Time Series Modelling for Real Estate Values

Overview

This project aims to empower real estate investors with sophisticated time series models using Zillow's dataset. By leveraging advanced forecasting techniques, the project provides accurate insights for strategic decision-making in real estate investments.

Contributors

• Carol Mundia
• Ian Nzangi
• Pascalia Maiga

Project Outline

1. Introduction
2. Problem Statement
3. Objectives
4. Data Understanding
5. Data Preparation
6. Exploratory Data Analysis (EDA)
7. Modelling
8. Recommendations
9. Conclusions

Introduction

The goal of this project is to provide investors with a time series model to forecast real estate values using historical data from Zillow. This enables investors to make informed decisions on potential investment opportunities.

Problem Statement

Real estate investment requires meticulous analysis and strategic decision-making. This project aims to identify the top zip codes offering promising investment opportunities by leveraging historical data from Zillow Research.

Objectives

1. Identify the top 5 zip codes and states with the best investment potential.
2. Analyze historical real estate value data to understand patterns over various time intervals (monthly, quarterly, semi-annual, and annual).
3. Create a time series model to predict future real estate values.

Data Understanding

• Dataset contains 14723 rows and 272 columns.
• Columns include RegionID, RegionName (zip code), State, City, SizeRank, and house values for each region.
• Date range: April 1996 to April 2018.

Data Preparation

• Removed missing values, reducing data to 13680 rows.
• No duplicate values, but outliers were retained as recorded events.
• Created two data frames: one for EDA and another for time series analysis and modeling.
• Added a "Returns on Investment" column.

Exploratory Data Analysis (EDA)

• Histogram showed normally distributed average return on investment (ROI).
• Most regions had an average ROI between 0.02 to 0.06.
• States with the highest number of houses: California, New York, Texas, Pennsylvania.
• Top zip codes for ROI: 11211, 11222, 11216 (Brooklyn, NY), 7302 (Jersey City, NJ), 11930 (Amagansett, NY).
• States with highest ROI: New York, New Jersey, Delaware, South Dakota, California.
• Overall upward trend in house values, except for a fall between 2006 and 2013 due to the 2008 recession.

Modelling

ARIMA Models

• 4 ARIMA models were tested.
• Model with the least error: 4th ARIMA model.
  • MAE: 84.47
  • RMSE: 117.55

Prophet Models

• 3 Prophet models were tested.
• Model with the least error: 3rd Prophet model.
  • MAE: 86.22
  • RMSE: 121.87

Model Comparison

• The 4th ARIMA model performed best overall.

Recommendations

Real Estate Investment

• Invest in real estate due to an upward trend in house values.

States and Regions

• Recommended states: New York, New Jersey, Delaware, South Dakota, California.
• Top zip codes: 11211, 11222, 11216 (Brooklyn, NY), 7302 (Jersey City, NJ), 11930 (Amagansett, NY).

Forecasting Model

• Use the developed model to forecast future real estate values to mitigate risks.

Conclusion

• A predictive model was built to assist investors in making informed real estate investment decisions in the United States.

Next Steps

• Collect more real estate value data to improve model accuracy.
• Continuously train the model for better predictive results.

Contact

For further information, please contact any of the team members listed above.