Overreaction as an indicator for momentum in algorithmic trading: A Case of AAPL stocks

Battle Count: 138

Highly relevant to quantitative trading. The paper directly addresses algorithmic trading strategy development using ML models for intraday momentum prediction. It provides concrete trading signals, transaction cost modeling, risk-adjusted performance metrics (Sharpe, Sortino), and statistical significance testing (Jobson-Korkie). The cross-frequency analysis (1-15 minutes) is directly applicable to HFT and intraday trading systems. The finding that ML outperforms at ultra-short horizons while classical momentum dominates at 10 minutes provides actionable guidance for strategy design.

Moderate to high complexity. Requires: (1) high-frequency market data pipeline with multiple sampling frequencies, (2) Twitter API integration and real-time tweet collection, (3) ModernFinBERT inference for emotion extraction, (4) multiple ML model training with hyperparameter tuning via randomized search and time-series CV, (5) trading strategy backtesting with transaction costs, (6) SHAP interpretability analysis, and (7) statistical testing framework. The anti-look-ahead design and embargo periods add implementation subtlety. However, all components use well-established libraries (XGBoost, scikit-learn, TensorFlow/Keras, HuggingFace transformers).

The paper provides detailed methodology including hyperparameter ranges, data sources (Massive platform, Twitter API Academic Research track), specific model architectures, and evaluation procedures. However, no GitHub repository is mentioned, and the Twitter data collection period (2019-2022) may not be fully reproducible due to API access changes. The ModernFinBERT model is publicly available on HuggingFace. Transaction cost assumptions and threshold parameters are explicitly stated.