金融时间序列基础模型应用方案设计师

You are a senior quantitative researcher and ML engineer specializing in financial time series modeling.

I want to leverage foundation models for financial time series analysis. Help me design an end-to-end solution.

Context:

• Asset class: [stocks / crypto / forex / commodities / bonds]
• Data frequency: [tick / 1min / hourly / daily]
• Target task: [price forecasting / volatility prediction / anomaly detection / regime classification / portfolio optimization]
• Data sources available: [OHLCV, order book, sentiment, macro indicators]
• Time horizon: [intraday / swing (1-5 days) / medium-term (1-3 months)]

Please provide:

1. Model selection analysis: Compare foundation model approaches (Kronos, TimesFM, Lag-Llama, Moirai, Chronos) vs traditional methods (ARIMA, Prophet, LSTM) for my specific use case
2. Data pipeline design: Preprocessing, feature engineering, tokenization strategy for financial data
3. Fine-tuning strategy: How to adapt the foundation model to my specific asset/market with limited data
4. Backtesting framework: Proper walk-forward validation, avoiding look-ahead bias, realistic transaction cost modeling
5. Risk management integration: How to use model uncertainty estimates for position sizing
6. Production deployment: Real-time inference pipeline, model monitoring, drift detection
7. Regulatory considerations: Model explainability requirements for financial applications

Include Python code snippets for key components. Highlight potential pitfalls specific to financial ML (overfitting, regime changes, non-stationarity).