Do Appraiser and Borrower Race Affect Valuation?


  • Authors: Brent W. Ambrose, James N. Conklin, N. Edward Coulson, Moussa Diop, and Luis A. Lopez
  • Discussant: Florian Oswald

Intro

  • Property appraisers are trained to asses property values.
  • Large body of evidence going back to the 60s of discriminatory practices in mortgage markets - HMDA.
  • Extremely relevant setting: In remortgaging setting, appraiser is the single agent determining the value of property (not market interactions) - Appraiser faces no competition.
  • This Paper: Is there racial bias in remortgaging appraisals?

Approach

  1. Use property level data on remortgage appraisals, where owner’s race is observed.
  2. Infer the appraiser’s race from first and last name.
  3. Compare appraisal to output of an industry model (Automatic Valuation Model, AVM).
  4. Assess and interpret differences.

Contribution

  • Novel dataset and insights in a contentious area of debate.
  • First paper to be able to assess impact of appraiser’s race.
  • Found effect does not disappear with race of appraiser.

Comment 1: The AVM Model

  • It would be good to know more about this black box model.
  • We should know what variables it uses, at which geographic scale.
  • How well does it perform?
  • Outcome $Y_i$ is difference between appraiser and some estimate we don’t know much about. (We know it’s a moving target.)
  • 🤔 Why do lenders send appraisers instead of relying on the model?

Comment 2: Unobserved Quality

$$Y_i = \delta_1 A_i + \delta_2 B_i + \delta_3 H_i + X_i \beta + \xi_i +\lambda_i +\omega_i + \epsilon_i$$

  • $X$: property type, investment properties, multi-unit properties, condominiums, and PUDs
  • Unobserved House Quality is potentially a big issue here.
  • Omitted variable bias could be severe. Notice that the appraiser FE does not deal with this.
  • If avg unobserved quality of group $g$ is relatively high, we will underestimate discrimination, and vice versa.
  • Here are a few candidates:

Potentially Omitted Variables

  1. Age of property. It may be that certain groups can afford only buildings of older vintage, which could imply a lower assessment.
  1. Investments in structure, maintainance and upkeep: Less affluent groups may find it harder to make timely investments (roof, heating, insulation…), which in turn impact the appraisal.
  1. Investments in home improvement: Less affluent groups may find it harder to install winter garden or swimming pools, with the same effect on valuations.

Potential Solutions

  • A building-level fixed effect. Vladimir Avetian can use one in documenting (overt) discrimination in Moscow’s housing market. Probably hard here - unless you have a bunch of properties in a single unit?
  • Get more data on the property to increase $X$? With current data seemingly impossible. What about ZTRAX, which has tons of characteristics (and First and Last name of Buyer and Seller), or something similar?
  • More info on borrowers? Controlling for income (wealth better) would be ideal - education?

Conclusion

  1. Very important topic.
  2. Exciting database with appraisers race marker.
  3. Getting better measures of quality and/or borrower characteristics would help.

Thanks!