Beyond project segments: An econometric evaluation of traffic network volume forecasting

Abstract

Misallocation of resources and suboptimal transportation projects often stem from inaccurate traffic forecasts, impacting cost-benefit analyses. This pioneering study employs econometric methods to analyze forecast accuracy across entire networks, including often-overlooked non-project segments that are critical to cost-benefit analysis like project segments. Despite an overall underestimation of 12.3% (Mean Percentage Error), project segments consistently exhibit a historical overestimation (−13.8%). Notably, unbiasedness varies among segments, with low-volume segments achieving it statistically. Our analysis reveals intriguing associations between forecast and calibration errors specific to segments. While future forecasting error is statistically linked to forecasts for all segments except project segments, a similar connection between base-year model calibration error and future forecasting error is observed for all segments except low-volume and project segments. Remarkably, low-volume segments, though lacking unbiasedness, demonstrate efficiency during base year calibration, presenting contrasting results. Policy suggestions highlight the potential for forecasting enhancement through improved base-year calibration and question the necessity for separate error tolerances for low-volume segments.