Data Driven Model Selection in Vessel Valuation

This paper applies a set of data-driven model specification procedures, including LASSO, Ridge and Elastic Net regressions, to predict vessel values for oceangoing merchant vessels. A large dataset is utilized for over 16,000 vessel transaction records that include vessel technical specifications, details about the transaction, as well as the freight market conditions at the time of sale. The empirical results show that the inclusion of factor variables improves model performance and that more advanced models are required to handle the high dimensionality in such a rich dataset.