金融时序数据Foundation Model微调与策略生成提示词

You are a quantitative finance AI researcher specializing in foundation models for financial time series. Help me design a complete workflow for fine-tuning a financial time series foundation model and generating trading strategies.

Task

Given a pre-trained financial time series foundation model (like Kronos or TimesFM), I need to:

Phase 1: Data Preparation

• Define the target financial instruments: [SPECIFY: stocks/crypto/forex/futures]
• Time granularity: [SPECIFY: 1min/5min/1h/daily]
• Feature engineering checklist:
  • Price features (OHLCV, returns, log-returns)
  • Technical indicators (RSI, MACD, Bollinger Bands, ATR)
  • Market microstructure features (bid-ask spread, order flow imbalance)
  • Cross-asset correlation features
  • Sentiment scores from news/social media

Phase 2: Model Fine-tuning

• Design the fine-tuning strategy (full vs LoRA vs prefix-tuning)
• Define loss functions suitable for financial forecasting
• Implement proper train/validation/test split respecting temporal ordering
• Anti-lookahead bias checks

Phase 3: Strategy Generation

• Convert model predictions into actionable trading signals
• Position sizing using Kelly criterion or risk parity
• Risk management rules (stop-loss, max drawdown, exposure limits)
• Backtesting framework with realistic transaction costs and slippage

Phase 4: Evaluation

• Metrics: Sharpe ratio, Sortino ratio, max drawdown, Calmar ratio, win rate
• Statistical significance tests (bootstrap, walk-forward analysis)
• Regime analysis (bull/bear/sideways market performance)

Provide complete Python code structure and configuration for each phase.