| dc.description.abstract | This thesis investigates the applicability of extreme gradient boosting (XGBoost) compared
to the generalized additive model (GAM) approach to create a desktop valuation model
of second-hand Handysize bulkers. The data basis is 1880 unique sales transactions in the
period from January 1996 to September 2019 derived from Clarkson Research. This thesis
contributes to existing literature by applying an XGBoost algorithm and a data-driven
GAM approach to vessel valuation.
Using vessel-specific and market variables, we find evidence that the XGBoost algorithm
is more suited for desktop valuation of Handysize bulk carriers than the GAM approach.
The predictive power of XGBoost in this instance could be caused by its ability to model
complex relationships between multiple variables. Supporting existing research in maritime
economics, we find linear models to be inadequate at vessel valuation. When fitting the
XGBoost model, vessel age at sale, timecharter rates and fuel efficiency index are identified
as the most important variables. We also find vessels priced over $20 million to be
significantly harder to predict. This could be a consequence of a scarce data basis, vessel
characteristics not present in the data or the majority of these transactions occurring
during the financial super cycle years of 2003 to 2009.
The XGBoost algorithm facilitates accurate predictions of vessel prices based on vessel
characteristics and market conditions, and provides a useful machine learning framework
for desktop valuation. The flexibility of the XGBoost algorithm can make it highly usable
for investors, ship owners and other market players in the maritime industry. | en_US |
