← Agent catalogue
AI Agent Profile · LendingIQ · Bengaluru
Stress Testing Agent AI
Function: Risk Stress Test AnalystInvoked via: risk governance pipeline — quarterly and on-demandRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens
DivisionRisk division
Resume
What this agent does
The Stress Testing Agent AI runs the macro scenario stress tests that LendingIQ's risk governance function requires — modelling the impact of adverse economic scenarios on the loan portfolio, simulating credit losses and capital requirements under stress, and producing the stress test reports for the Board Risk Committee and the CRO AI. It is the portfolio-level adversarial imagination function — asking what happens to the portfolio if things go badly, so the board can assess whether the current risk appetite is appropriate.
Primary functions
Macro Scenario Modelling
Quarterly and on-demandINVOKED WHEN: quarterly ICAAP cycle is due or CRO AI requests a specific scenario analysis
- Reads the scenario parameters from the scenario library — baseline (current conditions), moderate adverse (one standard deviation shock to GDP growth, unemployment, and property prices), severe adverse (two standard deviation shock), and any regulatory-prescribed scenarios from the RBI ICAAP guidelines — and applies them to the portfolio's risk parameters: the PD for each segment is stressed by the scenario's implied credit deterioration based on historical correlations between macroeconomic variables and segment NPA rates.
- Builds bespoke scenarios on request from the CRO AI or the Board Risk Committee: a sector-specific stress (what happens to the portfolio if the MSME construction sector experiences a 40% revenue decline), a geographic stress (what is the impact of a flood or drought on the portfolio in a specific region), or a rates stress (what happens to FOIR and repayment capacity if interest rates rise by 200 basis points). Bespoke scenarios require the CRO AI to specify the scenario parameters — the agent designs the portfolio impact model, not the macroeconomic scenario.
- Does not forecast which scenario is most likely to occur. The scenario library reflects the range of adversity the portfolio should be able to withstand — not a probability-weighted distribution of future states.
Output: Scenario parameter set — each variable stressed to scenario level, correlation assumptions applied, and sector-specific adjustments where applicable.
Portfolio Impact Simulation
Per scenario — run in sequenceINVOKED WHEN: scenario parameters are set and the portfolio impact needs to be computed
- Applies the stressed PD and LGD parameters to the current portfolio outstanding by segment, vintage, and product — computing the expected credit loss under each scenario, the resulting NPA rate by segment, and the aggregate provisioning requirement above the current provisioning level. The simulation shows not just the aggregate impact but the distribution across segments, so the CRO AI can identify which segments are most vulnerable to which scenarios.
- Computes the capital impact: the additional provisioning requirement under each scenario as a proportion of the current capital base, and the resulting capital adequacy ratio under stress. Where any scenario produces a capital adequacy ratio below the regulatory minimum, flags it as a capital adequacy concern requiring management action.
Output: Portfolio impact table — NPA rate by segment under each scenario, aggregate credit loss, additional provisioning requirement, stressed capital adequacy ratio, and capital adequacy breach flag where applicable.
Stress Report Generation
Per stress test cycleINVOKED WHEN: scenario modelling and portfolio impact simulation are complete and the formal stress test report is required
- Produces the formal stress test report in the format required for Board Risk Committee presentation and RBI ICAAP submission: executive summary of key findings, scenario descriptions, methodology note, portfolio impact results by scenario, capital adequacy analysis, management actions identified for scenarios that produce concerning results, and the overall conclusion on portfolio resilience.
- Produces the narrative interpretation of results — what the numbers mean for the portfolio's risk profile, which scenarios expose the most concentrated vulnerabilities, and where the portfolio shows resilience — so the board is reading a risk narrative, not a table of numbers without context.
Output: Formal stress test report — board-ready format with executive summary, scenario results, capital adequacy analysis, vulnerability narrative, and management action recommendations. ICAAP-formatted appendix for regulatory submission.
Hard guardrails
Will notProduce forecasts of NPA rates or macro events. Stress tests are hypothetical adverse scenarios — not probability-weighted forecasts of future portfolio performance.
Will notDesign new scenario parameters independently. Scenario design requires CRO AI specification and board approval for novel scenarios outside the standard library.
Will notRecommend specific management actions. Stress test results identify vulnerabilities; management actions are the CRO AI's and board's responsibility.
Known limitations
Stress test severity is bounded by the scenario library.A scenario outside the library — a genuinely novel shock that historical data has no precedent for — cannot be well-modelled with historical correlation parameters. The scenarios in the library represent the range of adversity the portfolio has been designed to withstand; tail risks outside that range require expert judgment overlays that the agent cannot provide.
PD and LGD parameters are estimated, not known.The stressed default probabilities applied to each segment are derived from historical correlations between macro variables and segment NPA rates — which may not hold in a structural break scenario. A macro shock that changes borrower behaviour in ways that differ from historical patterns will produce results that diverge from the model. The model is the best available tool; it is not the truth.
Important Reads
Learn more about how to deploy Stress Testing Agent AI to your lending workflow.
- Use case #0001How Stress Testing AI models a 200bps rate shock on your NBFC portfolioA 200 basis point rate shock is not a hypothetical — it is recent history. Between May 2022 and February 2023, the RBI raised the repo rate by 250 basis points over seven consecutive decisions. For an NBFC with a fixed-rate lending book funded by floating-rate borrowings, each of those hikes compressed the net interest margin in real time. The Stress Testing Agent AI models what that compression looks like across the specific composition of your portfolio — not the industry average, but your actual book.Read article →
- Use case #0002RBI-ready stress test reports: what the AI generates automaticallyThe RBI's stress testing framework for NBFCs does not prescribe a specific report format — it prescribes the content that must be demonstrably present: the scenarios tested, the methodology applied, the results produced, and the management actions identified. The Stress Testing Agent AI generates a report that satisfies every content requirement, in a format that a regulator can review without requesting supplementary information, in under 6 minutes of computation time.Read article →
- Use case #0003Stress Testing AI: quarterly vs on-demand test cadenceA quarterly stress test is a regulatory obligation. An on-demand stress test is a risk management tool. The difference matters because the events that create the most urgent need for stress test data rarely happen on a quarterly schedule. A 50bps RBI rate announcement, a sector NPA surge, a credit rating downgrade at a major counterparty — each of these events changes the stress picture materially and demands an updated test within hours, not at the end of the quarter.Read article →