Investigating the predictive ability of AIS-data : the case of arabian gulf tanker rates

Abstract

This thesis investigates whether information derived from AIS-data incorporates superior information about future freight rates. Specifically, we assess if such data improve the ability to predict TD3 spot rates between August 2015 and mid-February 2016. The ability to anticipate short-term fluctuations in freight rates is a key component to long-term profitability for both shipowners and charterers, making the purpose of this thesis an important objective. The AIS-data contain information about 81,728 individual shipments of crude oil between 2013 and mid-February 2016, and are reduced to 53,116 observations after proper cleansing. For analysis purposes, information deemed relevant are converted into weekly time series, ending up with 162 observations in total. Data-driven selection tools are then used to identify the most powerful predictors of future TD3 rates, and a multivariate VAR is specified in line with these results. In order to investigate the relative performance of information derived from AIS-data, a one-step-ahead forecast is conducted, and evaluated against an univariate ARMA and a multivariate VAR solely based on publicly available data. Our results suggest that multivariate models perform relatively better than univariate models to predict future freight rates. Further, comparing error measures from the two multivariate VAR models specified, we find weak evidence in favour of using information from AIS-derived data for predictive purposes.