Nifty VWAP-EMA Signal Bot

Automated trading bot for Nifty options based on 5 EMA crossing VWAP on 5-minute candles.

Strategy

Entry Signal:

• BUY: When 5 EMA crosses above VWAP → Buy CE (Call) option
• SELL: When 5 EMA crosses below VWAP → Buy PE (Put) option

Exit Rules (Fixed Target/SL, 1:2 R:R):

• Target: +50 points from entry
• Stop Loss: -25 points from entry
• No breakeven/trailing: trade closes at target, SL, or force close - nothing in between
• Force Close: 3:15 PM if still open

Philosophy: One trade per day, fixed 1:2 risk:reward. Inspired by Tom Hougaard's "Best Loser Wins" - cut losses early, but wins are capped rather than left to run.

Architecture

Fyers API (Candles) → Signal Detection → Kite API (Orders) → Telegram (Alerts)

Component	Purpose
vwap_ema_signal.py	Main bot - fetches candles, calculates indicators, detects signals
trade_manager.py	Enforces trade rules, monitors positions, tracks P&L
option_selector.py	Selects strike based on delta (~0.7 ITM) or premium mode
kite_api.py	Zerodha order execution

Key Findings

Findings from running the bot live (paper) and backtesting the trade log. Full detail in LEARNINGS.md.

The system is asymmetric by design, not by accident. Across 32 logged trades the win rate is ~25% (8W/24L), but the average win (+83.5 pts) is ~3.5× the average loss (−23.6 pts), giving a positive expectancy of ~+3.2 pts/trade. Frequent small stop-losses punctuated by rare large runners is the intended texture — a cluster of SL hits is normal, not a malfunction.

Crossover strength has no predictive edge — so I didn't add a filter for it. It was tempting to filter out "thin" EMA/VWAP crossovers after a losing streak, but the data refuted it: correlation between crossover separation and outcome was −0.03, and the three widest crossovers in the sample all lost. Threshold sweeps swung wildly on the small sample (classic overfitting), so the filter was rejected.

Time of day looked like an edge, but live results didn't hold up. Entries before 10:00 AM were the money pit in the initial backtest (−157 pts over 13 trades), so TRADE_START_TIME was pushed to 10:00 on Jun 16. The post-implementation tally told a different story: the filter blocked 3 would-be wins against only 1 avoided loss, so it was reverted back to TRADE_START_TIME=09:30 on Jun 27. The backtest isn't refuted — the before-10:00 window may still underperform over a larger sample — but the small live sample didn't support keeping the filter. See LEARNINGS.md §7b for the full tally.

Points-positive but rupees-negative pointed to position sizing as the bigger lever. The sample was +101 pts yet −₹23k, because size was largest during a drawdown. The signal wasn't the main problem — risk sizing was.

Caveat: 32 trades / 8 wins is a small sample, so these splits may be partly regime-driven. The takeaway is the process — test the hypothesis against the full log, reject what overfits, and change only what the data supports.

Setup

1. Clone and configure

git clone git@github.com:umeshkedimi/ema_vwap.git
cd ema_vwap
cp config.example.env config.env
# Edit config.env with your API credentials

2. Install dependencies

python3 -m venv venv
source venv/bin/activate
pip install requests pandas python-dotenv

3. API Credentials Required

API	Purpose	Get from
Fyers	Historical candle data	Fyers API
Kite Connect	Order execution	Kite Connect
Telegram	Trade alerts	@BotFather

4. Run locally

source venv/bin/activate
python vwap_ema_signal.py

5. Server deployment (optional)

See SERVER_SETUP.md for DigitalOcean VPS setup and cron configuration.

Configuration

Edit config.env:

# Trading mode
TRADING_ENABLED=true    # Enable/disable order execution
PAPER_TRADING=true      # true = simulate orders, false = live trading

# Trade timing
TRADE_START_TIME=09:30  # No trades before this (skip first 15 min)
TRADE_END_TIME=14:30    # No trades after this

# Trade parameters
MAX_TRADES_PER_DAY=1    # Max 1 trade per day
TARGET_POINTS=50        # Fixed target at +50 points
STOPLOSS_POINTS=25      # Fixed SL at -25 points
LOT_SIZE=195            # Quantity (1 lot = 65, this is 3 lots)

# Strike selection
STRIKE_MODE=delta       # delta or premium
MIN_PREMIUM=220         # Minimum option premium
ITM_OFFSET_FOR_DELTA=150  # ITM points for delta mode

Trade Rules

1. Timing: No trades before 9:30 AM or after 2:30 PM
2. One at a time: Must close current trade before taking next signal
3. Daily limit: Max 1 trade per day
4. Fixed exit: Target +50, SL -25 (1:2 R:R), no breakeven/trailing
5. Expiry: Weekly options (Tuesday expiry)

Option Symbol Formats

Weekly:  NIFTY{YY}{M}{DD}{strike}{CE/PE}  → NIFTY2651923600CE (May 19, 2026)
Monthly: NIFTY{YY}{MON}{strike}{CE/PE}   → NIFTY26MAY23600CE

Month codes (weekly): 1-9 for Jan-Sep, O/N/D for Oct/Nov/Dec

Files

ema_vwap/
├── vwap_ema_signal.py   # Main bot
├── trade_manager.py     # Trade rules & execution
├── option_selector.py   # Strike selection
├── kite_api.py          # Zerodha API client
├── review_signals.py    # Backtest/review tool
├── config.env           # Your credentials (not in git)
├── config.example.env   # Template config
├── trade_journal.csv    # Trade history with actual fills
└── SERVER_SETUP.md      # Deployment guide

Daily Token Update

Fyers and Kite tokens expire daily. First-time setup — copy the template and fill in your server IP and SSH key (the real update_tokens.sh is gitignored so credentials stay private):

cp update_tokens.example.sh update_tokens.sh
# edit SERVER and SSH_KEY at the top, then:
./update_tokens.sh

Or manually update FYERS_ACCESS_TOKEN and KITE_ACCESS_TOKEN in config.env.

Telegram Notifications

The bot sends alerts for:

• Bot startup and token validation
• Every candle scan (EMA/VWAP values)
• Signal detection
• Trade entry with option details
• Trade exit (target/SL hit)
• Daily summary

Version History

v3.0 (July 2026) - Fixed Target, 1:2 R:R

• Back to 1 trade per day, PAPER_TRADING=true
• Fixed target +50 / SL -25 (1:2 R:R), no breakeven/trailing
• LOT_SIZE=195 (3 lots × 65)

v2.1 (July 2026) - Two Trades

• MAX_TRADES_PER_DAY raised to 2 (LOT_SIZE=130, 2 lots × 65)
• Trade 2 only if Trade 1 closed with SL hit or breakeven (profit → day ends)
• Pre-10 AM block: max 1 trade before 10:00 AM

v2.0 (June 2026) - Trailing SL

• Replaced fixed +80 target with trailing SL starting at +75
• Philosophy: Cut losses early, let winners run

v1.0 (April 2026) - Initial Release

• Fixed target +80, SL -25, breakeven at +50
• 1 trade per day
• Delta mode strike selection

Disclaimer

This bot is for educational purposes. Trading involves risk. Past performance does not guarantee future results. Use at your own risk.