Risk

Risk Analysis

Explore risk characteristics and position sizing metrics for trading strategies.

Prerequisites

• TradeBlocks MCP server running
• Block with trade data (10+ trades minimum for meaningful metrics)

Process

Step 1: Understand User Goals

Risk analysis serves different purposes. Ask what the user wants to understand:

| Goal | Primary Tool | Also Consider | |------|-------------|---------------| | "What does Kelly suggest?" | get_position_sizing | Monte Carlo for drawdown context | | "What are worst-case scenarios?" | run_monte_carlo | stress_test for named scenarios | | "How did I do during COVID/bear markets?" | stress_test | Drawdown attribution | | "What caused my biggest drawdown?" | drawdown_attribution | Tail risk for future risk | | "How correlated are my strategies?" | get_tail_risk | Position sizing per strategy | | "What if I change allocations?" | what_if_scaling | Marginal contribution |

Ask: "What aspect of risk would you like to explore?"

Then use list_blocks to identify the target block.

Step 2: Position Sizing (Kelly Criterion)

Call get_position_sizing with the user's capital base.

Key parameters:

• capitalBase: Starting capital (required)
• kellyFraction: "full", "half" (default), or "quarter"
• maxAllocationPct: Cap per strategy (default: 25%)
• minTrades: Minimum trades for valid calculation (default: 10)
• sortBy: Sort by "kelly", "winRate", "payoffRatio", "allocation", "name"
• useMarginReturns: Use percentage returns based on margin (better for compounding strategies)

Tool returns:

• Win rate and payoff ratio (inputs to Kelly formula)
• Raw Kelly percentage (what the formula suggests)
• Adjusted allocations at full/half/quarter Kelly
• Per-strategy breakdown if multiple strategies exist
• Warnings (e.g., "Portfolio Kelly exceeds 25%", "negative Kelly")

Important context for Kelly:

• Full Kelly is mathematically optimal but assumes perfect knowledge of edge
• Half Kelly is commonly used to account for estimation uncertainty
• Negative Kelly indicates historical losses exceeded wins (Kelly formula doesn't apply)

Step 3: Stress Testing

Call stress_test to see how the portfolio performed during named historical stress scenarios.

Key parameters:

• blockId: Block folder name
• scenarios: Optional list of specific scenario names (omit for all built-in scenarios)
• customScenarios: User-defined scenarios with custom date ranges
• includeEmpty: Include scenarios with no trades (default: false)

Tool returns per scenario:

• Scenario name and date range
• Trade count, win rate, net P&L during that period
• Profit factor and max drawdown

Present as a stress test table:

| Scenario | Dates | Trades | Win Rate | Net P&L | Max DD | |----------|-------|--------|----------|---------|--------| | ... | ... | ... | ... | ... | ... |

Key insight: If a strategy had zero trades during a stress period, it had no exposure — that's important context (either it didn't exist yet, or its filters kept it out).

Step 4: Monte Carlo Projections

Call run_monte_carlo for probabilistic projections:

Key parameters:

• includeWorstCase: Enable worst-case injection (default: true)
• worstCasePercentage: Percentage of worst-case trades (default: 5%)
• worstCaseMode: "pool" (adds to resample pool) or "guarantee" (ensures worst appears)
• resampleMethod: "trades" (default), "daily", or "percentage" (for compounding)
• simulationLength: Number of trades/days to project forward
• normalizeTo1Lot: Normalize for fair comparison across position sizes

Focus on:

• 5th percentile outcome (valueAtRisk.p5)
• Probability of profit
• Mean and median max drawdown
• Distribution of terminal equity

Step 5: Drawdown Attribution

Call drawdown_attribution to identify what caused the worst drawdown.

Key parameters:

• blockId: Block folder name
• strategy: Optional filter
• topN: Number of top contributors (default: 5)

Tool returns:

• Drawdown period (peak date to trough date)
• Peak and trough equity values
• Per-strategy P&L attribution during the drawdown
• Percentage contribution to total loss

This answers "which strategies caused the most pain" during the worst period.

Step 6: Tail Risk (Multi-Strategy)

Call get_tail_risk for extreme co-movement analysis (requires 2+ strategies).

Key parameters:

• tailThreshold: Percentile for "tail" events (default: 0.1 = worst 10%)
• varianceThreshold: For effective factors calculation (default: 0.8)
• normalization: "raw", "margin", or "notional"

Tool returns:

• Joint tail risk matrix (do strategies fail together?)
• Effective factors (how many independent risk sources exist)
• Risk level: LOW (0.5)
• Copula correlation (statistical dependency structure)

Step 7: What-If Scaling

Call what_if_scaling to explore how allocation changes affect risk metrics.

Key parameters:

• blockId: Block folder name
• strategyWeights: Weight per strategy (e.g., {"StrategyA": 0.5, "StrategyB": 1.5})
• strategies: Multi-strategy mode with per-strategy block source and scale factor
• showUncapped: Also show results without maxContractsPerTrade ceiling

Tool returns:

• Before/after portfolio metrics comparison
• Per-strategy breakdown at new weights
• Profile-aware: respects contract ceilings from profiles

Common questions this answers:

• "What if I removed this underperforming strategy?"
• "What if I doubled my best performer?"
• "What if I halved everything during high VIX?"

Step 8: Present Findings

Synthesize findings into what the data reveals:

Position Sizing Metrics:

• Win rate: [value]% | Payoff ratio: [value]
• Kelly formula suggests: [value]% (based on historical data)
• At half Kelly: [dollar amount] of [capital base]
• [Any warnings from the tool]

Stress Test Results:

• [Worst scenario]: [P&L and drawdown during that period]
• [Best scenario]: [how the strategy held up]
• Scenarios with no exposure: [list]

Monte Carlo Projections:

• 5th percentile return: [value]
• Probability of profit: [value]%
• Mean max drawdown: [value]%

Drawdown Attribution:

• Worst drawdown: [peak] to [trough] ([magnitude])
• Primary contributor: [strategy] at [percentage] of total loss

Tail Risk (if applicable):

• Average joint tail risk: [value] ([LOW/MODERATE/HIGH])
• Effective factors: [value] of [strategy count]
• [Note any high-risk pairs]

What stands out:

• [Highlight notable findings]
• [Surface any warnings from the tools]
• [Note relationships between metrics]

Present these as insights from the historical data, letting the user decide what fits their situation.

Interpretation References

• [references/kelly-guide.md](references/kelly-guide.md) - Kelly criterion explained
• [references/tail-risk.md](references/tail-risk.md) - Understanding fat tails

Common Scenarios

"I have $100,000 - what does Kelly suggest?"

1. Run get_position_sizing with capitalBase: 100000
2. Review the Kelly percentages and warnings
3. Surface both raw Kelly and half-Kelly figures
4. Note that Kelly assumes independent trades and known edge

"Do my strategies fail together?"

1. Run get_tail_risk
2. Look at joint tail risk matrix for high values
3. Check effective factors (closer to 1 = more correlated risk)
4. High correlation means drawdowns may compound

"What's the worst realistic outcome?"

1. Run stress_test for historical named scenarios
2. Run run_monte_carlo with worst-case injection for probabilistic projection
3. Run drawdown_attribution to understand what drove the historical worst
4. Combine: stress tests show what happened, Monte Carlo shows what might happen

"Should I change my allocations?"

1. Run what_if_scaling with proposed weights
2. Compare before/after Sharpe, drawdown, net P&L
3. Check if the change improves risk-adjusted returns or just shifts risk

Related Skills

After risk analysis:

• /tradeblocks:health-check - Full metrics overview
• /tradeblocks:portfolio - Correlation and diversification analysis
• /tradeblocks:wfa - Test parameter robustness

Notes

• Kelly assumes independent trades; real trades may be correlated
• Historical volatility may underestimate future extremes
• Monte Carlo resamples history - it can't predict unknown risks
• Stress tests only cover known historical scenarios
• Negative Kelly means the formula doesn't apply (no positive edge in historical data)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: davidromeo
• Source: davidromeo/tradeblocks-skills
• License: MIT
• Homepage: https://tradeblocks.io

Install and usage instructions live in the source repository linked above.