← Back to blog

November 3, 2026 · 3 min read

IFRS 9 provisioning on Fineract: what the core computes and what you have to build

Fineract ships aging-bucket provisioning: day ranges, a percentage and GL accounts. That is incurred loss, not expected loss. What it covers, what it does not, and how to complete it without forking the core.

Accounting · IFRS 9 · Risk

It is the question that comes up in the first meeting with the risk team: “does Fineract calculate provisions?”. The honest answer is “part of them, and probably not the part your regulator requires”.

Settle it before signing the project scope, because the gap between what the core ships and what IFRS 9 asks for is an entire module.

What Fineract does ship

Fineract includes a provisioning engine based on arrears aging buckets. You define a set of criteria, and each criterion carries four things: minimum age in days, maximum age in days, provisioning percentage, and the pair of GL accounts — one liability, one expense — where it is recorded.

Those criteria are then attached to loan products, and a process walks the portfolio classifying each loan by its days in arrears and applying the matching percentage. The result is stored with its history, so you can query what provision was computed on what date.

It is a correct, auditable mechanism and it is sufficient for many local regulatory frameworks. For a credit union that must provision according to a supervisor’s table — 31 to 60 days in arrears, 10%; 61 to 90, 30% — Fineract solves it out of the box.

What IFRS 9 asks for, and this is not

The bucket model is incurred loss: it provisions on what already happened, observed arrears. IFRS 9 asks for expected loss, which is a different thing.

The practical gap is three pieces the bucket engine does not have.

Stage classification. IFRS 9 requires placing each loan in one of three stages depending on whether its risk has increased significantly since initial recognition. That does not derive from days in arrears: it depends on comparing today’s risk against the risk at disbursement.

Risk parameters. Expected loss is computed from probability of default, loss given default and exposure at default. Those come from statistical models over the portfolio’s history, not from a table of percentages.

Forward-looking information. IFRS 9 requires incorporating macroeconomic expectations. No core banking system ships that, and none should.

Put differently: Fineract solves the recording of the provision and part of the calculation. Expected loss is a modelling problem that lives outside.

How to complete it without touching the core

The pattern I recommend is the same as for any extension: a module of your own that reads and writes through the API, without modifying Fineract.

The module takes the portfolio from a read replica — never from the transactional database, because a provisioning run walks the whole book and should not compete with operations — applies the staging and parameter model, and returns the result as manual journal entries against the same liability and expense accounts the native engine already uses.

That last part matters: reusing the provisioning engine’s accounts rather than inventing parallel ones keeps accounting coherent and lets you reconcile both calculations during the transition.

And keep the native engine running even if its number is not what you report. Having both figures — regulatory bucket provision and expected loss — computed over the same portfolio is the best alarm you will get: when they diverge beyond the reasonable, something changed in the book or in the model.

What I would decide before starting

Which of the two figures is official for the balance sheet, and which is informational. They cannot both be.

Where history lives. A provision for a past date has to be reconstructible identically two years later, which means storing not just the result but the parameters it was computed with. A model that gets recalibrated and rewrites the past is the same audit problem a fund has when it rewrites its closes.

Who answers for the model. Buckets are defined by the regulator and there is little to argue about. An expected-loss model is defended by someone in front of an auditor, and it is worth knowing who that is before building it.