Bank stress testing without the Excel chaos

In many banks, the stress test is a collection of Excel files passed between treasury, risk and finance. Each department calculates its own piece, one person knows the formulas, and the final numbers are assembled by hand a few days before the deadline. A process like that is hard to repeat, check or explain. Here is how to build stress testing that works as a system.

Start with scenarios

A scenario is an agreed story about what could happen to the economy and markets. Several scenarios are usually in play.

• Baseline — the expected path of the economy.
• Moderately adverse — slower growth, a weaker currency, higher rates.
• Severe — a strong but plausible shock: sharp devaluation, a drop in remittances, deposit outflows.
• Reverse stress test — which event would push the bank below its capital or liquidity requirements.

In Central Asia, local factors matter: the national currency exchange rate, the share of foreign-currency loans, dependence on migrant remittances, commodity prices. Scenarios are approved by the bank’s management, not by risk management alone.

Link the macro picture to losses

The core technical task is translating macro variables into risk parameters. For credit risk, this usually means models linking probability of default by portfolio segment to GDP, the exchange rate, interest rates and unemployment. Where internal history is thin, expert overlays are used — but they must be documented.

Next comes the impact on provisions and capital: higher PD and LGD mean higher expected losses and provisioning charges, which reduce profit and capital. If the bank provisions under IFRS 9, it makes sense to use the same models and segments in the stress test so the numbers are comparable.

Don’t forget liquidity

Capital and liquidity are two different stories, and stress hits both. For liquidity, you model deposit outflows by client type and currency, reduced access to the interbank market, haircuts on securities sales, and changes in loan repayment schedules. The output is a maturity gap profile and a buffer of liquid assets that has to be enough to get through the scenario.

The scenario must be consistent: the same devaluation has to show up in credit risk, the FX position and liquidity at the same time.

The data foundation

The main cause of Excel chaos is the lack of a single data source. Every calculation starts with extracts from different systems, and the numbers don’t match between departments.

• Bring the loan book, deposits, securities and macro data into one repository.
• Fix reporting dates: the stress test runs on a data snapshot that can be reproduced.
• Reconcile source data to the general ledger before the calculation, not after.
• Keep versions of scenarios, model parameters and results.

A cloud data platform such as Snowflake fits this well: data snapshots and calculation history sit in one place, with controlled access for different departments. The calculations themselves can run in Python or a specialised tool — what matters is that they all read from the same source.

The calculation itself should become a repeatable pipeline: load the snapshot, check data quality, apply the scenario, calculate each block, produce summary reports. Each step runs as code, not as copying between files. The stress test can then run quarterly or on demand — say, when the exchange rate moves sharply or a new macro forecast comes out — rather than once a year when someone asks. Excel can stay, but as a way to view results, not as the place where they are calculated.

The value of a stress test is not in the precision of its numbers, but in management having discussed in advance what it will do in a bad scenario.

Governing the process

A technically sound stress test is useless if its results never reach decisions. You need a clear setup:

• a process owner — usually risk management — and named owners for each calculation block;
• a calendar: when scenarios are approved, when data is ready, when results go to committee;
• documented methodology and assumptions that internal audit can follow;
• independent model review;
• a link to action plans: limits, the recovery plan, the contingency funding plan.

What regulators expect

Supervisors generally expect a bank to run capital and liquidity stress tests regularly, use plausible scenarios, document its methodology, and have a board and management that understand the results and use them in planning. Specific requirements differ by country, so check them against your regulator’s current rules.

The good news: once the process runs on a single data source and is documented, answering a supervisory request becomes a matter of days rather than weeks.