Swarm Intelligence in Constructing Investment Strategies, Financial Crisis Warning and Currency Issuance Volume Forecasting Models

博士 === 國立臺灣科技大學 === 管理研究所 === 105 === To enhance management efficacy and maximize utilities, more effective forecasting methodologies are required in the context of continued economic and technological developments. Due to its computing power, versatility, learning capability and fault tolerance, th...

Full description

Bibliographic Details
Main Authors: Tsui-Hua - Huang, 黃翠華
Other Authors: Yungho Leu
Format: Others
Language:en_US
Published: 2016
Online Access:http://ndltd.ncl.edu.tw/handle/89816185703866196974
Description
Summary:博士 === 國立臺灣科技大學 === 管理研究所 === 105 === To enhance management efficacy and maximize utilities, more effective forecasting methodologies are required in the context of continued economic and technological developments. Due to its computing power, versatility, learning capability and fault tolerance, the neural network has been widely used in finance, electronic engineering and medicine since their inception in 1957. Neural networks are advantageous in that they do not require presumptions typically seen in multivariate analyses and have the ability to handle different data types. Due to the advances in computing technologies and computer networks, the swarm intelligence has become an important technology for problem solving. Algorithms mimicking ants, birds, bees and fruit flies have been developed to seek the optimal solution to a problem. This thesis refers to three studies to illustrate the contribution of swarm intelligence algorithms to the fields of finance and economics. In the first study entitled "A Mutual Fund Investment Method Using Fruit Fly Optimization Algorithm and Neural Network", an investing strategy was constructed in two stages. In the first stage, the data envelopment analysis (DEA), Sharpe ratio and Treynor ratio were used to select mutual fund portfolios. In the second stage, the Fruit Fly Optimization Algorithm, General Regression Neural Networks and traditional regression models were used to predict the closing net asset value (NAV) of a mutual fund based on the closing NAV of the previous day. Several experiments were conducted to compare the prediction accuracies and the accumulated return rates of different investment strategies. The results indicated that the investment portfolio constructed by Sharpe ratios outperformed the other portfolios. Furthermore, the investment prediction model built with the fruit fly optimization was superior to the other models. The second study is on "Constructing ZSCORE-based Financial Crisis Warning Models Using Fruit Fly Optimization Algorithm and General Regression Neural Network". First, the Fruit Fly Optimization Algorithm (FOA) was used to adjust the values of the coefficients of parameters in the ZSCORE model (FOA_ZSCORE model). Then, the difference between the forecasted value and the actual value of the dependent variable was calculated. Afterwards, the Generalized Regression Neural Network model (GRNN model), with the spread parameter optimized by the FOA (FOA_GRNN model), was used to forecast the difference to improve the forecasting accuracy. Various models, including ZSCORE, FOA_ZSCORE, FOA_ZSCORE+GRNN, and FOA_ZSCORE+FOA_GRNN, were trained and tested. The results showed that FOA_ZSCORE+FOA_GRNN model offered the highest prediction accuracy comparing to the others models. The third study is on "Swarm Intelligence and Neural Network in Constructing Prediction Models for Currency Issuance Volume: the US Experience". In this study, the Artificial Bee Colony Algorithm (ABCA) and Particle Swarm Optimization (PSO) were used to optimize the GRNN in constructing a predicting model for the volume of issuance of the United States. The constructed models include ABCA+GRNN, PSO+GRNN, GRNN and Multiple Regression. The experiments showed that the GRNNs optimized by the ABCA and the PSO, respectively, outperformed the non-optimized GRNN and the Multiple Regression model. The above three studies suggest that swarm intelligence algorithms can improve the prediction accuracy of a forecasting model. Managers in different industries can use the information as a reference to improve management efficiency and generate operational warnings. Governments can refer to the predictions of the volume of the currency issuance to promote efficiency in cash operation in their central banks. Finally, investors can utilize the swarm intelligence algorithm to construct investment strategies.