Evaluation of forecasting models for air cargo

Abstract

Purpose – Virgin Atlantic Cargo is one of the largest air freight operators in the world. As part of a wider strategic development initiative, the company has identified forecasting accuracy as of strategic importance to its operational efficiency. This is because accurate forecast enables the company to have the right resources available at the right place and time. The purpose of this paper is to undertake an evaluation of current month-to-date forecasting utilized by Virgin Atlantic Cargo. The study employed demand patterns drawn from historical data on chargeable weight over a seven-year-period covering six of the company's routes. Design/methodology/approach – A case study is carried out, where a comparison between forecasting models is undertaken using error accuracy measures. Data in the form of historical chargeable weight over a seven-year-period covering six of the company's most profitable routes are employed in the study. For propriety and privacy reasons, data provided by the company have been sanitized. Findings – Preliminary analysis of the time series shows that the air cargo chargeable weight could be difficult to forecast due to demand fluctuations which appear extremely sensitive to external market and economic factors. Originality/value – The study contributes to existing literature on air cargo forecasting and is therefore of interest to scholars examining the problems of overbooking. Overbooking which is employed by air cargo operators to hedge against “no-show” bookings. However, the inability of air cargo operators to accurately predict cargo capacity unlikely to be used implies that operators are unable to establish with an aspect of certainty their revenue streams. The research methodology adopted is also predominantly discursive in that it employs a synthesis of existing forecasting literature and real-life data for accuracy analysis.

Deep learning-based container throughput forecasting: a triple bottom line approach

“…According to Wang et al (2013), hybrid methods are observed to produce better forecasts. Many other similar studies also claim that the hybrid models yield higher accuracy as compared to a single model applied to the given time series (Hansen and Nelson, 2003;Khashei and Bijari, 2010;Klindokmai et al, 2014;Tseng et al, 2002;Zhang, 2003). The prediction results obtained in this study are also in accordance with this comprehension.…”