You are here

August 22, 2022

Are Appraisals and Assessments Biased?

Watch

Ruchi Singh
Ruchi Singh | Assistant Professor, University of Georgia
Norm Miller
Norm Miller | Hahn Chair & Professor of Real Estate Finance, University of San Diego

Racial bias in home appraisals and assessments is not just an anecdote. Norm Miller (University of San Diego and Homer Hoyt Institute), Ruchi Singh (University of Georgia), and Richard K. Green (USC Lusk Center for Real Estate) discuss the statistically significant racial and ethnic biases in appraisals and tax assessments. 

Miller details the validity of recent analysis on valuation gaps from Freddie Mac as well as the benefits of automated valuation models. He also cautions that using machine learning without human oversight of variables can result in a different set of biases. 

Singh shows how assessments are regressive, often resulting in a mismatch of a lower property value with higher property taxes. She also points out contributing factors, including why excluding information like nearby schools or the condition of the home can set the assessments in opposition to appraisals. 

More from the discussion:

  • How to make the assessment process fairer
  • The importance of loan-to-value ratios in underwriting
  • Pressure appraisers face in avoiding errors
  • Why short-term and long-term appraisal models will be required to avoid bias

Relevant links:

 

Listen via podcast

Listen

- My name is Richard Green and I am Director of the USC Lusk Center for Real Estate and I'd like to welcome you to the first Lusk Perspectives of the fall semester of 2022. This is a series we have been doing since we went underground in COVID times. Our first Lusk Perspectives was in March of 2020 and it was on the issue of how testing could be used as a way to manage COVID as we thought about returning, back in March 2020, we were thinking we would be done by I don't know, May 2020. We could do things like return to the office, return to shopping centers and so on. Since we've done more than 60 of these podcast webinars, and this is the first one. And as you can see, we have a fairly small but very high quality audience this morning and I think the reason for that is this is not an issue that is on the radar screen of most people which is the issue of, is there bias in residential real estate appraisal? Now, there was a story in the New York Times and I'm gonna share my screen for a moment. There was a story just last Friday about a couple in Maryland, Nathan Connolly and his wife Shani Mott who wanted to take advantage of the low, low mortgage interest rates that were available about a year ago. And in order to get a house refinanced, of course you need to get an appraisal, so that the lender knows there's enough equity behind the mortgage for it to be a safe mortgage. The appraiser came to look at the house and this couple had pictures of their kids, pictures of their kids' drawings, books relating to the stuff that Professor Connolly did in his own work which has been about redlining. And the appraisal came in as you see at $472,000. So Professor Connolly being the good social scientist that he is removed all traces of the home being owned by a Black family and replaced it with evidence that the home was owned by a white family. He had a white friend meet the appraiser at the door and when the appraiser was done he said the house was worth $750,000. Now those of us who care a lot about data will say, "Well this is an anecdote, it's an N of one." Unfortunately it actually is an anecdote that is supported by a lot of data that has been developed over the course of the last 25 years. And in 1998, I published a paper with Michael LaCour-Little in the Journal of Real Estate Finance and Economics where we showed that African Americans were more likely to be, what I'll call, low-balled on appraisers than white people were. And by that I mean if you're a white person and you offer a certain amount for a house, let's leave California for a moment and say $300,000 which is typical for the United States. The appraisal is gonna come in at 300,000 or more. So the appraisal is not going to screw up the loan. Whereas if you're a Black person, with some frequency, the appraisal comes in at less than the value of the house. Now there has been recently a, and that matters because basically, it doesn't matter what your appraised value is so long as it's at or above what you offered to pay for a house. If that happens, the loan is going to go through apart from the other underwriting issues but from a collateral standpoint, the loan is going to go through. But if it comes in at less than the offer price, what that means is that the potential purchaser of the house either will not get a loan at all or will have to pony up more equity, a bigger down payment in order to get the house. So it's really consequential, whether, it really is an on or an off switch whether the appraisal comes in at or above the offer price or below the offer price. The other thing is there's recently a study by Freddie Mac and we were gonna have Vivian Lee, one of the co-authors of that study participate today but for reasons that are not in any way her fault, she was not able to join us today. Where they look at about 12 million appraisals and what they find is that houses located in Census tracts with high levels of minorities are more likely to get low appraisals than other houses. With lots of controls in place, very thorough job that they do and again, Chase has put a link to that study in the chat. So at the same time, what do we know? We know the home ownership gap between African Americans and whites in the United States is on the order of 25 percentage points, more or less and there are a number of reasons for that. But clearly in all kinds of ways, access to credit has been one of the reasons for that. And if your appraisal outcomes are less favorable, it's less likely you're going to get access for credit. So it's a channel by which we could see this access to credit mechanism play itself out. So that's the one issue I want to bring up. The other issue is property taxes. So the property taxes you pay is determined based on an assessment. An assessment is basically an appraisal that your local unit of government does in order to determine how much property taxes you pay. And ironically, while there's evidence that when it comes to homebuying and refinancing, Black folks get lower appraisals than white folks do, when it comes to their property taxes, in part for mechanical reasons that one of our terrific guests is gonna talk about today, they get higher assessments which means they pay more in property taxes relative to the value of their house than what white people do. So there's a way in which things are sort of coming and going here. And again, just to explain quickly the difference. An appraisal is the thing that determines the amount of a loan that you get. An assessment is the thing that determines how much in property taxes you pay. And so with that introduction, let me briefly introduce my friends Norman Miller and Ruchi Singh. Norman is a person who needs no introduction in the field of real estate academics. If you took a class here at USC within the last 20 years in real estate, maybe it's 25 years Norm. I can't keep track of these things. But you almost certainly used the book by Norman Miller and David Geltner. Both, there's an undergraduate version, there's a graduate version. We use both books in our curricula and Norman has been very well known for his contributions. Particularly on the commercial real estate side of academia. For many, many years, he taught at the University of Cincinnati for many years and I remember meeting him at my first academic meeting as an assistant professor when Norm was already a legendary full professor. So the fact that he looks younger than I am should not in any way fool you. He has far more wisdom and experience than I. And then Ruchi Singh who is closer to the beginning of her career but still on a, has had a really remarkable trajectory over the course of her career. Teaches at the University of Georgia. Very well known applied econometrician in the field of real estate economics who is focused on assessment. So we have people who really know what they're talking about in discussing this comment. So I asked each of them to talk for about five to seven minutes and then we're going to get into a discussion with me and then we will open it up to the audience for questions. So Norman, as I said, I wanted to start with appraisal and then move on to assessment so please take it away.

- Thank you Richard and by the way, I did start my career at the same place that Ruchi is at. The University of Georgia. It was a good place to be for the beginning of my career. So Richard really set the stage very well here on a number of issues. The New York Times story that Richard just mentioned is important and the question is really, "How common are stories like that?" And so Richard also mentioned the Freddie Mac study. I think that one is a valid study, worth going into more detail. So I'd like to mention that. That was, as you said, 12 million appraisals that were evaluated and 7.4% of the time which is not much, white purchased homes were appraised at less than the purchase price. And for Black families, that was 12.5% below and for Hispanic, 15.5% below. That to me probably gives us some indication as to the extent of bias out there among appraisers and of course you could say that the majority are not biased. But this is statistically significant in systemic ways. There are other studies by the way that I don't think are as valid for a number of reasons which we can go into later on. I would like to mention too that AVMs are being used a little bit more than they used to. Partly because of this problem Richard mentioned of hitting the mark. That is, there's this incentive to hit the purchase price or the threshold for refinancing or something above it and when you end up having 90 to 95% of all the appraisals done in the traditional way, at or above the purchase price, well you're not stopping. You're not really mitigating the risk of overpaying very often. And for that reason, we've moved in the direction of AVMs. And now AVMs are also being questioned as are they biased? And for full disclosure, I've looked at a lot of AVMs. I've worked with collateral analytics or I have for many years with lots and lots of testing. And it is possible to show bias if you want to. As an example, if you did a regression and you put in Asian, Hispanic, Blacks, whites as dummy variables, you would, you would find significant loading onto each of those variables. Because Asians tend to buy homes that are much, much more expensive than do whites or Blacks or Hispanics. And the Black families tend to buy lower price homes. So you get this loading effect. So this is not an easy thing to investigate the bias which again is why I like the Freddie Mac study. So I would like to mention that all appraisals and all AVMs have error and that is another issue. With AVMs we report the error with the human appraisals at just a point estimate. And that's a problem. Because if you're agnostic about what kind of value you hit and your error is 50/50 on each side of the benchmark purchase prices, you're going to have a number of valuations that come in below the purchase price and in particular for AVMs that means that, certainty, more uncertainty. So when we go into property markets where there's more heterogeneity or more price dispersion which you do find with lower priced homes because they tend to be older and of mixed quality within the same neighborhood. So you tend to get more uncertainty. That in itself will actually work against Black households who tend to buy in lower priced homes. Because of that uncertainty and that greater probability that you're going to get a lower number than the purchase price and remember, lenders have to use the lower of the purchase price or the appraised value. So that's one of the issues that we have with AVMs. Let me mention just about one other thing. How we test it is another issue but the common assumption is to use Census blocks and when you see that the dominant ethnic group is 50% or more of one racial group, we assume that the borrower living in that same geography is of the same race. And that's typically how we test. So well mixed neighborhoods are left out of these studies. There are other methods to study the borrowers and sometimes you can identify them more explicitly but that's the typical way these studies have been done. So I think rather than take up more time, I'd love to go into more depth on how the AVMs work and the appraisals work. But I think I should stop there so we can have more of a discussion, so thank you.

- We'll circle back to that Norm. Thanks for the setup Ruchi, tell us about what you've been finding on assessment.

- So thank you for inviting me to contribute to the discussion on this important topic today. My comments are going to be based on some work that I have been doing on property taxes. This is joint work with Daniel McMillen who is at University of Illinois at Chicago. So assessments are regressive and this is a stylized fact that has been established in the real estate literate. Assessments are basically estimates of market value that serve as the base for property taxes. So the key role of these property tax assessors is to come up with an estimate of the most probable selling price of the house which we call the market value. Now these are regressive. What do I mean by regressive? So if you look at the graph here, assessment ratio which is nothing but assessed values divided by sale prices are higher for the low priced properties as compared to the high priced properties. If there was no inequality, you should have expected it to be around 10% because that's the statutory assessment ratio in Cook County and this is plotting actual data from Cook County for the year 2016. So all I'm seeing here is low priced properties are assessed at a higher rate than the high priced properties. Okay, we know assessments are regressive. So what? What we really care about is property taxes. Before I show you what happens in property taxes, let me show you theoretically what we should expect. So progressive tax is a tax which has a higher average tax rate for high income households. In terms of when it comes to property taxes, instead of measuring it as a function of income, we measure it as a function of house prices. Just for convenience because it's easier to observe house properties and property taxes at the same time. So tax, which is the property tax that we pay is a function of the tax rate multiplied by the assessed values minus exemption. The most common exemption is homestead exemption. In Cook County it's something like $10,000. In some places in Boston it's as high as 230,000. What it implies is that for the first 230,000 of your property value, you will not be paying any taxes. Some counties have an assessed value, they charge taxes on a fraction of assessed value. So for example, Georgia where I am, the assessment ratio is 40%. Cook County which I'm using as an example has an assessment ratio of 10%. So all that I'm saying here is you have this assessment ratio which is assessed values divided by price and therefore the tax would be a function of tax rate multiplied by rP minus exemption. If you look at effective tax rate, that's effective tax rate is tax divided by price. So that will be tr minus exemption price. As I mentioned, influence of exemption would eventually become negligible for high price properties and therefore what we should theoretically end up with is a progressive tax structure. T is constant by statute and assessment ratio should not vary by price. So as you can see from the graph, this is theoretically what we are expecting. As the house prices increases, the effective tax rate increases. Let me show you what actually happens in practice and this again is looking at data from Cook County for the year 2016. Effective tax rate which again is property taxes divided by price is higher for the low priced properties and is lower for the high priced properties. Now these low priced properties as we would expect is owned by the low income households which are more likely to be African Americans and minorities. So they are effectively paying a higher tax rate as compared to rich individuals. So the next obvious question is, why exactly is this happening? The first key reason because of which this happens is a mechanical relationship. Because in most counties, Hedonic regressions are used to assess residential property prices and when we are doing regressions, we are doing a regression to the mean and therefore that has an inverse regressivity in it. Some of the other reasons that have been discussed in the literature are similar to some of the points that Norm was raising earlier. When we think about house prices, they vary a lot by location. Characteristics of that particular neighborhood, they are very important sure. And as you would expect, high income individuals and low income individuals live in different kinds of neighborhoods. When assessed values are calculated, the neighborhood characteristics are not completely taken into account. And what that results in is over-assessment in neighborhoods which have negative externalities or negative neighborhood amenities and under assessment in neighborhoods which have positive neighborhood amenities. Moreover, what happens is these assessors don't have information on the quality or the condition of the house. That could lead to some regressive data. The second reason is appraise. So the property tax appraise are more likely to be done by individuals who are, who have these high priced properties, who are these richer individuals and there's also evidence which shows that conditional on a property tax appeal happening, the success rate is also higher for these high priced properties and that could again lead to regressivity. The other reason is in some counties, you have infrequent appraisal. So instead of the appraisal happening every year, it happens every two years or every five years and that could contribute to some regressivity. Some places have assessment caps which basically says that the assessment values can increase only by a certain percentage in a year. That could again be a problem. Lastly it is important to remember that property tax appraisers or property tax assessors are required to estimate the market value for each and every property. Irrespective of whether the transaction has taken place or not. And this becomes a very difficult problem if you have thin markets. I'll stop here and I look forward to your questions and comments, thank you.

- Thank you Ruchi. So let's circle back to Norm and you sort of set up the question that I wanted to ask you anyways. Could you, I don't think most of the audience, although this again is an unusually sophisticated audience really understands how AVMs work. So if you could take a few minutes to tell us how they are estimated and how they are applied and then the broader question is, given AVMs, do we even need appraisers anymore? And I'll come back to the issues with AVMs. I'm not suggesting they're a panacea. But are they better than appraisers and if you think they are better, do we need appraisers anymore at all?

- That's a lot. In brief, AVMs are automated valuation models. They use a variety of methods. Regression of course is one common technique but as an example, collateral analytics has 11 different AVM models. And over the last.

- And Norm and Ruchi, you both talked about regressions. But not everyone in the audience is a statistician so if you could just briefly.

- Yeah.

- Tell the group what a regression is.

- So we feed in the statistical relationship. The correlations between all the variables that we think affect value. Property characteristics such as living area and size, bedrooms, bathrooms, the size of a lot. The age is an indication of condition and we might have as many as 99 variables coming from property tax records that could be fed in. Now of course at some point they don't really explain much in the way of value. But we look at the average variation of these input variables. These independent variables and the variation in the selling price of the property which is our dependent variable. Now sometimes the relationships are non-linear. So that if you keep increasing lot size, at some point it has a marginally lower impact. Well the better statisticians who understand valuation know how to grab non-linearity. And then it gets into a big competition among bidders. Road noise, school quality, views. All of that can be modeled today because the data is so much better today than it used to be. So we've gotten to the point that residential appraisal AVM error is quite similar to human appraisal error. The common belief is if that we have very unique property, a human should do it. Because the sample would be too thin. But for your typical property, the reason why AVMs are favored by many lenders is simply because they're agnostic. They don't care about advising the value. Well on the other hand, we know from past studies that human appraisals are biased because well, the lender won't carry the appraiser back that keeps resulting in an appraisal that's too low for them to make the loan and sell it off to Freddie Mac or Fannie May. So they just don't get hired back. So there's a systematic reason why human appraisers tend to be biased. And that's a problem. So we are moving today, most HELOC loans, the home equity line of credits are done with AVMs and with an inspection many lenders will rely on AVMs today. Again, one of the problems with the AVMs is they are symmetric in their error on both sides. So they tend to be high and low symmetrically. 50%, 50% when they are developed and designed and tested and because of that, you're going to have 50% of the time that your appraisal results could be below the purchase price. Well with a human appraiser it's going to be 7% below for white people. 15% for Hispanics, 12% or so for Black families below. But there's a big bias on the right side. So I happen to believe in doing a lot of testing between AVMs and human appraisers, that for a fairly typical property, we really would be better off going with AVMs and this doesn't cut out appraisers. They can be interactive AVMs where an expert appraiser operates them but by doing an AVM we use so much more information than we do with a traditional appraisal. So I do think that an interactive AVM or an AVM by itself tends to be more objective but we need to change the lending policies so that we have a range of value that's acceptable instead of a pinpoint value. And that's one of the problems. We have to admit that there's error in appraisal. Both on the human side and the AVM side and I think I better stop there or I'll get too carried away.

- So I want to come back to this point. But I do want to bring in Ruchi at this point. So you sort of touched briefly on some of the solutions to this problem with assessment. But could you expand on that a little more? Are there ways that we could, what could we do to make assessment fairer?

- I think doing, the first simple thing that could be done is instead of doing it at very broad geographic level, doing it at a smaller geographic level. So we basically control for the location better and it is very important to control for the neighborhood characteristics. So one could do regressions at smaller geographic level which would reduce some of the errors at the two extreme pairs that we are observing. The other could be we as econometricians can't do it. But could be done is you could probably have a separate regression for the low-price properties and a separate regression for the high-price properties. Those would result in somewhat of a biased estimate and therefore as econometricians we don't, the traditional econometric textbook suggests not doing that but if the purpose is trying to come up with a good estimate, doing it separately for the different strata of properties would reduce the error to some extent.

- Excuse me, Richard. Richie on your graph, I saw .03 as being sort of the typical effect of assessment on the lower price homes. And .02 for the average. And just to be clear to the audience that might see this, that's 50% higher taxation right?

- Yep.

- Yeah.

- So that's very significant. It's not like two or three percent or five percent but 50% higher taxation. So I was surprised the results were that high.

- Yeah, it's looking at those two tails. There are fewer properties in the tails, but when we look at the tails, especially if you compare the high price property to the very low priced property. The difference could be huge.

- So Ruchi and I'm gonna get really in the weeds now but do you have any sense what kind of r-squared the models have on the assess? Are they just inferior models? Because one thing we know is while better models don't fix the problem, they do ameliorate the problem. And so maybe Norm's AVM would do a better job of assessing these properties than what is typically used by municipalities.

- Yeah, so we've spoken to some of the assessors and the assessors are aware of the problem in the regressivity. The r-squared on these models seem to be high. I don't know what the exact numbers would be but would be like around an 80% in some cases. But they're not putting in the very fine fixed effects as we put in our model. So the moment you start doing very fine, let's say a Census tract fixed effects. Your r-squared would be even higher and better. Now one of the reasons.

- And to the audience. Let me apologize for even bringing up r-squared in this conversation. But the idea is the more accurate the model is, the less kind of bias that Ruchi is talking about will exist. So making the model, and again as Norm says correctly, all models have errors. But every reduction in that error helps ameliorate this problem.

- Yeah, the only issue with that again is something that Norm was referring to and in some cases you really have thin markets. So if you go down to a Census tract where there are very few transactions, we don't have enough number of transactions to pick up the correct coefficients and that could be an issue. So there needs to be a balance between how small a geographic area we want to go to and whether we have enough number of transactions.

- Shows that the houses that African Americans buy tend to appreciate less than the houses that white people buy. And so so much evaluation is anchored on the initial price. That people, and that's fair right? If you know that somebody has spent, again I'll talk more nationally than California. 300,000 on a house yesterday, it is fair to value that house at 300,000. But you go two years out, and what is the model gonna do in essence with some controls? It's gonna make the overall markets increase five percent. It's gonna add five percent to all of those houses but those owned by African Americans may have only increased by three percent whereas white people it's seven percent. And so you get that, that the African American has they're overvalued for assessment purposes and vice versa for the houses bought by white people.

- And that problem really increases when you're not doing frequent appraisal. So some places do appraisals every three years or five years. So even if we did assUme they did a very good job in the first year of doing the appraisals and as you said, if the high priced properties appreciated by much more, then you know, you would see that, their taxes have not increased as much as a function of their price. So that again, infrequent appraisals would add to that kind of a regressivity.

- And then given that 30% or so of the property in the country by value is in California, Prop 13 and these caps, you have much more benefit to the very high priced, high appreciating markets than they do to the lower price. So that would also result in a, a property tax burden disparity.

- So just a little shameless advertising. We did a Prop 13 list perspective. This is something like a year or so ago. And yeah, I mean a lot of us like property taxes in principle because they are what we call an efficient tax. Since you can't move land, when you tax it, it doesn't have the impact on output that say taxing wages does or taxing capital does. So there's a guy named Henry George who would in fact argue you should only tax land and you should not tax buildings. But there's this issue, I've been sitting in my house for 14 years now and if I accept that okay, inflation, it should go up by the value of inflation and when I make improvements, yeah I should get that money back. It's gone up way, way more than those two things and I have done nothing productive in order for that to happen and that's the tendency for those of us who like property taxes is you want to claw back at least a little of that. Because that's society that's creating that value for you. That's not you that's creating that value for you. But the problem with the property taxes, you have these regressivity issues just in general. Right, because low income people spend more of their money on housing than high income people do. But this is, and to some extent that's why homestead exemptions exist. To mitigate that. But Ruchi and Dan's work shows that you have this enormous problem of how its value has an impact on what people pay. I'm trying to remember the very wise person who once said it's expensive to be poor. Because you just, there's stuff. Stuff just costs more for you to live life than other people. Now Ruchi on the, you made a really interesting point about appeals as well.

- Yeah.

- And so I'm guessing Norm, Ruchi, I will say this about myself. If I think my property tax evaluation is too high which is actually impossible for me in California because I have a Prop 13 valuation. But I'm gonna challenge it right? And not only challenge it, I'm gonna take a regression model and show where they're wrong.

- Yes.

- You know, do we have any sense and Norm, I'll open this to you. Who actually does do that? Do we know anything about the characteristics of people who'd challenge their property tax assessments?

- I haven't seen an empirical study but I think you're absolutely right. That the more educated professional will realize that it's possible to appeal. And I personally have appealed once in Ohio and I knew the evidence to put together and then I appealed once in Washington State and I knew the evidence to put together and won in both cases. And I can't imagine that process would be very easy for somebody that didn't have the awareness of the tools, the data, the professionals to hire. It would be very challenging. So I think you're right, but no I haven't seen a systematic study about it.

- Yeah there are some studies which suggest that you know, these individuals who own these high priced properties, just because they have more income and going back to what Norm is saying, they have the kind of resources to you know, hire a lawyer and even if they can't do it themselves or don't want to spend the time doing it themselves would make appeals. So the proportion of people who are making these appeals are more likely to be high income individuals and they're more likely to be successful when they make an appeal. And in fact there's one study which shows that conditional on being successful, the kind of re-evaluational discount that they get or the reduction in appraisal value is higher for these people. In fact I had a friend who actually, as Richard you were suggesting did this last year in Georgia and he had a regression-based model. He took it and he was really successful. He did not get exactly the price that he wanted but very close to that value.

- Yeah.

- So let me, so I should've said. So anyone in the audience who wants to pitch in, please just put something in the Q and A and we'll be happy to pay attention to you. I should've mentioned this earlier. But Norm, you said something really interesting in our pre-call about the problem with machine learning as a mechanism for using, for developing AVMs. Can you talk a little bit about that? And again, explain a little bit about what the difference is between machine learning.

- Yes.

- And other types of valuation methods.

- So let me begin. By way of background, the AVM vendors out there. There used to be 50 of them. Today there's no more than a dozen that cover the whole country and they're all much, much better than they used to be. They're all good and they all have multiple models. What we don't want to do is use a cascade model which means go to some vendor that says, "I'll use ABC's model and DHC's model "and so on and so on and put them all together." Because you don't know what's in there. You don't know what the inputs are and the variables are. And the same thing for machine learning. Machine learning is when we toss in a while lot of variables and we use it to explain something like selling price but we don't really have anything that we can interpret in terms of how it explains selling price because the variables could be interactive and non-linear and connected in strange ways. And those inputs can include many variables that are correlated with race and ethnicity. And that's the problem. Is we're not just explicitly putting in a variable, this is a Black household. If you put in variables on income and educational level and other socioeconomic variables, you're going to get correlations with race and you may in fact be using a racially-biased model without knowing it. So we really don't want to see machine learning models used, period. I don't personally want to see cascades used. I want to see models where I know all the inputs and I make sure that we work hard to make sure that we're choosing variables that are not correlated with ethnic and racial groups and then we have a better chance of doing a better job on valuing properties without bias. We're still going to have that problem of more error on Black household and Hispanic properties that are in lower-priced areas. And we're still gonna have that. So unless we change the lending policies, that's not really correctable statistically.

- And it's, there's so much in lending policy that needs to be changed. And let me just, indulge me mention two things. And one has nothing to do with appraisal but the use of debt to income ratios is really problematic from a couple of perspectives. First of all, they do a pretty bad job of predicting default. Which is what you care. If you're a lender, you legitimately care about, do people pay their loans back? And debt to income ratios have very poor track record of doing that unlike, we'll talk about equity. Equity in the house matters a lot, there's no question about that. And actually FICO, for all their problems with FICO actually does a pretty good job at predicting repayment as well. But the other thing is beyond the fact that it doesn't do a good job of predicting outcomes is lenders tend to discount contractor income. And without, I've seen no strong evidence that contractor income is less reliable than salaried income. And so they'll look at a borrower and they will say, "Okay you made $80,000 last year "but we're gonna count only 60 "because it was contractor income." Whereas if you have a salary, it's 80,000. So that takes people from being eligible for a loan to ineligible for a loan, even though they have the income and contractor jobs are disproportionately held by people of color. So that's an example of where, because lenders have just done things a certain way for a certain number of years, they continue to do it. On its face, it appears to be race-blind but it actually has an influence without improving the quality of the book of loans that they have. The other thing Norm, this is on the whole issue of appraisal. So if you use AVMs, you're gonna have 50% of people are gonna get low-balled. That's by construction, that's gotta be true. But I wonder if, and again, apologies for mentioning this. I've done some work on this. If you just look at how much money have people put down, is loan to value ratio really that relevant? So is 81 really different from 79? Now what does matter is people do put money into the deal. We know that money, that matters. But must it be the case that we care so much about these cliffs that we have basically created that determine whether you can get loans at the most favorable terms or not?

- Well the LT, and you mentioned before in terms of predicting default. The LTV matters a lot. Now go back to the loans made 2005 to 2007, a lot of those appraisals justified somebody paying, let's use easy numbers. $125,000 for a house in a modest, second tier city or third tier city and they got $100,000 loan and they were told the an 80% loan to value. In fact that home was only worth a little bit over $100,000 and instead of having 25,000 in equity, they only had a few thousand dollars in equity and they overpaid by $20,000 and nobody stopped it. Nobody mitigated it. That is a huge problem and that is a problem that we have of temptation to hit the mark. The AVM as you mentioned is going to have error on both sides. But 80% of the time, it's going to be within 10% which is going to be closer than the error that we just had when somebody justified 125,000 for a home that was worth more like 100. And so we need to have a range of value on the valuation that is acceptable. But we have to get rid of that bias of hitting the mark because we know in fact that during three or four years, when lending and underwriting standards decreased, that the amount of equity people had in their home was far less than they thought. And that does dramatically affect default.

- But it is interesting Norm that, I mean Laurie Goodman has shown that the most important predictor of default was the product. So the default rate on 30-year fixed rate mortgages, even for people who were upside down was remarkably low. I'll never forget talking to a guy at Fannie in 2009. And he said, "Our Las Vegas book is a disaster. "We have a 10% default rate or a 20% default rate." And I said, "It's a miracle that it's only 10 or 20% "'cause everybody is upside down in Las Vegas right now." Prices there fell 70% from peak to trough. So I'm just thinking of the interaction of these things really does matter. And sort of just having again a cliff, where if the model says you're above 80, you don't get the loan. I think that's problematic.

- It is problematic and also keep in mind that these markets get out of whack. We hit cascades of disequilibrium. So somebody overpays for a house, pays 125,000 for a teeny little house. And that becomes a comp because it's, the loan is made. It becomes a comp for the next house in the neighborhood and the next house in the neighborhood and so we get this cascading of having values that are not fundamentally sustainable in the long run. That is a problem. So having a little bit of error in the appraisal process is very problematic because it can cascade and compound. And make the whole neighborhood. We tended to see neighborhoods where there was a lot of predatory lending. Well because the first one justified the second one which justified the third one. And hopefully by using, even by using AVMs, you still have that problem a little bit. But not quite as much when you start.

- Well I was thinking, based on Ruchi's work is maybe what we need is, we need to have appraisers do assessment and assessors do appraisal.

- Well.

- That was a joke, I was just.

- Yeah, yeah.

- Yeah, yeah. So let me just turn back to the audience. We have about, I'm trying to find the time here. Hang on. We have about 12 minutes left. Let me just again ask the audience, is there any burning question out there that people. I mean these people know this stuff better than anybody. And I know it's a little bit, it's a little bit technical what we've been talking about this morning.

- And we did have a participant join us that I wouldn't mind if she would comment. If she's willing to.

- Yeah I think, I don't want to put her on the spot.

- Like I just did?

- Yes. But since you did it, I will wait for a second and then move on. So read Vivian's study though. It really is very, very well done. And I think Chase put it in the, in the text. So, so yeah go ahead Norm.

- If you wanted to hear some other points keep in mind that one of the problems with appraisal and I bet this is true for assessment as well. Most of the appraisers out there, particularly in residential, not commercial. 97% of them are white. 70%.

- That's the question I was gonna ask you.

- Yeah 70% are male and they're mostly older people. There's very few young people and in fact in the university classroom I would tell our students, if you're interested in evaluation go into commercial. Don't go into residential and one of the reasons is this pressure on the industry to hit the mark. And so we find that we don't really attract many professionals into appraisal right now in residential and so I think we have to rely on people that are good valuation experts but also statisticians who can come up with more accurate models and rely more on that. And more equitable models. I'm sure that on the property tax assessment, if we notice the effective tax rate is 50% higher for some neighborhoods or groups, there has to be a way of adjusting that and seeking equity. So we need, we need people that are statistically trained but also understand valuation involved in this process and I do believe that residential appraisal, as has been practiced the last 50 years is obsolete. And hoping the lending policies will adjust for that.

- Richard, do you have any thoughts about that?

- And now I'm going to get hate mail from anybody that watches this. You can handle it.

- Yeah.

- No, definitely. There is definitely scope for improvement and you know, using some of these more hybrid and complicated methods and econometrics method. The other solution could again, would be like if you have the higher homestead exemptions. And this is something that I'm interested in studying. Is places which have a higher homestead exemptions, probably the property tax is not as regressive in places which have a lower homestead exemption. Because if you have a very high amount, that would undo some of the problems that we have in assessments.

- Ironically Louisiana has one of the highest homestead exemptions in the country when you compare it to the value of the houses and one does not think of Louisiana as a particularly bastion of progressive thought but on the other hand, homeownership is something that is considered sacred there and so I'm pretty sure that's why the policy was put in place. Now again, at the time it was put in place, I'm also guessing there were maybe four Black homeowners in the entire, no. That's probably not true. But you get the idea.

- Yeah.

- It was probably hard to be a Black person and buy a home in Louisiana at the time that policy was put in place. But it is, sort of like what we're learning is raising the minimum wage is helpful to people of color getting paid better. Disproportionately so. And so it seems like again, a race-neutral policy but it actually has pretty profound implications across races. And so I think that is a good thought. But the other thing I get at is you know, awareness of disparity is key and you know, on the appraisal side as Norm said. 97% white. I think about, we had Bill Springs on a Lusk Perspectives a year or two ago and he talked about learning econometrics when he was a PhD student. At the University of Wisconsin which is where I went. And how any time there were race differences people would say, "Oh there must be some unmeasured variables "that explain the difference. "It can't be race per se that matters "'cause that's inefficient." Which shows that if you're, if your null hypothesis is that there's not a problem, it takes an awful lot of evidence. Maybe an infinite amount of evidence before you're going to admit that it's a problem. And so both on the appraisal side and on the analyst side. The sort of people Norm is talking about. The statistician side. It's really important to have more diverse participation than we've traditionally had in either of those areas. I think in order to move the needle on this. Clearly Ruchi you're helping on the diversity side if I may say so. But we need a lot more of it than we see. Okay I am getting some audience questions.

- Yeah just to add on to that, New York is trying to do some of these spatial regressions and trying to improve their assessment practices so at least the property tax assessors are aware and they're trying to move in the right direction. Which is always encouraging to see, yeah.

- So let me, so we have a few minutes left and we do have some audience questions. So I want to get to them. Georgetta Banks asks, says that in L.A. she's had two properties that were given low appraisals. There was no one I could go to for a reevaluation. So I guess that's more a comment. Georgetta, do you have a question that or are you just chipping in with your experiences? Guess I'm, from USC a PhD, Gene Berinski. What hasn't been touched on is how to break the cycle that Norman mentioned. Namely when purchase prices reflect discrimination, their subsequent use, this constant assessor and evaluation where thereby perpetuating the issue. So yeah this is key. Is data seems neutral but it actually has biases built into it. So either one of you, do you have ideas about how we break that cycle?

- This is a tough question. I'll start. The typical dependent variable when we're trying to use statistical methods is selling price. But what happens if everybody in the neighborhood before you overpaid and it was justified and they got the loan because the appraiser hit the mark? So how do you break that cycle? So my belief is that we need longer term equilibrium value models as well as short term price prediction models. So I think the best models are the ones that help predict default. And so I would love to use valuation models that predict default along with models that just hit the benchmark selling price. But I know that in the appraisal industry, I got into a nasty debate with one of the leading vendors about this questions. And the point was that in my book, just hitting selling price 99% of the time was a useless model. But that's, but you have to hit it much less than that in order to break the cycle. So we need a model that is not gonna be too reliant on a short term price indicator when somebody makes a mistake. So it has to be a little bit more longer term, fundamentally driven as a valuation model. That's tough, that's tough and I don't know if that would be accepted and right now the acceptance is just to hit selling price is your benchmark but if everybody did that, they'd be totally useless. So Ruchi?

- Yeah no that's a tough question and yeah, I don't really know how to break that cycle, so.

- Well it's, the fundamental problem with appraisal is you're basically always using information that's stale. Right, even. I mean think about the market today relative to three months ago, the housing market. It's a completely different market.

- It's not, it's true for 90% of the models out there. However I do know models that look at recent price concessions and list price seeling price ratios and the trend and they incorporate that into the model and they do that to try to use the current market dynamics and not just the historical information on selling prices. So some of the better models do take that into account and in fact if listing prices in your neighborhood are dropping down below the old appraised values, that should be your asymptote for how high you appraise a property. So we do need to allow the models to use market dynamic information and again, I think the better experts would figure out ways of doing this.

- So Vivian does have a comment which is, speaking of appraiser diversity. Appraisal Institute has an appraiser diversity initiative which commits to building a more diverse next generation of appraisers and Freddie Mac is a core partner in the initiative. And Chase just put a link to that. Andrea will, "Appraisers already take courses in statistics "and evaluation method. "Have a subjective element to it. "However, I am unsure what we can do to change it." Well one thing Andrea, I will say this and I'm gonna be a little mean here. I see the textbooks that the Appraisal Institute puts out and that the International Association of Assessment Officer puts out and the statistical methods discussed in those textbooks are quite rudimentary and not sufficient to really think in terms of the kind of AVMs that, or develop the kind of AVMs that Norm and the assessment models that Ruchi is thinking about. I think you need people and it's not arguing the point that statistics by themselves don't, aren't. Don't have errors. I mean Norm correctly pointed out that they all do have errors. But one thing on the subjectivity. We're doing things like scraping data on how nice people's lawns looked from Google Maps view. And incorporating that into data. So we're really getting more and more information. And in a sense, much more objective information about the characteristics of houses than we had even two or three years ago. So.

- Yeah we are. We're doing text mining, we're doing image data mining and the great thing is we're using all of the market information. Not just three to five observations.

- So, so with that we have come to the top of the hour. Ruchi Singh, Norman Miller, thank you very much for spending an hour with us today. Thank you for joining us this morning.

 

 

 

 

Related Episodes

Erika Poethig
Erika Poethig
Diane Slemmer
Diane Slemmer
Ben Metcalf
Ben Metcalf
Susan Geddes Brown
Susan Geddes Brown
Samar Jha
Samar Jha
Meredith Stowers
Meredith Stowers