Sources and Nonlinearity of High Volume Return Premium: An Empirical Study on the Differential Effects of Investor Identity versus Trading Intensity (2020-2024)

Battle Count: 92

Highly relevant for quantitative trading strategies. The paper identifies actionable signals: (1) Institution-led abnormal volume events with high conviction (Q4, market cap normalized) predict +12.12% CAR over 50 days; (2) Foreign-led high-conviction events predict +13.30% CAR; (3) Retail-led events provide zero predictive signal regardless of intensity. The key insight that market cap normalization (conviction) outperforms trading value normalization (participation) is directly applicable to signal construction. The mid-cap dominance finding (4.00% CAR vs 1.43% large-cap) provides a universe selection criterion. The threshold effect (Q4 >> Q3) suggests binary signal construction rather than continuous intensity measures. However, the Korean market focus and specific data requirements (investor-type net buying data) limit direct applicability to other markets.

Moderate complexity. The core methodology (event study with quartile sorts) is straightforward to implement. Key requirements: (1) Investor-type-specific net purchase data (available via pykrx for Korean markets); (2) Market capitalization data for normalization; (3) ADTV calculation with 20-day moving average; (4) Event classification using argmax rule; (5) CAR calculation over [-50, +50] windows. The main challenge is data acquisition (investor-type trading data) and ensuring proper event independence (50-day exclusion). No complex ML models required - primarily statistical analysis with quartile sorting and correlation testing.

The paper provides detailed methodology descriptions, data sources (Daishin Securities Creon Plus API, pykrx library), event definitions (ADTV > 5), and sample exclusion criteria. However, data availability is restricted due to data provider terms of service. The pykrx library is open-source, but the Creon Plus API requires a subscription. No GitHub repository is provided. The event study methodology is well-documented and follows established practices (An et al. 2006, Lyon et al. 1999, Petersen 2008).