Twenty Seven (27) years ago “A Matched Pairs Analysis Program in Compliance with FHLBB Memorandum R41B/C” article was published in the January 1987 edition of the Appraisal Journal.
The article topic: Regression modeling applicable to the direct sales comparison approach for matched pair sale analysis to comply with the FHLBB memorandum regarding communicating a self-contained analysis in an appraisal report.
In 1992-94 I sat on an ad hoc committee “National Property Data Services” (NPDS) that included: FNMA, FHLMC, CMDC, several MI companies and the who’s who of the loan origination lenders at the time. Most of those lenders have either been acquired or are out of business due to the last crisis with both GSE’s going under government conservatorship.
NPDS objective: to develop an appraisal property data co-op modeled after the California Market Data Cooperative (CMDC) to share property data that could be used by cooperative participants to improve valuation quality.
For the next 27 years the valuation industry struggled to deliver a platform to meet the NPDS objective.
Back to the future: 2012 the UAD codes kicked in. In 2015 Collateral Underwriter kicks in. Now what, obviously the GSEs have determined the valuation industry isn’t going to get it together.
The appraisal industry has been adopting emerging technology tools as they become cost effective or as they get shoved down our throat; thanks to the internet and big data, what we could not even fathom doing in 2000, let alone in 1987 is now a cell phone app.
Regression modeling is not new to the valuation profession, it is the basis for most assessor mass appraisal applications and Automated Valuation Models (AVM). All have a variation of a regression model embedded in the algorithm. Thanks to innovators like Bradford Technology, SAVVI Analytics, Agility 360 CURE, NCCI REDS, Collateral Analytics… the list is endless; the valuation process has become efficient and more complex at the same time.
Hedonic Step Wise Forward and Backward regression is a matched pair emulation process using statistics, the same matched pair process taught in entry level appraisal courses on the Sale Comparison Approach to Value. In 1987 the FHLBB required appraisal reports to have a self-contained analysis, in 1987 the internet, big data, MLS feeds, GPS Geo Coding, point and click software programs did not exist, therefore compliance was cost prohibitive.
Commercialization of regression models to meet the needs of all users is a complex process considering there are over 850 MLS boards each with proprietary file layout, membership and rules of use.
Regression models work best in the hands of subject matter experts (appraisers) that can interpret and simulate the results through sensitivity analysis to develop credible correlations statistically derived from sales comparison that are then explained and communicated in the report. This is not a commodity type process, it takes time, expertise, education and experience. Customary fees will certainly not be reasonable for the effort involved.
However this is exactly what the GSEs are expecting appraisers to do just as the FHLBB R41B/C memorandum did. Except now it is feasible, the tools exist. Any appraiser can develop a credible regression model using the Excel Data Analysis Tool as taught in the Appraisal Institutes classes “Quantitative Analysis” and “Real Estate Finance Statistics & Valuation Modeling”. Although it is much more efficient to use one of the Commercial Versions embedded in the appraisal forms software (Comp Cruncher), or downloaded to a desk top (Real Stat) or a browser based application (SAVVI). Just make sure you can explain and defend the model output and adjustments for USPAP compliance.
So where does this take us? Appraisers that adopt the emerging tools will find their cheese. Collateral Underwriter will hide the cheese from those that don’t, of course this assumes the GSEs can let go of their legacy underwriting process, because the 10%, 15%, 25% line item, net and gross adjustment rules are going to get blown apart by regression model driven adjustments. Regression modeling draws unbiased correlations between actions in the market between buyer and seller, not biased underwriting risk rule sets.