Forecasting and trading in the crude tanker FFA market : forecasting and applying trading strategies on crude tanker forward freight agreements using neural networks and AIS-data
Abstract
The objective of this thesis is to forecast derivative prices of Forward Freight Agreements (FFAs) using machine learning techniques and investigate the profitability of implementing quantitative trading strategies. The thesis concentrates on two dirty tanker routes: TD3C, transporting oil from Ras Tenura in the Middle East to Ningbo in the Far East, and TD20, transporting oil from Nigeria in West Africa to Rotterdam in Europe.
The machine learning model predicts the future daily price movements of the individual FFA contracts, and the daily price spread of the FFA pair using a Long-Short-Term Memory (LSTM) Neural Network (NN) machine learning methodology. The model benefits from Automatic Identification System (AIS) and voyage contracts data when constructing proxies for supply, demand and geographical distribution. To capture economic development, the model utilizes macroeconomic and financial data. With a forecasting horizon of one day, the findings suggest that the LSTM model outperforms Vector Autoregressive (VAR) and Random Walk (RW) benchmark models.
To generate profitable trading signals, the forecasted individual routes and the directly forecasted price spread make use of two quantitative trading strategies: A Simple Long Short strategy and a Bollinger Band strategy. The strategies compare trading signals generated from the VAR and LSTM model with a Buy-and-Hold benchmark strategy (B&H). The results suggest that the Simple Long Short trading signals generated from the LSTM model is profitable when implemented on individual FFAs, but not profitable when implemented on the FFA pair. Conversely, the Bollinger Bands strategy combined with LSTM model is profitable when implemented on the FFA pair, but not profitable when implemented on the individual FFAs. The LSTM model combined with the two strategies outperforms the VAR model and the B&H benchmark.