temporal, geographic) for each set of PINs, the pool of subjects, and the pool of potential comparisons. -
Optional: Temporal adjustment - rather than simply having hard bounds on the time frame for you comparables pool, consider some sort of adjustment to bring all prices to a particular point in time. -
Scoring comps: For now, I would suggest a very rudimentary comparison finder. See residential/code.legacy/41_model.R, line 53 for an example. For now, since we are using hard thresholds, your comparable_score will be binary. -
Using comps to value subject properties: Develop a simple framework for leveraging the distribution of identified comparisons to estimate the value of each subject property. -
Optional: Give measures of statistical accuracy of these values. -
Evaluating the comp-finder: Once you have predicted values for each subject property, subject those predictions to the IAAO battery. -
Human evaluation of comp-finder: Output 20 randomly selected subject PINs and their comps w/addresses to a well-formatted spreadsheet for human review.