optimize-strategy

name	optimize_strategy
description	guide user to optimize a strategy for msd backtest
tags	["msd","strategy","backtest","optimize"]
version	1
author	elsejj

Workflow

1. Follow the user request to understand the strategy idea, write a initial basic strategy.
2. Do backtest
3. Review the metrics, analyze the shortcomings and make improvement suggestions.
4. If the strategy is not satisfied, start a new round (increment ${round_id}), follow the suggestions to improve and write strategy again, and goto 2.
5. After optimize, write a full report

Notes

• Don't create plan before review, all optimize steps are guided by backtest results.
• If user had forgot to provide a symbol list, ask to provide one, then create or copy it to workspace's symbols.txt.
• All variable like ${var} in command should be replaced with actual values.
• Don't read/write files out of the workspace, except reading .env file, or files user explicitly point out.

Terms

• workspace: the sub-directory for this strategy under optimize folder. All files you created, saved to this folder. Reference it with ${workspace}.
• round_id: the id of the round, starts from 1, increase by 1 each round, zero-padded to 2 digits, e.g. 01, 02. Reference it with ${round_id}.
• symbols: a list of symbols to backtest. symbols.txt is the default file name, it can be changed by user. It's placed in the workspace. Reference it with ${symbols}.
• strategy_file: The name of the strategy file, it should be strategy.${round_id}.py. Reference it with ${strategy_file}.
• backtest_output_file: The name of the backtest output file, it should be backtest.${round_id}.json. Reference it with ${backtest_output_file}.
• report_file: The name of the report file, it should be report.${round_id}.md. Reference it with ${report_file}.

Action details

Write strategy

1. Create a strategy_file for this round. Follow previous round's report_file to improve.
2. You can freely experiment from multiple perspectives, add constraints, use more data, test different parameters, while keeping the user's core needs in mind.
3. Using write-strategy skill to write it.

Do backtest

• Backtest the strategy with uv run python -m mbt.run.cli ${strategy_file} -s ${symbols} -o ${backtest_output_file}. This will be generate backtest_output_file, which contains detailed performance metrics for each symbol.
• You can use uv run python -m mbt.run.cli --help to get more information about the command.
• You can also edit .env to modify some fixed arguments

Review the metrics

• Because backtest_output_file may be very large, you should use uv run python -m mbt.run.review ${backtest_output_file} to explore the data instead reading the file directly. There are 3 sub command of uv run python -m mbt.run.review ${backtest_output_file}:
  • stat: python -m mbt.run.review ${backtest_output_file} stat to review the statistics of the metrics.
  • detail: python -m mbt.run.review ${backtest_output_file} detail -s ${symbol} to review the metrics of a specific symbol.
  • sort: python -m mbt.run.review ${backtest_output_file} sort -m ${metric} to show the metric sorted. the metric can be any of:
    • profit : the final profit of the strategy.
    • profit_original : the final profit during the test but without apply the strategy.
    • alpha : the alpha of the strategy.
    • beta : the beta of the strategy.
    • sharpe : the sharpe ratio of the strategy.
    • information_ratio : the information ratio of the strategy.
    • ir : the information ratio of the strategy.
    • sortino : the sortino ratio of the strategy.
    • treynor : the treynor ratio of the strategy.
    • max_drawdown : the maximum drawdown of the strategy.
    • calmar : the calmar ratio of the strategy.
    • win_rate : the win rate of the strategy.
    • avg_win : the average win of the strategy.
    • avg_loss : the average loss of the strategy.
    • profit_factor : the profit factor of the strategy.
    • fee_rate : the fee rate of the strategy.
    • trade_count : the trade count of the strategy.
• All review result are JSON format and printed to stdout.
• You may run the review commands multiple times to explore the data.
• Write a report to analyze the metrics to report.${round_id}.md using chinese, mention the key findings and insights.
• Point out the shortcomings and suggests to improve.
• Use the report, rewrite a strategy file for next round

Full Report

• Use chinese when writing the full report, write it to report.md.
• The report should include the following sections:
  • Strategy Introduction
  • Backtest results summary, with top 10 better and worse symbols.
  • Optimization process and key insights
  • Explanation of advantages
  • Reflection on the shortcomings and limitations
  • Conclusion and outlook