wfa

Walk-Forward Analysis

Test whether strategy parameters hold up when applied to data the optimizer never saw.

What is Walk-Forward Analysis?

Walk-forward analysis (WFA) tests parameter robustness by:

1. Dividing history into segments
2. In-Sample (IS): The data used to find "optimal" parameters
3. Out-of-Sample (OOS): Data the optimizer never saw, used to test those parameters
4. Rolling forward: Repeat across the entire history

|------ IS Period 1 ------|-- OOS 1 --|
              |------ IS Period 2 ------|-- OOS 2 --|
                            |------ IS Period 3 ------|-- OOS 3 --|

If OOS performance is close to IS performance, the parameters may be capturing real patterns. If OOS significantly underperforms, the parameters may be fitting to noise.

Prerequisites

• TradeBlocks MCP server running
• Block with sufficient trade history (50+ trades for meaningful analysis)

Process

Step 1: Select Strategy

Use list_blocks to show available blocks.

Ask:

• "Which backtest contains the strategy you want to analyze?"
• "Do you want to test a specific strategy or the full portfolio?"

Note the block ID and optional strategy filter for subsequent steps.

Step 2: Understand User Goals

Walk-forward analysis answers different questions:

Goal	What to Look For
Test if parameters are robust	Overall WF efficiency, OOS vs IS degradation
Check for potential overfitting	High IS but low OOS performance
Evaluate consistency	How many OOS periods were profitable
Understand parameter sensitivity	Parameter stability across windows
Test strategy weight combinations	Use parameterRanges with strategy weights
Test position sizing approaches	Use parameterRanges with Kelly/fraction params

Ask: "What are you trying to understand about this strategy?"

Step 3: Run Analysis

Call run_walk_forward with the selected block.

Core parameters:

• blockId: Block folder name
• strategy: Filter to specific strategy
• isWindowCount: Number of in-sample windows (default: 5)
• oosWindowCount: Number of out-of-sample windows (default: 1)
• optimizationTarget: Metric to optimize (default: "sharpeRatio")
  • Options: "netPl", "profitFactor", "sharpeRatio", "sortinoRatio", "calmarRatio", "cagr", "avgDailyPl", "winRate"
• minInSampleTrades: Minimum trades in IS period (default: 10)
• minOutOfSampleTrades: Minimum trades in OOS period (default: 3)
• normalizeTo1Lot: Normalize trades to 1-lot (useful for pct_of_portfolio sizing)