Research by: John Francis Diaz & Thanh Tung Nguyen
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 (BPN), recurrent neural network (RNN), and radial basis function neural network (RBFNN) to capture non-linear characteristics of CSR indices for a better forecasting accuracy. The research identifies which ANN model has stronger predictive power compared with the other models, based on the ranking of the grey relational grades (GRGs).
CSR initiatives and SRI channels have shown remarkable growth in the recent decades. The possibility of being both ethical and financially operational as a corporation has attracted a growing number of stakeholders around the globe. The strong influence of SRI interest motivated this study to utilize GRA and ANN models to CSR indices, which monitor the performance of major global benchmarks with higher-than-average ESG ratings.
The paper found that the BPN model outperformed the RNN and RBFNN models by having the lowest values of MAE and RMSE in modelling CSR index returns. The model was also relatively more powerful for having used 33% testing data set 58.33% of the time than the other ANN models. On one hand, both RNN and RBFNN models favored 50% testing data, also 58.33% of the time. These findings, related to the first and second objectives of this study, uphold the assertion of Moshiri et al. (1999) on the forecasting power of the BPN model and that portfolio management companies can better rely on its predicting power—especially in larger data sets.
Based on the GRA rankings, the US Dollar Index and the S&P 500 index were the 1st and 2nd ranking variables, respectively, 67% of the time, while the Volatility Index, Trade Index, CRB Index, and Brent Crude Oil Index were the more consistent ranking variable, placing 3rd, 4th, 5th, and 6th, respectively, for the twelve CSR index returns. These results are related to the third objective of this study. For the BPN and RNN models, the All Variables group dominated the High GRG and Low GRG groups.
The study observed the lowest MAEs and RMSEs when using the complete set of predictor variables. Compared to the first two results, however, the High GRG variables dominated the data 83.33% of the time. This means that MAE and RMSE are at its lowest when using three of the top variables. These findings are related to the last objective of this study. Considering the claims of Zhang and Xiao (2000) regarding the RNN model being good at forecasting when using fewer data sets, the findings of this study concludes that traders and fund managers can rely on the said model not only on fewer data sets but also on larger ones. Overall, traders and fund managers have stronger chance of achieving more accurate forecasting using the BPN model, and it is more beneficial to use larger amount of data when it comes to forecasting financial time-series data.
To cite this article: Diaz, J. F. T. & Nguyen, T. T. (2021). Application of grey relational analysis and artificial neural networks on Corporate Social Responsibility (CSR) indices. Journal of Sustainable Finance & Investment. https://doi.org/10.1080/20430795.2021.1929805
To access this article: https://doi.org/10.1080/20430795.2021.1929805
About the Journal
The Journal of Sustainable Finance & Investment is an international quarterly journal publishing peer-reviewed articles on radical and reformist initiatives for social responsiveness in global financial markets. The Journal specifically focuses on environmental, developmental, social and governance principles as formulated in the financial markets, managed investment, banking, micro-finance, project finance and philanthropy.
Journal of Sustainable Finance & Investment [ABS1]