| Summary: | 碩士 === 中原大學 === 國際商學碩士學位學程 === 105 === This research examines return and volatility predictability of Corporate Social Responsibility (CSR) Indices through the grey relational analysis (GRA), and also applies three types of artificial neural networks (ANN) model, namely, back-propagation perceptron, recurrent neural network, and radial basis function neural network to capture nonlinear tendencies of CSR indices for a better forecasting accuracy. The research will find which ANN model has stronger predictive power compared with the other models, based on the ranking of the grey relational grades (GRGs). This research aims to first apply the GRA model in determining which among the 6 variables of stock and volatility indices, US dollar index, Trade index, CRB index, and Brent crude oil futures index have the strongest influence based on their relevant ranks. Then, the relatively more powerful forecasting tool, ANN models, will be used to predict CSR index returns, based on the lowest values of mean absolute error and root mean square error. And lastly, to check the robustness of GRA results, this paper divides the six variables in half depending on their relevant ranks based on their GRGs between those with high GRGs and low GRGs. This study will try to suggest to financial market players in determining appropriate ANN models in trying to forecast CSR index returns. The result in this paper showed the comparison of three ANN models, BPN had the best predicting power compared with RNN and RBFNN, we also learned that RNN and RBFNN model also got good performance with predicting accuracy. This paper also separated the data to 10%, 33% and 50% testing data level to test the proficiency of the available forecasting information in the time-series of the predictors. The result with CSR indices showed 66.6% of the data from the BPN model had the lowest values of MAEs compared with RNN model had only 33.3%. The predicting power of BPN model also showed with Non-CSR indices, 60% of the non-CSR were best by BPN model, 30% by RNN model and only 10% by RBFNN model. Traders, investors and fund manager can rely on BPN predicting power with large or even small data set. The result also suggests the predicting power of RNN and RBFNN model with a small set of data. Overall, with the best forecasting ability by using BPN model, we can say that, traders and fund managers have stronger chance of achieving more accurate forecasting. The GRAs table showed the dominance of High GRG of the influence toward CSR and Non-CSR indices. In summary, traders and fund managers can benefit by combining all six variables to get better forecasting accuracy.
|
|---|
