What’s behind your Milwaukee property tax bill and how machine learning could make it fairer
One of the most common ways of calculating assessments — the “sales comp” approach — has come under scrutiny as more and more researchers identify biases in the underlying data.
For example, most of the historical data used to assess neighborhood value was based on redlining — a practice that automatically devalued neighborhoods with a majority Black or Hispanic population.
“If the information being fed into these databases has historical biases … it’s going to produce biased results," said Chad Venne, one of the instructors at the University of Wisconsin-Milwaukee’s real estate certificate program.
Even though these assessments are an integral part of how property taxes are calculated, the process itself is complex and often feels opaque to homeowners. Those calculations are further complicated by turbulent housing markets and a reliance on factors such as neighborhood desirability that have roots in racially biased data.
With property tax bills arriving in Milwaukee residents’ mailboxes this month, the question of whether everyone is paying their fair share, based significantly off property assessments, will be top of mind.
Homeowners who believe they have received an assessment lower than the home’s worth worry about losing equity. Homeowners who believe they have an assessment higher than their home’s worth worry about what the likely increase in property taxes could mean.
Need to appeal an assessment?The deadline to appeal Milwaukee property assessments is May 16. Here are five tips to know beforehand.
SAS, an analytics and technology company founded in the 1970s as the Statistical Analysis System, has created its own assessment model which is supposed to correct for historical bias.
In anticipation of property taxes being mailed out this month, we asked SAS and real estate experts what residents should know about the assessment process and what systems may have the best chance to overcome biased data.
How are assessments calculated?
According to Venne, there are three different ways assessed values are typically determined:
Replacement costs, or what it would cost to rebuild the property
An income approach, typically only used with investment properties
Sales comparison, often shortened to “sales comp”, approach which looks at the sales of similar homes in the area.
The sales comp approach, Venne explained, is the most common and also the most dependent on factors outside of the home.
“If you’re an existing homeowner, you’re sitting there at the mercy of what is happening around you,” he said.
Factors such as higher crime rates or vulnerabilities to weather events (such as a neighborhood that becomes chronically prone to flooding) can cause homes to sell for lower than they may actually be worth, lowering the value of homes assessed in that neighborhood.
What do property assessments have to do with taxes?
When multiplied by the local tax levy, assessments are used to help calculate a homeowner's property tax bill.
Higher-assessed properties tend to have higher property taxes.
“The way property taxes work is the municipality needs to raise a certain amount of money, that’s a dollar amount, and then they need to figure out what the property tax needs to be for each property. If I have $200,000 property and you have a $100,000 property, and they’re using the same rate, my taxes will be twice as yours,” explained Anthony Pennington-Cross, the Bell chair of real estate at Marquette University.
That’s why Jennifer Robinson, the global government strategic advisor at SAS, said it is important assessments are done accurately and without bias — to ensure tax burdens are equitable.
“That is one of the largest forms of revenue that governments have, so it’s really important that we get this right,” she said.
How does Milwaukee calculate assessments?
In Milwaukee, property assessments are done on an annual basis.
According to a spokesperson from the assessor’s office, the city uses a combination of market-adjusted costs and sales comp data to develop assessments. Factors considered include age/condition, house features, square footage and neighborhood desirability.
However, Pennington-Cross noted models such as the city’s, tend to work best in areas where the homes are very similar.
“If there’s a lot of variety, then they’re going to struggle. If a new house (is) next to an old house, a really well-maintained house next to a poorly maintained house, then the model will be off,” he explained. “It basically is looking at, here in the neighborhood, the quality of the housing stock is good. If you’re lower income and your home is not in great shape, you will likely be over appraised.”
How does a turbulent real estate market impact assessments?
Venne said the recent volatility of the market has made any assessment method factoring sales comp data more difficult.
“We have rarely seen so much fluctuation at the magnitude we have over the last 18 months,” he said. “If you’re looking at the sales comp approach, every sale that happened changed the market because the prices were increasing so rapidly so quickly that the appraisers had a hard time trying to understand what truly was the market value.”
This was particularly problematic in areas where gentrification was taking place.
“You find if you are a homeowner in those areas, that value of the real estate surrounding you might be dramatically increasing based on the process of new development coming in or new buildings happening in the neighborhoods,” Venne said.
That, in turn, leads to higher property taxes which can sometimes displace residents who can no longer afford the taxes on their home.
However, there is good news.
Venne noted the market is cooling and those in the real estate industry are expecting enough lag from the previous year that averaging out the data for several years will get them closer to actual value.
What could make assessments fairer?
One company believes it has found the solution to overcoming the bias in historically racist data.
Reggie Townsend, the director of data ethics at SAS, said their assessment system uses machine learning, which is, as defined by MIT, “artificial intelligence that gives computers the ability to learn without explicitly being programmed.”
That machine learning can detect whether a model is exhibiting bias.
“In our platform, (we) look at sensitive variables, and we can flag those variables to determine if those might create a degree of bias in the outcome of the model,” Townsend said.
Appraisers then go over the results for accuracy. The goal is for the model to learn accuracy overtime.
In general, Pennington-Cross said companies that use machine learning are much better at picking up outliers, such as a dilapidated home among a block of remodeled and updated homes. “They’ve got a good technology which should be a little more flexible in how it views the world than traditional methods,” he explained.
But Pennington-Cross said assessments are rough by nature and the industry is still likely to struggle between a home’s actual value and value based on other factors, such as comparable sales.
“My guess is this is a step, but not a massive step,” he said.
How likely is it the SAS system could be used here?
The company has been using their machine learning assessment tool in a few North Carolina counties, such as Mecklenburg and Wake County. Robinson said the counties are running it alongside their traditional system.
Robinson noted the technology only takes roughly three months to get off the ground.
However, experts such as Pennington-Cross say it’s an investment most municipalities will have to think long and hard about before adopting it.
“And if I’m a town, I don’t know if I can afford to do this,” he said.
Need more help with lead questions? The Milwaukee Resource Guide is here to help. Have something you want answered? Submit a question.
Talis Shelbourne is an investigative solutions reporter covering the issues of affordable housing, environment and equity issues. Have a tip? You can reach Talis at (414) 403-6651 or tshelbourn@jrn.com. Follow her on Twitter at @talisseer and message her on Facebook at @talisseer.
This article originally appeared on Milwaukee Journal Sentinel: What a better model could mean for property assessments
