Second-hand vessel valuation: an extreme gradient boosting approach
Roar Ådland, Haiying Jia, Hans Christian Olsen Harvei, Julius Jørgensen
Abstract
We investigate the efficacy of the Extreme Gradient boosting (XGBoost) machine learning technique in desktop vessel valuation and compare it to benchmark models consisting of a LASSO regression, a Generalized Additive Model (GAM) and a Generalized Linear Model (GLM). Our data consists of of 1880 sale and purchase transactions for Handysize bulkers between January 1996 and September 2019. Using vessel-specific and market variables, we find that the XGBoost algorithm outperforms the GAM approach in its ability to model complex non-linear relationships between multiple variables. When fitting the XGBoost model, we find that vessel age, timecharter rates and fuel efficiency are the most important variables. Our findings are important for investors, shipowners and ship financiers in the maritime industry.