Crude Oil Tanker Freight Rate Forecasting Based On ARIMA And Artificial Neural Network

Crude oil tanker freight rate is one of the most important economic indexes in crude oil shipping market. It is considered as barometer reflecting crude oil market's variation. Crude oil tanker freight rate forecasting has become a hot issue for academic researchers and practitioners. The crude oil tanker freight rate has the characters of high volatility, nonlinearity and irregularity. Since the establishment of OPEC, the fluctuations in the crude oil tanker market are more volatile than ever. Furthermore, some important events have a significant impact on tanker freight rate fluctuation. Meantime, these characteristics also make it difficult to predict crude oil tanker freight rate. In the past decades, various methods and models were presented to predict tanker freight rate, but their forecasting performance is not satisfactory.The analysis and prediction of crude oil tanker freight rate are from two points of view. First, the thesis concentrated on the crude oil demand market and tanker shipping market to analyze the crude oil shipping market. The impact factors which are proved reserves, production, and consumption and trade movements of crude oil are discussed. While the routes of crude oil tankers and tanker fleets are also introduced.The Baltic Dirty Tanker Index published by Baltic Mercantile and Shipping Exchange is chosen as the predictive object. Two kinds of models are introduced. Firstly, a study of simulation and forecasting performance of ARIMA time series model is conducted to BDTI. In the process of BDTI forecasting with ARIMA, stationarity test should be firstly conducted. After estimating the parameters and testing, an ARIMA forecasting model is established. Secondly, using the ANN analysis, the thesis proposes a BP nonlinear model to forecast BDTI. After making original data processed, the next step is establishing the network structure and defining the learning functions. A BP forecasting model is obtained after training. Comparing forecasting results from the two models, the thesis concludes that the BP forecasting model performs better in the long term prediction. Both of the two models didn't take the risk factors into consideration, such as the economy crisis, which affected the real market a lot.