Lenders need to have more skin in the game by holding greater financial liability for their decisions. One of my greatest disappointments of the 2008 financial meltdown was that none of the major lenders were forced to either recapitalized or economically euthanized with the parts sold off to well managed institutions. The opposite happened with the good players penalized by not being Ridley and the bad players rewarded with a bailout.
So, how to correct this? Good question. This is not my area of expertise but here’s a few thoughts. More liability for loans they make. Have management bonuses that don’t vest for 3-5 years with clawback provisions (to avoid what happened in 2005-2007 where bonuses were based for loan volume, not loan profitability). Require the lender be responsible for the first x% of loan losses for any reason when the loan is sold; hire an inept appraiser or shoddy AMC and the lenders eats that x%. In short, better alignment of public risk to lender performance.
However, such claims can be contested in court, which would make them expensive to implement. You need to avoid the court; otherwise, your method will be inefficient and not really possible to implement across the board.
What you need are straightforward methods that are clearly either implemented or not implemented, at the ground level. And if clear protocols are not implemented, it is very evident that it is a black-white issue, and the appraisal is rejected before it gets off the ground.
For example:
1. Requiring adjustments to only be calculated as differences between feature value contributions, where we have the very clear constraint that all value contributions have to add up to the net sale price for each comparable and the estimated value conclusion for the subject property.
2. All subjective value adjustments, as well have to be calculated from the corresponding subjective feature value contributions, where those value contributions have to add up to the residual difference between each comparables net sale price and MARS estimate of sale price or in the case between the subject estimated residual based on a comparison with a ranked ordering of comparable residuals (per the RCA method).
However, it will be acknowledged that to utilize this particular RCA method, the appraiser must be trained in using MARS, which is non-trivial and somewhat comparable in difficulty to learning how to create neural network models (although building neural network models is a much larger subject and requires the modeler to exert contol over the process at a lower level, - and work with usually much larger data sets). But, given that neural networks are needed for ranking properties based on photos, we might as well require that skill from appraisers. So, that, all of a sudden, appraisers will need to have some understanding of:
1. Matrix Algebra
2. Minimal calculus and partial differential equations.
3. Programming in R, Python, and perhaps a little C/C++.
4. Non-parametric statistics.
5. Mulivariate Adaptive Regression Splines (MARS) - which is of CENTRAL importance in appraisal.
6. Neural Networks (How to train neural networks to discern property appeal based on photographs for specific market areas).
And the big problem (1) is automating the workflow, and (2) getting the market to pay people to do this kind of work, which has to compete with high AI salaries.
It's difficult to predict how this will ultimately turn out. I suppose that with time, it will happen, because appraisal is not only an appraisal, but also a control mechanism to stabilize the economy, to prevent systemic overvaluation that leads to economic crashes every 20 years or so.
