Development of Net Asset Value Forecasting Models using Soft and Bio Inspired Computing Strategies for Indian Mutual Funds

Abstract

newlineThe purpose of developing a time series forecasting model is to predict the future value of the particular observation in the specified domain. In this research work, attempts have been made to develop Net Asset Value (NAV) forecasting model using soft computing techniques and bio inspired optimization algorithms for the Indian mutual funds. Neural Network (NN) models have been proposed as promising soft computing techniques for time series forecasting in various domains. In this thesis, NN models are explored for the NAV forecasting. newlineIn the first phase of this thesis, Artificial Neural Network (ANN) model is used for the short term and long term NAV forecasting of two of the Indian mutual funds. To avoid the limitations of the Backpropagation learning, a newly developed Elephant Herding Optimization (EHO) algorithm is used to train the ANN model. For training the proposed ANN-EHO model, the NAV dataset is divided into train dataset and test dataset. The other bio inspired optimization algorithms i.e. Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Differential Evolution (DE) have also been used with ANN model for training purpose in this work. During testing the model, The Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) are taken as the performance measures.. The forecasting performances of the proposed ANN-EHO model has been compared with the ANN, ANN-GA, ANN-DE and ANN-PSO models. From the simulation study, it is found that the prediction ability of the proposed ANN-EHO model outperformed the other four models for the two Indian mutual funds. newlineIn the second phase of the thesis, the Extreme Learning Machine (ELM) model is used to predict the NAV of two of the Indian mutual funds. But in ELM, the connection weights between the input layer and hidden layer are assigned randomly and the output weights are mathematically calculated using Moore-Penrose pseudo inverse. The random assignment of weights and biases in the hidden layer affects prediction accuracy and these weights may not be always optimal. In this work, Dolphin Swarm Algorithm (DSA), one of the recently developed bio inspired optimization algorithm is used for determining the initial weights of ELM model. The other bio inspired algorithms i.e. PSO, Artificial Bee Colony newlinevii newline(ABC) algorithms are also used for determining the initial weights of the ELM model. The performance of the DSA-ELM model is compared with basic ELM, PSO-ELM and ABC-ELM model in terms of RMSE and MAPE. The empirical results indicate that the DSA-ELM model produced better results as compared to the other three models for the two Indian mutual funds. newlineThe third phase of our work is based on the development of an ensemble NAV forecasting model combining Adaptive Moving Average (AMA) model, ANN model and Functional link ANN (FLANN) model for two of the Indian mutual funds. The AMA model is the linear one where as the ANN and FLANN models are the nonlinear models. In this work, the outputs are linearly combined using some weights. The modified Whale Optimization Algorithm (WOA), one of the newly developed bio inspired optimization algorithm is used to optimize the ensemble weights. The performance of the proposed ensemble model is compared with the independent models as well as the ensemble model optimized with other established bio inspired algorithms i.e. GA and PSO. The forecasting performance comparison in terms of RMSE and MAPE values of the proposed model and the other models are presented at the end of the simulation study to validate the model. newlineFrom this research work, three novel NAV forecasting models have been developed using NNs and the recently developed bio inspired algorithms. These models may also be useful in predicting other financial time series data.