How Basel IRB Credit-Risk Algorithms Turn PD and LGD into RWA: Asset Correlation, Maturity Adjustment, the Vasicek Model and Model Limits

Reader question: If a bank has already estimated a borrower’s probability of default and loss given default, how does the Basel internal ratings-based framework turn those inputs into regulatory risk-weighted assets — and what assumptions can make that calculation misleading?

The short answer is that the IRB framework does not simply multiply probability of default by loss given default. It uses a nonlinear credit-portfolio model. A borrower’s PD, facility LGD, EAD, effective maturity and prescribed asset-correlation relationship are combined to estimate an unexpected-loss capital requirement. That capital requirement is then converted into risk-weighted assets. The calculation is mathematically elegant, but its output is only as meaningful as its inputs, classification, assumptions and supervisory constraints.

What this page does — and does not do

This is a mathematics and computational-risk article. It explains the mechanism behind a major bank-capital algorithm. It is not a guide to lending decisions, not a credit recommendation, and not a substitute for the Basel Framework or a jurisdiction’s binding prudential rules. National implementation can differ, and the Basel Committee’s standards are implemented through local law and supervision.

The five inputs that matter first

• PD — probability of default: the estimated probability that the obligor defaults over the relevant horizon.
• LGD — loss given default: the proportion of exposure expected to be lost if default occurs, after recoveries and relevant costs under the applicable framework.
• EAD — exposure at default: the exposure amount to which the capital calculation is applied.
• M — effective maturity: a maturity measure used for many non-retail exposures.
• R — asset correlation: a prescribed relationship that represents how strongly the obligor’s asset value is assumed to move with the common systematic risk factor.

These are not interchangeable. PD answers “how often?”, LGD answers “how severe if it happens?”, EAD answers “how much is exposed?”, and maturity and correlation change how strongly systematic stress affects the capital requirement.

The mathematical engine

For a standard corporate exposure, the Basel IRB risk-weight function is built around a one-factor Gaussian credit model. In simplified notation, the capital requirement per unit of exposure can be written as:

K = [LGD × N((1−R)^−1/2G(PD) + (R/(1−R))^1/2G(0.999)) − PD × LGD] × maturity adjustment.

Here, N is the standard normal cumulative distribution function and G is its inverse. The appearance of G(0.999) is important: the formula is not asking what happens in an average year. It is placing the obligor inside a severe systematic-risk state in the model and asking how much unexpected loss must be absorbed beyond expected loss.

For corporate exposures, the prescribed correlation is itself a function of PD. In the classic Basel expression:

R = 0.12 × A + 0.24 × (1−A), where A = (1−e^−50PD)/(1−e⁻⁵⁰).

The maturity adjustment uses another PD-dependent term, commonly written b = [0.11852 − 0.05478 ln(PD)]², and then scales K according to effective maturity. Different exposure classes use different relationships, so this corporate formula should not be copied blindly onto retail, mortgage, bank or other exposure classes.

Once K has been calculated, the familiar conversion is:

RWA = 12.5 × K × EAD.

The factor 12.5 is the reciprocal of the 8% minimum capital ratio used in this conversion step. It converts a capital requirement into an equivalent risk-weighted-asset amount. This does not mean a bank’s final capital requirement is simply 8% of that number: buffers, other risk types, leverage constraints, output floors and jurisdiction-specific requirements can also matter.

A worked numerical example

Take an illustrative corporate exposure with PD = 1%, LGD = 45%, EAD = $1,000,000 and effective maturity M = 2.5 years. Using the Basel corporate correlation relationship gives R of about 19.3%. The PD-dependent maturity coefficient b is about 0.1375. At M = 2.5, the numerator of the maturity adjustment is neutral relative to its reference point, while the denominator still matters.

Substituting the values into the risk-weight function gives K of roughly 7.39% of EAD. Multiplying by 12.5 gives illustrative RWA of about $923,000.

This example is useful because it exposes a common mistake: expected loss is not the same quantity as regulatory unexpected-loss capital. PD × LGD here is only 0.45% before EAD is applied. The capital formula is much larger because it is conditioning on a severe common-risk state and then removing the expected-loss component.

Why correlation changes the answer so much

If defaults were independent, a very large diversified portfolio would become unusually predictable: idiosyncratic losses would average out. Real credit portfolios are not independent. Recession, financing conditions, property prices, commodity shocks and other common forces can make many obligors deteriorate together.

The IRB model represents that dependence using a common systematic factor. Higher assumed asset correlation means a bad systematic state can pull more borrowers toward default simultaneously. The capital requirement therefore rises even if the individual borrower’s PD and LGD stay unchanged.

That makes correlation a structural assumption, not a cosmetic parameter. It is one reason a single-obligor interpretation of the formula is incomplete: the equation is a portfolio-capital construction being applied at exposure level.

What the Vasicek-style model is assuming

• A very granular portfolio so that much idiosyncratic risk can diversify away.
• A single systematic risk factor captures the dependence structure used by the formula.
• Defaults can be represented through a latent asset-value threshold model.
• The normal-distribution machinery is an adequate approximation for the regulatory purpose.
• PD, LGD, EAD and maturity are measured consistently with the framework.
• Portfolio concentration and dependence not captured by the formula are addressed elsewhere through supervision, stress testing or additional controls.

These assumptions explain both the strength and weakness of the model. It is tractable enough to produce consistent regulatory calculations across huge books, but it is not a literal simulation of every firm, industry, country and feedback loop in a real crisis.

Failure modes and counterexamples

Input error: a beautifully coded formula cannot rescue a badly calibrated PD or an optimistic LGD. A 1% PD entered where the defensible estimate is 2% is not a rounding error; it changes both the direct default term and the nonlinear systematic-risk transformation.

Wrong exposure class: using the corporate correlation function for a retail exposure is a category error. Similar-looking formulas can have materially different calibration.

Concentration: a portfolio consisting of a few highly connected borrowers violates the intuition behind asymptotic granularity. The formula may remain the regulatory calculation, but concentration analysis is still needed.

Regime break: a model estimated on a period with stable recoveries can fail when collateral markets seize up, legal recovery times lengthen or a new shock changes default relationships.

False precision: reporting RWA to the nearest dollar can create an illusion of certainty. The input parameters are estimates, and some are constrained by supervisory floors or prescribed relationships.

Diagnostics: how to test the weak links

• Recalculate independently. Reproduce K and RWA from a second implementation and compare intermediate values, not only the final RWA.
• Unit-test the tails. Test very low and high PDs, maturity boundaries, LGD limits and exposure-class transitions.
• Shock one input at a time. Plot RWA against PD, LGD and maturity. Unexpected non-monotonic behaviour may reveal coding or classification errors.
• Backtest the inputs. Compare realised defaults with PD grades and realised recoveries with LGD estimates over suitable periods.
• Check population drift. A model can remain mathematically correct while the borrower population changes underneath it.
• Compare with alternative views. Standardised RWA, stress-loss estimates, concentration measures and portfolio simulations should not be identical, but unexplained divergence deserves investigation.

What would falsify confidence in the result?

Confidence should fall if realised default rates persistently miss the calibration range, if recovery outcomes contradict LGD assumptions, if exposure classifications are wrong, if the calculation cannot be reproduced from documented inputs, or if the model produces unstable changes that cannot be explained by economics or rules. A regulatory formula is authoritative only when the correct rule, version, inputs and scope have been applied.

How this connects to the surrounding mathematics

The IRB formula sits between several other topics already explored on Bukit Timah Tutor. Start with credit-scoring probability and calibration, then examine loss-given-default estimation. The resulting risk parameters connect to the broader bank-capital constraint. Finally, even internally modelled RWA can interact with the Basel output floor.

Verification and update triggers

A robust implementation should record the Basel Framework version, national implementation, exposure class, PD/LGD/EAD source, maturity rule, input floors, model approval status and calculation date. Re-check the article if the Basel Committee changes the consolidated credit-risk chapter, if a jurisdiction changes its IRB implementation, if supervisory input floors or exposure eligibility change, or if a material model redevelopment changes the risk parameters feeding the equation.

Primary and high-quality references

• Basel Committee on Banking Supervision, An Explanatory Note on the Basel II IRB Risk Weight Functions.
• Basel Committee, Basel III: Finalising post-crisis reforms, now integrated into the consolidated Basel Framework.
• European Banking Authority, Guidelines on PD estimation, LGD estimation and treatment of defaulted assets.
• Basel Committee, Basel III implementation dashboard and RCAP material for jurisdictional implementation context.

Educational boundary: This article explains regulatory mathematics. It does not tell a bank how to obtain model approval, tell a lender whether to extend credit, or provide investment or financial advice.