Problem Description
Model Descriptions
Use the nonlinear response function approach to model sales price and list price as a function of properties such as geographic location, house size, lot size, number of bedrooms, and number of bathrooms.
Because full time-course data does not exist for this application (homes are typically not sold more than 2-3 times over the course of 10-12 years), time-course profiles are constructed using data from houses that are comparable in terms of the input properties used.
Results
A nonlinear response function is built for all homes within a given area (MLS number).
For a input set of properties (geographic location, house size, lot size, number of bedrooms, and number of bathrooms), time-course profiles of both sales price and list price are calculated using the nonlinear response function.
A list of homes comparable to the home defined by the input set of properties is shown in table format.
A map showing all the homes comparable to the input set of properties is shown.
The time-course profiles of sales price and list price, table of comparable homes, and map showing comparable homes are all displayed in a single page shown here.
Conclusions
The nonlinear response function approach is able to show not only a model prediction of a home’s value at the time in which the model is built, but shows the historical valuation trend through an entire time-course of sales price and list price over a given period of time.
Through the machine learning process used to build the nonlinear response functions, predicted sales and list price trends for a given home take into account known information (geographic location, house size, lot size, number of bedrooms, and number of bathrooms) as well as information that is not explicitly defined (underlying market trends).
Comparable homes (comps) are identified using the sales price trend line.
Comps are not just linked to a given home through information at a single point in time, but are identified as having a connection to the given home over the entire range of time being modeled.
Machine-learned nonlinear dynamics allows for more accurate time-course profiles of sales price and list price, and comps that can take into account more information, even that which is not explicitly defined.