orkas-ai/orkas/portfolio-risk

Portfolio Risk

Read with market-data for a portfolio review, stress test, or rebalance question. Keep account values and holdings private and use only what is needed.

Choose the calculation from the supplied data: --mode historical (default) requires aligned daily decimal returns; --mode scenario uses weights and declared shocks, returns no historical statistics, and needs no return series. Do not manufacture a series to make a scenario request fit historical mode.

Historical input uses weights that sum to one and aligned daily decimal returns:

{"positions":[{"symbol":"700.HK","weight":0.4,"currency":"HKD","sector":"Technology","returns":[0.01,-0.02]}],"scenarios":{"tech_down":{"Technology":-0.15}}}

Run:

"$ORKAS_NODE" "$ORKAS_PC_DIR/bin/run-skill.cjs" market-data stock -- portfolio-risk --input portfolio.json --confidence 0.95

The core calculates portfolio return series from aligned observations, annual volatility, max drawdown, historical VaR/CVaR, HHI and top-name concentration, sector/currency exposure, and declared linear scenario losses. VaR is a sample statistic, not a loss ceiling.

For weight-only concentration or declared stress, use the same command with --mode scenario. For simultaneous local-price and FX shocks, supply weights already measured in base_currency NAV and a compound scenario:

{"base_currency":"USD","positions":[{"symbol":"700.HK","weight":1,"currency":"HKD"}],"scenarios":{"stress":{"local":{"700.HK":-0.1},"fx":{"HKD":-0.05}}}}

The core multiplies local and FX returns and also reports their separate contributions. Unspecified shocks are zero assumptions; the base currency cannot move against itself. Prepare this structured input from supplied facts without changing the source or inventing currency conversions. If the weight basis is unknown, resolve it before a cross-currency calculation.

Review missing positions, stale prices, derivatives, leverage, FX translation, liquidity, tax lots, and correlation regime change before recommending action. Frame rebalance ideas as constraint-aware options with trade-offs. Do not silently normalize invalid weights, fabricate correlations, or assume an investor's suitability, tax status, or loss capacity.