The Fairness of Fair Market Rent

Time Series Analysis of HUD's Fair Market Rent (FMR) Rates

Corey Baughman, Mack McGlenn, Aaron Moore, Adriana Nuncio

Codeup: O'Neil Cohort
June 2023

---

Project Overview

Every year, the Department of Housing and Urban Development releases a list of Fair Market Rents (FMRs) for over 2,600 markets across the United States. The FMR is produced for each combination of property location and number of bedrooms from 0-4. FMR is used as the baseline for various government programs that provide housing assistance to low-income families, such as the Housing Choice Voucher Program (2.7 million units)and the Project-Based Voucher Program (0.3 million units). As such, the accuracy of these values can't be overstated for the sake of the millions of Americans who depend on these programs. Our project's goal is to analyze the difference between HUD's FMR and 3rd party data for rent prices to determine the reliability of FMR data in an actual market. If FMRs are too high, public housing agencies (PHAs) will be unable to maximize the number of families who receive assistance. If they're too low, it becomes more difficult for voucher users to acquire adequate housing.

---

How to Read Our Work

1. Start with this readme document for an overview of the project and how to reproduce the work
2. Check out our slide deck presentation here: https://www.canva.com/design/DAFlGI_3Ciw/SobAGAVg8JKFyQm2ETh0Sw/view?utm_content=DAFlGI_3Ciw&utm_campaign=designshare&utm_medium=link&utm_source=publishsharelink
3. Look in the final_project folder of this repo and open the final_report.ipynb notebook to see the work through the data science pipeline step-by-step
4. To further examine code used in final_report modules, check out the .py files in the final_project folder.
5. Individual folders contain workbooks showing the unpolished groundwork and insights that lead to the final notebook.

---

Project Plan

1. Planning 25-28 May 2023
2. Acquire Data 29-30 May 2023
3. Exploration and Modeling Phase 31 May - 03 Jun 2023
4. MVP Notebook and Slide Deck 5 Jun
5. Add feature scope 6-7 Jun
6. Presentation prep 8 Jun
7. First presentation 9 Jun
8. Presentation refinement 10-15 Jun

Data Dictionary

Columns	Description
diff	The difference between MMR and FMR. This is calculated as MMR - FMR
fmr	The HUD established Fair Market Rate for the entity. We chose to use the rate for two-bedroom properties.
mmr	The Median Market Rent. This is the median rent price of two-bedroom rental properties in the San Antonio/New Braunfels area as aggregated from 3rd party data (see the Acquire section for retrieval methodology) We cannot confirm that the boundaries of this area exactly match the boundaries of the federally defined MSA, but it should be close and any mismatch should remain consistent over time.
percent_diff	The percentage difference between MMR and FMR expressed in terms of FMR. This is calculated as (MMR - FMR) / FMR.

Glossary

Word	Definition
CBSA	Core Based Statistical Area (CBSA) is an Office of Management and Budget(OMB) defined area containing a large population nucleus, or urban area, and adjacent communities that have a high degree of integration with that nucleus. There are two types of CBSAs: Metropolitan statistical areas (MSAs) and micropolitan statistical areas (µSAs).
Entity ID	An entity id is a code the federal government uses to uniquely identify Metropolitan Statistical Areas. In the case of the San Antonio - New Braunfel MSA, the entity id is 'METRO41700M41700'
Fair Market Rent (FMR)	FMRs are rates established by HUD that are supposed to reflect the 40 percentile of gross rent prices in a given gegraphic area. These are established for large areas such as MSAs and rural counties as well as for individual ZIP Codes where they are referred to as Small Area FMRs or SAFMRs.
Gross Rent	The gross rent represents the entire housing cost. It is calculated by adding the rent to the owner and the utility allowance for the unit. If all the utilities are included in the rent, the rent to the owner and the gross rent will be the same. exclude telephone and internet service
Housing Choice Voucher (HCV)	HCVs are income-based subsidies that are HUDs primary tool to help provide housing for low-income Americans. They are paid directly to landlords wherever voucher-holders can qualify for a lease. HCVs are capped at the FMR for and supplement a co-payment made by the voucher-holder that is based on 30% of their income.
HUD	The U.S. Department of Housing and Urban Development (HUD) was created in 1965 as part of President Lyndon Johnson's War on Poverty. It is a cabinet-level Department that administers federal housing and urban development laws.
MSA	Metropolitan Statistical Area. This is a federally defined area that is used by the government to aggregate statistics for metropolitan areas. Each MSA is identified by an entity id code.

---

Table of Contents:

1. Initial Questions
2. Acquire & Prepare Data
3. Exploration/ Correlation Analysis
4. Statistical Analysis
5. Modeling
6. Conclusion

---

1. Initial Questions

1. Does the HUD FMR rate accurately track 3rd Party market rental data over time?
2. Are there noticeable seasonal trends in the 3rd party data?
3. Are there seasonal trends in the difference between the FMR rates and 3rd party data?

---

2. Acquire

---

Acquire Actions

Steps Taken:

1. HUD Data

   a. HUD Fair Market Rent (FMR) Acquisiton for Aggregated Metropolitan Statistical Area Data:

   • The HUD FMR data was downloaded as a csv from https://www.huduser.gov/portal/datasets/fmr.html#history .csv was acquired on 31 May 2023
   • After filtering to entity id, two-bedrooms, and years 2017-2023, there were 75 row and 4 columns.
   • Each row (obervation) represents a month between January 2017 and March 2023.
   • The features are the date index, Median Market Rate (MMR), Fair Market Rate (FMR), Difference between MMR and FMR in dollars (diff), and the percentage difference between MMR and FMR in terms of FMR (percent_diff)

   b. HUD Fair Market Rent Acquisition for ZIP Code Tablulation Area (ZCTA) level data:

   • Retrieved using the HUD API at https://www.huduser.gov/portal/dataset/fmr-api.html
   • After filtering for only the the 39 months (April 2020-Jun 2023) for which we had monthly third-party market data, and then

2. Market Rent Data

   a. The Market Rent data was acquired from Bigger Pockets, a media company and social network focused on commercial and residential real estate investing. This particular dataset was created by Kaylin Cooper, an investor who has been tracking market trends in the top 100 cities in the U.S. since 2017.

   • Size at Acquisition: 300 rows x 77 columns
   • Source: https://www.biggerpockets.com/files/user/leahd42/file/webinar-bonus-2022-rental-data

3. Realty Mole Data a. Accessed Realty Mole data using rapid api b. Saved data as a csv file

Link to functions in acquire.py file:

3. Prepare

---

Prepare Actions:

Steps Taken:

1. HUD Data a. filter rows to only those in our entity_id a. chose columns for two bedroom FMRs in years 2017-2023 a. reassign column names as years and transpose to make years rows a. group counties sharing the same entity_id into a single value for the entity a. resample the annual rates to monthly rates to align with 3rd party data frequency. a. time shift the monthly rates to line up with the federal fiscal year. a. there were no null values in the HUD data.

2. Market Rent Data a. For the sake of reaching MVP, we will only be looking at two bedroom units in San Antonio. We merged this median market rent (MMR) data with HUD's Fair Market Rent (FMR) data into one dataframe. The step by step comments for preparation can be found in the wrangleHUDpro_amr_data.py file. Here are the steps that we took to make sure the dataframes matched for joining:

   • Set the dates as the index for the dataframe
   • Isolate the market rent for the San Antonio market only
   • Keep only the values for 2 bedroom units

3. Realty Mole Data a. Converted string to dictionary b. Added zipcode and date to dictionaries c. Converted dictionary to a dataframe d. Set date as the index

https://github.com/HUD-FMR-Time-Series-Analysis/project_repo/blob/main/mvp_folder/wrangle.py

4. Exploratory Data Analysis (EDA)

We performed Time Series Analysis on the data using Jupyter Notebooks with python, pandas, matplotlib, seaborn, and statmodels.api. Further analysis was done in Tableau Public. After some initial analysis, we decided not to split the data for exploration as there appeared to be some trend reversal in the newest data and we felt that visual analysis would be impaired by looking at a truncated training dataset.

Here's what we found:

a. There has been a general widening of the gap between the FMR and 3rd Party Rents during the period from 2017-2023 for the San Antonio - New Braunfels MSA. b. There appear to be seasonal disparities between the two rates: When the federal rate is updated in October, the gap between rates tends to close, but by summer and early fall, the gap tended to widen.

Modeling

---

Features:

• Modeling will be done on all features in the dataset, with a focus on trying to predict the difference between MMR and FMR.

Model Evaluation Metric:

• Root Mean Squared Error will be used as the evaluation metric for its applicability to continuous variables.

Target:

• While the difference is the target, we will also be modeling and predicting MMR and FMR due to the difference being reliant on both.
• We will also be modeling and predicting MMR and FMR due to the difference being reliant on both.

Baseline Models:

• Three baseline models were used, with 1 month rolling average performing the best. The model below is the metrics below are the metrics to beat for the models.

Results:

• Only two models provided useful predictions, Holts Linear Trend with optimization and Holts Seasonal Trend (ExponentialSmoothing) with a seasonal period of 12 months, additive seasonality, additive overall trend, and damped hyperparameters set.
• When comparing to baseline predicitons, both models outperform the baseline, with Holt' Seasonal Trend performing the best. We will continuw on with Holt's Seasonal Trend into testing.

---

Conclusion & Reccomendations

---

Summary

• There has been a steady increase in the difference between MMR and FMR, with occasioanal decreases in the difference occurring in October, when the FMR voucher amount is adjusted.
• The best time to use a voucher is between October and March.
• Modeling was successful when using the baseline RMSE as a benchmark, but we do not believe it has a low enough RMSE to be useful for consumers.

Recommendations

• There is an adjustment period of 1 year that the FMR takes to account for changes in the market. We recommend doing semi-annual adjustments, with October being the official adjustment and March holding a preliminary adjustment. This recommendation is based on the average shift in difference between March and October.

Next Steps

• Acquire MMR data from before 2017
• Acquire and explore more locations
• Take a granular look into zip code level data instead of MSA

Steps to Reproduce

---

1. Clone this repo.
2. The final_report.ipynb in the final_project folder.
3. The data files are the .csv files included in the repo.