Credit Cost in Loan Pricing: A Practical, End-to-End Explanation

At a high level, loan pricing looks like this:

We have fair idea about all the components except for credit cost.

Let’s assume everything except credit cost is already known.

This article is only about how to think about, estimate, and sanity-check credit cost — in a way that actually works in real lending.

What credit cost really represents

Credit cost answers one very boring but very important question:

“Out of all the money I lend, how much will I not get back?”

It is not:

GNPA,
write-offs,
stress loss,
worst-case loss.

It is the average, expected loss baked into pricing.

If you don’t price this correctly, everything else in loan pricing becomes meaningless.

The irreducible formula (don’t fight it)

Credit cost has only two moving parts:

Credit Cost = PD × LGD

Where:

PD (Probability of Default)
Out of 100 similar loans, how many will default at least once?
LGD (Loss Given Default)
If a loan defaults, what % of the outstanding amount will I ultimately lose?

That’s it.

All the “science” is about estimating these two numbers honestly.

Start from terminal loss, not annual numbers

Most confusion starts because people jump straight to “annual credit cost”.

Don’t do that.

Start with the terminal outcome.

Terminal loss (or cumulative default)

Ask:

“Out of all loans I originate, how many eventually become bad loans?”

Example:

1,000 MSME loans originated
65 loans eventually cross 90+ DPD at least once. (I'm assuming 90+DPD as default definition)

This 6.5% is the anchor.
Everything else is derived from this.

Why timing matters: enter the vintage curve

Knowing how many loans fail is not enough.

You also need to know when they fail.

That’s because:

losses that happen early are far more damaging,
interest income is earned slowly,
recoveries take time.

This is why lenders use vintage curves.

What a vintage curve shows

A vintage curve answers:

“After origination, how does default accumulate over time?”

Example (illustrative):

Two things immediately stand out:

Most damage happens early
After a point, the curve flattens

This shape is extremely common in retail MSME lending.

The idea of a “risk window” (this is critical)

Notice something important in the table above:

By 24 months, ~90% of lifetime defaults have already happened
After that, the curve barely moves

So even if the loan tenure is 3 or 4 years, the economic risk window is much shorter.

This leads to an important shift in thinking:

PD for pricing should be aligned to the risk window, not the contractual tenure.

Annualising credit risk the right way

Now let’s convert the vintage curve into something usable for pricing.

From the example:

Terminal default ≈ 6.5%
~6% happens by month 24

So we say:

Risk window ≈ 2 years
Annualised PD ≈ 6.5% ÷ 2 = 3.25% per year

This is not pessimism.
This is matching pricing to when capital is actually at risk.

A common mistake is to divide by full tenure just because the loan is 3 or 5 years long.
That assumes defaults are evenly spread. They almost never are.

Where LGD comes in (and why it’s often wrong)

PD tells you how often things go wrong.
LGD tells you how painful it is when they do.

LGD depends on reality, not documentation:

How fast can you recover?
How enforceable is the collateral?
How disciplined is your recovery process?
How cooperative are borrowers in stress?

Typical MSME heuristics (very rough):

LGD should usually be product-level, not borrower-level.
Changing LGD borrower-by-borrower often creates false precision.

Putting PD and LGD together (worked example)

Let’s combine everything.

Assume:

Terminal default = 6.5%
Risk window = 2 years
Annual PD = 3.25%
LGD = 75%

This 2.44% is the “X” that goes into loan pricing.

Once this is clear, pricing stops being emotional.

Different borrowers, different credit cost

Not all borrowers behave the same.

Instead of pretending everyone has the same PD, lenders use risk bands.

Example:

Notice:

LGD remains same at product level but you may verify it in your dataset.
PD does most of the work

This keeps the system simple and explainable.

Where does the data come from?

Three common sources:

1) Your own portfolio (best)

clean,
aligned to your behaviour,
but takes time to mature.

2) Bureau / industry data (useful, but dangerous)

reflects peers’ borrower mix and peers’ collection strength,
silently assumes you can replicate their execution.

If your collections are weaker, your realised credit cost will be higher.

3) Peer disclosures / conversations

helpful for anchoring,
but often optimistic in public narratives.

Never use any one source blindly.

The hidden assumption inside every vintage curve

A crucial warning:

A vintage curve is not just borrower behaviour.

It is borrower behaviour × lender behaviour.

If your collections are:

slower,
less consistent,
less credible,

your realised PD will be worse than the borrowed curve, even with identical borrowers.

So when using industry data, you must ask:

Can we match their reminder speed?
Can we match their escalation discipline?
Can we match their recovery seriousness?

If not, adjust credit cost upward.

Why buffers exist (and why they’re not greed)

Even good models are wrong sometimes.

Buffers exist to absorb:

data gaps,
operational slippage,
macro shocks,
learning-phase mistakes.

Buffers are not profits.
They are humility in numeric form.

Final sanity check before locking credit cost

Before you freeze credit cost into pricing, ask one question:

“If defaults are 25% worse than expected for one year, does the business survive without hurting capital?”

If yes, your credit cost assumption is probably sane.
If no, it isn’t.

Remember

Credit cost is not about being aggressive or conservative.

It is about being honest:

about borrowers,
about timing,
about recoveries,
and about your own execution capability.

Once you are honest about that, loan pricing becomes boring.

And boring is exactly what you want in lending.

Search This Blog

the-playbook.in