wrtrade 2.0.0

Advanced Portfolio Trading Framework with N-dimensional portfolios, Kelly optimization, and multi-broker support

wrtrade

Ultra-fast Python backtesting and trading framework. Built with Polars for speed, designed for simplicity.

For Quantitative Traders

You have a trading idea. You want to test it, validate it's not overfit, optimize position sizing, and deploy it to live trading. wrtrade makes this workflow simple and fast.

The Journey:

1. Research - Develop your strategy and backtest it
2. Validate - Use permutation testing to ensure it's statistically significant
3. Optimize - Apply Kelly criterion for optimal position sizing
4. Paper Trade - Deploy to paper trading to verify in live markets
5. Go Live - Deploy to live trading with confidence

Install

pip install wrtrade

For broker integrations:

# With Alpaca support
pip install wrtrade[alpaca]

# All broker integrations
pip install wrtrade[all]

Step 1: Research & Backtest

Start with an idea. Let's say you think a simple moving average crossover might work.

Your First Backtest

import wrtrade as wrt
import polars as pl
import numpy as np

# Get some price data (or use your own)
np.random.seed(42)
prices = pl.Series([100 * (1 + r) for r in np.cumsum(np.random.normal(0.001, 0.02, 252))])

# Define your signal
def ma_crossover(prices):
    fast = prices.rolling_mean(10)
    slow = prices.rolling_mean(30)
    return (fast > slow).cast(int) - (fast < slow).cast(int)

# Build portfolio
builder = wrt.NDimensionalPortfolioBuilder()
component = builder.create_signal_component("MA_Cross", ma_crossover)
portfolio = builder.create_portfolio("MA_Strategy", [component])

# Backtest
manager = wrt.AdvancedPortfolioManager(portfolio)
results = manager.backtest(prices)

# See the results
print(f"Total Return: {results['total_return']:.2%}")
print(f"Sortino Ratio: {results['portfolio_metrics']['sortino_ratio']:.2f}")
print(f"Max Drawdown: {results['portfolio_metrics']['max_drawdown']:.2%}")

# Detailed performance report
wrt.tear_sheet(results['returns'])

Output:

Total Return: 15.3%
Sortino Ratio: 1.45
Max Drawdown: -8.2%

=== Performance Metrics ===
Volatility (Annual): 18.5%
Sortino Ratio: 1.45
Gain-to-Pain: 2.31
Max Drawdown: -8.2%

Great! Your strategy looks promising. But is it real or just luck?

Step 2: Validate with Permutation Testing

The Problem: Many strategies look great in backtests but fail in live trading. Why? Overfitting.

The Solution: Permutation testing. This scrambles your price data thousands of times while preserving statistical properties. If your strategy only works on the exact historical sequence, it's probably overfit.

# Test if your strategy is statistically significant
tester = wrt.PermutationTester(wrt.PermutationConfig(
    n_permutations=1000,
    parallel=False  # Set to True for faster testing
))

results = tester.run_insample_test(
    prices,
    lambda p: portfolio,
    metric='sortino_ratio'
)

print(f"Your Strategy: {results['real_performance']:.2f}")
print(f"Random Average: {results['permutation_mean']:.2f}")
print(f"P-value: {results['p_value']:.4f}")

if results['p_value'] < 0.05:
    print("✓ Strategy is statistically significant!")
    print("  Your edge is unlikely due to chance.")
else:
    print("✗ Strategy may be overfit")
    print("  Performance could be due to luck.")

What does this mean?

• P-value < 0.05: Your strategy performs significantly better than random chance
• P-value > 0.05: Your strategy might just be lucky - keep researching!

Walk-Forward Testing

Even better - test your strategy on data it's never seen:

# Split data into train/test periods
results = tester.run_walkforward_test(
    prices,
    lambda p: portfolio,
    train_window=252,  # Train on 1 year
    metric='sortino_ratio'
)

print(f"Out-of-Sample P-value: {results['p_value']:.4f}")

If your strategy passes both in-sample AND out-of-sample tests, you have a robust edge.

Step 3: Optimize Position Sizing

You have a validated strategy. Now optimize how much to bet using the Kelly criterion.

The Kelly Criterion tells you the optimal fraction of capital to allocate for maximum long-term growth.

# Apply Kelly optimization
optimizer = wrt.HierarchicalKellyOptimizer(wrt.KellyConfig(
    lookback_window=252,      # Use 1 year of data
    max_leverage=1.0,         # No leverage (conservative)
    rebalance_frequency=21,   # Rebalance monthly
    risk_free_rate=0.02       # 2% risk-free rate
))

kelly_results = optimizer.optimize_portfolio(portfolio, prices, rebalance=True)

print("=== Kelly Optimization ===")
for level in kelly_results['level_optimizations']:
    print(f"Strategy: {level['portfolio_name']}")
    print(f"  Optimal Allocation: {level['kelly_weights']}")
    print(f"  Expected Sharpe: {level['metrics']['sharpe_ratio']:.2f}")

What changed? Your portfolio now uses Kelly-optimal position sizing instead of equal weights. This maximizes long-term growth while controlling risk.

Step 4: Combine Multiple Strategies

Don't put all your eggs in one basket. Combine multiple strategies for diversification.

# Define multiple signals
def trend_following(prices):
    fast = prices.rolling_mean(10)
    slow = prices.rolling_mean(30)
    return (fast > slow).cast(int) - (fast < slow).cast(int)

def momentum(prices):
    returns = prices.pct_change(20)
    return (returns > 0.05).cast(int) - (returns < -0.05).cast(int)

def mean_reversion(prices):
    sma = prices.rolling_mean(20)
    std = prices.rolling_std(20)
    z_score = (prices - sma) / std
    # Buy when oversold, sell when overbought
    return (z_score < -2).cast(int) - (z_score > 2).cast(int)

# Build hierarchical portfolio
builder = wrt.NDimensionalPortfolioBuilder()

# Create individual components
trend = builder.create_signal_component("Trend", trend_following, weight=0.4)
mom = builder.create_signal_component("Momentum", momentum, weight=0.3)
mean_rev = builder.create_signal_component("MeanRev", mean_reversion, weight=0.3)

# Combine into master portfolio
portfolio = builder.create_portfolio("Multi_Strategy", [trend, mom, mean_rev])

# View structure
manager = wrt.AdvancedPortfolioManager(portfolio)
manager.print_structure()

Output:

=== Portfolio Structure: Multi_Strategy ===
Multi_Strategy (weight: 1.000)
  Trend (weight: 0.400)
  Momentum (weight: 0.300)
  MeanRev (weight: 0.300)

Now validate and optimize this multi-strategy portfolio the same way!

Step 5: Deploy to Paper Trading

You've researched, validated, and optimized. Time to test in live markets - but safely.

Create a Deployable Strategy File

# my_strategy.py
import wrtrade as wrt
import polars as pl

def trend_signal(prices: pl.Series) -> pl.Series:
    """Your validated trend following signal."""
    fast = prices.rolling_mean(10)
    slow = prices.rolling_mean(30)
    return (fast > slow).cast(int) - (fast < slow).cast(int)

def momentum_signal(prices: pl.Series) -> pl.Series:
    """Your validated momentum signal."""
    returns = prices.pct_change(20)
    return (returns > 0.05).cast(int) - (returns < -0.05).cast(int)

def build_portfolio():
    """
    Required function for deployment.
    The CLI calls this to get your portfolio.
    """
    builder = wrt.NDimensionalPortfolioBuilder()

    trend = builder.create_signal_component("Trend", trend_signal, weight=0.6)
    mom = builder.create_signal_component("Momentum", momentum_signal, weight=0.4)

    return builder.create_portfolio("My_Strategy", [trend, mom])

# Backtest when run directly
if __name__ == "__main__":
    import numpy as np

    # Load your data
    prices = pl.Series([...])  # Your historical prices

    portfolio = build_portfolio()
    manager = wrt.AdvancedPortfolioManager(portfolio)
    results = manager.backtest(prices)

    print(f"Sortino: {results['portfolio_metrics']['sortino_ratio']:.2f}")

    # Validate
    tester = wrt.PermutationTester(wrt.PermutationConfig(n_permutations=1000))
    perm = tester.run_insample_test(prices, lambda p: portfolio, 'sortino_ratio')

    if perm['p_value'] < 0.05:
        print("✓ Ready to deploy!")

Test Locally First

python my_strategy.py

Make sure everything works before deploying.

Set Up Broker Credentials

# Get API keys from Alpaca (alpaca.markets)
# Start with paper trading account (free)

export ALPACA_API_KEY="PK..."
export ALPACA_SECRET_KEY="..."

Deploy to Paper Trading

# Deploy your strategy
wrtrade strategy deploy my_strategy.py \
    --name my_strategy \
    --broker alpaca \
    --symbols AAPL,TSLA,MSFT \
    --max-position 1000 \
    --risk-per-trade 0.02 \
    --cycle-minutes 5 \
    --kelly-optimization

# Start trading (paper money)
wrtrade strategy start my_strategy

What's happening?

• wrtrade runs your strategy in the background
• Every 5 minutes, it fetches new prices
• Generates signals using your portfolio
• Places trades with your broker (paper account)
• Automatically restarts if it crashes
• Logs everything

Monitor Your Paper Trading

# Check status
wrtrade strategy status my_strategy

# Watch logs in real-time
wrtrade strategy logs my_strategy --follow

# See all strategies
wrtrade strategy status

Output:

🟢 my_strategy: RUNNING (2h 15m)
   Status: RUNNING
   PID: 12345
   Uptime: 2h 15m
   Trades: 12 (P&L: $145.32)
   Broker: alpaca
   Symbols: AAPL, TSLA, MSFT

Let It Run

Paper trade for at least 30-60 days. Watch for:

• Does it perform as expected?
• Are slippage and fees accounted for?
• Does it handle market volatility?
• Any execution issues?

If everything looks good after paper trading, you're ready for live deployment.

Step 6: Go Live

After successful paper trading:

# Stop paper trading
wrtrade strategy stop my_strategy

# Update credentials to live account
export ALPACA_API_KEY="your_live_key"
export ALPACA_SECRET_KEY="your_live_secret"

# Deploy to live (same command, but with live credentials)
wrtrade strategy deploy my_strategy.py \
    --name my_strategy_live \
    --broker alpaca \
    --symbols AAPL,TSLA,MSFT \
    --max-position 500 \
    --risk-per-trade 0.01 \
    --kelly-optimization

# Start live trading
wrtrade strategy start my_strategy_live

Start small! Use lower position sizes and risk limits when going live.

Examples

Simple MA Crossover

# Test the example
python examples/simple_strategy.py

# Deploy it
wrtrade strategy deploy examples/simple_strategy.py \
    --name simple_ma \
    --broker alpaca \
    --symbols AAPL

wrtrade strategy start simple_ma

Multi-Strategy Portfolio

# Test the example
python examples/multi_strategy.py

# Deploy it
wrtrade strategy deploy examples/multi_strategy.py \
    --name multi_strat \
    --broker alpaca \
    --symbols AAPL,TSLA,MSFT

wrtrade strategy start multi_strat

Complete Workflow Example

Here's the entire journey in one script:

import wrtrade as wrt
import polars as pl
import numpy as np

# 1. RESEARCH - Define your strategy
def my_signal(prices):
    sma = prices.rolling_mean(20)
    return (prices > sma).cast(int) - (prices < sma).cast(int)

# 2. BACKTEST
builder = wrt.NDimensionalPortfolioBuilder()
component = builder.create_signal_component("SMA20", my_signal)
portfolio = builder.create_portfolio("SMA_Strategy", [component])

prices = pl.Series([...])  # Your data
manager = wrt.AdvancedPortfolioManager(portfolio)
results = manager.backtest(prices)

print(f"Backtest Sortino: {results['portfolio_metrics']['sortino_ratio']:.2f}")

# 3. VALIDATE - Permutation testing
tester = wrt.PermutationTester(wrt.PermutationConfig(n_permutations=1000))
perm = tester.run_insample_test(prices, lambda p: portfolio, 'sortino_ratio')

if perm['p_value'] >= 0.05:
    print("✗ Strategy not significant - back to research!")
    exit()

print(f"✓ P-value: {perm['p_value']:.4f} - Strategy is significant!")

# 4. OPTIMIZE - Kelly criterion
optimizer = wrt.HierarchicalKellyOptimizer()
kelly_results = optimizer.optimize_portfolio(portfolio, prices, rebalance=True)
print(f"✓ Kelly optimization complete")

# 5. DEPLOY - Save to file and use CLI
print("\nReady to deploy! Use:")
print("  wrtrade strategy deploy my_strategy.py --name sma20 --broker alpaca --symbols AAPL")

Why wrtrade?

Fast

Built on Polars - 10-50x faster than pandas-based frameworks. Backtest millions of data points in seconds.

Validated

Built-in permutation testing catches overfitting before you lose money.

Optimized

Kelly criterion optimization for maximum long-term growth.

Production-Ready

CLI deployment manages strategies as background processes with monitoring, logging, and auto-restart.

Simple

Clean API. The entire workflow is straightforward:

builder = wrt.NDimensionalPortfolioBuilder()
component = builder.create_signal_component("MySignal", my_func)
portfolio = builder.create_portfolio("MyStrategy", [component])
manager = wrt.AdvancedPortfolioManager(portfolio)
results = manager.backtest(prices)

API Reference

Core Components

# Basic backtesting
portfolio = wrt.Portfolio(prices, signals, max_position=float('inf'))
portfolio.calculate_performance()

# Component-based portfolios (recommended)
builder = wrt.NDimensionalPortfolioBuilder()
component = builder.create_signal_component(name, signal_func, weight=1.0)
portfolio = builder.create_portfolio(name, [components])

# Backtesting
manager = wrt.AdvancedPortfolioManager(portfolio)
results = manager.backtest(prices)

Validation

# Permutation testing
config = wrt.PermutationConfig(n_permutations=1000, parallel=False)
tester = wrt.PermutationTester(config)

# In-sample test
results = tester.run_insample_test(prices, strategy_func, 'sortino_ratio')

# Walk-forward test (recommended)
results = tester.run_walkforward_test(
    prices, strategy_func, train_window=252, metric='sortino_ratio'
)

Optimization

# Kelly criterion
config = wrt.KellyConfig(lookback_window=252, max_leverage=1.0)
optimizer = wrt.HierarchicalKellyOptimizer(config)
results = optimizer.optimize_portfolio(portfolio, prices, rebalance=True)

Metrics

# Performance metrics
wrt.tear_sheet(returns)  # Detailed report
metrics = wrt.calculate_all_metrics(returns)
rolling = wrt.calculate_all_rolling_metrics(returns, window=252)

# Available metrics:
# - volatility, sortino_ratio, sharpe_ratio
# - gain_to_pain_ratio, max_drawdown

CLI Reference

# Initialize workspace
wrtrade init

# Deploy strategy
wrtrade strategy deploy <file> \
    --name <name> \
    --broker alpaca \
    --symbols AAPL,TSLA \
    --max-position 1000 \
    --risk-per-trade 0.02 \
    --cycle-minutes 5 \
    --kelly-optimization \
    --auto-restart

# Manage strategies
wrtrade strategy start <name>
wrtrade strategy stop <name>
wrtrade strategy status [name]
wrtrade strategy logs <name> [--follow]
wrtrade strategy remove <name>

# Broker info
wrtrade broker list

Strategy File Requirements

Your strategy file must have a build_portfolio() function:

import wrtrade as wrt

def build_portfolio():
    """
    Required function - CLI calls this to get your portfolio.
    Must return a CompositePortfolio.
    """
    builder = wrt.NDimensionalPortfolioBuilder()
    # ... build your portfolio
    return portfolio

if __name__ == "__main__":
    # Optional: backtest when run directly
    pass

Testing

Run the test suite:

pytest tests/ -v

113 tests passing - Full coverage of backtesting, validation, optimization, and deployment.

Best Practices

1. Always Validate

Don't trust a backtest alone. Use permutation testing to verify statistical significance.

2. Walk-Forward Test

In-sample tests can still overfit. Walk-forward testing simulates real trading.

3. Start with Paper Trading

Always paper trade for at least 30 days before going live.

4. Use Kelly Optimization

Kelly criterion maximizes long-term growth while controlling risk.

5. Diversify

Combine multiple strategies with different return profiles.

6. Start Small

When going live, use smaller positions than in paper trading.

7. Monitor Everything

Watch logs, check performance daily, especially in first weeks of live trading.

Common Pitfalls

❌ Overfitting: Testing too many parameters until something works ✓ Solution: Use permutation testing and walk-forward validation

❌ Survivorship Bias: Only testing on stocks that survived ✓ Solution: Include delisted stocks in your data

❌ Look-Ahead Bias: Using future information ✓ Solution: Careful with rolling calculations, ensure proper alignment

❌ Ignoring Costs: Not accounting for slippage and fees ✓ Solution: Paper trade first, be conservative in backtests

❌ Over-Leverage: Taking too much risk ✓ Solution: Use Kelly criterion, start with max_leverage=1.0

Performance

• 10-50x faster than pandas-based frameworks
• Parallel permutation testing for faster validation
• Memory efficient with Polars columnar format
• Scales to millions of data points

Contributing

Contributions welcome! Please:

1. Keep it simple
2. Add tests
3. Follow existing style
4. Update docs

License

MIT License - see LICENSE file for details.

Support

• Issues: https://github.com/wayy-research/wrtrade/issues
• Examples: See examples/ directory
• Questions: Open an issue

---

Remember: Past performance doesn't guarantee future results. Always start with paper trading. Never risk more than you can afford to lose.