Return on Quality

The title and central theme of this blog is “return on quality”, which we broadly define as the benefits to be gained from an intelligent and continuous approach to improving mortgage loan quality. 

We said in an earlier post that we would try to formulate “return on quality” and as a step in that direction, we offer a Cogent white paper called “Quality Performance Benchmarking” that was originally developed for an audience of mortgage quality control professionals. 

Image by jacob earl

In this paper, we talk about the prevalence in the mortgage industry of a production maximization mentality, in which metrics and compensation are centered on volume; the potential hazards of this mentality; guidelines for estimating the costs of poor quality, (the inverse of the return on quality); how to reward good quality; and how to craft appropriate performance metrics, or benchmarks.  The second part of the paper talks in depth about one of the most powerful tools for benchmarking performance, control charts.

This white paper was written in 2002.  Nothing has changed in the methodology.  But in the last couple of years, the eyes of most of us in the industry have been opened to the dangers of focusing exclusively on volume, volume, volume.  We welcome your comments.

Statistics is baffling enough without being footloose with terminology.  So let’s clear up what we mean by a statistical sample.

The term “Statistical Sample” has a very specific meaning in the Cogent system.  It refers to a sample that is randomly selected from the entire population of loans eligible for a particular sample type (aka “audit shell”).  The suggested sample size is calculated every period by the system and is designed to yield a 95% confidence and 2% precision over 12 months.  This is the standard originally established by FNMA, FHLMC, and HUD for lenders who qualify to substitute ’statistical sampling’ for the traditional 10% random sample.

The generic term ’statistical sample’ is not very meaningful, in and of itself.  It simply refers to a sample in which some statistical principle has been employed, without defining which principle.  For example, it could refer merely to a randomly drawn sample, without specifying what population is being drawn from or how much precision will be achieved across what period.

Image by johnbullas
Rick Astley reference 

To illustrate: most Cogent ProductionQC clients have at minimum a “Production” sample type, for which all loans originated in a particular period (typically a month) are eligible.  When a Statistical Sample (in the Cogent definition) is randomly drawn from this population, all loans have the same probability of being selected.  No distinction is made between loan type, loan source or any other loan characteristic.  It is intended  to establish a baseline of overall loan quality across the organization.

In order to achieve a 95% confidence and 2% precision for a particular category of loan, it is necessary to go beyond the “Statistical Sample”.  For example, in the Cogent system, to achieve this standard for all FHA loans originated, define a Targeted Query (Loan Type = FHA) and run the query.  The resulting screen displays all qualifying loans, including qualifying loans  that were randomly drawn previously in the “Statistical Sample” or any other samples in this period.  These count towards the total required.  Use the embedded Cogent Statistical Calculator to calculate the required (”suggested”) sample size for the period.  From the suggested sample size, subtract the number of qualifying loans that have previously been sampled and enter the result in the Sample Size box. The Cogent system will then randomly select the entered number of loans from the qualifying loans.

The Cogent “Stratified Sample” is in effect a pre-defined Targeted Sample.  Most typically, the Cogent system stratifies originations by Source or Channel and automatically tracks and calculates the sample size required for each stratum (Source or Channel), net of qualifying loans randomly drawn in the “Statistical Sample.”  Over 12 months, the Stratified Sample achieves 95% confidence and 2% precision for each stratum.  In Targeted Samples, this automated operation is performed by the user, using Cogent’s embedded tools.

Thus, in order to leverage the Cogent system’s sampling optimization, clients should begin sampling from the broadest category (all loans eligible) to the most narrow category (e.g., individual underwriters).  In this way, all loans selected in previous broad categories are counted towards ever narrower categories, minimizing the number of loans to be sampled and audited.

One of our favorite books is “How to Lie with Statistics”, a tongue-in-cheek primer on using statistics to make just about any argument you like.  It illustrates how easy it is to mislead people who are unaware of basic statistical principles.

For instance, what do you make of the headline “Median home price in Jefferson County falls by 27%”?  Sounds pretty dire when you read it in bold headline on a newspaper as you’re walking by.  You might think that home prices in Jefferson County have all fallen by 27%.  But it’s worth digging deeper.  What period are we talking about?  What are we comparing to?  What does ‘median’ mean (compared to ‘mean’, for instance)? How many homes were sold and does that make a difference in the statistic?  Are we including detached homes and condos?  None of this is clear from the headline, and there are many other questions to ask.

Among other things, basic statistical understanding gives you a sense of context, scale and precision.  In the knowledge economy, where information is coming at you from a variety of sources, that’s pretty important.  Whenever you see a statement involving statistics - or indeed, any measurement - it’s worth asking whether there is any ambiguity in the statement.  And if there is, dig deeper.

To help you get comfortable with some basics, we’ve compiled a short list of statistical concepts for mortgage quality control.  If there’s anything else you’d curious about, let us know.

Key Statistical Concepts for Mortgage Quality Control

CONTENTS:

• Sample inferences & statistical precision
• Random Selection
• Sample size estimation
• Qualitative analysis & defect rates
• Random variation & statistical control
• Sampling error & non-sampling error
• Correlation vs. Causation
   

Sample inferences & statistical precision

The fundamental purpose of sampling for Quality Control is to render judgments regarding quality of the overall loan portfolio, i.e., to infer general conclusions from the sample’s findings.  The degree to which those conclusions can be reliably inferred is measured by statistical precision. Keep in mind that the goal of Quality Control is to focus on the forest, not the trees. Accordingly, your objective is not to identify and correct errors or defects in specific loan files, but to use the incidence of such errors to infer conclusions about your loan origination process.

Critical issues:

1. Statistical inference must be based on random selection; the most common error is to draw conclusions from a non-random sample. To avoid this error, you should eliminate all non-random selections from any group used to make statistical inferences to the population.
2. Statistical precision (e.g., of two percent) must be demonstrated on the actual sample defect rate (i.e., the number of loans with defects divided by the number of loans reviewed). If you were unable to review some of your randomly sampled loan files, then the precision achieved by your process will be degraded.
    

Continued…

It’s encouraging to see the adoption of statistical methods in the world of mortgage quality control.  Done right, it can lead to enormous returns on your investment in qualiy control - what we call ’return on quality’.  But you have to do it right.  There is plenty of misinformation about statistics on the Internet and a non-expert may have difficulty sifting through what’s right or wrong, especially as applied to mortgage quality control.  

So we’d like to present the principles and methods that Cogent’s statisticians and QC experts have honed over the past 15 years.  We invite your comments.  We begin with an overview of statistical sampling in mortgage quality control, which is available as a white paper (PDF) here.

______

Statistical Sampling in Mortgage Quality Control
By Hakki Etem, CEO, Cogent Economics

Introduction

Statistical methods are well-established tools for efficiently measuring and improving product quality in a variety of industries.  Unfortunately, statistical analysis has been slow to gain acceptance in the mortgage industry, although the ability to originate the best quality product at the lowest possible cost is just as valuable to mortgage originators as it is to automobile manufacturers.  There are many reasons why the mortgage industry has avoided statistical methods, but surely one reason is the subject itself:  few disciplines can be as mind-numbing as statistical theory.

Nevertheless the most effective way for QC managers to measure and improve loan quality — at the lowest possible cost — is to employ statistical methods.  Although this means that QC managers must necessarily become familiar with basic statistical concepts, with the right tools and professional support the process can be greatly simplified. 

Continued…