Portfolio Stress Test

Portfolio Stress Test

Read-only across the user's live account(s). reference/READ-ONLY-POLICY.md (hard rule) and reference/RESEARCH-STANDARDS.md (data efficiency, completeness checklist, disclosure) apply. Steps name capabilities — resolve each against reference/BROKER-CAPABILITIES.md.

This is sensitivity analysis, not forecasting: it answers "how much would this portfolio lose if X happened", never "will X happen". No broker exposes beta or correlation, so the inputs are derived or web-sourced — state the method with the number, every time.

Steps

1. Load the plan's risk side. PORTFOLIO-PLAN.md — Risk limits (especially max drawdown the user would hold through, single-stock and sector caps, F&O and leverage rules), Target allocation, horizon, the fixed-income inventory (no MCP can see direct bonds/FDs/SGBs, and they are the ballast the whole exercise turns on), the SIP register, and Deployable capital — dry powder and known upcoming outflows decide whether a drawdown is survivable or forced. Missing "max drawdown" → say so: without it there's nothing to judge the result against, only a number.

2. Pull live state. Equity holdings from every active broker; open F&O/intraday positions and available margin where any exist — leverage is where a scenario stops being academic. One batched LTP call for valuation.

3. Map exposure before modelling anything — weights by sector, market cap, asset class, and single name, plus overlap (two funds and a direct holding in the same top names is one exposure). The two or three largest concentrations will drive every scenario, so identify them and keep the depth there.

4. Derive the sensitivity inputs honestly. Beta and correlation are on no broker's tool list. For the top holdings only: derive beta from historical candles (name vs. index, daily, ~1Y) via the candles capability, or take a web-sourced beta and label it. Often more useful than beta — actual peak-to-trough drawdown in past shocks from the same candle data, which needs no model. Debt-sleeve duration comes from the plan inventory or bond-analysis. Name the method and its window alongside every figure, and don't derive a beta for a small position.

5. Run the scenario set — these, plus any the user names:

   • Broad equity −10% / −20% / −30%, beta-weighted.
   • The largest sector −25%.
   • The largest single position −40% (single-stock risk is usually the sharpest number in the report).
   • Rates +100bps — duration × Δyield across the debt sleeve, plus the second-order hit to rate-sensitive equity.
   • INR −10%, only where it actually bites: US holdings, importers, exporters, commodity-linked names.
   • F&O stress — margin adequacy against available margin, and gap risk on any short-option leg. Flag it as the fastest path to a realised loss the rest of the portfolio can't cushion.

6. For each scenario report the estimated ₹ and % portfolio hit, the allocation after the shock (a crash rebalances the portfolio without asking), and whether the drawdown exceeds the plan's stated tolerance. That breach — not the loss figure — is the actionable finding.

7. Two checks the scenarios miss.

   • Exit liquidity: quote + depth on the largest and thinnest positions. A position that can't be sold in size is a bigger risk than its beta says.
   • Forced-selling risk: known outflows and expenses against cash and the fixed-income sleeve — months covered without selling equity into the fall.

8. Present. Assumptions box first: single-factor shocks, correlations rise in a real crash so treat every figure as a floor not a ceiling, and no probability is being assigned to any scenario. Then the scenario table, then breaches of stated limits, then positioning notes — scenario-framed with both branches, never a market prediction. Data as-of near the top; disclosure block. Formal version: reference/templates/portfolio-stress-test.md.

9. Hand off for the fix: rebalancing-planner to size any reduction, fno-analysis for the derivatives leg, bond-ladder-planner for ballast and rate-watch for the rate leg, tax-capital-gains before anything is trimmed pre-emptively.