Retail Investor Horizon and Earnings Announcements

Battle Count: 71

Highly relevant for quantitative trading. The paper provides a concrete, implementable trading strategy (zero-cost long-short within earnings-surprise bins) with documented risk-adjusted alphas of 30-40 bps per month. The horizon-based signal is distinct from standard factor exposures (value, profitability, momentum). The strategy is monthly rebalanced, making it practical for implementation. The paper also provides detailed sample construction, event study methodology, and factor-adjusted performance metrics. The NLP-based sentiment and information content analysis adds an alternative data dimension. However, the reliance on StockTwits data and the need for self-reported holding periods may limit scalability.

Moderate complexity. The core regression analysis uses standard econometric techniques (OLS with fixed effects, propensity score matching, Fama-French factor models). The NLP component requires BERT-based sentence transformers and sentiment extraction pipelines. Data acquisition from StockTwits, I/B/E/S, CRSP, and COMPUSTAT requires institutional access. The trading strategy construction (sorting into 11 bins, forming long-short portfolios, monthly rebalancing) is straightforward. The main complexity lies in the data pipeline: merging StockTwits posts with earnings announcements, classifying users by self-reported holding period, and computing sentiment metrics.

The paper uses publicly available data (StockTwits, I/B/E/S, CRSP, COMPUSTAT) and well-documented methodologies (DellaVigna and Pollet 2009 approach, Fama-French factor models). However, StockTwits data access may be limited, and the specific NLP pipeline (Vamossy and Skog 2025) for sentiment extraction is referenced but may require access to the authors' code. The BERT model used (bert-base-nli-mean-tokens) is publicly available. No GitHub repository is mentioned. The sample construction steps are detailed in Table 1, and regression specifications are fully described.