순환신경망 모델을 활용한 벌크 운임지수 예측

Forecasting Bulk Market Indices with Recurrent Neural Network Models

One of the characteristics of the shipping market is extreme volatility. The play in the volatile market requires players to make cautious decisions based on scientific analysis. Market risk management is of utmost importance in the shipping market and market forecasting is an important element in the management process. This paper deals with forecasting issues in the dry bulk shipping market. Despite the fact that bulk shipping is dominated by spot trading, there are few papers that apply scientific models to short-term forecasting. This paper employs recurrent neural networks that currently draws attention in time-series forecasting. More specifically, Elman neural networks and Jordan neural networks were applied to improve the forecasting performance over traditional econometric models and simple multi-layer perceptron models. Monthly observations of the BDI, BCI and BPI were used for spot forecasting. The result showed that the proposed two models outperformed the ARIMA model and the MLP model. The Elman model performed better for the time series with high volatility and the Jordan model demonstrated better performance for the time series with a modest volatility. The BDI is composed of sub-indices with varying levels of volatility. Hence in the case of the BDI forecast, the Jordan networks performed better than the Elman networks. The results will provide scientific grounds for chartering managers to make better decisions concerning the most active spot transactions.