Scope of Work is required in legacy appraisal practice. Data Science (and Evidence Based Valuation) are more based on principles of good science.
Editor’s Note: This is Standards, part 3.11 of George Dell’s series on How Do I Move to EBV? Links to the earlier posts are here.
To sum up, the goal of traditional appraisal is to be credible (worthy of belief). The goal of EBV© is measurable reliability.
USPAP Scope requires six points: the client, intended use, value definition, date, subject, and “assignment conditions.” A reliability estimate of the “supported opinion” is not required. The “reconciliation” only seeks to subjectively explain why the three-approach numbers do not match. The USPAP scope is to include: 1) property identification, 2) inspection, 3) data researched, and 4) “type and extent of analysis applied.”
EBV requires one more “scope” point, the seventh. EBV requires the analytic path description, recognizing the role of expert judgment. And also the expected process and algorithms to be applied, and interaction with the user. In addition to the six USPAP elements, EBV requires specifying the actual data analyzed. Data science is said to be 80% on getting the data objectively right. This means specific requirements of the amount and quality of data.
The user-reader is presented with the full exposition of the process, the logic, and visual presentation of the complete market segment, and the position of the subject in that identified market segment.
The process clearly connects the interaction of computer algorithm and human judgment.
- The initial hypothesis of the subject and the problem. This involves abductive reasoning, and the application of the analyst-appraiser’s prior knowledge, experience, and education. At times, the nature of the problem may change because of new discovery during the process.
- The data requires an initial judgment of the overall data frame, which includes all the data which might possibly be needed for the analysis. The action of turning raw data into useful information has two components:
- The size of the data set to actually be analyzed. This comprises directly-and indirectly-competitive sales. Judgement is required for outlier and bi-modal data handling, as well as inclusion decisions, which is called the bias-variance tradeoff (where to stop adding sales);
- The key predictor variable selection (“elements of comparison”). This typically requires overall judgment, as well as the ability to identify special situations – which may not be picked up strictly by data shown (or not shown), or may only be picked up upon the actual property inspection. There are several useful similarity algorithms which are easy to use, once the underlying theory is understood.
- The predictive algorithms (“adjustment support”). There are only three basic predictive algorithms. Simply put, they can be called: 1) Association, typically using simple regression; 2) Contrasting, a refined form of “grouped-pair comparison;” and 3) sequencing, a manner of handling qualitative data by transforming the ordinal data via “non-parametric” methods.
- Delivery is via electronic data-stream. Properly organized, different dashboards are able to be customized to convenience different users (such as reviewers, auditors, regulators, portfolio and risk managers, and executive administrators).
The EBV (Data Science approach) scope of work is similar to the traditional, but is broader and more specific as to function, along with the interaction of expert judgment and computation. EBV recognizes that the intended/expected scope-of-work will usually differ from the completed work scope. Evidence Based Valuation© also recognizes that the role of the expert analyst (appraiser) is different at different points in the valuation process.