The Quarter-Hour Effect: Periodic Algorithmic Trading and Return Predictability in Cryptocurrency Futures

Battle Count: 81

Highly relevant to quantitative trading. Documents a systematic, out-of-sample predictable pattern in cryptocurrency futures at quarter-hour boundaries. Provides actionable insights for scheduled algorithmic traders (condition execution timing on forecasts or randomize to reduce synchronization) and liquidity providers (adjust quotes and inventory ahead of quarter-hour openings). The LASSO-based forecasting framework with 40 predictors is directly implementable. The finding that boundary order imbalance predicts returns over 4-12 hours has clear implications for medium-horizon trading strategies. The ACM methodology is a novel diagnostic tool for detecting phase-specific dependence in any high-frequency dataset.

Moderate complexity. The core forecasting model uses standard LASSO regression with 40 predictors (12 lags + 28 technical indicators) on 15-minute OHLCV bars, requiring no low-latency infrastructure (predictors fixed 15+ minutes before boundary). The ACM requires custom sign-based autocorrelation computation across clock phases. The two-stage decomposition with moving-block bootstrap (B=1000) is computationally intensive but straightforward. Main challenges: constructing the 28 technical indicators correctly, implementing the rolling-window walk-forward protocol, and handling the nested clock-time interactions in predictive regressions. Data processing of millisecond-level aggregate trades is non-trivial but well-documented.

Data is publicly available from Binance's data portal. Replication code promised per journal policy. Independent data validation conducted by Wade Kimbrough. Detailed methodology, variable definitions, and technical indicator formulas provided in appendix. Minor missing-data episodes in official files noted but deemed immaterial. Funding from Ripple UBRI acknowledged with no role in research design.