Skip to content

Backtesting Framework

Overview

Simulate trading strategies on historical data to evaluate performance before risking real capital.

Quick Start

require 'sqa'

# Load stock data
stock = SQA::Stock.new(ticker: 'AAPL')

# Create backtest
backtest = SQA::Backtest.new(
  stock: stock,
  strategy: SQA::Strategy::RSI,
  initial_cash: 10_000,
  commission: 1.0
)

# Run simulation
results = backtest.run

# View results
puts "Total Return: #{results.total_return}%"
puts "Sharpe Ratio: #{results.sharpe_ratio}"
puts "Max Drawdown: #{results.max_drawdown}%"
puts "Win Rate: #{results.win_rate}%"

Configuration

Basic Parameters

backtest = SQA::Backtest.new(
  stock: stock,                    # SQA::Stock object
  strategy: SQA::Strategy::MACD,   # Strategy class
  initial_cash: 10_000,            # Starting capital
  commission: 1.0,                 # Per-trade commission
  position_size: 100               # Shares per trade
)

Results Object

The backtest.run method returns a results object with:

Return Metrics

  • total_return - Total percentage return
  • annualized_return - Return per year
  • benchmark_return - Buy-and-hold comparison

Risk Metrics

  • sharpe_ratio - Risk-adjusted return
  • sortino_ratio - Downside risk-adjusted return
  • max_drawdown - Largest peak-to-trough decline
  • volatility - Standard deviation of returns

Trading Metrics

  • num_trades - Total number of trades
  • win_rate - Percentage of profitable trades
  • avg_win - Average winning trade
  • avg_loss - Average losing trade
  • profit_factor - Gross profit / gross loss

Portfolio Metrics

  • final_value - Ending portfolio value
  • portfolio_value - Value over time (array)
  • trades - Full trade history

Complete Example

# Load data
stock = SQA::Stock.new(ticker: 'AAPL')

# Test multiple strategies
strategies = [
  SQA::Strategy::RSI,
  SQA::Strategy::MACD,
  SQA::Strategy::BollingerBands
]

strategies.each do |strategy|
  backtest = SQA::Backtest.new(
    stock: stock,
    strategy: strategy,
    initial_cash: 10_000
  )

  results = backtest.run

  puts "\n#{strategy.name} Results:"
  puts "-" * 40
  puts "Return: #{results.total_return.round(2)}%"
  puts "Sharpe: #{results.sharpe_ratio.round(2)}"
  puts "Max DD: #{results.max_drawdown.round(2)}%"
  puts "Trades: #{results.num_trades}"
  puts "Win Rate: #{results.win_rate.round(2)}%"
end

Advanced Usage

Custom Date Range

# Backtest specific period
stock = SQA::Stock.new(
  ticker: 'AAPL',
  start_date: '2020-01-01',
  end_date: '2023-12-31'
)

backtest = SQA::Backtest.new(stock: stock, strategy: strategy)
results = backtest.run

Parameter Optimization

# Test different RSI thresholds
(20..40).step(5).each do |oversold|
  results = test_strategy_with_params(oversold)
  puts "RSI #{oversold}: Return = #{results.total_return}%"
end

Walk-Forward Testing

# Train on first 70%, test on last 30%
train_end = (stock.df.height * 0.7).to_i
train_stock = stock.df[0...train_end]
test_stock = stock.df[train_end..-1]

# Optimize on training data
best_params = optimize_on(train_stock)

# Validate on test data
results = backtest_with_params(test_stock, best_params)

Visualization

# Plot equity curve
require 'gruff'

g = Gruff::Line.new
g.title = "Portfolio Value Over Time"
g.data("Portfolio", results.portfolio_value)
g.write('equity_curve.png')

Best Practices

1. Use Enough Data

  • Minimum: 252 trading days (1 year)
  • Recommended: 1260+ days (5 years)
  • Include different market conditions

2. Account for Costs

backtest = SQA::Backtest.new(
  commission: 1.0,        # Per trade
  slippage: 0.001         # 0.1% slippage
)

3. Avoid Overfitting

  • Use walk-forward analysis
  • Test on out-of-sample data
  • Limit parameter optimization

4. Compare to Benchmark

if results.total_return < results.benchmark_return
  puts "⚠️  Strategy underperforms buy-and-hold"
end

5. Check Multiple Metrics

Don't rely on return alone: - Sharpe ratio (risk-adjusted) - Max drawdown (worst case) - Win rate (consistency) - Profit factor (quality)

Common Pitfalls

❌ Survivorship Bias: Only testing stocks that still exist
❌ Look-Ahead Bias: Using future data
❌ Curve Fitting: Over-optimizing parameters
❌ Ignoring Costs: Unrealistic commission/slippage
❌ Small Sample: Too few trades for statistical significance