pd lgd ead credit risk

PD, LGD, EAD: The Three Numbers That Decide Whether a Bank Says Yes

The credit officer spent twenty minutes explaining why the facility was declined. Revenue was solid, the business had been trading for four years, management had industry experience. Everything looked bankable on the surface. Then she said it: "Your PD came in at 8.2%, and we can't price that profitably at the rate you're willing to pay."

The founder had no idea what that meant. Neither did his advisor. They had been thinking about credit as a binary decision—either you qualify or you don't. But banks don't work that way anymore. They model the probability that you won't pay them back, the fraction they'll lose if you default, and the size of the exposure when it happens. Those three numbers—PD, LGD, and EAD—are the mechanical core of every credit decision a regulated lender makes under Basel capital rules, and most founders never see them until a deal falls apart.

If you're raising debt, restructuring a facility, or trying to understand why one lender said yes and another said no at completely different rates, this is the framework that's running in the background. It's not subjective. It's not vibes. It's a model, and the model has three inputs.

A shorter version of this post appears on LinkedIn, but the mechanism deserves more room than a feed allows. This is how the model actually works, what each parameter means in practice, and where the edge cases live that most explainers skip.

What PD, LGD, and EAD Actually Measure

Probability of Default (PD) is the likelihood, expressed as a percentage, that a borrower will fail to meet their contractual obligations over a defined time horizon, typically twelve months. A PD of 3% means the model estimates a three-in-one-hundred chance that this borrower defaults within the next year. It's forward-looking, not historical, though it's calibrated on historical default data and updated with current financials, sector stress indicators, and macro overlays.

Loss Given Default (LGD) is the percentage of the exposure the lender expects to lose if the borrower does default, after accounting for recoveries from collateral, guarantees, or restructuring. If a lender has a £500,000 exposure and expects to recover £300,000 through asset sales and workout processes, the LGD is 40%. It's always a fraction between zero and one hundred percent, and it varies wildly depending on the quality and liquidity of the security.

Exposure at Default (EAD) is the total amount the lender expects to be owed at the moment of default. For a fully-drawn term loan, EAD is just the outstanding balance. For a revolving credit facility, it's trickier—the borrower might draw more between now and default, so banks apply a credit conversion factor (CCF) to the undrawn portion. If you have a £1 million revolver with £400,000 drawn and the bank applies a 50% CCF, the EAD is £400,000 plus (50% of £600,000), which equals £700,000.

These three parameters feed into the expected loss calculation that drives capital allocation, pricing, and approval:

That expected loss, in turn, determines how much regulatory capital the bank must hold against the exposure under Basel III, which directly affects whether the deal is profitable at the rate you're willing to pay.

How PD Is Built (And Why It's Hard to Move)

PD estimation starts with a scorecard or a statistical model trained on years of default data, segmented by industry, size, and financial profile. The model ingests your financials—leverage ratios, interest coverage, liquidity measures—and outputs a score that maps to a probability band. A strong score might land you in a 1.5% PD bucket; a weaker one might put you at 6% or higher.

But the model doesn't stop there. Banks layer on qualitative adjustments: management quality, market position, sector outlook, concentration risk. If you operate in a sector under stress (say, commercial real estate in a rising-rate environment), your PD gets an overlay even if your standalone financials look clean. If your revenue is heavily concentrated in one customer, that's another adjustment. These overlays are where judgment enters the process, but they're applied systematically, not arbitrarily.

The reason PD is hard to move once assigned is that it's anchored in audited financials and observable sector data. You can't talk your way out of a high PD with a compelling narrative if the numbers don't support it. The only reliable way to improve PD is to improve the underlying credit fundamentals: reduce leverage, diversify revenue, build liquidity buffers, extend your trading history. All of those take time.

One edge case worth naming: PD is typically estimated on a through-the-cycle (TTC) basis for regulatory capital purposes, meaning it's calibrated to smooth out short-term volatility and reflect long-run default rates. That makes it less responsive to sudden improvements in performance than you might expect. A single strong quarter won't move your PD materially. A sustained trend over multiple reporting periods will.

LGD: The Recovery Assumption That Changes Everything

LGD is where security structure matters most. A senior secured lender with a first charge over tangible assets—property, equipment, inventory with a liquid secondary market—might model an LGD of 25% to 35%. An unsecured lender, or one secured against intangible assets like goodwill or IP with uncertain realisable value, might model 60% to 80%.

The difference is enormous. Take a £1 million exposure with a 5% PD. If LGD is 30%, expected loss is £15,000. If LGD is 70%, expected loss is £35,000. That gap flows straight through to capital requirements and pricing. A lender can't profitably price a 70% LGD exposure at the same margin as a 30% LGD one, even if the PD is identical.

LGD estimation involves modelling the recovery process: how long it takes to enforce security, what the assets fetch at auction or in a distressed sale, what the legal and workout costs are, and what other creditors rank ahead of you. For property-backed lending, banks often apply haircuts to current valuations to reflect forced-sale discounts—30% to 40% haircuts are common. For receivables-based lending, they model collection rates and dilution.

One practical edge case: LGD can vary depending on the default scenario. A borrower who enters an orderly administration with cooperative management might yield a much higher recovery than one who experiences a sudden liquidity failure and ceases trading abruptly. Some banks model a downturn LGD (the loss in a stressed scenario) separately from a base-case LGD, and the downturn figure is what drives capital calculations under the Basel advanced internal ratings-based (A-IRB) approach.

If you're negotiating a facility and the lender is asking for additional security or a guarantee, they're trying to lower LGD. That's not a signal that they don't trust you—it's a signal that the expected loss at the current LGD makes the deal uneconomic. Offering better security can sometimes unlock a lower rate or a larger facility size, because it changes the math directly.

EAD and the Credit Conversion Factor Problem

EAD is straightforward for term loans: it's the outstanding balance at default, which you can estimate with reasonable confidence. But for revolving facilities, overdrafts, and undrawn commitments, EAD becomes a forecast problem. How much of the undrawn line will the borrower draw before they default?

This is where the credit conversion factor (CCF) comes in. The CCF is an empirically-derived percentage that estimates the proportion of undrawn commitments that will be drawn by the time of default. If historical data shows that borrowers in financial distress tend to draw down 50% of available undrawn lines before defaulting, the bank applies a 50% CCF.

For example, consider a £2 million revolving credit facility with £800,000 currently drawn. The undrawn portion is £1.2 million. If the bank applies a 40% CCF, the EAD is:

That £1.28 million is the figure used in the expected loss calculation, not the £800,000 currently drawn. The difference matters because the bank has to hold capital against the full EAD, not just the drawn balance.

CCFs vary by facility type and borrower quality. Committed facilities typically have higher CCFs than uncommitted ones, because borrowers know they can draw on a committed line even in distress. Higher-quality borrowers sometimes get lower CCFs because they're less likely to draw aggressively as a last resort before default.

One edge case that catches people: if you have multiple facilities with the same lender (say, a term loan and a revolver), the bank models EAD across the whole relationship, not facility by facility. If you're in distress, you might max out the revolver to service the term loan, which increases total EAD even if the term loan balance is amortising. Relationship-level EAD can be higher than the sum of individual facility EADs, and that affects pricing and structure.

How Expected Loss Drives Pricing and Approval

Once the bank has PD, LGD, and EAD, it calculates expected loss and compares it to the revenue the deal will generate. The margin you pay has to cover expected loss, the cost of funds, operating costs, and a return on the regulatory capital the bank must hold.

The regulatory capital requirement is calculated under the Basel framework, either using a standardised approach (which applies fixed risk weights to different exposure classes) or an internal ratings-based approach (which uses the bank's own PD, LGD, and EAD estimates). Under A-IRB, the capital requirement is a function of PD and LGD, adjusted for maturity and correlation assumptions. Higher PD or higher LGD means more capital, which means the deal needs a higher margin to generate the same return on equity.

Here's the practical implication: if your PD is high and your LGD is high (because you don't have strong security), the expected loss might be so large that no margin you're willing to pay makes the deal profitable for the bank. That's when you get declined, not because the bank thinks you're going to default, but because the probability-weighted loss, multiplied by the capital charge, doesn't pencil out.

This is also why different lenders quote wildly different rates for the same borrower. A bank with a lower cost of capital, a more optimistic sector view, or a different PD model might price the same risk at 200 basis points less than a competitor. The underlying credit quality hasn't changed—the model inputs and the capital efficiency have.

Where the Model Breaks (And What to Do About It)

PD, LGD, and EAD are powerful tools, but they're backward-looking models applied to forward-looking decisions. They struggle with structural changes—new business models, sector disruption, one-off events that don't fit historical patterns. A software business with high gross margins and recurring revenue might get penalised by a PD model trained on lower-margin industrial comparables, even though the cash generation profile is fundamentally different.

The model also assumes rational, continuous operation up to default. It doesn't handle well the case where a borrower could service debt comfortably but chooses to default for strategic reasons (rare in SME lending, more common in distressed real estate). And LGD estimates can be wildly wrong in a systemic downturn when asset markets freeze and recovery timelines stretch.

If you're on the borrowing side and you think the model is mispricing your risk, your options are limited but real. You can improve the inputs: strengthen the balance sheet, add security, bring in a guarantor, provide more frequent reporting to reduce information lag. You can also shop the deal to lenders with different model calibrations—challenger banks and specialist lenders sometimes have more granular scorecards that capture nuances the big banks' models miss.

If you're advising a client through a credit process and the pricing or structure doesn't make sense, ask the lender what PD band they've assigned and what LGD they're assuming. Most won't disclose the exact figures, but they'll often tell you the broad category (investment grade, sub-investment grade, high default risk) and whether the issue is PD or LGD. That tells you where to focus: if it's PD, you need to improve credit fundamentals over time; if it's LGD, you need better security or structural protections.

Why This Matters More Than Ever

Basel III and its successors have made PD, LGD, and EAD the universal language of credit risk in regulated lending. Every bank above a certain size is running some version of this model, even if the implementation varies. That means credit decisions are more consistent, more transparent in theory, and more mechanical than they used to be. It also means they're harder to influence with relationship capital or a compelling story if the model says no.

For borrowers, the implication is that improving your credit profile is now a quantifiable exercise. You're not trying to impress a credit officer—you're trying to move a model. That requires knowing which inputs matter most (usually leverage and interest coverage for PD, security quality for LGD) and focusing your efforts there.

For advisors and finance teams, it means you can reverse-engineer why a deal was priced the way it was, or why it was declined. If a lender quotes a margin that seems high, you can estimate the implied expected loss and work backwards to see whether they're assuming a high PD, a high LGD, or both. That gives you a negotiating position: if you think their LGD assumption is too conservative, you can offer additional security and ask for a re-price.

The model isn't perfect, but it's the model. Understanding PD, LGD, and EAD won't guarantee you get the facility you want at the rate you want, but it will tell you exactly why you didn't, and what you'd need to change to get a different answer next time.

If you're structuring a raise, renegotiating a facility, or trying to understand why your credit pricing doesn't match your risk profile, this is the framework to work within. The numbers are built from assumptions, and assumptions can be challenged, but only if you know which ones are driving the outcome.

If this is the conversation you need to have about a live deal, I keep time available for exactly this kind of diagnostic—calendly.com/muhammed-adediran/30min.