Double Calendar Analysis

Comprehensive health check for double calendar strategies. Each DC responds differently to filters and exits depending on its DTE spread, delta selection, and underlying. This skill surfaces those differences.

Prerequisites

• TradeBlocks MCP server running
• Block with DC trade data loaded
• Strategy profile recommended (will prompt to create if missing)
• Market data (SPX daily + VIX context) for regime analysis

Process

Step 1: Select Block and Load Profile

1. Ask which DC to analyze. Use list_blocks if needed.
2. Check for a profile. Call get_strategy_profile with the block and strategy name.
   • If profile exists: load it and summarize the structure (DTE spread, deltas, entry/exit rules, underlying).
   • If no profile: ask the user for the OO settings (screenshots work) and create one via profile_strategy. Key fields needed:
     • Underlying, DTE spread (short/long), put/call deltas
     • Entry filters (day, time, S/L ratio min, VIX, RSI)
     • Exit rules (profit target, time exit, S/L ratio exit, delta exits)
     • Position sizing (allocation %)

Display the profile summary before continuing:

Structure: [underlying] [short DTE]/[long DTE] DC, [put delta]/[call delta] delta
Entry: [day], [time], [filters]
Exits: [list exit rules]
Sizing: [allocation]%

Step 2: Baseline Performance

Run get_statistics for the block.

Present the core metrics:

Step 3: Exit Attribution

Run get_performance_charts with charts: ["exit_reason_breakdown"].

This is critical for DCs. Build a table:

Key insight from DC research: Exit types play DIFFERENT roles depending on the DC structure.

• For some DCs (like SPX 8/10), S/L ratio exits are the profit engine — they catch winners early.
• For others (like QQQ 2/4), time exits are the money maker and delta exits cause all the damage.
• If time exits are big losers, the DC may need tighter management before expiry.

Step 4: S/L Ratio Analysis

Run find_predictive_fields to check if openingShortLongRatio correlates with P&L.

Interpretation guide based on DTE/delta:

• Short DTE + high delta (2/4 at 20Δ): S/L ratio typically irrelevant (correlation ~0). Don't filter.
• Mid DTE + mid delta (5/7 at 25Δ): S/L ratio is the strongest predictor. Sweet spot typically 0.5-0.7.
• Long DTE + lower delta (21/28 at 30Δ): Weak but positive correlation. Some value in minimum threshold.

If correlation > 0.05, run filter_curve on openingSLRatio to find the optimal entry threshold.

Also check if closingShortLongRatio has strong negative correlation (it usually does) — this confirms S/L ratio as a structural health indicator for the trade.

Step 5: Structure Fit Analysis

If the strategy has a profile, run analyze_structure_fit with the block and strategy name. This provides a multi-dimensional breakdown of how well the strategy fits various market conditions, using the profile to contextualize results.

Tool returns:

• Performance by Vol_Regime, day-of-week, time-of-day
• Profile-derived dimension analysis (entry filters, DTE, deltas)
• profile_update_hints when data shows patterns diverging from profile settings
• Thin-data warnings for small buckets

Surface any profile_update_hints — these are actionable suggestions where the data contradicts the profile's assumptions.

If no profile exists, fall back to analyze_regime_performance with segmentBy: "volRegime" instead.

Step 6: VIX Regime Performance

Run analyze_regime_performance with segmentBy: "volRegime".

Build a regime table and flag the sweet spot vs danger zones:

Interpretation guide:

• Longer DTE DCs (9+) tend to thrive in Normal-Elevated VIX (more vega to harvest).
• Shorter DTE DCs may do better in low VIX (less gamma risk, cleaner theta decay).
• If a regime has <10 trades, flag as thin data.
• Compare to profile's expectedRegimes if set.

Step 7: Edge Decay Check

Run analyze_edge_decay for the block.

Focus on:

1. Yearly trends: Is win rate, profit factor, or Sharpe declining?
2. Recent vs historical: Is the recent window diverging from full history?
3. 2026 YTD performance: Has the strategy held up in the current environment?
4. Worst consecutive losing months: How bad does it get?

Flag any of these:

• Win rate slope < -3%/yr
• Profit factor slope < -0.5/yr
• Recent Sharpe < 50% of historical Sharpe
• Current streak is worst-ever losing streak

Step 8: Predictive Fields Deep Dive

From Step 4's find_predictive_fields results, surface the top 5 actionable fields beyond the output-derived ones (exclude netPl, plPct, rom, isWinner, maxProfit, maxLoss, etc.).

Actionable fields to look for:

• openingShortLongRatio — entry quality
• closingShortLongRatio — exit health
• durationHours — holding period effect
• movement — underlying movement sensitivity
• premium — entry pricing
• openingVix / closingVix — VIX sensitivity
• vixChangePct — VIX direction during trade
• gap — gap sensitivity
• numContracts — position size effect

For any field with |correlation| > 0.1, note it as a potential filter candidate.

Step 9: Curve Fit Detection

Before making any recommendations, test whether the current setup is robust or overfit. Run these checks in order — each one builds on the previous.

9a. Entry Filter Validation

If the strategy has a profile with entry filters, run validate_entry_filters with the block and strategy name. This is the most direct test of whether filters help or hurt.

Tool returns:

• Per-filter comparison: entered trades vs filtered-out trades (full stat suite for both groups)
• Ablation study: removes one filter at a time and tests all pairs
• profile_update_hints when filters appear counterproductive

What to look for:

• A filter is helping if entered trades outperform filtered-out trades on profit factor and win rate
• A filter is hurting if filtered-out trades actually perform better (the filter is excluding winners)
• The ablation study shows which filters interact — some only work in combination

Also run suggest_filters to see if there are market-based filters the data suggests but the profile doesn't use. This tool analyzes losing trades and suggests filters that would have improved performance. Pass strategyName to cross-reference suggestions against existing profile filters.

9b. Baseline Sanity Check

The strategy should be profitable WITHOUT entry filters. Use the validate_entry_filters ablation results to see what happens when all filters are removed.

Ask: "Is this strategy only profitable BECAUSE of the filters?" If removing all filters makes it a net loser, the filters are creating edge — that's curve fitting. Filters should IMPROVE an already-positive baseline, not rescue a broken structure.

9c. Parameter Sensitivity (Adjacent Value Test)

For each numeric filter or exit threshold in the profile, run filter_curve to check if small changes destroy the result:

Interpretation:

• Robust: Performance degrades gradually as you move away from the chosen value. A plateau or wide sweet spot is ideal.
• Overfit: Sharp cliff at the exact chosen value. Performance is great at X but terrible at X+1 or X-1.

Present the filter_curve results as a table showing performance at each threshold step. Look for smooth gradients, not cliffs.

9d. Walk-Forward OOS Check

From Step 7's edge decay data, extract the walk-forward efficiency ratios:

Interpretation:

• OOS efficiency > 1.0 = outperforming in-sample (unlikely to be overfit)
• OOS efficiency 0.7-1.0 = reasonable decay, probably genuine edge
• OOS efficiency < 0.5 = significant IS/OOS gap, possible overfitting
• OOS efficiency declining over time = edge may be decaying or original fit was to a specific regime

9e. Year-over-Year Consistency

From Step 7's period metrics, check: does the strategy work in EVERY full year, or only some?

Interpretation:

• Profitable in ALL full years = strong evidence of genuine edge
• Profitable in 3/4 years = acceptable, check if the bad year has an explanation (regime shift, etc.)
• Profitable in only 1-2 years = high curve fit risk — the "edge" may be a specific market period

9f. Trade Count Check

After all filters are applied, how many trades per year remain?

If filters cut more than 50% of trades, the question is: does the P&L per trade improve enough to justify the reduced sample?

9g. Curve Fit Verdict

Summarize the findings:

• 5 PASS: Strong confidence the strategy is robust
• 4 PASS, 1 WARN: Likely robust, monitor the warning
• 3 or fewer PASS: Curve fit risk is significant — recommend simplifying filters
• Any FAIL on baseline: The structure itself may not have edge — filters are papering over a broken thesis

Step 10: Synthesis and Recommendations

Bring it all together. Answer these questions:

1. Is this DC healthy? Sharpe, win rate, profit factor trends
2. What's driving profits? Which exit type makes the money?
3. What's causing losses? Which exit type or regime is the biggest drag?
4. Is the edge decaying? Yearly trends and recent performance
5. Is it overfit? Curve fit verdict from Step 9
6. What would you change? Based on the data:
   • Add/remove entry filters?
   • Adjust exit thresholds?
   • Change position sizing for specific regimes?
   • Consider pausing in certain VIX environments?

DO NOT recommend specific threshold values without supporting data. If S/L ratio correlation is 0.001, don't suggest an S/L ratio filter. Let the data speak.

Reference the DC workshop learnings:

• "Adding little bits at a time" — don't stack filters, test one at a time
• Each filter should have explanatory power — you should know WHY it works
• If removing a filter doesn't hurt much, the filter may be noise
• If 27% threshold works but 30% doesn't, that's curve fitting territory

Optional: Exit Simulation (when replay data available)

If the user wants to test exit rule changes, use batch_exit_analysis with the profile's exit rules translated to trigger configs:

Note: Delta thresholds use decimal format (0.70, not 70). OO uses whole numbers.

Reference

• For DC mechanics, term structure, curve fitting principles, and common failure modes, see references/dc-mechanics.md

What NOT to Do

• Don't assume S/L ratio filtering works for all DCs — test it first
• Don't stack VIX + S/L ratio + VIX9D/VIX filters — they're often correlated, pick one
• Don't recommend delta exit thresholds without testing adjacent values (65 vs 70 vs 75)
• Don't present a single "optimal" configuration — show the tradeoffs
• Don't ignore thin data warnings — 10 trades in a bucket is suggestive, not conclusive