Risk rating validation, explained.

Why every business loan carries a risk grade, and who checks that the grade is right

The life of one loan grade, in five steps

1. The lender proposes a grade at approvalBrinmoor Freight Lines is graded on the bank's 1-to-10 scale using its scorecard: coverage, leverage, liquidity, management, industry, guarantor.
2. A credit officer agrees, or the lender sets it aloneIn the example bank's Mountain region, loans under $20 million are graded by the lender alone. Elsewhere a credit officer co-signs every grade.
3. The grade is refreshed each year, or when news arrivesBrinmoor's annual review in March 2026 affirmed grade 4, two notches better than its scorecard, with no reason code.
4. The grades are checked across the bookAgainst the scorecard, against look-alike borrowers graded by other lenders, against rating changes and against defaults. This is where Brinmoor's 4 and Coulter's 7 meet.
5. Loan review and examiners sample, and judge the systemAbove 5% of loans or 3% of dollars graded wrong, examiners look for the root cause and decide whether to widen their sample.1

Step 4 is the subject of this guide. The bank, borrowers, people and figures in this example are fictional.

Undefined pass grades, optimistic lenders: rating surprises

Grade, dual rating, classified, override defined

Fourteen terms are defined in the one-page summary.

What a commercial loan risk rating measures, and the jobs one number is asked to do

In shortA grade measures the borrower's expected ability to repay over at least the next year, mostly from its cash flow. Banks grade the borrower as it is today, while the rating agencies grade through the cycle, so a bank grade should move faster. The same number then drives approval authority, pricing, review frequency, the allowance and the reports the board reads, which is why one wrong grade travels so far.

The OCC asks banks to rate "based on the borrower's expected performance, i.e., the likelihood that the borrower will be able to service its obligations in accordance with the terms", evaluated "over the foreseeable future", which it sets at "not less than one year". Examiners look at "at least the upcoming 12 months". The primary question is the strength of the main source of repayment, "a sustainable source of cash"; collateral and guarantors matter more as that source weakens.1

The jobs a grade does

Today's condition, or condition under stress?

The rating agencies grade "through the cycle": they try to look past a downturn. Banks do the opposite. In a Federal Reserve study of the fifty largest US bank holding companies, "almost all banks have chosen to grade to current condition". Asked over what period they judge risk, the banks interviewed split three ways, and half had no fixed answer.4

The difference matters when a bank compares itself with public ratings. A bank grade 4 is a statement about this borrower today; an agency BBB is meant to survive a recession. The OCC says banks that use public ratings "must ensure that their internal credit risk ratings change when risk changes, even if there has been no change in the public rating".1

How many grades a scale needs

Most of the work happens in the pass grades, which regulators leave undefined. The OCC: "most rating systems require several pass grades", and a community bank may manage with a few, for example one for loans secured by cash-like collateral, a "watch" grade and one or two others.1 Among the fifty largest banks the Federal Reserve studied, pass grades ranged "from two to the low twenties"; the median was five, six among the ten largest, plus "three or four regulatory problem-asset grades".4

The example bank uses ten: grades 1 to 6 are pass, with 6 as pass/watch, then 7 special mention, 8 substandard, 9 doubtful and 10 loss.

Which supervisory texts govern loan risk ratings, from the 2001 OCC booklet to the 2026 exam rules

In shortThere is no federal rule that sets a bank's rating scale. Supervisors expect accurate and timely grades, a written scale that translates into their five classes, independent review, and evidence that worse grades default more. The OCC's 2001 booklet on rating credit risk is still the core text. In 2026 the model risk guidance was replaced, the OCC issued a new lending booklet, the 18-month exam cycle widened to banks under $6 billion, and a new rule, effective November 2, 2026, raises the bar for formal exam findings at OCC and FDIC banks.

Dual risk rating: a borrower grade for the chance of default, a facility grade for the loss

In shortA single grade has to mix two questions: will this borrower default, and how much would the bank lose on this loan if it did? A dual system answers them separately, with a borrower grade tied to a probability of default (PD) and a facility grade for each loan tied to loss given default (LGD) or expected loss. The OCC accepts either design, as long as both questions are answered. The common failure is a borrower grade that quietly absorbs the collateral.

The OCC describes the design in one sentence: "Dual rating systems typically assign a rating to the general creditworthiness of the obligor and a rating to each facility outstanding." The facility rating "considers the loss protection afforded by assigned collateral and other elements of the loan structure in addition to the obligor's creditworthiness". Banks adopted it "because a single rating may not support all of the functions that require credit risk ratings": borrower grades serve structuring and administration, facility grades serve the allowance and capital.1

The OCC "does not advocate any particular rating system" and accepts "a single rating or dual ratings".1

The three numbers behind a dual grade

The borrower grade carries the probability of default. The facility grade carries the loss given default, the share of the loan the bank would not recover. Their product with the amount owed at default is the expected loss. The regulators' classes do not use these words, but the OCC notes that the concepts sit inside them: substandard loans have "a high probability of payment default", and a doubtful loan "a high probability of total or substantial loss".1

Illustrative arithmetic with assumed rates: a 3.0% one-year default probability is the Federal Reserve example's grade 5, and the loss rates are assumptions to replace with your own. The point is the shape: the two loans share one borrower grade, and the revolver's higher loss and its tendency to be drawn as the borrower weakens belong in the facility grade.

Four rules that keep two scales honest

Mapping a 1-to-10 scale to pass, special mention, substandard, doubtful and loss

In shortEvery bank grades on its own scale, and every bank reports in the regulators' five words. The 2020 interagency guidance asks a bank with its own scale to keep documentation that translates it into the agencies' classes, so examiners can reconcile the totals. The map is easy to draw and easy to blur: the risk sits at the line between the last pass grade and special mention, where a watch grade can become a place to park loans.

The interagency guidance on credit risk review says a bank whose framework "differs from the framework for loan classifications used by the Federal banking agencies" should "maintain documentation that translates their risk ratings into the regulatory classification framework". The documentation lets examiners "reconcile the totals for the various loan classifications or risk ratings under the institution's system to the Federal banking agencies' categories".6 The agencies' definitions are the same for every bank and every kind of credit.1,5

Classified means substandard, doubtful or loss. "Non-pass" or "criticized" usually adds special mention. Special mention comes from an interagency statement of June 1993; the classified definitions from April 1991.1

Where mapping goes wrong

The watch grade is the line most worth testing. In the example, Brinmoor's scorecard grade is 6, pass/watch, and its assigned grade is 4. Three of its Mountain neighbors with coverage near or below 1.0x carry pass grades of 5 while matched peers elsewhere sit at 7, special mention. Those three loans, $31.4 million together, are on the wrong side of the line the board and the examiners count.

Scorecards, rating drivers and overrides: where a grade comes from and how it bends

In shortMost commercial grades start from a scorecard: a handful of factors such as coverage, leverage, liquidity, earnings trend, industry, management and guarantor support, each with a weight. Then a person may override the result. Overrides are legitimate, because a scorecard cannot see everything, but each one needs a reason that someone else can check. An override with no reason code, set by the lender alone, is the pattern validation looks for first.

The OCC asks for criteria "clear and precisely defined using objective (e.g., cash flow coverage, debt-to-worth, etc.) and subjective (e.g., the quality of management, willingness to repay, etc.) factors", and for ratings that reflect "both the borrower's expected performance and the transaction's structure".1 A scorecard writes those criteria down with weights, so two people scoring the same borrower land on the same number.

Why overrides happen, and when they are right

Supervisors and researchers name the same reasons lenders lean toward better grades. The Federal Reserve's study of large banks found that "at banks that use ratings in computing profitability measures, establishing pricing guidelines, or setting loan size limits, the staff may be tempted to assign ratings that are more favorable than warranted".4 The OCC lists compensation that rewards production, relationship structures that encourage lenders to "hide" problems "for fear of losing control over a customer relationship", and "unfounded optimism" as impediments examiners should look for.1 Its 2026 lending booklet warns against pay that ignores "risk-rating recognition".3

What a sound override policy says

• Who may override, and by how much. For example, a lender may move one notch with a credit officer's agreement; two notches needs the regional credit officer; never the lender alone.
• A reason code from a fixed list, plus the memo passage that supports it, so a reviewer can test the reason.
• Joint grades. The OCC: "Many banks, therefore, find risk rating accuracy improves by requiring ratings to be a joint decision of lenders and credit officers (at least one person from each function)."1
• Override reporting by lender and region, with the share that carry no reason. The OCC's MIS list includes "Rating changes by line of business, loan officer, and location".1
• A review of the scorecard itself when overrides cluster on one factor. In the example, trucking overrides cluster on spot-rate exposure, which the bank's scorecard does not score.

Do two lenders give the same borrower the same grade? What the evidence shows

In shortNot reliably. Federal Reserve researchers who compared the risk estimates different banks reported for the same syndicated loans found wide dispersion, and found that banks short of capital reported lower default risk. Nobody publishes the same test inside one bank, but the causes are the same: incentives, local habits and stale inputs. Comparing look-alike borrowers graded by different lenders is the fastest way to see it, because it needs no defaults.

Both studies use loans shared by several banks in a syndicate, which is the cleanest test there is: the same borrower, the same loan, the same date, different graders. Plosser and Santos found the link to capital stronger "for larger, riskier, and more opaque credits".12 Firestone and Rezende found "Banks differ substantially in PDs, but only a few set PDs systematically higher or lower than the median bank", while loss estimates differed systematically.13

The same test inside one bank: matched pairs

A bank cannot put one borrower in front of two of its lenders, but it can find two borrowers that look the same on paper and were graded by different people. The method is simple to state:

1. Define "alike". Same industry at a fine level, revenue within a band, coverage and leverage within a tolerance. The example uses the same three-digit NAICS industry, revenue within 25%, coverage within 0.10x and leverage within 0.5x.
2. Take peers from other lenders. A lender's own loans never serve as each other's peers; most peers come from other regions.
3. Compare the grades given and the grade the scorecard gives, and flag pairs two notches or more apart.
4. Read the memos behind both grades. A gap with a good reason is an override doing its job.
5. Add the gaps up by lender, region and industry. One pair is an anecdote; a cell of 38 loans leaning the same way is a pattern.

In the example, the bank's 2,305 rated loans fall into 45 cells of nine industries by five regions. Four cells lean more than half a notch from their matched peers. The widest is freight trucking in the Mountain region: 38 loans, $286.4 million, graded on average 1.2 notches better than trucking borrowers elsewhere, where the gaps stay within 0.2. Inside that cell, one lender's 17 loans lean 1.6 notches, a second's 1.0 and a third's 0.7.

Back-testing grades against defaults, and reading a migration matrix

In shortBack-testing asks whether the grades came true: did the worse grades default more, and did each grade default about as often as the bank assumed? Migration asks how grades moved: how many stayed put, how many fell one grade and how many fell two at once. A test across the whole bank can pass while one region fails it, so read both by region and lender. Expect few defaults, and treat a small cohort's surprise as a reason to look closer and gather more evidence.

The OCC sets two standards. Systems that state default probabilities or expected loss rates "should be back-tested", and back-tests "should show that the definitions' default probabilities and expected loss rates are largely confirmed by experience". Systems with no stated probabilities should still "rank-order risk": "banks should be able to show that credits with more severe ratings exhibit higher defaults and losses".1 The Basel Committee's study separates the same two questions: "discriminatory power", meaning ranking, and "calibration", meaning the level.2

The shape matters more than the levels. Default risk grows slowly at the top of a scale and very fast at the bottom, so a one-notch error among the strong grades costs little, while the same error near special mention can hide a loan that will default.

A back-test that passes for the bank and fails for one region

Across the example bank, each grade defaults close to its expectation and the order holds. In the Mountain region, grade 4 loans defaulted at 1.9% over three years, against 0.8% expected and above the region's own grade 5. That is a rank-order break: the region's 4s behave like 5s, which is exactly what a habit of grading one notch better would produce. Averaged into a book of 2,305 loans, it disappears.

The one-in-five figure is our arithmetic (Poisson, mean 0.8): chance of two or more defaults = 1 − e^−0.8(1 + 0.8) ≈ 19%.

The migration matrix

A migration matrix sets each loan's grade at the start of a year against its grade at the end. Each row adds to 100%. The diagonal is loans that stayed put; right of it, loans that got worse; two or more columns right, the "double downgrades" the OCC asks banks to count, along with "seasoning of ratings (the length of time credits stay in one grade)", the "velocity of rating changes" and "The ratio of rating upgrades to rating downgrades".1

What the Shared National Credit reviews of 2024 and 2025 say about large-loan grades

In shortEach year the Fed, the FDIC and the OCC re-grade a sample of the largest shared loans themselves. In the 2025 review, released January 12, 2026, 8.6% of $6.9 trillion of commitments were special mention or classified, down from 9.1%, but the agencies said the fall came from new lending, "rather than an underlying improvement in credit quality". Nonaccrual commitments rose 30%. The review is the closest public view of how supervisors grade, and its definitions are the ones every bank maps to.

The Shared National Credit (SNC) program covers any loan or commitment of "$100 million or more" shared by "three or more unaffiliated federally supervised institutions".11 Examiners review a risk-based sample twice a year and the agencies publish one statement a year after the third-quarter review.14 The 2025 review covered 6,857 borrowers and $6.9 trillion of commitments, up 6%.15 No 2026 report had been released by the end of October 2026.

In 2024 the agencies described "a modest increase in the severity of risk due to a migration of commitments from special mention to a classified rating".14 From 2025 the program document keeps the tables and drops the commentary.11

How supervisors argue about a grade

An SNC appeal from the third quarter of 2024 shows the reasoning a validation team should expect. The bank that runs the loan for the lending group appealed a substandard rating on a revolving credit, arguing for special mention because the industry downturn and an operational problem were temporary and results had recovered from the low point. An interagency panel upheld substandard: operating cash flow did not cover fixed charges, free cash flow was negative, leverage was high, and "the company had not achieved satisfactory and sustained performance to warrant an upgrade".17

It is the OCC booklet's rule in practice: only sustained numbers lift a grade, however good the plan or the recovery story.1

Relationship grades, stale statements and parked watch loans: the ten errors behind drifting risk ratings

In shortGrades go wrong in four places: in the people who set them, in the inputs they read, in the scale itself and in the checks around it. Most of the errors lean the same way, toward a better grade, and most look reasonable one loan at a time. They show up as patterns: by lender, by region, by industry, and as double downgrades.

From Barnes Banking to First NBC: what failed-bank reviews found when weak loan grading let problem loans pile up

In shortWhen a bank failure costs the deposit insurance fund enough, the agency's inspector general writes a material loss review. The ones that deal with commercial credit tell a consistent story about grades: weak or unreliable grading, loan review that missed problem credits, examiners repeating the same criticism, and problem loans that appear all at once when examiners finally re-grade them. The 2024 and 2025 failures reviewed so far add a harder lesson: grades can be falsified.

What the reviews found

The lesson for validation: check grades against independent facts (statements, payment records, collateral), and treat a rating history that never moves with as much suspicion as one that moves all at once.

The reviews from 2010 judge supervisors as well as bankers, and several of their findings concern the agencies' CAMELS ratings of the bank as well as the bank's loan grades. Their point about loan grades is the one quoted here: weak internal grading let problem loans accumulate until examiners re-graded them.

Is a rating scorecard a model? AI in rating validation under the April 2026 guidance

In shortSoftware can re-score every loan on the bank's own scorecard, find look-alike borrowers, read the memo behind each grade, build the migration and back-test reports and draft the calibration note. A statistical scorecard may be a model under the April 2026 guidance; arithmetic and fixed rules are not, and generative AI falls outside that guidance, under the bank's own controls. People keep the grade: no supervisory text lets software decide one.

The 2026 guidance defines a model as "a complex quantitative method, system, or approach that applies statistical, economic, or financial theories to process input data into quantitative estimates", and excludes "simple arithmetic calculations, such as those found within spreadsheets, as well as deterministic rule-based processes". Its footnote puts generative AI, and AI that acts on its own, outside the guidance as "novel and rapidly evolving", while a bank's own governance "should guide the determination of appropriate governance and controls".7

One sentence fits judgmental rating systems closely: "When a model's design relies substantially on expert judgment, quantitative outcomes analysis helps to evaluate the quality of that judgment."7 Back-testing and matched pairs are that outcomes analysis.

What supervisors have said about technology in rating work

General AI frameworks apply too. NIST's framework organizes the work under GOVERN, MAP, MEASURE and MANAGE.24 The Treasury's 2024 report warns that generative AI "can be susceptible to 'AI hallucinations,' producing confidently stated but incorrect output".25 For memo reading, the defense is the same as for any reviewer: every finding quotes the passage it rests on, so a person can check it in seconds.

Who still sets the grade when software checks ratings

• The grade itself, and every regrade. A regional credit officer, never the lender on the loan, should confirm a proposed change.
• The scorecard's design and weights, and the decision that a scorecard change goes to model validation before use.
• Judgments a scorecard cannot score: a verified guarantee, a contract with escalators, a customer about to leave.
• The independent sample: software can point loan review's sample at the loans where pairs disagree, and the reviewer still sets an independent grade.6

How banks validate risk ratings by hand today, and why re-spreading look-alike borrowers eats the most time

In shortLenders and credit officers set grades; a credit risk or portfolio team reports on them; loan review samples and re-grades; the chief credit officer owns the scale and answers to the board. Comparing like borrowers across regions means re-spreading and re-scoring them by hand, so it is done for a handful of loans a quarter. Most of the days go to gathering and re-scoring, leaving little for judging.

Five people who set, check and sign a grade

A quarterly calibration, hand-off by hand-off

Where the days of a hand-built calibration go

Twenty working days for one region explains why most banks compare like borrowers for a few loans a quarter and rely on loan review's sample for the rest. In the example bank, loan review sampled 401 of 2,305 rated loans in the third quarter, about 17%. A region and an industry can drift for a year before a sample lands on them.

Spreadsheets, external loan review or rating software

Graded 4 in Denver, 7 in Omaha: how the Brinmoor and Coulter pair became a calibration note signed for committee

In shortThis is the example from the top of the page, inside the product. Overnight, six agents re-score every rated loan on the bank's own scorecard, pair each borrower with look-alikes graded by other lenders and read the memo behind each grade. On Monday Rachel Okafor, senior credit risk officer, sees the pair, then the pattern, takes the evidence to the regional credit officer and the chief credit officer, and has a signed note for Senior Credit Committee a week ahead of the meeting.

1

Same numbers, different grades

Monday morning, the home screen plots the bank's 195 rated freight trucking borrowers by debt service coverage (up) and senior debt to EBITDA (right), colored by the grade each was given, over shaded bands for the grade the bank's scorecard gives at each point. Two dots sit almost on top of each other and are circled: Brinmoor Freight Lines, graded 4 in Denver, and Coulter Haulage, graded 7 in Omaha.

The panel beside them says why the pair matters: coverage within 0.03x, leverage within 0.1x, both scored 6 by the scorecard. Brinmoor's 4 is an override with no reason code; Coulter's 7 is a documented covenant downgrade. The rail below lists 22 look-alike pairs graded two or more notches apart.

2

Against the scorecard, one region lights up

Rachel switches the colors to notches off the scorecard and picks the Mountain region. Fourteen loans, $146.8 million, are graded two or more notches better than the bank's own scorecard. Eleven of them belong to one lender, Brett Kalani. None of the region's 38 trucking loans is graded stricter than the scorecard.

The pattern answers the first question a credit officer asks about any single pair: is this bad luck, or a habit?

3

Brinmoor beside its three peers

She opens the pair side by side. Brinmoor (loan C-25-04417, $18.5 million: a $12.0 million term loan and a $6.5 million revolver) sits beside its three matched peers: Coulter in Omaha, Thorne Freight Systems in Milwaukee and Brazos Line Carriers in Waco. Revenue, coverage and leverage all fall within the matching tolerances, each marked with a check.

Assigned 4, affirmed March 14, 2026 by the lender alone. Scorecard 6. Peers 7, 6 and 5, each set by a lender and a credit officer. Peer median 6.

4

The scorecard and the two memos

Lower on the same screen, the scorecard shows Brinmoor factor by factor: coverage 1.14x against the grade 4 guide of 1.25x, leverage 3.8x against 3.0x, earnings down 18%, a guarantor statement from June 2024. Weighted 5.6, grade 6. Then the lender's override: −2 notches, reason code none.

Beside it, the memos. Brinmoor's annual review affirms grade 4 on "a long-standing relationship since 2014" and a sponsor with "substantial outside liquidity", and scores 34 of 100. Coulter's covenant review explains its one-notch stricter grade with the covenant test, the ratios and an upgrade path, and scores 82. Each highlighted passage opens the memo at its page.

5

Confirmed, then signed for committee

The calibration drafter writes the finding and its cause, every line cited, and proposes a grade for each of the 14 loans: 12 regrades, 3 of them into special mention ($31.4 million), and 2 kept, Cache Valley Freight and Snake River Carriers, whose memos show a verified guarantee and a contract. Rachel sends it for sign-off.

Kevin Brandt, the Mountain regional credit officer, confirms the regrades; the lender on a loan can never confirm a regrade of it. Daniel Whitford, chief credit officer, signs the note for Senior Credit Committee on Tuesday, October 20, 2026. Nothing changes in the loan system until committee approves.

In the example the solution compares all 2,305 rated loans with their look-alikes every night, against a loan review sample of 401 loans, 17% of the book, so drift can surface within the quarter it starts, ahead of the next exam. Measure both on your own book.

The drift map, loan review disagreement, memo scores and lender comparisons on the remaining screens

In shortOne pair is the story. These screens show the rest: the map of industries by regions that points to the drifting cell, the overnight comparison step by step, loan review disagreement against the examiners' lines, migration and the default back-test, memo scores, each lender's grades against their peers, the policy changes in the note, and the settings that hold the bank's own rules.

Select any screen to open it larger. The bank, borrowers, people and figures are fictional.

Who may change a risk grade: a regional credit officer confirms, the chief credit officer signs, six agents prepare the evidence

In shortEach agent has one job and limits it does not cross. None of them changes a grade, a scorecard weight or the allowance, and none sends a note unsigned. A regional credit officer, never the lender on the loan, confirms each regrade; the chief credit officer signs the note for committee. Every finding cites the rating history, the spread, the scorecard line or the memo passage it rests on.

Six agents that gather rating history, spreads and look-alikes

Risk officer prepares, regional officer confirms, CCO signs

Scorecard and threshold changes are versioned. Agent steps and human decisions are kept apart on the activity trail with who, what and when. A memo that scores below the bank's minimum goes back to the lender with the tests it missed.

The calibration team behind 2,305 rated loans

What rating validation software reads from your loan system and memos, and the drift map and calibration note it leaves behind

In shortIt reads what the bank already keeps, the grades and their history, the spreads, the approved scorecard and the memos, and leaves behind a comparison of every rated loan with its look-alikes, a drift map by lender, region and industry, scored memos and a calibration note with its regrades confirmed and signed. The loan system stays the system of record.

Loan system, spreading tool, scorecard and loan review tracking

For each, check whether it has an API, a file feed, or only screens, and who owns it.

Your rating scale, scorecard and loan review stay as they are

Like with like, cited evidence, people grade

What a flattering risk grade costs you: an understated allowance, an exam finding on the rating system, and double downgrades later

In shortThe direct cost is analyst days spent comparing a handful of loans by hand. The bigger costs are harder to count: an allowance too small for the loans you hold, a rating system that becomes an exam finding, and problem loans that arrive all at once as double downgrades. No public study prices any of these, so the numbers that matter are yours.

Calibration days and unsampled loans, from your own records

Rating-system findings, mispriced loans and a misled board

Comparing every rated loan each quarter: the analyst days and sample coverage automated calibration could return

In shortPut in your own numbers. The time saved on calibrations rarely pays for a tool on its own; the case usually rests on what the time buys: every loan compared every quarter beyond loan review's sample, drift found within the quarter, and evidence ready when examiners ask. The calculator values the time and shows how many grades a sample never reaches.

Six places automated rating checks pay back

Value the analyst days a calibration takes, with your numbers

A bank that calibrates one or two segments a year by hand will not cover a tool's cost with time alone, and should buy only if it wants to compare every loan every quarter. A bank that already calibrates every region each quarter is closer to the first line.

Score your rating system's readiness: scale and map, scorecard, overrides, rating history and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

What your score says about starting rating validation

What to ask a rating validation vendor about look-alike matching, your scorecard, memo evidence and who changes a grade

In shortAsk every vendor the same questions, us included. The answers show whether the tool compares the grades you actually gave or only calculates new ones, whose scorecard and rules it uses, what evidence you keep, who decides, and how you would measure it on your own book.

We found no published independent test of any rating validation vendor, as of October 2026. A pilot on one industry, checked against your own loan review, is the fairest comparison, and the same test measures your current process.

Test rating validation on one industry across every region, and judge it against loan review's own re-grades

In shortPick one industry with enough loans in every region, measure your current process, compare every loan with its look-alikes, check the findings against loan review and your credit officers, and decide with four numbers. None of these steps commits you to a vendor.

1. Pick one industry with at least 30 loans in most regions, and write down the scale, the map and the override rule as they stand.
2. Measure your baseline: loan review disagreement by region, overrides without a code, double downgrades and the days your last calibration took.
3. Compare every loan in the industry with its look-alikes on your scorecard, and score the memos behind the grades.
4. Check the findings against loan review's own re-grades and against your regional credit officers' decisions on each flagged pair.
5. Decide with four numbers: pairs flagged, share confirmed by credit officers, false alarms, and analyst days per calibration.

What to measure in a rating validation pilot, and how

Dual risk ratings, the 5% and 3% lines, back-testing and SR 26-2: what credit teams ask about rating validation

In shortThe questions credit teams and buyers search for most, each answered on its own.