Backtest Expert

Backtest Expert — Indian Market Strategy Validation

Core Philosophy

"Find strategies that break the least, not profit the most."

A strategy that survives stress testing across multiple market regimes, transaction cost assumptions, and parameter perturbations is far more valuable than one that shows spectacular returns on a single optimized parameter set. Overfitting is the silent killer of trading accounts.

6-Step Backtesting Workflow

Step 1: State the Hypothesis (1 Sentence Edge)

Before writing a single line of code, articulate why the strategy should work in one clear sentence.

Good hypotheses:

• "Stocks that gap up >3% on above-average volume after consolidation tend to continue higher for 2-5 days on NSE."
• "Nifty 50 stocks that revert to their 20-day mean after RSI drops below 30 produce positive expectancy within 5 trading sessions."
• "Selling strangles on Bank Nifty on Wednesday expiry with delta <0.15 captures time decay faster than gamma risk materializes."

Bad hypotheses:

• "This indicator combination looks good on the chart." (no edge articulated)
• "I saw someone on Twitter making money with this." (no reasoning)

Ask yourself:

• What behavioral or structural edge am I exploiting?
• Why would this edge persist? (Structural > Behavioral > Statistical)
• Who is on the other side of this trade, and why are they losing?

Step 2: Codify Rules (No Ambiguity)

Every rule must be binary — a computer must be able to execute it without interpretation.

Rule Categories

India-Specific Rules to Consider

• Circuit limits: Stocks hitting upper/lower circuit cannot be exited. Define handling.
• F&O ban period: Stocks crossing 95% MWPL cannot add fresh F&O positions.
• T+1 settlement: Cash equity settles next trading day (changed from T+2 in 2023).
• Pre-open session: 9:00-9:08 AM orders, 9:08-9:15 AM matching. Define if you use pre-open.
• Muhurat trading: Special Diwali session — include or exclude?
• Corporate actions: Adjust for splits, bonuses, dividends, rights issues.

Step 3: Run Initial Backtest

Minimum Requirements

Indian Market Regimes to Cover

Key Metrics to Record

Returns: CAGR, total return, monthly returns distribution
Risk: Max drawdown, average drawdown, drawdown duration, Calmar ratio
Efficiency: Sharpe ratio (use 6% risk-free for India), Sortino ratio
Trade quality: Win rate, avg win/loss, profit factor, expectancy per trade
Consistency: % profitable months, worst month, longest losing streak

Step 4: Stress Test (Spend 80% of Your Time Here)

This is where most backtests fail — and where the real value lies.

4a. Parameter Sensitivity

Perturb every parameter by +/-20% and check if performance degrades gracefully or collapses.

Rule of thumb: If the strategy only works with exact parameter values, it is overfit. You want a "plateau" of profitability, not a "peak."

4b. Execution Friction (India-Specific Costs)

Apply realistic transaction costs:

Total round-trip cost estimates:

• Delivery large-cap: ~0.3-0.5%
• Intraday large-cap: ~0.1-0.2%
• F&O (options): ~0.15-0.4%
• Small-cap delivery: ~0.5-1.0% (wider spreads)

4c. Time Robustness

• Split data into 3-year rolling windows. Is the strategy profitable in each?
• Check year-by-year returns. Is any single year driving total performance?
• Remove the best month. Is the strategy still positive?

4d. Sample Size Validation

• Minimum 30 trades for any statistical claim (even this is weak)
• 100+ trades: Moderate confidence
• 200+ trades: Good confidence
• Use the t-test: Is average trade return significantly different from zero?

Step 5: Out-of-Sample Validation (Walk-Forward Analysis)

Never skip this step.

Walk-Forward Method for Indian Markets

1. In-sample period: Train on 5 years of data (e.g., 2015-2019)
2. Out-of-sample period: Test on next 1-2 years (e.g., 2020-2021)
3. Roll forward: Move window, retrain on 2016-2020, test on 2021-2022
4. Combine: Aggregate all out-of-sample periods for true performance estimate

Walk-Forward Efficiency (WFE):

WFE = Out-of-Sample Return / In-Sample Return

• WFE > 50%: Good — strategy generalizes
• WFE 30-50%: Acceptable — some overfitting present
• WFE < 30%: Poor — likely overfit

Paper Trading Validation

Before deploying capital, paper trade for at least:

• 30 trades minimum
• 2 months minimum
• Cover at least one volatile period (expiry week, results season, RBI policy)

Step 6: Evaluate Results (Deploy / Refine / Abandon)

Use the evaluation script to get an objective score:

python3 evaluate_backtest.py \
  --total-trades 150 \
  --win-rate 62 \
  --avg-win-pct 1.8 \
  --avg-loss-pct 1.2 \
  --max-drawdown-pct 15 \
  --years-tested 8 \
  --num-parameters 3 \
  --slippage-tested

Decision Framework

Before Deploying

• Strategy has positive expectancy after ALL costs
• Survived parameter sensitivity testing
• Walk-forward efficiency > 50%
• Maximum drawdown is psychologically tolerable
• Sample size > 100 trades
• No more than 3-4 free parameters
• Slippage and transaction costs included
• Paper traded for 30+ trades
• Written trade plan with exact rules
• Risk management plan for live trading (position sizing, max daily loss, max drawdown circuit breaker)

Using Broker MCP Tools for Backtesting Support

While the MCP tools are not backtesting engines, they support the process. Use whichever broker is connected:

Groww MCP (if connected)

• fetch_historical_candle_data: Fetch OHLCV data for strategy development and spot-checking
• get_historical_technical_indicators: Calculate indicators (SMA, EMA, RSI, MACD, Bollinger, SuperTrend, etc.) on historical data
• get_historical_candlestick_patterns: Identify candle patterns in historical data
• fetch_stocks_fundamental_data: Screen for universe construction (PE, ROE, market cap filters)
• fetch_fundamentals_screener: Natural language screening for universe building
• fetch_technical_screener: Technical screening for strategy ideas
• get_ltp: Current price for live validation
• fetch_market_movers_and_trending_stocks_funds: Discover momentum and volume patterns

Zerodha Kite MCP (if connected)

• get_historical_data: Fetch OHLCV candle data for strategy development
• get_ltp / get_quotes: Current prices for live validation
• search_instruments: Find instruments for universe construction
• get_holdings / get_positions: Verify live portfolio against strategy signals

Quick Reference: Red Flags

Files in This Skill

• scripts/evaluate_backtest.py — CLI scoring tool for backtest evaluation
• references/methodology.md — Comprehensive backtesting methodology for Indian markets
• references/failed_tests.md — Common failure patterns and documentation framework