You are on page 1of 27

Affordable Housing in Kent County, MI

Title Slide 2

MADS Milestone 1 Project


Alan Fehsenfeld | Koon Leong Ho | George Thio
February 3, 2021
Table of Contents
1. Executive Summary
2. Data Sources
3. Data Manipulation
4. Results Summary
5. Analysis and Visualization
6. Conclusions
7. Problems, Enhancements, Future Development
8. Appendix
Executive Summary
Background Motivation
A study by the The Terner Center for Housing Innovation The viability of affordable housing projects depends
at UC Berkeley found that housing stability leads directly critically on securing adequate funding. As MSHDA
to social improvement by allowing underprivileged allocates funds on the basis of the LIHTC multi-factor
members of society to focus on education, employment, points scoring system, the ability of affordable housing
and other dimensions of economic mobility, while also developers to identify high-scoring projects is an
promoting investment in professional development that important determinant of success.
reduces involuntary relocation through greater job stability.
Unfortunately, no tools exist to help developers in this
The Federal Low Income Housing Tax Credit Program endeavor and they typically rely on word of mouth and
(LIHTC) provides tax credits to promote development of ad-hoc site visits to find potential properties for
affordable rental housing. In Michigan, this is administered development. This is inefficient, unsystematic, and often
by the Michigan State Housing Development Authority produces suboptimal and uneconomical decisions.
(MSHDA) which evaluates projects according to overall
quality and the location-driven LIHTC Scoring Criteria: Our project fills this gap with an integrated search and
visualization tool to identify and compare ideal
• Proximity to Transportation development sites by aggregating multiple data sources
into a single database and calculating the MSHDA score.
• Proximity to Amenities
• Enterprise 360 Opportunity 360 Score The geographic scope of this project is limited to Kent
• Walk Score County, but it can be easily expanded state, region or
• Distance from Employment Centers nation wide if the required data sources are available.
• Located in Neighborhood Investment Activity Areas
• Affordable/Market Rent Differential
• Located in Opportunity Zone or Rising Tide
Communities
Data Sources
Given numerous location-based criteria used in the MSHDA scoring system, one of the most challenging and time consuming tasks for
developers seeking high-scoring properties is accessing, extracting, verifying, and integrating data from many disparate sources.
Our tool instantly performs these tasks and stores the results on a unified table which enables filtered user searches across 227,210
properties and 547 apartments in Kent County. Once data aggregation is complete, MSHDA scores for every property can be calculated.

Data Type Source Extraction method


Proximity to Transportation https://transitfeeds.com/p/the-rapid/380 Text file download
Proximity to Amenities Google Maps Platform Google Maps search
https://www.michigan.gov/mshda/0,4641,7-141-5
Enterprise 360 Opportunity 360 Score CSV file download
587-60010--,00.html
Walk Score Walk Score® Walk Score API
Distance from Employment Centers Not included due to insufficient data N/A
Located in Neighborhood Investment Activity Areas Not Included due to insufficient data N/A
Affordable/Market Rent Differential Rentometer Rentometer API
Located in Opportunity or Rising Tide Communities https://www.cdfifund.gov/opportunity-zones CSV file download
Geospatial shapefiles https://www.accesskent.com/Departments/GIS/ Shapefiles download

Note: Please refer to this page throughout this presentation. Given the number and amalgamation of data sources used, they will not be cited individually for each chart.
Data Manipulation
All property feature files are tagged with Assessor's Parcel Numbers (APN) which are used as keys to merge features in a
unified table on which data cleaning is performed. Meaningless data (e.g. rows without a city) are dropped, fields with NaN
are replaced with ‘0’, and calculated features are created. The final processed data is saved to a pickle file to improved read
performance with geospatial files organized by municipality. All files are then merged into a single set for the entire country.

Data Type Manipulation


Properties within ¼ mile of a bus stop will have its indicator set to ‘5’. This is calculated based on the
Proximity to Transportation
lat/lon of the property and lat/lon of bus stops using GTFS transit data.
Properties within 1 mile of a amenities will have its indicator set to ‘1’ for each amenity type. This is
Proximity to Amenities
calculated based on the lat/lon of the property and the lat/lon of the amenity using Google Maps.
‘0’ or ‘1’ score for each measure based on census tract of each property and the census tract of
Enterprise 360; Opportunity 360 Score
properties within ¼ mile.
Walkscores were calculated for the centroid of each census block. Properties receive 5, 7 or 12
Walk Score
points according to proximity to centroids with certain walk scores.
Distance from Employment Centers Not included as employment data was not readily available.
Located in Neighborhood Investment Activity Areas Investment data is available through building permits, but the system is not easily searchable.
Pass property lat/lng and unit size to API to provide rent estimate based on other properties in close
Affordable/Market Rent Differential
proximity, This is determined by a search radius from the property.
Located in Opportunity Zone or Rising Tide Indicator set to ‘3’ for being in a Opportunity Zone Census Tract. There are no Rising Tide
Communities Communities in Kent County.
Property tax SEV was extracted from CoreLogic database available through the University of Michigan Library
Geospatial files were downloaded from Kent County GIS and US Census. Geospatial references were
Geospatial shapefiles converted to EPSG:4326 (lat/lon) format. All Kent County municipalities were merged into a county
wide set of files.
Data Manipulation
● The bulk of our data originated from vetted and reviewed government sources that were almost completely error-free
with Kent County GIS data being the notable exception as it contained duplicate values which we carefully handled to
ensure legitimately duplicated IDs on a single property were retained (e.g. condo units in the same building).

● The largest source of unavailable data was the SEV field which was not available for every parcel. This was not
problematic as SEV was informational only, and did not contribute to the MSHDA score, so NaNs were not changed.

● Another source of uncertain data is the market/affordable differential generated by evaluating each property to
determine if is within ¼ mile of a high-rent apartment building. Unfortunately, we could only find rent estimates for
apartment buildings rather than for individual units. To address this, we created a rent threshold variable that
awarded full or zero points (in line with the MSHDA) at various rent levels, to reflect the inherent uncertainty of the
estimate. We chose not to include this in the MSHDA score due to the variability of this proxy measure.

● The biggest data challenge was joining disparate geographic dataset which entailed the following steps:

1. Converting geographic data to a common projection.


2. “Buffering” one of the datasets to add a ¼ mile radius around each geospatial observation.
3. Comparing the buffered dataset to a dataset containing all parcels to determine proximity to at least one high-rent
property of interest. This was done with the geopandas sjoin function.

● Overall, our primary challenge was finding good datasets that contained all the necessary information needed to
produce the MSHDA scores. For example, we could not find good datasets for community investment, apartment rental
rates, or large employers. As such we had to scale down our original plans, but we are encouraged by our initial efforts
and endeavor to continue searching for these datasets.
Summary of Results
Highlights: Efficient all-in-one search and visualization tool
In the following slides we will demonstrate the efficiency
and information gains available to affordable housing
developers who use our tool to search for right-sized
properties in Kent County, MI that have high pro-forma
MSHDA scores, and thus a greater chance of receiving
competitive MSHDA funding.

Based on the professional working knowledge of one of


our team members, we know that even the more
sophisticated developers in this specialized real estate
domain typically conduct their searches with only a single,
often proprietary database.

By automating sourcing, manipulation and linking of


records from different data sources using geospatial
coordinates, we created a robust set of reporting
features to facilitate customized data views and maximize
probability of project success by capturing all relevant
scoring criteria, and overall search efficiency.
Analysis and Visualization
Score distribution by city: Size matters ... but is not everything
The absolute number of higher scoring properties are in larger, more highly populated cities such as Grand Rapids and
Wyoming (left chart), but many smaller cities also have high percentage of properties with high scores (right chart) while in
Grand Rapids, the vast majority of properties have low scores highlighting the overall scarcity of high-scoring properties.
Analysis and Visualization
Size matters (again) … but so does money
Budgets for affordable housing projects are effectively capped by
funding limits set by MSHDA. Properties cannot be too large or
too expensive while small properties will not have enough space
for buildings or parking. Based on experience with past projects,
plots ares categorized as follows:
● Small (< 1 acre)
● Ideal (1-10 acres)
● Large (> 10 acres)
Kernel Density Estimates of MSHDA scores show ideal sized
plots have low scores while many properties over one acre are
suburban or rural which is suboptimal for affordable housing.
Visualizing MSHDA Score and Cost Per Acre (with SEV as a cost
proxy) across the three categories for a score range of 0-30 that
captures 95% of the dataset, we see that the vast majority of
ideally sized properties (red dots) are lower cost (quadrants 1
and 2) although more of these have lower scores (quadrant 2)
than higher scores (quadrant 1).
We also see no higher priced (quadrants 3 and 4) large
properties (blue dots) and almost all large properties having
lower scores (quadrant 2) reflecting the greater challenge of
profitably developing large scale projects, which is true for
residential property development in general.
Analysis and Visualization
Plotting for potential: Scoring is inclusive and agnostic
It takes less time and effort to identify properties and LOW score property in Grand Rapids: 1422 Preston Ridge NW
prepare MSHDA submissions based on conversion of
existing multi-family dwellings (e.g. apartments)
available for sale in the market.
However, developers that limit searches to this universe
overlook attractive opportunities among properties not
currently for sale and/or not currently categorized as
multi-family which require more time and money to
identify, acquire and redevelop, but may offer a greater
return on investment and effort though a higher MSHDA Description: Active, apartment building, 3-storeys, 222-units, 43 acres
score and higher probability of funding success.
Our search tool scores all properties solely on the basis
of their location-driven MSHDA score irrespective of
property type or whether the property is available for sale. HIGH score property in Grand Rapids: 1450 Fulton St E
It also allows users to create sensitivity bands around the
rent estimates used to determine the binary affordability
component of the MSHDA score (0 or 5 points).

In Grand Rapids, we have identified two properties for


comparison to demonstrate the efficacy of our inclusive
and agnostic search tool, one of which is a live project that
has been submitted to MSHDA for funding.
Description: Vacant, seniors living and care facility, 5.5 acres
Analysis and Visualization
Ready to rent, but score says ignore 1422 Preston Ridge NW, Grand Rapids, MI: MSHDA Score = 8

A first glance of the website for 1422 Preston Ridge


NW would suggest it to be a promising plug-and-play
candidate for affordable housing conversion given its
“abundant wildlife, scenic trails, picnic areas, outdoor
pool, fitness center, dog park, community room, and
community garden” and its location “just minutes from
I-96, US-196, US-131 and the Alpine Avenue retail
corridor”.

However, good looks do not deceive our 100%


data-driven search tool which assigned a MSHDA
score of just 8 points and would save an unsuspecting
developer the considerable time and effort of gathering
disparate data on the property and making a site visit.

MSHDA Scorecard for 1422 Preston Ridge NW

Photo: https://www.richmondhillsapts.com/photogallery.aspx
Analysis and Visualization
Redevelopment diamond in the rough 1450 Fulton St E, Grand Rapids, MI: MSHDA Score = 25

The status of this property as the subject of a currently


live MSHDA affordable housing proposal1 by Hope
Network Housing and Community Development
(“Hope”) is an encouraging validation of our tool which
assigned the site a strong score of 25.

Discussions with a key team member revealed that Hope


does not have the wherewithal to access, aggregate, and
process the many data sources required to produce a
pro-forma MSHDA score, and it is believed this vacant
property was discovered through informal channels and
was not the result of a systematic search process.

MSHDA Scorecard for 1450 Fulton St E

1
See Transforming Fulton Manor to Eastpointe Commons in the Appendix
for more details and a presentation of our featured subject property
Photo: LoopNet
Conclusions
⌂ Our project premise that it is difficult and time ⌂ Accessibility and availability of shapefiles is a
consuming to find right-sized properties for prerequisite for producing our visualizations and
affordable housing development that are both low showing a property in its neighborhood. Tabulated
cost and have high pro-forma MSHDA scores was MSHDA scores are useful for filtered searches and
affirmed by the 6+ weeks we spent on diverse data do not require shapefiles, but interpretation of
acquisition, manipulation and integration to produce results is greatly enhanced by the ability to
the final dataset on which scores were calculated and visualize properties in their local context allowing
visualizations created. users to make qualitative assessments and draw
valuable insights that data alone cannot provide.
⌂ With over 220k plots in Kent County (94k in Grand
Rapids alone) this search process is not only costly ⌂ Neighborhood safety is an important concern for
and inefficient but also inherently low-success in the every home dweller. Our plan to include a safety
absence of an integrated, systematic and fully score (not a MSHDA criterion) was thwarted by the
MSHDA-score compliant approach that will leave no unavailability of crime statistics at the block level
stone unturned. Our tool fills this gap and immediately necessary to differentiate between properties. We
increases productivity and probability of success of will incorporate this scoring factor in our tool when
developers heretofore reliant on ad-hoc methods and the data becomes available.
drawing upon a narrow set of search resources.

⌂ Our search and visualization tool offers high utility for property developers seeking to identify sites in Kent County,
Michigan with the highest probability of receiving MSHDA-distributed LIHTC tax credits which contribute significantly to
the financial viability of affordable housing projects. However, it is important to note that the MSHDA evaluation
includes many non-scorable factors such as developer experience, property manager experience and whether a subject
property has received a tax abatement from city government. As such, developers still need to play an active role in the
planning and preparation of MSHDA submissions, but we believe our tool will be an invaluable asset in this process.
Problems, Enhancements, Future Direction
Problems Future Direction
Notable problems we encountered during the project: A study commissioned by Kent County2 projects an
affordable apartment supply gap of 2,892 units in Grand
• High network latency to the hosted PostgreSQL server Rapids by 2025. With average household size of 2.57, this
made it unusable so Pickle file format was used. represents 3.75% of the city’s metro population or 7,432
affected persons which is below the 9,228 count of
• Plotly plots all properties regardless of plot view homeless people in 2019. Supply shortages such as this
produces long response times and browser crashes that are growing in many cities nationwide resulting in high
made interactive filtering impractical. A strategy to rents and a vicious circle of unaffordability.
reduce the number of data points to plot is required.
As such, upon implementation of proposed enhancements,
• Interactive Dash filtering results in a full refresh of a we hope to expand the geographic scope of our tool and
map view. A method to preserve the last location and explore opportunities to collaborate with stakeholders at
zoom settings is required to make interactivity useful. different levels (e.g. Kent County) to improve data
accessibility, functionality and accuracy.

Enhancements Given limited public funds and higher profit sensitivity to


occupancy level for affordable housing projects, we are
• Develop demand-side modelling capability. keen to develop localized demand-side modelling
capability that will allow us to incorporate a score for
• Find unit count and size data for multi-family buildings. projected occupancy which should be invaluable to
• Find solutions to database access and plotting MSHDA and its LIHTC administrator peers in other states
problems (e.g. geofeather). to reduce risk of funding projects that may score highly on
supply factors, but meet with low market demand.
• Add Dash interactivity to plots (e.g. sliders, radio
buttons, HoloViews) to allow live user search filtering.
2
See Transforming Fulton Manor to Eastpointe Commons in the Appendix
for more details and a presentation of our featured subject property
Statement of Work
The evolution of our search tool concept and realization ⌂ Project Concept was the brainchild of Alan
of this project was the culmination of the combined and Fehsenfeld who is also the subject matter expert
coordinated efforts of all the team members.
⌂ Primary Author for the project proposal was Alan
This statement of work does not imply that work was Fehsenfeld
completed exclusively by those mentioned, rather it lists
the team members that assumed lead responsibility for ⌂ Data Extraction was performed by Alan
research and defining the best approach to meet the Fehsenfeld and George Thio
requirements of the project.
⌂ Data Cleaning was performed by Koon Leong Ho

⌂ Geospatial Plots were completed by Alan


Fehsenfeld, and Koon Leong Ho

⌂ Visualizations for Exploratory Data Analysis


were done by Koon Leong Ho

⌂ Project Report was the combined effort of all the


team members.

⌂ Project Administrator was George Thio


References & Acknowledgements
References Acknowledgements
We used the following resources to develop our domain knowledge We thank the following people for their guidance and support:
and support data manipulation and analysis:
• Mike Lapsley, President at Rentometer
• Plotly Mike provided us with a free two-week Pro Standard
https://plotly.com/python/getting-started/ account inclusive of 1,000 API calls which he generously
• Seaborn extended and topped up multiple times to allow us to
https://seaborn.pydata.org complete our exploratory data analysis and obtain rent
estimates for the 547 buildings in our Kent County
• GeoPandas apartments reference dataset
https://geopandas.org/gallery/plotting_with_geoplot.html • Anthony Whyte, Lecturer at UMSI
Anthony provided valuable feedback and guidance that
• European Petroleum Survey Group (Coordinate Systems) addressed many of concerns and uncertainties around the
https://epsg.org/crs_2253/NAD83-Michigan-South-ft.html?sessi form and function of the final product
onkey=sheapb9657
• Chris Teplovs, Lecturer and Research Investigator at UMSI
• Kent County, Michigan A timely consultation with Chris at the start of the project
https://www.accesskent.com/GISLibrary/checkAgree.do#Parcels helped us to quickly come together as a team, and draft a
&Streets project timeline that established a baseline for productivity
and accountability
• Reid, C. K. (n.d.). The Links Between Affordable Housing and • Catherine Morse, University of Michigan Library
Economic Mobility. Retrieved February 01, 2021, Granted us access to the CoreLogic files containing various
https://ternercenter.berkeley.edu/wp-content/uploads/pdfs/Links_ property level data including APN (Assessor’s Parcel
Between_Affordable_Housing_and_Economic_Mobility_.pdf Number) and SEV (State Equalized Value)
Appendix
1. Python libraries used in this project
2. Transforming Fulton Manor to Eastpointe Commons - Presentation
by Hope Network Housing and Community Development
Python Libraries
Libraries Usage
Numpy and Pandas Data manipulation

GeoPandas Data manipulation of geospatial information

Matplotlib Backend plotting library that supports Plotly and Seaborn

Plotly Plotting library used for geospatial visualizations

Seaborn Plotting library used for non-geospatial Exploratory Data Analysis


Transforming Fulton Manor to Eastpointe Commons

A Hope Network Housing and Community Development Project


Why Affordable Housing?
• A study commissioned by Kent County found that Grand Rapids found that over the next five years, the City of
Grand Rapids will have a 2,892 low-income apartment “housing gap”
• The authors of the study surveyed 78 low income apartment complexes in Kent County and only found 11
vacancies
• The lack of housing supply means rents are high.
• 33,222 households are paying more than 30% of their household income for rent
• This leaves very little room for other necessities and unexpected financial crises
• Lack of housing supply and high rents put families at risk of homelessness
• In 2019, Kent County recorded 9,228 homeless people, including:
• 2,820 children
• 1,931 survivors of domestic violence
• 2,922 people with a disability
Why Affordable Housing?
From the Headlines:
What is Affordable Housing?
Affordable housing is built to the same standard as market rate housing, but rents are limited.
What is Affordable Housing?
Affordable Rents Low Income Rents Compared to Average Eastown Rents
Affordable housing is for two groups: Bedrooms Affordable Rent Average Eastown Rent Difference
• The most vulnerable, which includes: 1 $738 $1,130 $392
• Homeless families
• Survivors of domestic violence 2 $893 $1,439 $546
• People with disabilities 3 $1,023 $1,794 $771
• Families with low incomes

For all groups, rents are limited


according to a formula based on
household income.
What Is the Eastpointe Project?
• Fulton Manor is a 200,000 square foot former home for the elderly.
• The Eastpointe project will convert Fulton Manor into a 118 unit
affordable apartment complex.
What Is the Eastpointe Project?
• 118 Apartments
• 77 One Bedroom
• 36 Two Bedroom
• 5 Three Bedroom
• Four floors with as many as
34 apartments per floor
• The total cost to renovate
the building will be
approximately $40 million.
• Building interior will be
completely renovated
• Building exterior will be
preserved according to
historic standards.
• The building will be LEED
certified.
How soon will the project be done?

Conservative Project Timeline

Financing Construction Most apartments


Apply for funding complete/begin Complete occupied
2/1/2021 construction
2/1/2022 9/1/2023 2/15/2024
And finally….

Thank you for


your attention
and support!

You might also like