| Summary: | Each section of the shipping market including the Newbuilding, Freight, Second-hand and Demolition markets has its own unique structure and individual internal parameters. Internal parameters can influence one or more parameters in their own and other markets. This makes the shipping markets, and each of their sections, a complex environment. Additionally, some external elements, such as inflation, political issues and economic policies, will affect certain outcomes. In such an environment, the main problem for creation of a "market model" is to recognise the most effective and influential input parameters on a set of desired outputs whilst considering the time-dependant nature of the data. In this study, the traditional multivariate analysis methods have been implemented to try and create the best model of the Demolition market and use the created model to forecast the market. However, the accuracy of the model is poor. Then a new approach, based on the Artificial Neural Networks (ANN) methodology, has been implemented to model the market and consequently forecast the market. Both static and dynamic ANNs were implemented, trained and tested for various internal and external inputs and the desired outputs of the Demolition market to find out the best combination of various elements. Performance of the network, in terms of Mean Square Error (MSE) and correlation coefficient (r), has been measured and compared for every individual structure and consequently the best functional relationship has been identified. In addition, the sensitivity of different parameters has been identified and the effectiveness of the input parameters demonstrated. The results of the studies indicate that it is feasible to implement a suitable Neural Network architecture to map the inputs and outputs accurately and establish a usable "Ship Demolition Model". The model produced good results and can explain the complex structure of the Demolition market and identify and validate the main inputs which can alter market trends. The performance of the model has also been measured for forecasting three months ahead of the market and it shows a reasonable accuracy.
|
|---|
