Using Regression for Identifying Opportunities in Real Estate
0 likes91 views
The document presents a regression report for a real estate agency outlining a model created to estimate rental prices based on key property factors, utilizing data from idealista.com. It recommends targeted marketing campaigns, an internal alert system for identifying opportunities, and a personalized agent compensation structure to enhance competitiveness in the market. The model validation indicates a predictive accuracy improvement with an R² value ranging from .83 to .87, focusing on properties priced below 10,000 or exceeding 10,000 sq. meters.
Using Regression for Identifying Opportunities in Real Estate
- 1. Presented by: Bruno Cervantes Quino Bruno Gobbet Gianini Federico J. Garcia Lopez George Kofi Akanza Kamal Nandan Melody Ucros Peder Viland Identifying New Opportunities REGRESSION REPORT 2017 To: Sales Director
- 2. Executive Summary ...….….…………….…………….………………..…………….……….…. 3 Recommendations …….……………………………………….………………………………...... 4 Targeted Campaigns…………………....………….……….…………….………..... 4 Internal Alert System ………………..…………………….…………….………...... 4 Agent Compensation ..…………….…….....……………….………………………. 4 Analyst Approach ………...………..….……………….………….…………………………..…... 5 Data Preparation ……………………………………………………………….……..... 5 Variable Selection ………………………………………………………………….…… 5 Model Creation …………………………………………………………….......……… 5 Model Validation ………………………………………………………….…….…..... 5 Annex …………………………….…….……………..……………………………………….……..…. 6 Table of Content Regression Report 2017 2
- 3. Executive Summary As a Real Estate Agency, identifying good opportunities to help clients buy or sell a home is a key task of our value proposition. The team of analysts has created a model with which we can estimate rental prices based on a number of factors and gain a competitive advantage with our offers in the market. To create this tool, we used a dataset from idealista.com. The dataset consisted of numerous descriptive variables that could influence price but, based on their statistical significance, the team has only kept the following variables and some of their interactions: • Sq. Meters of the property • Area where the property is located • Bedrooms that the property has • If it is an Outer Property • If it is a Penthouse or a Duplex Note that certain descriptors have more influence over others when estimating price. For example, key observations regarding the models predictive ability is that: • When a property is a penthouse, or a property which we know the sq.mt. in the Area of Retiro, our models estimation is the most accurate. • When we know the sq.mt of properties in the Areas of Salamanca, Tetuan, Chamartín, Centro and Hortaleza, our estimations can also be highly trusted. • Our model works best for properties with rental prices below 10,000 or when the property is above 10,000 Sq.Mts. Therefore, these are the three initiatives that could help the agency gain a competitive advantage when compared to competitors: • Create Targeted Marketing Campaigns for these specific areas informing tenants about the potential of their property in regards to buying, selling or renting prices. • Build an internal tool where properties that meet a certain sq.mt threshold and marked at a price below area standards, are automatically added to the pipeline of opportunities. • Establish Quarterly Goals for the agents to meet in these areas, since we can significantly improve our offense strategy when finding properties now. In conclusion, we can expect these initiatives to improve the ROI of the agents time, the proactivity with which we approach clients, and the offers that we can negotiate in the market. Regression Report 2017 3
- 4. Recommendations Targeted Campaigns Branding ourselves as helpful and knowledgeable about the market can help us attract new clients and grow our business. The best way to do this is through targeted Facebook Campaigns, with an informative video of different neighborhoods, a walking tour, the average prices, and how to maximize the value of properties. Another way to do this is through an email or mailbox campaign, but it would be a bit harder to measure our efforts and allocate our resources. Therefore, if the second option in preferred, the best strategy would be to partner with specific building owners or local businesses in these different areas. The information from the model will be used in the video but also once these new clients approach us, so that we can more accurately assess the opportunity in hand. Internal Alert System Creating an internal tool to keep track of good opportunities is essential. This can be done by using software that can be configured to automatically discover and crawl real estate websites. The initial configuration usually allows you to set data parameters like city, state, zip code, selling price, rent price, address, property size and characteristics. Based on these parameters, listings can be saved to database with normalized fields for easier access and search capabilities. The information from the model will be used to create those specific parameters per area, in order to automatically issue an alert once a good opportunity is identified. Agent Compensation Real Estate is a people-business, and that means treating agents and clients at a superb level. The information on rental prices that the model provides can be used to create a more personalized compensation structure for finding those “anomalies” in the market, and closing them. We would already be providing them the tool to identify some, but it is up to them to bring that business to our agency. This compensation structure can be driven by quarterly goals that agents should fulfill in different areas, and the recruitment of clients that might not be listed online. Regression Report 2017 4
- 5. Analyst Approach Data Preparation The original dataset had the following information: ID, Area, Address, Number, Zone, Rent, Bedrooms, Sq. Mt. , Floor, Outer, Elevator, Penthouse, Cottage, Duplex, and Semi-Detached. We removed the columns that were un-factorable, were too narrow for our analysis, or had too many missing variables. The dataset wasn’t too big, so we used Excel to clean it manually. We then imported the prepared dataset into R Studio and factored several fields to be able to work with them. Variable Selection In order to choose which variables to keep, we used Stepwise Forward method in R, which simply kept adding variables that were significant until it could no longer be improved. This was the final output: Rent ~ Sq..Mt. + Area + Outer + Bedrooms + Penthouse + Duplex. This was giving us an R2 of .76, which meant that only 76% of the variance in rental prices could be explained by our model. Stepwise method doesn’t test for interactions though, so we manually tested different interactions of the variables and kept the ones that improved the R2. Model Creation To compare the original model with the manually modified one, we used an Anova test. Because the p-value resulted in < 5%, then the variables added did matter and the new model was significantly better. This was the final model used: Rent ~ Sq. Mt. + Area + Outer + Bedrooms + Penthouse + Duplex + Area: Sq. Mt. Model Validation To validate our model we used K-Fold Cross Validation, dividing our dataset 80/20 for training and testing. When testing, our final model yielded an R2 that ranged from .83 to .87. We also created some graphs to visually assess the models prediction abilities. One helped us see the average price per area, and another one which values might be having an impact in the curve of our model. Another important conclusion from one of the graphs was that our model works best for properties with rental prices below 10,000 or when the property is above 10,000 Sq.Mts. Regression Report 2017 5
- 6. Annex Average Rent by Area: Regression Report 2017 6 Estimating Rent:
- 7. Annex Identifying Opportunities: Regression Report 2017 7 Stepwise Output:
- 8. Annex Final Model: Regression Report 2017 8