Loan Prepayment Prediction: Features, Maths and RBI Rules
Last reviewed: November 2026
Loan prepayment prediction is a model that scores each loan for the chance it will be part-prepaid or foreclosed in the next few months, so a lender can plan cash and protect yield before the money leaves. Since the Reserve Bank of India (RBI) bars foreclosure charges on most floating-rate retail loans, the warning has to come from data rather than from a penalty clause.
This guide is for chief financial officers (CFOs) and heads of analytics at banks, non-banking financial companies (NBFCs) and housing finance companies. It gives a risk-band decision matrix, a feature list, the conditional prepayment rate (CPR) formula, yield and retention maths with worked numbers, and the build approach used by Digiqt's artificial intelligence (AI) development team.
Key Takeaways
- Score every loan monthly for the probability of prepayment in the next six months.
- Act only where the maths says the retention offer earns more than it gives away.
- Respect the RBI's ban on pre-payment charges for covered floating-rate loans; retention must be value-led.
- Keep reason codes with every score so relationship managers can explain the call.
What Should Lenders Do With Each Loan Prepayment Risk Band?
Lenders should map each loan prepayment risk band to one action and one owner. The probabilities in this decision matrix are illustrative and should be set on your own book.
| Risk Band (6-Month Probability) | Typical Signals | Action | Owner |
|---|---|---|---|
| Very high (above 50%) | Foreclosure statement requested, other lenders' bureau enquiries | Retention call within 48 hours; reprice within policy | Retention desk |
| High (30–50%) | Rate gap above 1%, seasoning past 24 months | Proactive rate review or top-up offer | Branch relationship manager |
| Medium (10–30%) | Rising account credits, past part-prepayments | Service call; no price change | Customer service |
| Low (below 10%) | Recent loan, small rate gap | No action | None |
| Any band, prepayment from asset sale or claim | Property sale, insurance payout | Process smoothly; no retention spend | Operations |
Feed the bands into treasury's liquidity forecast every month, not only into sales follow-up. Review the band cut-offs every quarter against what actually prepaid.
What Is Loan Prepayment Prediction?
Loan prepayment prediction estimates two events: part-prepayment, where the borrower pays down principal, and foreclosure, where the borrower closes the loan early, often through a balance transfer. Both cut future interest income.
Loan prepayment prediction vs loan foreclosure processing: processing handles the closure once the borrower asks, while prediction flags the risk months earlier. The most common trigger is a competitor's offer, covered in balance transfer loan processing.
Is early repayment eating into your yield plan?
Discuss a Prepayment Model With Digiqt's AI Team
Share last year's foreclosure list and ask us to test prepayment signals on your book.
What Do RBI Rules Say About Pre-payment Charges?
RBI rules ban pre-payment charges on most floating-rate loans to individuals, so prediction is now the main defence for yield. Paragraphs 33 and 34 of the RBI's NBFC Responsible Business Conduct Directions, 2025 (RBI/DOR/2025-26/362, 28 November 2025) set the rules.
- Loans sanctioned or renewed on or before 31 December 2025: no pre-payment charges on floating-rate term loans to individuals for non-business purposes.
- Loans sanctioned or renewed from 1 January 2026: no pre-payment charges on any floating-rate loan to individuals for non-business purposes.
- Floating-rate business loans to individuals and micro and small enterprises (MSEs): no charges at an NBFC-Upper Layer, and none at an NBFC-Middle Layer up to a sanctioned amount of ₹50 lakh.
- These apply whatever the source of funds and without any lock-in period; dual-rate loans follow the rate type at the time of pre-payment.
- Any charge that remains allowed must be disclosed in the sanction letter, the loan agreement and the key facts statement (KFS).
Banks follow the parallel RBI (Pre-payment Charges on Loans) Directions, 2025, issued on 2 July 2025. For a floating rate home loan or loan against property to a salaried borrower, that means the exit is free, so retention has to rest on price, service and speed.
Which Features Predict Loan Prepayment?
Ten features predict most loan prepayment, and nearly all are already in your loan management system (LMS) or credit bureau pulls.
| Feature | Why It Predicts Prepayment | Source |
|---|---|---|
| Rate gap: borrower rate minus current offer rate | Bigger gap, stronger reason to switch | LMS, pricing grid |
| Seasoning: months on book | Prepayment rises after the early years | LMS |
| Remaining tenor | Long tenors give more time and reason to refinance | LMS |
| Bureau enquiries by other lenders | Direct sign of shopping for a balance transfer | Credit bureau |
| Foreclosure statement requested | Strongest short-term signal | Service request log |
| Past part-prepayments | Habit repeats | Repayment history |
| Growth in account credits | Borrower can afford to pay down | Bank statement, mandate data |
| Top-up eligibility | A top-up can retain or trigger a switch | Credit policy rules |
| Product and rate type | Floating-rate retail loans carry no exit fee | Loan master |
| Sourcing channel | Some channels re-pitch borrowers to rivals | Origination data |
Leave out religion, caste, gender and similar personal attributes, and test whether any feature acts as a proxy for them before go-live.
Combine features with vintage analysis of the loan portfolio to see how prepayment curves differ by cohort.
How Does a Simple Loan Prepayment Prediction Model Work?
A simple loan prepayment prediction model works in six steps that a small analytics team can follow.
- Snapshot: take one row per loan per month with the features above as of that month-end.
- Label: mark 1 if the loan closed or prepaid at least 10% of principal in the next six months, else 0.
- Baseline: fit a logistic regression for a transparent first model.
- Challenger: fit gradient-boosted trees and keep it only if it beats the baseline.
- Validate: use out-of-time validation on later months, checking top-decile lift and calibration.
- Explain: attach the top three reason codes to every score, such as "rate gap 1.4%" or "two bureau enquiries in 30 days".
The RBI's FREE-AI (Framework for Responsible and Ethical Enablement of Artificial Intelligence) Committee Report of 13 August 2025 lists "Understandable by Design" among its seven sutras, which is why reason codes belong in the first release.
How Much Yield Does Unexpected Loan Prepayment Cost?
Unexpected loan prepayment costs yield through repricing, idle cash and faster write-off of acquisition costs, and the numbers below are illustrative. Start by converting monthly prepayment into an annual rate.
Formula: CPR = 1 − (1 − SMM)^12, where SMM, the single monthly mortality, is prepaid principal in a month ÷ principal outstanding at its start.
| SMM | CPR |
|---|---|
| 1.0% | 11.4% |
| 1.5% | 16.6% |
| 2.0% | 21.5% |
Now take a ₹1,000 crore book yielding 11% that planned for a 15% CPR but ran at 22%.
- Extra prepaid principal: ₹1,000 crore × (22% − 15%) = ₹70 crore.
- Repricing loss when replaced at 1% lower yield: ₹70 crore × 1% = ₹0.70 crore a year.
- Idle cash for one month at 6% instead of 11%: ₹70 crore × 5% ÷ 12 = ₹0.29 crore.
- Unamortised acquisition cost written off early at 0.6%: ₹70 crore × 0.6% = ₹0.42 crore.
- Total first-year hit: about ₹1.41 crore of net interest income and fee income.
Run the yield impact separately for each product, because home loans, business loans and vehicle loans prepay on very different curves. Item 4 follows the effective interest rate method described in EIR amortisation for NBFCs.
When Is a Retention Offer Worth Its Cost?
A retention offer is worth its cost when the spread kept from borrowers it saves exceeds the concession paid to borrowers who would have stayed anyway.
Formula: Net value = (p0 − p1) × B × (s − c) × T − (1 − p0) × B × c × T, where p0 and p1 are prepayment probabilities without and with the offer, B is balance, s is spread, c is the rate cut and T is years (undiscounted).
| Borrower | p0 → p1 | Spread Kept | Concession Given Away | Net Value |
|---|---|---|---|---|
| High risk, ₹40 lakh, s 3%, c 0.5%, T 3 | 60% → 30% | ₹90,000 | ₹24,000 | +₹66,000 |
| Low risk, same terms | 10% → 5% | ₹15,000 | ₹54,000 | −₹39,000 |
The same offer earns money on the first borrower and loses it on the second, which is the whole case for scoring.
How Does Digiqt's AI Development Team Build Prepayment Prediction?
Digiqt's AI development team builds prepayment prediction as a custom machine learning model trained on your own loan, repayment and service history. The AI development services cover custom machine learning modelling, explainable AI and analytics, starting with a discovery sprint and a dataset assessment.
The team then runs model selection and validation, wires scores and reason codes into dashboards or the retention desk's queue, and sets up monitoring and retraining. For a ready reference design, see Digiqt's prepayment risk forecasting AI agent.
Pricing, retention offers and policy exceptions stay with your credit and finance teams.
What Do CFOs Ask About Loan Prepayment Prediction?
Can We Add a Fee to Offset Prepayment Losses?
You cannot add a fee where the RBI bans pre-payment charges, and paragraph 34 also bars charges not disclosed in the sanction letter, agreement and KFS. Fixed-rate and larger business loans may carry disclosed charges under your board-approved policy.
How Often Should a Prepayment Model Be Retrained?
A prepayment model should be checked monthly and retrained at least once a year, or sooner when the rate cycle turns. Rate cuts change borrower behaviour faster than any other factor.
Does Prepayment Prediction Help With Securitisation?
Prepayment prediction helps securitisation because investors price pools on expected CPR. Pool-level scores strengthen the assumptions in your securitisation data tape.
Ready to protect yield without pre-payment charges?
Build an Explainable Prepayment Model With Digiqt
Bring two years of loan history, or visit Digiqt to plan a prepayment risk pilot with our analytics team.
References
- RBI: Non-Banking Financial Companies – Responsible Business Conduct Directions, 2025
- RBI: Pre-payment Charges on Loans Directions, 2025, 2 July 2025
- RBI: FREE-AI Committee Report, 13 August 2025
- NIST: AI Risk Management Framework
- RBI: Non-Banking Financial Companies – Credit Facilities Directions, 2025


