Skip to content

Latest commit

 

History

History
52 lines (36 loc) · 2.19 KB

optimizing.md

File metadata and controls

52 lines (36 loc) · 2.19 KB

Optimizing

Passivbot's configuration can be automatically optimized through iterative backtesting to find optimal parameters.

Usage

python3 src/optimize.py [path/to/config.json]

Defaults to configs/template.json if no config specified.

Results Storage

Optimization results are stored in `optimize_results/`` with filenames containing date, exchanges, number of coins, and unique identifier. Each result is appended as a single-line JSON string containing analysis and configuration.

Analysis

The script automatically runs src/tools/extract_best_config.py after optimization to identify the best performing configuration, saving the best candidate and the pareto front to optimize_results_analysis/.

Manual analysis:

python3 src/tools/extract_best_config.py path/to/results_file.txt

Performance Metrics

Based on daily equity changes: daily_eqs = equity.groupby(day).pct_change()

Key Metrics:

  • adg: Average daily gain (daily_eqs.mean())
  • mdg: Median daily gain
  • gain: Final gain (balance[-1] / balance[0])
  • drawdown_worst: Maximum peak-to-trough equity decline
  • drawdown_worst_mean_1pct: Mean of the 1% worst drawdowns on daily equity samples
  • expected_shortfall_1pct: Average of worst 1% losses (CVaR)

Risk Ratios:

  • sharpe_ratio: Risk-adjusted return (adg / daily_eqs.std())
  • sortino_ratio: Downside risk-adjusted return (adg / downside_eqs.std())
  • calmar_ratio: Return to max drawdown ratio (adg / drawdown_worst)
  • sterling_ratio: Return to average worst 1% drawdowns ratio (adg / drawdown_worst_mean_1pct)
  • omega_ratio: Ratio of gains to losses
  • loss_profit_ratio: Absolute loss sum to profit sum ratio
  • equity_balance_diff_neg_max: greatest distance between balance and equity when equity is less than balance
  • equity_balance_diff_neg_mean: mean distance between balance and equity when equity is less than balance
  • equity_balance_diff_pos_max: greatest distance between balance and equity when equity is greater than balance
  • equity_balance_diff_pos_mean: mean distance between balance and equity when equity is greater than balance

Suffix _w indicates weighted mean across 10 temporal subsets (whole, last_half, last_third, ... last_tenth).