Improving Transport Demand Forecasting: Ex-post Evaluation Using Smart Card Data

Abstract

Transport infrastructure investments represent some of the largest public expenses, yet systematic ex-post evaluation of demand forecasts remains uncommon, perpetuating systematic biases and limiting institutional learning. This study develops a systematic ex-post evaluation framework using smart card data to investigate multimodal transport demand forecast accuracy, focusing on the Hoekse Lijn metro conversion in Rotterdam. Using OV-chipkaart data from 2020-2025, we compare 2015 forecasts with observed ridership through a four-step methodology: demand reconstruction, systematic comparison, diagnostic analysis via cluster-based station typologies, and quantitative attribution analysis. After correcting for COVID-19 impacts (-12%), results reveal substantial systematic deviations: eight of ten stations underperformed forecasts by 9-40%, with westbound overestimation exceeding 50% at most locations. Despite lower station-level boardings, passenger kilometres exceeded forecasts by 18\% due to longer trip lengths, while beach-related travel generated up to 27 times normal weekday boardings on peak days. Attribution analysis identifies that 60% of residual deviations stem from network representation errors (30%), outdated parameters (20%), socioeconomic forecast errors (5%), and unmodelled e-bike competition (10%), revealing four systematic biases: spatial asymmetry, directional bias, trip purpose misalignment, and underestimated trip length distributions. The study proposes four improvement strategies: systematic ex-post validation using automated data sources, adaptive parameter calibration, scenario-based uncertainty management, and integrated network planning. This requires shifting from "predict-and-forget" to "predict-and-learn" approaches, transforming forecasting into continuous cycles of planning, evaluation and improvement for more accountable, evidence-based transport planning practices.