dc.description.abstract | This thesis focuses on the application of machine learning for vessel valuation. In the following paper, we present four different models and conclude that supervised machine learning models such as Catboost exhibit predictive prowess in estimating vessel prices. The CatBoost model is compared against a PLS/PCA model, Lasso and a traditional linear regression model. We find conclusive evidence that linear regression is not effective in predicting vessel prices.
Furthermore, CatBoost proves to be an ideal solution to vessel valuation due to its natural ability to encode categorical variables efficiently. The model found that the most important variables that affect price are age at sale, freight rates and one-year yield bond prices. The findings support previous research in this topic. Another reason why CatBoost can be very useful for vessel valuation is that the algorithm uses an extreme gradient boosting approach that makes it immune to multicollinearity between predictors. The results from CatBoost exhibit the lowest measurement errors and do not indicate any signs of overfitting.
The data used in this thesis is provided by the Clarkson World Fleet register. The data set contains more than 17,700 observations focused on five different vessel types: bulk, container, gas carrier, tanker, tanker chem. There are roughly 22 numerical variables and 20 categorical ones. Other macro variables such as interest rates, freight rates and exchange rates were added into the models to gauge the overall effect that macroeconomics has on vessel prices.
Machine learning algorithms facilitate accurate predictions by analyzing numerous independent variables and presenting the top factors that mostly affect the dependent variable (i.e. vessel price). Considering the efficiency and accuracy that machine learning algorithms like CatBoost offer, we suggest that CatBoost is extremely useful for asset valuation in the maritime industry.
Keywords – Vessel Valuation, Machine Learning, CatBoost, PCA, Lasso Regression | en_US |