How to Simulate a Loan Portfolio Using Behaviour (Roll Rates)

 This is not a textbook explanation.

This is how I actually think about portfolios when someone asks:

“What will my book look like after 6 months?”
“Why are my NPAs rising even though disbursements are stable?”
“What happens if my collection efficiency drops slightly?”

Let’s start from first principles and build our way up.

Step 1: The easy part. Total AUM forecasting

If all you care about is overall AUM, life is simple.

You take:

• Opening AUM

• Add fresh disbursements

• Subtract normal amortisation

• Subtract prepayments

And you get closing AUM.

For example:

Particulars	Amount (₹ Cr)
Opening AUM	100
Disbursements	+20
Normal amortisation	−8
Prepayments	−2
Closing AUM	110

This works fine if you don’t care where the AUM is sitting.

But lending is never that simple.

Step 2: Why bucket-wise AUM actually matters

Most real decisions depend on where the AUM sits, not just how much.

Bucket-wise AUM is required for:

• ECL provisioning

• Understanding delinquency build-up

• Estimating future write-offs

• Planning collection team capacity

• Stress testing the portfolio

A ₹110 crore book where:

• ₹108 crore is current

• ₹2 crore is overdue

is very different from one where:

• ₹95 crore is current

• ₹15 crore is overdue

Same AUM. Completely different risk.

This is where portfolio simulation comes in.

Step 3: Think in buckets, not loans

Instead of thinking loan by loan, think in behaviour buckets.

A very simple structure:

At any point in time, your entire AUM is distributed across these buckets.

Step 4: What actually moves a portfolio? Behaviour.

Every month, loans do one of four things:

1. Stay in the same bucket

2. Roll forward (get worse)

3. Roll backward (get better)

4. Exit the system (amortisation or prepayment)

We capture this behaviour using roll rates.

Step 5: A simple roll rate matrix (intuition first)

Let’s assume this behaviour for one month:

Exit includes normal EMI amortisation plus prepayments.

This table alone tells you how your portfolio breathes.

Step 6: One-month live simulation (do this yourself)

Let’s start with a clean book.

Opening AUM (Month 0)

Now apply the roll rates for one month.

Month 1 movement from Bucket 0

From ₹100 crore in bucket 0:

• Stays current: 94 → ₹94

• Rolls to 1–30: 4 → ₹4

• Exits: 2 → ₹2

So after Month 1:

Already, the portfolio has “aged”.

Step 7: Add disbursements (important)

Real portfolios are not static.

Assume fresh disbursement of ₹10 crore, all entering bucket 0.

Updated Month 1 closing:

Exited ₹2 crore is gone forever.

Step 8: Month 2 simulation (now it gets interesting)

Now you apply roll rates bucket-wise.

Bucket 0 (₹104)

• 94% stays → 97.76

• 4% to bucket 1 → 4.16

• 2% exit → 2.08

Bucket 1 (₹4)

• 30% cures to 0 → 1.20

• 50% stays → 2.00

• 15% worsens to 2 → 0.60

• 5% exit → 0.20

Now aggregate:

This is actual portfolio physics, not accounting math.

Step 9: Extend this logic to 12 months

If you repeat this process every month:

• Apply roll rates to each bucket

• Add fresh disbursements

• Remove exits

• Track bucket balances

You get a full AUM trajectory.

Below is a simplified 12-month illustration (rounded):

Notice something important.

Step 10: A key insight most people miss

A portfolio can grow and still deteriorate.

Why?

Because:

• Older cohorts dominate behaviour

• Delinquencies from mature loans outweigh fresh inflow

• Disbursement growth is not fast enough to “dilute” bad behaviour

This is why portfolio age matters as much as portfolio size.

Step 11: What you can do with this simulation

Once you can simulate bucket-wise AUM, you can:

• Apply PD and LGD to estimate ECL

• Forecast collections and shortfalls

• Simulate write-offs beyond 90 dpd

• Stress test roll deterioration

• Evaluate growth vs stability trade-offs

All without fancy tools.

Just logic and discipline.

A quick note on secured vs unsecured portfolio simulation

So far, everything we discussed works cleanly for unsecured portfolios.

But secured portfolios behave differently once loans start slipping beyond early delinquency.

Let’s understand how and where the logic changes.

What changes in a secured portfolio?

In unsecured lending, once an account hits 90+ dpd, the economic outcome is simple:

You either recover some cash later or you don’t.

In secured lending, there is another layer in between:

• Settlement

• Repossession

• Sale of collateral

• Legal and time costs

That means 90+ dpd is not the end state.

It is just the beginning of another process.

Conceptually, add one more stage after delinquency

For secured portfolios, your simulation should mentally look like this:

Current → Delinquent → NPA → Resolution → Loss / Recovery

Instead of assuming:

“90+ dpd = X% loss”

You break it down further.

Adjusting the simulation logic (without changing the structure)

You do not need a new roll-rate framework.

You only need to add resolution assumptions once accounts enter deep delinquency.

For example, assume this for loans entering 90+ dpd:

Now attach economics to each.

How losses actually occur in secured lending

Let’s say ₹10 crore enters 90+ dpd.

1. Settlement losses

• Settlement at 70% of POS

• Loss = 30%

₹4 crore × 30% = ₹1.2 crore loss

2. Repossession losses

• Recovery after sale = 65%

• Includes haircut, legal, time costs

₹3.5 crore × 35% = ₹1.225 crore loss

3. Legal / unresolved

• Assume conservative recovery of 40%

₹2.5 crore × 60% = ₹1.5 crore loss

Total expected loss

So instead of assuming a flat LGD, you’ve derived it from behaviour.

Effective LGD here = ~39%, not a guessed number.

Where this plugs into your simulation

In a forecast:

• Roll rates take you up to 90+ dpd

• Resolution assumptions take you from 90+ dpd to loss

• Loss timing depends on settlement and repossession cycles, not EMI schedules

This is why secured portfolio forecasts often:

• Look stable early

• Then suddenly show lumpy losses

• Even when GNPA looks under control

Why most simulations get this wrong

Most models either:

• Treat secured loans like unsecured ones, or

• Assume collateral magically converts to cash

In reality:

• Time kills value

• Legal friction is real

• Behaviour matters even after default

If you don’t model this explicitly, your forecasts will always feel “off”.

Closing thought

If you can simulate:

• Bucket-wise behaviour

• Growth versus ageing

• And resolution economics

You are no longer “forecasting”.

You are thinking like a lender.

Want to go deeper?

If you want to learn:

• How secured and unsecured simulations differ in practice

• How to model settlement haircuts and repossession losses

• How these flows hit P&L, balance sheet, and capital

• And how to build a full NBFC model end to end

You can join my NBFC financial modelling course, where we build all of this step by step.