1. Predicting the Market Value of the Property Using JMP® Pro 11
Girish Shirodkar1 and Gaurav Pathak1
1Management Information Systems, Oklahoma State University, Stillwater, OK 74078
Introduction
Many individuals and corporations often end up paying more or getting less in a property deal due to limited ken of factors that decide the market value of the Property. For any individual or corporation it is therefore important to understand how particular parameters/characteristics of a property drive its ultimate market value. Very few concrete studies have been done to find which factors eventually decide the value of the property. This study, which is based on the New York City property valuation data, is an attempt to figure out governing factors for the actual market value of the property. Based upon this data and socio-demographic data derived from zip codes, JMP ® Pro 11 is used for prediction of the market value of the property in New York City.
Methods Data Preparation
The New York City property valuation dataset consists of observations and 39 variables. It is a multivariate dataset which has missing values. The target variable is a continuous variable explaining the market value of a property. Following steps were taken during data preparation and exploration:
Identification of multi-collinear data using multivariate methods
Missing Value imputation using appropriate methods
Outlier analysis using Mahalanobis distance statistic
Fig. 2: K-means clustering
Fig. 1: Scatterplot matrix and Ellipsoid 3D Plot showing correlation amongst variables
Segment
Description 1
Group of properties in NYC where the number of floors are between 3 and 7 2
Group of properties in NYC having actual total value of property between $12 - $22 million and having 10 to 25 floors 3
This segment comprises of the properties whose monthly maintenance is $20000 and above and land cost of the property is greater than $350000 4
The properties whose actual total value of property is greater than $ and the front dimensions of the building are greater than meter. 5
This segment contains of properties whose dimensions are greater than 92m X 118m and whose land cost is greater than $ 6
This segment contains of properties whose number of floors are in between and whose land cost is greater than $
Clustering and Segment Profiling
To segment the properties of New York City in distinct groups several clustering methods were applied on the imputed and transformed data. The newly formed segments were then profiled and characteristics of each segment were understood. After segments were created, different predictive models such as linear regression and decision trees were applied and compared to predict the market value of a property in New York City.
Following steps were taken during this phase:
Creation and evaluation of clusters using hierarchical and k-means clustering methods (Ref, Fig. 2)
ANOVA testing to compare means of variables across segments (Ref. Fig. 3)

2. Predicting the Market Value of the Property Using JMP® Pro 11
Girish Shirodkar1 and Gaurav Pathak1
1Management Information Systems, Oklahoma State University, Stillwater, OK 74078
Street Map Functionality
In order to get real sense of property value distribution across the New York City, We used the newly introduced street map functionality of JMP® Pro 11.
From Fig 4 it is clear that the commercial and rented properties are spread evenly across the lengths and breadths of Manhattan.
The residential properties are concentrated near the areas like SOHO, Flatiron building, Canal street and Lower Manhattan.
Fig 4 shows, the distribution of tax classes properties across all the zips and the width of the color gives the dominating property class by the full value of properties.
It can be inferred from the fig 5 that the most expensive residential properties are located in Chelsea, Greenwich and SOHO localities of Manhattan. The most expensive commercial and leased properties are located in the lower Manhattan (Financial District), Times Square and Lower east side of the Manhattan.
Fig. 3: Segment profiling for FULLVAL and AVTOT2
Predictive Modeling
The data has been cleaned and prepared by adding demographic variables, computation of new variables and transformation of skewed variables. Predictive models such as Forward Linear Regression Model, Backward Linear Regression Model, Stepwise Linear Regression Model, Decision tree and Neural Network have been used and competing models were analyzed and compared with each other. Based on R-squared criterion, the forward regression model with r-squared value of outperformed other models. Along with the property characteristics such as Extended land cost, front and depth measurements of the building in which the property is located, land costs; the demographic variables such as major industry prevailing in the area of the property came out to be important factors which ultimately derive the market value of the property.
Conclusion and Discussion
Variable transformation improves the performance of clustering algorithms as compared to the usage of skewed variables.
Liner regression model outperformed other models such as decision trees and neural network, hence it was selected as a final model
Along with the property characteristics such as Extended land cost, front and depth measurements of the building in which the property is located, land costs; the demographic variables such as major industry prevailing in the area of the property came out to be important factors that influence the property value.
The properties whose actual total value of property is greater than $ and the front dimensions of the building are greater than meter, are the highest market value fetching properties.
Reference
Acknowledgements
Prof. Dr. Goutam Chakraborty, founder of SAS and OSU Business Analytics Program at Oklahoma State University, for his continued support and guidance
Fig. 4: Liner Regression model results and parameter estimates