tradeblocks-compare

Performance Comparison

Explore differences between strategies, execution modes, or time periods.

Prerequisites

• TradeBlocks MCP server running
• At least one block with trade data loaded
• For backtest vs actual: Both trade log and reporting log for the same strategy
• For strategy vs strategy: Two blocks or multiple strategies in one block

Process

Step 1: Identify Comparison Type

Ask the user what they want to compare:

Ask: "What would you like to compare?"

Step 2a: Backtest vs Actual Comparison

Use compare_backtest_to_actual to explore how theoretical performance compares to live execution.

Key parameters:

• blockId: Block folder name
• scaling: How to compare P&L fairly (see below)
• strategy: Optional filter to specific strategy
• dateRange: Optional date filter
• matchedOnly: Only include trades where both backtest and actual exist

Scaling modes (see references/scaling.md):

Tool returns:

• Per-date comparison with backtest vs actual P&L
• Slippage calculation (actual minus backtest)
• Match status (whether both sides exist for each date)
• Summary totals and average slippage percentage

Present findings from the data:

• Total backtest P&L vs Total actual P&L (at selected scaling)
• Matched trade count (how many dates have both)
• Average slippage (percentage deviation from backtest)
• Unmatched trades (missed fills or extra trades)

Step 2b: Strategy vs Strategy Comparison

For comparing two different strategies:

1. Use get_statistics on each block (or with strategy filter)
2. Use get_correlation_matrix to understand how they move together

Key parameters for correlation:

• method: "kendall" (robust, rank-based), "spearman" (rank), "pearson" (linear)
• alignment: "shared" (only days both traded) or "zero-pad" (fill missing with 0)
• timePeriod: "daily", "weekly", or "monthly" aggregation

Present side-by-side metrics:

Correlation context:

• Very low (<0.2): Strategies move independently
• Low (0.2-0.4): Some independence
• Moderate (0.4-0.6): Shared behavior
• High (>0.6): Similar movements, less diversification

Step 2c: Period vs Period Comparison

For analyzing performance across time:

Use get_period_returns with period type and optional date filters.

Key parameters:

• period: "monthly", "weekly", or "daily"
• dateRange: Optional filter to specific time range
• normalizeTo1Lot: Normalize for fair comparison across different position sizes

Present period breakdown:

• Performance by month/quarter/year
• Best and worst periods
• Trends or regime changes

Questions to explore:

• "How does recent performance compare to earlier results?"
• "Are there periods that stand out?"
• "Does performance vary by time of year?"

Step 3: Present Findings

Synthesize the data into what stands out:

Comparison Summary:

• What was compared: [backtest vs actual / strategy A vs B / period X vs Y]
• Key observation: [Most notable difference from the data]
• Magnitude: [Size of the divergence]

What the data shows:

• [Notable finding 1 from tool output]
• [Notable finding 2 from tool output]
• [Any patterns or anomalies]

Context for interpretation:

• [Possible explanations for observed differences]
• [Factors that may affect the comparison]

Present these as observations from the historical data. The user can decide what meaning to draw from the findings.

Interpretation Reference

For detailed explanation of scaling modes, see references/scaling.md.

Related Skills

After comparison analysis:

• /tradeblocks-health-check - Deep dive into either strategy
• /tradeblocks-portfolio - Explore correlation and diversification
• /tradeblocks-wfa - Test parameter robustness

Notes

• Backtest results typically look better than live (ideal fills, no slippage)
• Some degradation from backtest to live is expected
• High correlation between strategies means they may draw down together
• Short comparison periods have more noise than signal
• Scaling mode choice affects what story the data tells