Long-term Memory and Forecast for The Bond ETFs

• Main Authors: Yi-Chen Tsai, 蔡易成
• Other Authors: Jo-Hui Chen
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/45999953001716113739

碩士 === 中原大學 === 企業管理研究所 === 103 === With the changing of the economic environment and financial instruments, original investment market is no longer applicable. Investors prefer to the low cost, high formation transparency and high reward instruments. Comparison to conventional financial products, ETF has above characteristics and even Nobel laureate in economics Robert F. Engle in 2003 once praised ETF is a great innovation . There are two main themes. First, this study provides additional evidence of nonlinearities in economic time-series from the long-term memory properties in return and volatility by using ARFIMA-FIGARCH which could reveal that long memory parameters are non-integer values or not. Second, this paper uses neural networks such as Back propagation Neural Network (BPN), Recurrent Neural Network (RNN) and Time-delay Recurrent Neural Network (TDRNN) models to predict bond ETFs with the six variables that including stock price, volatility indice (VIX), Put-Call Ratio, exchange rate, LIBOR and commodity research bureau (CRB) index. The result shows that only iShares J.P. Morgan USD Emerging Markets Bond ETF (EMB) is completed with long-term memory. The long-term memory is worth to not on part of integration process associated with a sequence of recession velocity and impulse response coefficients. In the results of neural network, the best forecasting performance is BPN, while except BUND and AGG have better predictions by using TDRNN. The result of this paper will to provide an investment strategy in the future to be a reference for investors or issuers. And the results can also provide the academic community potential avenues for research that will benefit the investing community in creating potential opportunity to create profit.