From a Red Cell to a Defensible Answer
The order the work actually happens in, on real data.
A percentage gap is not a finding. A significance level is
Any two groups will differ. The question a regulator asks is whether the difference is larger than chance explains, and that is a calculation — not a judgement call about whether 43% versus 30% looks like a lot. Computing it per row, per product, turns a wall of percentages into a short list of things to actually investigate.
- Significance per row so the 99.87% cell separates itself from the 11.15% one.
- FFIEC ethnicity and race categories as the agencies define them, including the not-available rows.
- Segmented by product because a disparity in home purchase is a different conversation from one in refinancing.
The applications behind the number, without rebuilding a filter
This is the step that decides whether a fair lending programme is sustainable. If getting from “this cell is flagged” to “here are the 120 applications” means exporting and re-filtering, it happens once a year under duress. If it is a right-click, it happens whenever somebody is curious.
- FICO, LTV and DTI alongside APR so the obvious credit explanations are visible immediately.
- Create File Review from the cell the comparative review is scoped by the finding, not rebuilt around it.
- Add to Group to keep a population under observation across future cycles.
Regression an examiner recognises, not a black box
A model is only useful in an examination if its specification can be shown and defended. Forward selection, stepwise and backward elimination with explicit p-value thresholds, locked variables you insist on keeping, and standardized residuals and Cook’s distances calculated as standard — these are the terms the conversation is conducted in.
- Logistic for underwriting, linear for pricing with values transformed and dummy-coded rather than hand-built.
- Standardized and studentized residuals, Cook’s distances so influential applications are identifiable.
- Locked variables for the controls you will always include regardless of the entry method.
The point sitting on its own is the one you will be asked about
Residual tables tell you an application is an outlier. A plot tells you what kind — a genuine exception, a data error, or a cluster with something in common. Plotting any field against any other, with correlation and dispersion computed alongside, is how you tell those apart before somebody else does.
- Any field on either axis APR against amount, FICO, LTV or anything else on the record.
- Correlation, covariance and rank correlation next to the picture, not in a separate report.
- Mean crosshairs and per-axis dispersion so “far out” has a scale attached to it.
- 39years, since 1987
- Hundredsof institutions filing
- 40%+of applicants decline to self-report
- BISGCFPB-approved proxy fills the gap
Fair Lending Features
Last updated: August 13, 2026
Scorecards across all 12 FFIEC risk factors
Click any cell to drill straight into application-level data, no rebuilding filters. Add fields, comparisons and conditional formatting, or start from an examiner-tested template. Screenshot.
Regression built the way examiners run it
Logistic regression for underwriting, linear for pricing, with field values transformed and dummy-coded automatically rather than by hand, and consistency-verification tools to catch relationships between variables before they distort a model. Screenshot.
Residuals are data, not just output
Every application-level regression result is filterable, sortable and usable in a scorecard — set a rule to automatically group every application with a standardized residual past ±3, and go straight to the ones worth a second look. Screenshot.
BISG proxy for non-reporting borrowers
Over 40% of applicants don't self-report race or ethnicity. CFPB-approved BISG estimates it from surname and geography, using the same methodology regulators apply, so coverage is complete rather than partial. Screenshot.
Workspaces bring it together
Application groups, Data Mines, file reviews and regression models sit on one dashboard, organized at whatever granularity you choose, with reports and interactive charts a click away. Screenshot.
Field Explorer
Browse application data at a summary level, drill to a focal point, then create a Data Mine, file review, application group or regression model directly from what's selected. Doubles as a general exploratory statistics tool. Screenshot.
Filter, sort and sample
Filters built on the fly, saved and reloaded against any application set. Save a filtered or hand-picked selection as an application group and reuse it as the source for any analysis, automated or interactive.
Analyze results, then act on them
Turn a regression model into a scorecard identifying pricing disparities, and compare reports across time periods, assessment areas, underwriters or brokers — not just one snapshot in isolation. Screenshot.
Integrate with any external package
Export data, run an external command and reimport the result with one click. Those fields then behave like any other Comply field, including inside scorecards and file reviews. Screenshot.
Design once, reuse everywhere
Every view and process saves as a reusable "Definition" — scorecards, regression models, file reviews and imports included — and Auto-Pilot runs any of them on a schedule, offline or on.
Run any process on your terms
Edit checks, geocoding and field updates run individually or in batch, rather than one undifferentiated "update" that touches everything whether you asked it to or not.
Integrated compliance-grade geocoding
Batch the whole dataset or geocode a single application, plus the ZOOM interactive geocoder for checking results against a map across multiple data sources. Screenshot.
One relational database, enterprise security
A single SQL Server database and a .NET client, not a separate siloed database per dataset. User- and group-level security, offline synchronization, and Windows or SQL authentication throughout. Screenshot.
Unlimited custom fields, tab-based multitasking
Any data type, usable anywhere — a filter, a regression variable, a scorecard statistic — shown or hidden per dataset. Keep several datasets, reports or analyses open in tabs, each with its own history.
And the rest
Import-time filtering of any fixed-length or delimited file, five regulatory sampling methods, official LAR/MicroData file creation with encryption, AutoUpdates for system data, HMDA/CRA disclosure and performance tables, and a built-in mapping component.
Fair Lending Resources
FFIEC Interagency Fair Lending Examination Procedures
This overview provides a basic and abbreviated discussion of federal fair lending laws and regulations. It is adapted from the Interagency Policy Statement on Fair Lending issued in March 1994.
http://www.ffiec.gov/pdf/fairlend.pdf
OCC Comptroller's Handbook (Fair Lending)
Examiners use these procedures to evaluate a national bank's compliance with the Fair Housing Act (FH Act), Equal Credit Opportunity Act (ECOA), and the Federal Reserve Board's Regulation B. This booklet contains the Federal Financial Institutions Examination Council's (FFIEC) "Interagency Fair Lending Examination Procedures," and appropriate OCC supplemental material.
http://www.occ.treas.gov/publications/publications-by-type/comptrollers-handbook/fairlend.pdf
http://www.chicagofed.org/digital_assets/others/region/community_reinvestment/bankers_guide_to_risk_based_fair_lending.pdf
GAO Data Limitations and the Fragmented U.S. Financial Regulatory Structure Challenge Federal Oversight and Enforcement Efforts
GAO analyzed fair lending laws, relevant research, and interviewed agency officials, lenders, and consumer groups. GAO also reviewed 152 depository institution fair lending examination files. Depending upon file availability by regulator, GAO reviewed all relevant files or a random sample as appropriate.
http://www.gao.gov/new.items/d09704.pdf
Anatomy Of A Fair-Lending Exam: The Uses And Limitations Of Statistics
In this paper, we consider the role of statistical analysis in fair-lending compliance examinations. We present a case study of an actual fair-lending examination of a large mortgage lender, demonstrating how statistical techniques can be a valuable tool in focusing examiner efforts to either uncover illegal discrimination or exonerate an institution so accused.
http://www.federalreserve.gov/pubs/feds/2000/200015/200015pap.pdf
How Low Can You Go? An Optimal Sampling Strategy for Fair Lending Exams
Empirical researchers face a tradeoff between the lower resource costs associated with smaller samples and the increased confidence in the results gained from larger samples. Choice of sampling strategy is one tool researchers can use to reduce costs yet still attain desired confidence levels.
http://www.occ.treas.gov/publications/publications-by-type/economics-working-papers/2008-2000/wp2001-3.pdf
CRA And Fair Lending Regulations: Resulting Trends In Mortgage Lending
In this article, we examine the evolution of the fair lending regulations and the CRA. We then summarize the economic literature that pertains to these regulations.
http://www.chicagofed.org/digital_assets/publications/economic_perspectives/1996/epnd96b.pdf
FDIC Side by Side: A Guide to Fair Lending
In this guide, which we first distributed in August of 1994, we provide alternative means that an institution may use to discover uneven customer service or inconsistent lending practices that may be discriminatory. This guide is not about finding discrimination, that is, violations of the fair lending laws. It's about tools that a lender can use to compare the treatment of loan applicants, identify differences and correct potential problems.
http://www.fdic.gov/regulations/resources/side/side.pdf
Frequently Asked Questions
The seven we are asked most often about Fair Lending analysis, answered properly rather than in a sentence. If yours is not here, it is a better use of your time to ask a person.
Ask us directlyWhat is BISG in Fair Lending analysis?
BISG (Bayesian Improved Surname Geocoding) is a proxy method to estimate borrower race and ethnicity when not self-reported. Comply Fair Lending includes BISG analysis using CFPB-approved surname and geocoding data to support comprehensive fair lending reviews and ECOA compliance testing.
What statistical methods does Comply Fair Lending support?
Comply Fair Lending provides logistic regression for underwriting analysis, linear regression for pricing analysis, comparative file review (matched-pair analysis), and risk scorecard generation. All methods follow FFIEC fair lending examination procedures and include built-in validation tools to verify model quality and statistical significance.
How does risk scoring work in Fair Lending analysis?
Risk scorecards aggregate statistical results to identify high-risk areas requiring file review. Comply calculates disparities between control and protected class groups across underwriting decisions, pricing, and terms. Scorecards prioritize areas by statistical significance and practical impact, helping focus limited review resources on applications most likely to reveal fair lending concerns.
Can Comply Fair Lending help prepare for regulatory exams?
Yes, Comply generates exam-ready reports following FFIEC procedures. Run fair lending analysis quarterly or annually to identify and remediate risks before examiners arrive. The software produces documentation examiners expect: regression model summaries, comparative file reviews, scorecard results, and corrective action tracking—demonstrating your proactive fair lending monitoring program.
How does RATA identify Fair Lending issues across FFIEC risk factors?
Risk scorecards cover all 12 FFIEC fair lending risk factors, and clicking any cell drills straight into application-level data with no report to close or filter to rebuild. Use an examiner-tested template or build a custom scorecard for your institution's specific risk areas. Interactive scorecards cut analysis time from weeks to hours, so an anomaly gets investigated immediately instead of waiting on a manual report.
How does RATA normalize data across origination systems?
Comply identifies missing or invalid data and lets you focus on applications with incomplete information or entry errors, then update or remove outlying data from the analysis workspace with automated tools. Scatter plots and other interactive charts help compare fields visually to catch a trend or an integrity issue before it reaches a model.
How does RATA support comparative file reviews for examiners?
Build a matched-pair pool with whatever partitioning and matching criteria you choose, then create a file review from any set of applications or analysis elements. Document each pair at the file-review, application-pair, application and individual-field level, print the pairs that need further attention, and group and annotate the ones where no risk was found — then generate an executive summary or a detailed field-level report from the whole review.
Comply Fair Lending vs Manual Analysis
See how software-driven analysis reduces risk and saves time
| Analysis Task | With Comply Fair Lending | Manual Analysis |
|---|---|---|
| BISG Proxy Analysis | Automated Bayesian Improved Surname Geocoding with Census Bureau surname list and tract demographics. Complete proxy analysis in minutes. | Manual surname lookup in Census tables, manual geocoding, complex spreadsheet calculations. Hours of work, high error risk. |
| Regression Analysis | Built-in logistic regression models for pricing and credit decisions. Automatic control variable inclusion, statistical significance testing, and results interpretation. | Requires statistical software expertise (SAS, STATA, R). Manual data preparation, model specification errors common, difficult to replicate. |
| Comparative File Review | Automated matched-pair selection based on your criteria. Side-by-side comparison reports with highlighted disparities and risk scoring. | Manual file selection prone to bias. Time-consuming individual file reviews. Difficult to document methodology and ensure consistency. |
| Pricing Analysis | Scatter plots, pricing disparity reports, and statistical tests for prohibited basis pricing differences. Visual identification of outliers and patterns. | Manual Excel charts, limited statistical testing capability. Difficult to control for legitimate pricing factors. Results may not withstand examiner scrutiny. |
| Risk Scoring | 12-factor risk scorecard with weighted assessment. Quantifies fair lending risk across multiple dimensions with actionable remediation priorities. | Subjective risk assessment. Inconsistent evaluation criteria. Difficult to track risk changes over time or justify to examiners. |
| Documentation | Exam-ready reports following FFIEC procedures. Complete audit trail, methodology documentation, and statistical validity evidence. | Manual report creation. Inconsistent documentation. Examiners question methodology. Difficult to demonstrate systematic monitoring. |
| Time to Complete Analysis | Hours for comprehensive fair lending review | Days or weeks for partial analysis |
| Regulatory Updates | Automatic updates to BISG surname list, FFIEC procedures, and regulatory guidance. No additional work required. | Manual research and implementation of methodology changes. Risk of using outdated procedures or data sources. |
What Our Customers Say
Trusted by financial institutions nationwide for compliance excellence
“The support has been exceptional. The program is very easy to use. RATA has been one of the best companies I have dealt with in my 25 years of banking. Their service has exceeded my expectations.”
“I wanted to let you know that RATA and Comply is the absolute best!!! My submission is complete and was accomplished in record time. I more than doubled the amount of activity I reported last year and it took me less than half the time. Thank you for being there for us and the industry! You are and always will be number one to me!!!”
“Would highly recommend both the product and the personnel!! Being able to accomplish what I did in such a short period of time is a testimonial to the product's user friendliness as well as the support I received from the great folks at RATA. Could not have done it without them!!!”
… and hundreds of others since 1987Methodology & Sources
The risk factors and regression tests on this page follow FFIEC fair lending examination procedures; the BISG (Bayesian Improved Surname Geocoding) proxy method follows the CFPB's published methodology. See the risk scorecard and regression testing sections above for how each is applied. RATA Associates maintains this page and updates it when FFIEC or CFPB guidance changes.
You may republish the figures and methodology on this page with attribution and a link to rataassociates.com/comply-fair-lending/.
