DeepQuant Strategy Comparison Assistant

Problem

When iterating quantitative strategies, a single backtest report shows only one strategy's performance. It does not explain which strategy is more robust, where the difference comes from, or what should be changed next. This assistant places multiple completed backtests on the same evaluation frame, comparing returns, risk, risk-adjusted returns, position structure, rebalancing timing, and transaction cost rather than listing isolated metrics.

How It Works

It reads the report directory generated by deepquant-backtest, which should include portfolio.csv, trades.csv, and stock_positions.csv; summary.xlsx is optional. The workflow includes:

• Metric comparison: reports total return, annualized return, maximum drawdown, Sharpe ratio, win rate, trade count, and transaction cost, highlighting the best-performing strategy.
• Overlay charting: draws cumulative return curves on the same axis, shades maximum drawdown intervals, and optionally overlays a benchmark curve.
• Difference analysis: explains differences through return source, risk profile, and transaction cost, such as rotation-driven diversification versus single-asset beta exposure.
• Optimization notes: provides actionable directions for each strategy's weaknesses, including cash buffers, position persistence thresholds, and cost constraints.

Boundaries

The skill depends on local Python 3.8 through 3.12 and the deepquant SDK, and requires existing readable backtest reports. It does not create new backtests; it compares and diagnoses completed results. Authentication, permissions, and credential handling are security-bound: initialization checks must not be bypassed, and sensitive values such as appKey or TOKEN must not be exposed or written into files.