Distributional Fitting and Tail Analysis of Lead-Time Compositions: Nights vs. Revenue on Airbnb

Battle Count: 50

The paper is primarily about tourism/hospitality revenue forecasting and booking behavior, not financial markets or trading. However, the methodological tools (compositional data analysis, structural break detection, GPD tail analysis, Wasserstein distance, CRPS scoring) are transferable to quantitative finance contexts such as order-flow composition analysis, regime detection in market microstructure, and tail risk modeling. The volume-revenue divergence concept loosely parallels volume-price divergence in trading, but the paper does not address any trading or financial market questions directly.

Moderate complexity. Requires: (1) handling compositional data on the 365-simplex; (2) interval-censored likelihood optimization with BFGS and log-transformed parameters; (3) Bai-Perron structural break detection with HAC standard errors; (4) GPD fitting with hierarchical fallback strategy (evd, ismev, PWM, MOM); (5) GAM fitting with REML and basis dimension selection; (6) CRPS computation; (7) block bootstrap for uncertainty. Multiple R packages needed (mgcv, evd, ismev, strucchange, sandwich). The synthetic sampling procedure for GPD adds computational overhead. No code repository provided.

Data is proprietary Airbnb data from a single North American region, not publicly available. Methods are well-described with specific parameter choices (thresholds, block sizes, basis dimensions). Code is not mentioned as available. The synthetic sampling procedure (1,000 draws/day) is described but not reproducible without the underlying data. No GitHub repository is referenced.