backtest-expert

name	backtest-expert
description	Expert guidance for systematic backtesting of trading strategies on Indian markets (NSE/BSE). Use when developing strategies, testing robustness, avoiding overfitting, or validating trading ideas.

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

Category	What to Define	Example
Universe	Which stocks/instruments	Nifty 200 constituents, F&O stocks only, market cap >5000 Cr
Entry	Exact trigger conditions	Close > 20 EMA AND RSI(14) crosses above 40 AND volume > 1.5x 20-day avg
Exit — Target	Profit-taking rule	Close 3% above entry OR trailing stop of 1.5 ATR
Exit — Stop	Loss-cutting rule	Close below entry-day low OR 2% fixed stop
Exit — Time	Maximum holding period	Exit after 10 trading sessions if neither target nor stop hit
Position Sizing	How much capital per trade	5% of equity per position, max 10 concurrent positions
Filters	When NOT to trade	Skip if stock is in F&O ban period, skip 2 days around results

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

Parameter	Minimum	Recommended
Time period	5 years	8-10+ years
Number of trades	100	200+
Market regimes covered	2 (bull + bear)	4+ (bull, bear, sideways, high-vol)
Data quality	Adjusted for corporate actions	Survivorship-bias-free universe

Indian Market Regimes to Cover

Regime	Period Examples	Characteristics
Bull market	2014-2017, 2020-2021	Nifty trending up, broad participation
Bear market	2008, 2020 (Mar), 2022 (Jun)	Sharp drawdowns, high correlation
Sideways/Range	2018-2019, 2023 H1	Nifty in 10% range, stock-specific moves
High volatility	2008, 2020, Budget days	India VIX > 25
Low volatility	2017, 2021 H2	India VIX < 15
Pre/Post Budget	Every Feb 1	Gap moves, policy-driven sectors
Election cycle	2014, 2019, 2024	Uncertainty then rally pattern
Monsoon impact	Jun-Sep annually	Agri, FMCG, rural economy impact
RBI policy shifts	Rate hike/cut cycles	Banking, NBFC, rate-sensitive sectors
Global crude shock	2018, 2022	INR weakness, OMC impact, inflation

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.

Parameter	Base	-20%	-10%	+10%	+20%	Verdict
EMA period	20	16	18	22	24	Stable if all profitable
RSI threshold	40	32	36	44	48	Fragile if only 40 works
Stop loss %	2%	1.6%	1.8%	2.2%	2.4%	Check drawdown impact

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:

Cost Component	Delivery (CNC)	Intraday (MIS)	F&O
Brokerage	~₹20/order or 0.03%	~₹20/order or 0.03%	~₹20/order
STT	0.1% (buy+sell)	0.025% (sell only)	0.0125% (sell, options)
Exchange charges	0.00345% (NSE)	0.00345% (NSE)	0.05% (options)
GST	18% on brokerage+exchange	18% on brokerage+exchange	18% on brokerage+exchange
Stamp duty	0.015% (buy)	0.003% (buy)	0.003% (buy)
SEBI charges	0.0001%	0.0001%	0.0001%
Slippage	0.05-0.1% large-cap	0.1-0.2% mid-cap	0.1-0.3% options

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

Score	Verdict	Action
80-100	Deploy	Size small initially (25% of intended), scale up over 50+ live trades
60-79	Refine	Identify weakest dimension, address it, re-test
40-59	Refine with caution	Multiple issues — may not be salvageable
0-39	Abandon	Fundamental edge likely does not exist. Document lessons and move on.

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

Red Flag	Why It Matters
CAGR > 50% with no drawdowns	Too good to be true — check for look-ahead bias
Win rate > 80%	Likely not accounting for slippage or adverse fills
Only works on specific parameters	Overfitting — no edge, just noise
< 50 trades in backtest	Statistically meaningless
No losing months in 5+ years	Data error or survivorship bias
Strategy stops working after 2020	Market structure may have changed (T+1, algo proliferation)
Uses > 5 parameters	Degrees of freedom too high — curve-fitted
No transaction costs modeled	Real returns could be negative
Tested on Nifty 50 only	Survivorship bias in universe selection

---

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