美國股債報酬相關性與總體經濟因素之分量迴歸分析

The Stock-Bond Return Relation and Macroeconomic Factors：A Quantile Regression Approach

紀怡婷(Yi-Ting Chi)

指導教授 ： 李美杏

國立臺北大學/商學院/統計學系/碩士(2011年)

摘要

本研究目的在於探討總體經濟因素的落後期是否能夠預測美國股債報酬相關性，所選取的總體經濟因素有短期利率、長短期利差及波動率指數，分別進行Granger因果關係檢定及建構分量迴歸模型作為實證分析的基礎。本研究實證結果發現：
一、Granger因果關係檢定及OLS迴歸均顯示，平均而言，前一期短期利率可以用來預測股債報酬相關性，且為同向影響，即短期利率上升，將使股債報酬相關性增加。
二、在未考量相關性的自我相關項時，分量迴歸模型的解釋能力相當低，且對相關性左右兩尾的影響因素也是不同的。在美國股債報酬為負相關(0.1~0.4分量)時，對相關性最具影響性的解釋變數為前一期短期利率，為同向影響；而在美國股債報酬為正相關(0.5~0.9分量)時，對相關性最具影響性的解釋變數為前一期工業生產指數月增率，一樣為同向影響，這些總體經濟因素對相關性的影響方向皆與過去文獻實證結果一致。
三、在考量相關性的自我相關項後，分量迴歸模型的解釋能力可高達七成，此與股債報酬相關性已反映當期整體景氣經濟變化有關，此時反而是市場投資人對於未來股市的預期心理顯得重要，尤其是在美國股債報酬為負相關(0.1~0.3分量)時，當前一期波動率指數增加(減少)，將使股債報酬相關係數變小(變大)，也就是負相關程度會提高(降低)，表示當股債走勢相反時，若市場避險情緒高漲，投資人”flight-to-quality”傾向易產生股市的恐慌性賣壓，造成棄股買債的現象更加嚴重，導致負相關程度又提高，此時股債同時配置可達到更好的風險抵減效果。

關鍵字

股債報酬相關性 ； DCC ； Granger因果關係檢定 ； 分量迴歸

並列摘要

This paper examines the impact of the short rate, the yield spread, VIX, the industrial production growth and Index of Consumer Sentiment on the U.S. stock-bond return correlation. We use the Granger causality test and quantile regressions to investigate if and how the dynamics in the stock-bond correlation are different at all quantiles. Our results are summarized as follows. First, the Granger causality test and the OLS regressions show that, on average, the major trend in stock-bond correlation is determined primarily by the lagged short rate. The positive effect from the short rate will lead stock-bond return correlation increased. Secondly, the explanatory power of model without AR (p) terms added to the regression specifications is fairly low. And we find that the behavior of the stock-bond correlation differs when the correlation is negative (from 0.1 to 0.4 quantile ) as opposed to when it is positive (from 0.5 to 0.9 quantile ). The short rate has strong positive influences upon the stock-bond correlation at the left tail. But at the right tail, the industrial production growth is important and positively related to the stock-bond correlation. The empirical findings are consistent with previous literature. Finally, with model considering AR (p) terms we are able to explain 70% of the variation in the correlation across the different quantiles. The reason is the conditional correlations produced by the DCC model should account for the changes in the overall economy. Instead of the short rate, VIX is more important and negatively related to the stock-bond correlation when the correlation is negative (from 0.1 to 0.3 quantile ). This implies that when the stock-bond correlation is negative, there is a negative effect from the high stock market uncertainty leads to a decoupling between stock and bond prices consistently with the “flight-to-quality” phenomenon. The diversification benefits of combined stock-bond holdings tend to be particularly higher during times of extreme negative correlations.

並列關鍵字

stock-bond return relation ； DCC ； Granger causality test ； Quantile Regression

參考文獻

8.陳安琳(2007)，「動態條件相關係數-台股與美股的實證」，國立台灣大學財務金融學研究所碩士論文。

16.莊家彰、管中閔(2005)，「台灣與美國股市價量關係的分量迴歸分析」，經濟論文，第33卷第4期，頁379-404。

17.張怡文、江穎慧、張金鶚(2009)，「分量迴歸在大量估價模型之應用-非典型住宅估價之改進」，都市與計劃，第36卷第4期，頁281-304。

1.Andersson, M., Krylova, E. and Vähämaa, S. (2004) , “ Why does the correlation between stock and bond returns vary over time? ”, ECB Working Paper.

2.Andersen, T. G., Bollerslev, T., Diebold, F. X. and Vega, C. (2004) , “ Real-time price discovery in stock, bond and foreign exchange markets ”, Center for Financial Studies Working Paper (19).