Financial Modeling Using Bio Inspired Algorithms

Abstract

newlineThe basis for this research originally stemmed from my passion for developing better and efficient methods to predict the time series financial data. As the world moves further into globalization and in this digital age, generating vast amounts of financial data and born digital content, there will be a greater need to access accurately the financial information about a country, so that it will help in economic growth of that country. Previously it is very difficult to get the parameters and technical indicators that affects the economy of a country. In most of the research works the researchers have used technical indicators as the parameters to predict the stock index and exchange rate of any country. These data are biased so they affect the prediction performance. It has been observed from the analysis of global market that the exchange rate and stock index of any country depends on the major stock indices and exchange rates of developed countries. Therefore, we have designed datasets by considering major stock indices of the world and exchange rates of developed G-7 countries to predict the future values of stock index and exchange rate of another country. In this research work, we have experimentally concluded that we can use the major stock indices of the world and exchange rates of developed countries as predictors. newlineMoreover, from the deep analysis, it has been observed that radial basis function neural networks are capable of universal approximation and are performing better than the other traditional prediction models for predicting the financial data. However, in many cases/instance, it is difficult to obtained the optimal parameters for the radial basis function neural network. Therefore, we have concentrated on designing and improving the efficiency of radial basis function neural networks by using bio-inspired algorithms. In this globalization era the economy of most of the country depends on the financial stability of other country. The prediction of financial data can be done more accurately if we could use better algorithms for prediction purpose. Researchers have suggested that neural networks based algorithms are performing better than traditional statistical algorithms and all most all the researchers are agreed that radial basis function network can be used as a universal approximator. Therefore, in our research work we have used radial basis function neural network as our prediction algorithm and then, we have improved its performance by fine tuning the parameters of the radial basis function neural network by using bio-inspired algorithm. One of the most popular bio-inspired algorithm is particle newlinevii newlineswarm optimization algorithm. It is widely used for solving optimization problems due to its simplicity and less number of parameters. Hence, we have considered canonical particle swarm optimization algorithm to fine tune the parameters of radial basis function neural network. From the experimental results we have observed that the performance of particle swarm optimized radial basis function neural network is performing better than the traditional radial basis function neural network algorithm. However, in this approach we have selected the particles randomly and the initial weights are updated by using the random number generator function. Further, we have analyzed that chaotic functions have better statistical and dynamical behavior than the random number generator function, which basically follows the normal distribution. Therefore, to improve the performance of the above model we have considered chaotic function instead of random number generator function to fine tune the inertia weights. Finally, based on the experimental results, we have compared our proposed model with other models. We have applied our proposed model to the three different areas in financial sector such as stock index prediction, exchange rate prediction and credit risk analysis. From the experimental evidence we have drawn the conclusion that our proposed chaotic improved particle swarm optimization tuned radial basis function neural network model is performing better in all the three financial areas as compared to the other traditional statistical and neural network models.