delta-f/deltafstation/backtest

Backtest Skill

Context

Server injects this block for backtest-related user messages (keyword gating is outside this file).

Data

Market data uses yfinance; pass symbol as a yfinance ticker (e.g. 000001.SS, AAPL, GLD).

Dates

For relative ranges (e.g. last 2 years), use the Server date line in the system message as end_date, compute start_date, both YYYY-MM-DD.

Goal

1. Input can be as little as a trading symbol (uses default BOLLStrategy if it exists under data/strategies).
2. For a new or custom strategy (e.g. RSI, KDJ): first call `ensure_strategy` with class_name and full source_code, then call `run_backtest_auto` with symbol and strategy_id equal to that class name.
3. If the user only asks for a quick demo and BOLLStrategy is present, you may skip ensure_strategy and only call run_backtest_auto.

Tools

1. ensure_strategy (when you need new code)

• Required: class_name, source_code (complete file).
• Optional: file_basename, overwrite.
• Strategy must subclass BaseStrategy from deltafq.strategy.base and implement generate_signals(self, data: pd.DataFrame) -> pd.Series with values in {-1, 0, 1} (or as engine expects).

Indicator and signal policy

• Do not default to deltafq.indicators.TechnicalIndicators or deltafq.strategy.SignalGenerator. That package does not guarantee every indicator you need; treat it as optional, not the standard path.
• You (the model) implement whatever math is required—typically pandas / numpy on data["Open"], High, Low, Close, Volume—and derive the signal series yourself. Same for any custom or niche factor.

Minimal shape (structure only — you fill in all indicator and signal logic):

import numpy as np
import pandas as pd
from deltafq.strategy.base import BaseStrategy
class MyStrategy(BaseStrategy):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)
    def generate_signals(self, data: pd.DataFrame) -> pd.Series:
        close = data["Close"]
        # Example: roll your own rules; replace entirely for RSI/KDJ/…
        fast = close.rolling(window=5).mean()
        slow = close.rolling(window=20).mean()
        raw = np.where(fast > slow, 1, np.where(fast < slow, -1, 0))
        return pd.Series(raw, index=data.index, dtype=int)

For interface-only reference (uses deltafq helpers internally): `data/strategies/boll_strategy.py`.

2. run_backtest_auto

• Required: symbol
• Optional: strategy_id (defaults to BOLLStrategy), dates, capital, fees, trade_preview_count
• If strategy_id cannot be loaded from data/strategies, the tool returns `strategy_not_found` — use ensure_strategy first, then retry.

Tool behavior expectations

run_backtest_auto should:

1. Ensure/refresh local CSV data for symbol.
2. Resolve strategy: load strategy_id or default BOLLStrategy; do not auto-generate placeholder files.
3. Return structured result JSON with status, result_id, resolved, summary_metrics, trade_preview.

Response style

After tool returns:

• Summarize key metrics in concise Chinese (include summary_metrics.total_trades and summary_metrics.avg_trades_per_day when present).
• If ensure_strategy was used, mention the saved class name and file.
• If strategy_not_found was returned, explain that the model must write the strategy with ensure_strategy first.