PD Model - Why So Many Banks Land at "A"

The observation

On the Q1 2026 production scoring run (9,284 FDIC-insured commercial banks scored under v1.2), the PD letter distribution looks like this:

Letter	Share
AAA / AA+ / AA	0.00%
AA-	0.43%
A+	14.08%
A	37.95% (modal)
A-	23.99%
BBB+	8.56%
BBB	4.11%
BBB-	2.54%
BB+	1.20%
BB	1.02%
BB-	1.24%
B+	1.10%
B	1.70%
B-	2.08%
CCC	0.00%

Roughly 76% of banks land in A+/A/A-, ~92% are investment-grade (BBB- and above), ~8% are speculative-grade. The first question every credit analyst asks: "Why is the whole population piled up in the A bucket? Are the ratings broken?"

Short answer: no, the ratings are working correctly. The distribution reflects two real facts about banks - they fail less often than corporate bond issuers, and the letter scale used was originally calibrated to corporate bonds, so when applied to banks the population shifts toward the safe end. This document unpacks both.

Reason 1: Bank failures are intrinsically rarer than corporate-bond defaults

FDIC-insured commercial bank failures in normal cycles run roughly 0.1–0.5% per year (5–20 failures across ~4,500 active commercial banks). Compare to corporate-bond default rates from S&P / Moody's data:

Cohort	Typical annual default rate
FDIC-insured commercial banks (normal cycle)	0.1 – 0.5%
Investment-grade corporate bonds (BBB or better)	0.05 – 0.5%
Speculative-grade corporate bonds (BB and below)	1 – 5%
Distressed (CCC)	25%+

The median bank PD on a typical production run is 0.18%, consistent with the FDIC's historical annual failure rate. The mass of the distribution is correctly at low PD because most banks really do have low failure probability.

Four structural reasons banks are intrinsically safer than corporate-bond issuers:

Regulatory capital requirements, Tier 1 leverage minimums and Basel III floors force a baseline level of solvency that corporate issuers have no equivalent for.
FDIC deposit insurance + Fed liquidity backstop, materially reduces failure modes that come from deposit-run dynamics (which would otherwise account for a large share of bank failures).
Prompt corrective action regime, regulators are required to intervene before insolvency; many "would-have-failed" banks instead get sold or merged.
Universe coverage, every FDIC-insured bank is scored. Corporate-bond rated universes are self-selected (only large, established, debt-issuing companies get rated), which biases their populations toward middle-of-the-distribution credits.

Reason 2: The master scale was calibrated to corporate-bond defaults

The 17-letter overlay (AAA → CCC) on the 1–80 PD score is the "Master Scale" framework adapted from S&P / Moody's. Those agency scales were originally calibrated against corporate-bond default histories. The score-to-letter table is:

Letter	Score range	Width (notches)	Approx PD range
AAA	1 – 10	10	< 1 bps
AA+	11 – 24	14 (widest)	1 – 5 bps
AA	25 – 27	3	5 – 8 bps
AA-	28 – 31	4	8 – 12 bps
A+	32 – 34	3	12 – 18 bps
A	35 – 37	3	18 – 27 bps
A-	38 – 41	4	27 – 45 bps
BBB+	42 – 44	3	45 – 70 bps
BBB	45 – 47	3	70 – 110 bps
BBB-	48 – 51	4	110 – 180 bps
BB+	52 – 54	3	180 – 280 bps
BB	55 – 57	3	280 – 430 bps
BB-	58 – 61	4	430 – 700 bps
B+	62 – 64	3	700 – 1100 bps
B	65 – 67	3	1100 – 1800 bps
B-	68 – 73	6	1800 – 4500 bps
CCC	74 – 80	7	4500+ bps

Two design choices from this table to be aware of:

The AA+ band is unusually wide (14 notches). The agency scale reserves a lot of "safe end" space for AA+ specifically, because in the corporate-bond universe that's where most highly-rated issuers (sovereigns, top-tier corporates) cluster. Almost no banks reach AA+ on this scale because the AA+ PD range (1–5 bps) requires a near-perfect balance sheet.

A is a narrow 3-notch band centered on the bank-population mode. A median bank with PD ~18 bps lands at score ~35–37, which is the A band. With a band only 3 score notches wide and a population whose log-PD is normally distributed around that point, the A band naturally captures the modal mass. A+ and A- pick up the shoulders.

So the combination of (a) bank PDs clustering low, plus (b) the master scale carving narrow A-tier bands at exactly where bank PDs cluster, produces ~76% of banks landing in A+/A/A- and ~92% landing investment-grade overall.

Why corporate-bond letter distributions look completely different

Rated corporate-bond universes typically look roughly bell-shaped around BBB:

Letter tier	Corporate bonds (rated universe)	Banks (PD model)
AAA / AA	~10–15%	< 1%
A	~30–35%	~76%
BBB	~35–40%	~15%
BB	~10–15%	~3%
B / CCC	~5–10%	~5%

Three differences explain this:

Corporate universes are self-selected. Only companies that issue rated debt get ratings, and those tend to be mid-to-large-cap with multi-cycle track records. Marginal issuers either don't get rated or stop issuing debt entirely. The PD model scores the full FDIC-insured universe.
Agency rating methodology rewards a normal-distribution shape. Agencies actively manage rating actions to roughly balance the population across letter buckets, that's part of what makes ratings useful as a relative-ordering tool. The PD model has no such constraint; it produces calibrated absolute PDs.
Corporate defaults span a wider range of PD. Junior unsecured corporate debt can default at 5–10% annual rates in distressed cycles; banks rarely reach those PD levels because regulatory intervention pre-empts them.

The takeaway: comparing the shapes of the two distributions isn't apples-to-apples. The agency letter labels are the same, but the underlying populations they're applied to are structurally different.

Common questions

"Why is everything A?" The 1–80 score is empirically calibrated to FDIC failure rates. The letter is a label on top of that score, using the corporate-bond Master Scale convention. Banks are intrinsically safer than corporate-bond issuers (regulators + FDIC + capital rules) so the population shifts toward the safe-end letters by construction. The actual signal is in the 1–80 score and the calibrated 12-month PD; letters are for at-a-glance familiarity.

"Is the distribution defensible from an MRM standpoint?" The 0.18% median PD aligns with the FDIC's historical annual failure rate of 0.1–0.5%. The shape of the letter distribution is a consequence of (a) bank base rates being low and (b) the Master Scale band widths. The parallel risk_tier field (Severe/High/Elevated/Moderate/Low) uses empirically-anchored band edges, each band's lower edge equals the realized 12-month forward failure rate observed in that band over the v1.2 walk-forward test cohort. Both fields ship in the bundle so the methodology that fits a given use case can be selected.

"How does this compare to a corporate-rating product?" The PD model produces a comparable letter overlay so the output can be read the same way as a corporate bond rating. The underlying population is different (banks are not corporates) so the distribution shape will look different too. The same A rating means roughly the same forward PD whether it's a bank or a corporate bond; what differs is what share of each population sits in each letter.

What "risk_tier" provides as a sanity check

The 5-tier empirical overlay carries direct calibration meaning that letters don't:

Tier	Realized failure rate floor (from v1.2 test set)	Q1 2026 share
Severe	≥ 25%	4.17%
High	≥ 5%	5.75%
Elevated	≥ 1%	18.31%
Moderate	≥ 0.1%	71.76%
Low	< 0.1%	0.01%

For a direct "how risky is the population" claim, the tier shape is the right answer: ~9.9% of banks have meaningful (≥5%) annual failure probability, ~71.8% are in normal operating risk, ~18.3% are watch-list candidates. The letter distribution is a presentation overlay; the tier distribution is the calibrated empirical reading.