Modeling Trade Durations under Temporal Granularity Effects in Forex Markets

Battle Count: 88

The paper addresses a fundamental market microstructure issue in FX markets - the accurate modeling of trade durations. This is directly relevant to algorithmic execution, optimal trade timing, transaction cost estimation, and liquidity assessment. The finding that ignoring heaping leads to biased parameter estimates and distorted inference has practical implications for any quantitative strategy relying on duration-based models in FX. However, the paper is primarily econometric/methodological rather than strategy-focused.

The model requires implementing a two-component mixture distribution with a complex score function involving truncated normal distributions and generalized gamma components. The score-driven dynamics add recursive updating of the scale parameter. Maximum likelihood estimation involves numerical optimization of 7 parameters jointly. The paper notes computational intensity for datasets >10M observations. The gasmodel R package provides some infrastructure but requires modification for the specific mixture distribution.

The paper references the gasmodel R package (Holý, 2026) modified for the proposed mixture distribution. However, the empirical data comes from Refinitiv Eikon (proprietary). The model specification, estimation procedure, and simulation design are described in detail. Full parameter values for simulations are provided. The R package is publicly available but the specific modification for the GA-ACD model is not explicitly linked to a repository.