外溢效果、結構改變和預測的研究:以消費性 ETF為例

The Study of Spillover Effect, Structural Breaks and Forecasting: Evidence from Consumer Exchange-Trade Funds

黃美玉(MAYA MALINDA)

指導教授 ： 陳若暉

中原大學/商學院/商學博士學位學程/博士(2015年)

https://doi.org/10.6840/CYCU.2015.00177

摘要

本研究的目的為增添有關於外溢效果、長期記憶、波動性和預測消費者指數型基金(ETF)相關研究不足處，分為三部分分析。
第一部分側重於消費者指數型基金(非必需性消費和必需性消費)和生產者相關指數型基金的外溢和槓桿效應。本研究採用GARCH-M-ARMA模型，結果發現消費者ETF和生產者相關ETF分別與其追蹤股票指數間存有雙向影響關係。本文利用EGARCH-M-ARIMA 模型分析，結果顯示生產相關ETF之波動外溢效果較消費ETF效果為低。此外，消費者和生產者相關ETF均存在顯著之負面槓桿效應｡
第二部分預測消費者ETF，並以國別區分，如美國、美國境外國(EX-US)、新興市場國家、巴西、中國和印度。根據灰色關聯分析(GRA)的分析結果顯示,前四名影響消費者ETF的因素，計有紐約證券交易所綜合指數、商品研究局商品指數、歐元兌美元之匯率、和賣權/買權比率。在涵蓋所有的資料和變數分析，並導入人工神經網絡(ANN)模式，結果顯示倒傳遞感知模型(BPN)可以更有效地進行預測。然而，根據不同的樣本，研究發現延時反饋神經網絡(TDRNN)和徑向基函數神經網絡(RBF神經網絡)均可提供一致的結果。與其他國家相比，透過人工神經網絡模型亦發現，巴西和中國之消費者ETF較容易預測。

在第三部分採用ARFIMA模型，發現與傳媒體、消費服務、食品及消費品產業相關之消費者ETF報酬較容易預測。另外，ARFIMA-FIGARCH模型顯示，長期記憶波動性只存在於遊戲和消費品行業。此外，透過迭代累加平方測試模型(ICSS)，本研究發現消費者ETF出現存有多重結構性改變之不對稱的效果｡
本研究之結果將不僅為發行人、投資者和基金經理提供相關經濟意涵，以利其規劃交易策略，實同時證結果和為學者和研究人員提供與消費者ETF相關之新趨勢看法｡

關鍵字

消費者ETF ； 預測 ； 槓桿效果 ； 外溢效果 ； 波動性

並列摘要

The aim of this research is to close the gap in the literature of the spillover, the long memory, volatility and forecasting for consumer exchange-traded funds (ETFs). This research is divided in to three parts.
The first part focuses on spillover and leverage effects of Consumer ETFs (Consumer Discretionary and Consumer Staples) and Producer Related ETFs. This study used Generalized Autoregressive Conditional Heteroskedasticity-in-Mean Autoregressive Moving Average (GARCH-M-ARMA) and found a bilateral correlation between Consumer ETFs and Producer Related ETFs tracing fundamental indexes. With Exponentially Generalized Autoregressive Conditional Heteroskedasticity-in-Mean Autoregressive Moving Average (EGARCH-M-ARMA) models, this paper found that Producer Related ETFs have less spillover effects for compared with Consumer ETFs. There were strongly negative leverage effects of both consumer and Producer Related ETFs.
The second part forecasts consumer exchange-traded funds (ETFs) which classified by country, such as the United States (US), excluding the United States (EX-US), Emerging Markets, Brazil, China, and India. The findings of Grey Relational Analysis (GRA) showed that there are top four ranking to influence Consumer ETFs, such as New York Stock Exchange Composite Index, Commodity Research Bureau, exchange USD/EUR and Put/call ratio. Artificial Neural Network (ANN) approach connected with all data and variables revealed that Back-Propagation Perception (BPN) can be much more effective for prediction. However, based on different sample this paper found that Time-Delay Recurrent Neural Network (TDRNN) and Radial Basis Function Neural Network (RBFNN) provide consistent results. ANN model also found that Brazil and China Consumer ETFs can be easier to predict, comparing with others countries.
The methods used in the third part is known as Autoregressive Fractionally Integrated Moving Average which found that media, consumer service, food and beverage and consumer goods industries of Consumer ETFs returns can be a good prediction. Fractionally Integrated Generalized Autoregressive Conditional Heteroskedasticity (ARFIMA-FIGARCH) model revealed that the long memory in volatility existed only for gaming and consumer goods industries. Moreover, there are multiple structural breaks for asymmetrical effects in Consumer ETFs by applying the Iterated Cumulative Sums Squares Test (ICSS).
The outcome of this research will not only offer economic meaning for issuers, investors and fund managers to plan for their trading strategies, but also provide for academicians and researchers stepping stones in having empirical results and a new perceptive from Consumer ETFs.

並列關鍵字

Forecasting ； Volatility ； Consumer Exchange-Traded Funds (ETFs) ； the long memory ； Spillover and Leverage Effects

參考文獻

Brandt, M. W., and Jones, C. S. (2006). Volatility forecasting with range-based EGARCH models. Journal of Business & Economic Statistics, 24 (4), 470-486.

Hsia, K. H., Chen, M.Y., and Chang, M. C. (2004). Comments on data pre-processing for Grey Relational Analysis. Journal of Grey System, 7 (1), 15-20.

You, M. L., Wang, C.W., and Yeh, C. K. (2006). The development of completed grey relational analysis toolbox via Matlab. Journal of Grey System, 9 (1), 57-64.

Box, G. E. P., and Jenkins, G. (1970). Time series analysis, forecasting and control. Holden-Day, San Francisco.

Engle, R. F., Lilien, D. M., and Robins, R. P. (1987). Estimating time varying risk premia in the term structure: The ARCH-M Model, Econometrica, 55 (2), 391-407.