international Financial Market

Journal of International Money and Finance 49 (2014) 401e424

Contents lists available at ScienceDirect

Journal of International Money and Finance

journal homepage: www.elsevier.com/locate/jimf

Investor induced contagion during the banking and European sovereign debt crisis of 2007e2012: Wealth effect or portfolio rebalancing?*

Dimitris Petmezas a, *, Daniel Santamaria b

a Surrey Business School, University of Surrey, GU2 7XH, UK b Canterbury Christ Church University Business School, Canterbury, Kent CT1 1QU, UK

a r t i c l e i n f o

Article history: Available online 5 July 2014

Keywords: Contagion Wealth effect Portfolio rebalancing Banking crisis Sovereign debt crisis Stock-bond relationships

* We would like to thank an anonymous referee a Crisis: European Financial Markets and Institution * Corresponding author. Tel.: þ44 1483 686 376

E-mail addresses: [email protected] (D.

http://dx.doi.org/10.1016/j.jimonfin.2014.06.005 0261-5606/© 2014 Elsevier Ltd. All rights reserved

a b s t r a c t

This study investigates the way a crisis spreads within a country and across borders by testing the investor induced contagion hypothesis through the liquidity channel on stock-bond relationships of the US and five European countries before and during the global banking and European sovereign debt crisis of 2007e2012. We provide ev- idence consistent with the wealth effect as a source of contagion for the majority of countries. Nevertheless, we uncover evidence of investor induced contagion sourced by the portfolio rebalancing effect for correlations involving Spanish and Italian bonds during the debt crisis. Further, we find that tight (narrow) credit spreads reduce (magnify) the wealth and portfolio rebalancing effects, which are offset by the opposite effects of risk aversion amongst investors, a dynamic that is not restricted to crisis periods.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

One of the most prominent characteristics in the recent financial crisis is the global loss of confi- dence in the financial system with unprecedented levels of risk aversion amongst investors, the lack of

nd participants at the 2013 International Conference on the Global Financial s for helpful comments and suggestions. ; fax: þ44 1483 686 301. Petmezas), [email protected] (D. Santamaria).

.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424402

liquidity and freezing of the credit markets. The European sovereign debt crisis that followed was not entirely unexpected (Reinhart and Rogoff, 2009). In fact, the contagion effect of past crises, which is defined as the transmission mechanism that occurs during a financial crisis, continues to dominate the views of economists, academics and policy makers (Caporale et al., 2006 on the Asian financial crisis; Yang et al., 2006 on the Russian crisis; and Ravichandran and Maloain, 2010 on the recent financial crisis).

The objective of this study is to investigate what causes a crisis to spread by testing the investor induced contagion hypothesis on the relationship between stock and bond markets. In a recent paper, Boyer et al. (2006) suggested that investor induced contagion stems from either the wealth effect or the portfolio rebalancing hypothesis. Therefore, motivated by Boyer et al. (2006), we investigate whether investor induced contagion between stock and bond markets sources from the wealth effect or port- folio rebalancing hypothesis before and during the banking and European sovereign debt crisis over the period 2007e2012. Moreover, given that a prominent characteristic of the recent financial crisis was the collapse in confidence in the financial system, the third aim is to uncover cross regional evidence on the effect of credit market liquidity and level of risk aversion on the transmission channel of the stock- bond relationship. Both sources of risk provide a barometer on the liquidity in the financial system and the degree of investor uncertainty associated with taking on high risk assets. A crucial feature of the wealth effect and portfolio rebalancing hypothesis, especially in light of credit market liquidity and investor risk aversion, is the underlying assumption that a crisis spreads through the liquidity channel. This envisages a scenario where a shock causes liquidity in the financial system to decline. Hence, the wealth effect in a stock-bond relationship stipulates that during a crisis, a shock to the funding con- straints of investors causes risk aversion, a liquidation of positions, and an increase in volatility in both markets (Brunnermeier and Pedersen, 2009); similarly, under the portfolio rebalancing hypothesis, risk aversion and the associated decline in liquidity in the stock market from a shock encourages in- vestors to sell high risk assets in search for quality (Chordia et al., 2005; Baur and Lucey, 2009).

To test for investor induced contagion, we use the Dynamic Conditional Correlation (DCC) EGARCH to test the wealth effect versus portfolio rebalancing hypothesis on stock index and benchmark bond yield data at different times to maturity for the UK, US, France, Germany, Spain and Italy. A key feature of this approach is that proxies of credit market liquidity conditions and investor risk aversion are modelled within the conditional mean as the means of investigating the transmitting effects of both risk barometers on the transmission channel of the stock-bond relationship. The appealing feature of this framework is that we propose an asymmetric DCC estimator along the lines of Cappiello et al. (2006) that allows for conditional asymmetries to govern the dynamics of the stock-bond relation- ship. Additionally, we introduce a DCC(Z)-EGARCH risk factor model using interactive dummy variables on DCC estimates to identify whether the wealth effect and portfolio rebalancing hypothesis is sen- sitive to credit market stress and risk aversion under different market conditions. As a result, this al- lows us to draw fruitful conclusions on the drivers that cause a crisis to spread within a country and across borders.

In brief, we report that stock-bond correlations strengthened considerably during the banking crisis for all countries. The increase in correlation is consistent with the investor induced contagion hy- pothesis caused by the wealth effect. However, for stock-bond correlations involving Spanish and Italian bonds, we report a dramatic decline in correlation and diminishing investor induced contagion immediately after the banking crisis up to 2009. However, for both countries' stock-bond relationships, we report investor induced contagion sourced by the portfolio rebalancing effect throughout the sovereign debt crisis from 2010 to 2012. The inclusion of proxies for credit market liquidity conditions and investor risk aversion reveals further interesting results. We find that high levels of risk aversion magnify investor induced contagion sourced by the wealth effect during the banking crisis and by portfolio rebalancing effects during the debt crisis for correlations involving Spanish and Italian bonds. Despite this, the manifestation of investor induced contagion due to risk aversion is countered by the opposite effects of changes in credit market conditions, a finding that is not necessarily limited to crisis periods.

This paper makes a number of contributions to the literature. Firstly, this study is the first to our knowledge to formally test the wealth effect versus portfolio rebalancing hypothesis on stock-bond relationships as a transmission mechanism that could explain how the banking crisis and sovereign

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 403

debt crisis spread within a country and across borders. Therefore, our study deviates from the prior literature that focuses on contagion as a by-product of the arrival of negative innovations (Billio and Pelizzon, 2003; Kim and Moshirian, 2004; Baele, 2005). Secondly, we differentiate the banking and sovereign debt crisis as separate stages of the financial crisis and in doing so, we are able to extract information and draw fruitful conclusions on how stock-bond relationships not only change during a crisis but also in the dynamics that cause a crisis to spread. Finally, we investigate the feed through effects of key financial market indicators of credit market stress and investor risk aversion in the markets on the transmission mechanism before the crisis and during the banking and sovereign debt crisis. This is particularly important when investigating how a crisis spreads given that these variables contain useful information on credit market liquidity in the real economy and investor confidence (Brunnermeier, 2009; Melvin and Taylor, 2009; Andersson et al., 2007). Therefore, this paper has important implications in terms of how monetary authorities should respond in maintaining liquidity levels in the financial system especially during a crisis.

The rest of the paper is structured as follows; Section 2 reviews the related literature. Section 3 in- troduces the DCC-EGARCH model and the transmission channel of credit market stress and risk aversion. Section 4 discusses the data and provides descriptive statistics. Section 5 presents the empirical results. Finally, a discussion of the results is presented in Section 6, and Section 7 concludes the paper.

2. Related literature and hypotheses development

2.1. The stock-bond relationship

In early studies, Shiller and Beltratti (1992) and Campbell and Ammer (1993) hypothesized that the stock-bond relationship is constant over a period of time. However, the former study reported a strong negative relationship between stock prices and long maturity bond yields, which was inconsistent with the findings of Campbell and Ammer (1993). Although later studies have documented a time varying stock-bond relationship (Gulko, 2002; Connolly et al., 2005; Jones and Wilson, 2004), they did not test for investor induced contagion between the stock and bond markets during normal market conditions and crisis periods. As a result, first and foremost, this paper tests the existence of investor induced contagion by identifying the time varying nature of correlation before and during the banking and debt crisis.

Furthermore, little attention has been paid about the factors that drive this relationship. One factor that has been suggested by previous studies is the expected rate of inflation in the long run (Li, 2004). The intuition behind the use of this macroeconomic variable is that uncertainty surrounding an in- crease in forecasted inflation causes an increase in discount rates, thus leading to a fall in bond prices. Indeed, according to Li (2004), a stronger correlation between stock and bond yields is observed following an increase in expected inflation. In a departure from prior literature, this paper introduces proxies for credit market liquidity and investor risk appetite as two potential drivers that represent barometers of liquidity in the interbank market and confidence in the financial markets.

Given the role of credit market liquidity and the level of risk aversion as potential drivers of the stock-bond relationship, we depart from previous studies that find the source of contagion as a by- product of innovations (King and Wadhwani, 1990; Koutmos, 1996; Wongswan, 2006) and risk pre- mium (Acharya and Pedersen, 2005). Instead, this study postulates that the stock-bond relationship is governed by the amount of liquidity in the financial system in which a shock leads to an overall decline in liquidity in the asset markets. One of the leading proponents of the liquidity channel in the stock- bond relationship is the study by Chordia et al. (2005). They hypothesized a relationship between liquidity induced by the microstructure of the market and macro induced liquidity as a means of providing important inferences on the main drivers of liquidity in both markets and hence, the stock- bond correlation. They reported a significantly positive relationship between stock and bond market volatility caused by shocks, with volatility being a major driver of liquidity.

2.2. Wealth effect versus portfolio rebalancing

Investor induced contagion and the liquidity channel that causes a crisis to spread, in view of credit market liquidity and investor risk aversion, nest neatly with two competing theorems by which

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424404

this study tests the stock-bond relationship. The first one is the wealth effect developed by Kyle and Xiong (2001) and later tested by Boyer et al. (2006). In their paper, Kyle and Xiong developed a contagion model based on changes in risk appetite that is determined by the wealth effect of convergence traders. The wealth effect arises when convergence traders incur losses due to negative shocks, thereby causing high levels of risk aversion and a liquidation of positions in the financial markets. According to the liquidity channel, this causes liquidity levels to decline and magnify the impact of the initial shock which inevitably transmits from one asset to the next. Therefore, under the wealth effect, the correlation between assets increases during a crisis and, thus, is consistent with investor induced contagion.

However, the Kyle and Xiong's model assumed the existence of two risky assets, which at a time of falling (rising) stock prices, leads to a decline (increase) in liquidity, an increase (decrease) in volatility and, hence, increased correlation between assets. One implication of their model is that, when faced with high and low risk assets, investors at a time of falling stock prices will substitute growth for income. However, as income falls with declines in yields due to rising bond prices, convergence traders will be inclined to liquidate their positions in the bond market, thus causing a decline in liquidity. The Brunnermeier and Pedersen's (2009) liquidity-funding model suggests that such a scenario is likely given that shocks to investors' funding constraints encourage a liquidation of po- sitions causing a “liquidity spiral” between the markets due to the fact that the risk of hitting capital limits increase. The funding illiquidity causes a decline in liquidity in both markets leading to an increase in volatility in both asset classes and, hence, an increase in the stock-bond correlation during a financial crisis.

The alternative hypothesis tested is the portfolio rebalancing effect. This postulates a scenario where investors faced with falling (rising) stock prices, embark on a “flight to (from) quality” from (to) stocks. In such a scenario, investor induced contagion sourced by the portfolio rebalancing hypothesis is characterized by a decline in the correlation estimate in the event of a financial crisis. Like the wealth effect, proxies for credit market liquidity and investor risk aversion complement the portfolio reba- lancing hypothesis through the liquidity channel. In times of a crisis, a lack of liquidity in the trans- mission mechanism coupled with a high level of uncertainty will cause investors to look for low risk assets during major falls in stock prices. The prior literature in support of the portfolio rebalancing hypothesis within the context of a financial crisis is limited at best. Hartmann et al. (2004) report some evidence in favour of the portfolio rebalancing hypothesis when analysing the stock-bond relationship amongst the G-5 countries; however, this is countered by contagion effects between both asset classes at different periods of time. On the other hand, Connolly et al. (2005) find evidence that negative relationships between stock and bond markets coincide with major stock market falls as investors rebalance their portfolios from high risk to low risk assets. However, focusing on the stock-bond re- lationships for eight major countries, Baur and Lucey (2009) report more conclusive evidence that “flight to quality” is not only frequent during crisis periods, but it crosses borders, thus indicating the existence of “cross-country contagion”.

3. Methodology

The Exponential-GARCH (EGARCH) model, first introduced by Nelson (1991), has been amongst the most commonly used GARCH specifications that capture asymmetries in asset returns.1 For the purpose of this study, the EGARCH model is used to derive the standardized residual returns for stocks and bonds as inputs for the dynamic correlation model. We start with the EGARCH (p,q) specification for the conditional mean and variance:

1 Early advocates of the EGARCH model point out the studies of French et al. (1987) and Schwert (1990) as uncovering an asymmetric effect in the data in addition to other stylized features such as volatility clustering, serial dependencies and long memory (Hentschel, 1995; Andersen et al., 2001). Later studies continue to highlight the usefulness of EGARCH models in generating reliable volatility forecasts that is attributable to a well-specified model (Brandt and Jones, 2006). According to Alberg et al. (2008), the forecasting performance of the EGARCH model outperforms other asymmetric GARCH model speci- fications when applied to. Tel Aviv stock market.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 405

Rt ¼ g1 þ εt� � �

εt ��ðεt�1; εt�2:::::::Þ$N 0; s2t

s 2 t ¼ u0 þ

( Xq k¼1

a1 ln � s 2 t�1

� þ a2

Xp k¼1

ct�1

) (1)

where s2t denotes the conditional variance stock market returns and bond yields, and ct�1 is the innovation term that comprises the theta q1 coefficient that captures the asymmetric component.

2

Using the standardised residual returns st ¼ ct=s2t generated by equation (1) as inputs, we propose the dynamic correlation coefficient (DCC) structure, first introduced by Engle (2002) to test the wealth effect versus the portfolio rebalancing hypothesis. The DCC model has the appealing structure that allows conditional correlation estimates to be time varying, a notion that requires decomposing into the following:

Ht ¼ DtRtDt (2) where Dt ¼ diagf

ffiffiffiffiffiffi s2t

q g and is the correlation matrix of time varying standardized residual returns that

leads the structure of the DCC model for assets i and j

qi;j;t ¼ ð1 � v1 � v2Þq�þ v1 � si;t�1sj;t�1

� þ v2qi;j;t�1 (3)

where the term denotes the matrix of unconditional covariance of the standardised errors from equation (1) that leads to the DCC matrix Rt ¼ q*�1qi;j;tq*�1i;j;t from which the components of Rt are computed using

ri;j;t ¼ qi;j;tffiffiffiffiffiffiffiffiffiffiffiffiffi qi;tqj;t

p (4) where ri,j,t represents the DCC estimates for assets i and j respectively. An increase in ri,j,t during a crisis period is consistent with investor induced contagion, sourced by the wealth effect. Conversely, a decline in ri,j,t suggests support for investor induced contagion sourced by the portfolio rebalancing hypothesis. Combined with the EGARCH model, the DCC structure of equations (3) and (4) is remi- niscent of the asymmetric DCC estimator of Cappiello et al. (2006) in that it allows for conditional asymmetries to govern the dynamics of the stock-bond relationship. Furthermore, the DCC model proposed can be adapted to capture the transmission effects of credit market liquidity and investor risk aversion from the univariate EGARCH model by expanding the conditional mean of equation (1) so that the following transmission channel εt/st/q*t /Rt/ri;j;t is envisaged.

A major implication of the DCC-EGARCH model is that the maximum likelihood function needs to be modified to take into account the DCC estimator so that the following quasi-maximum likelihood estimates (QMLE) is maximised:

ln2ðwjfÞ ¼ � � T 2

XT t¼1

n n lnð2pÞ þ lnðjDtRtDtjÞ þ εTt D�1t R�1t IkD�1t εt

o

¼ � � T 2

PT t¼1

n n lnð2pÞ þ 2 lnðjRtjÞ þ cTt R�1t ct

o

¼ U0 � � T 2

XT t¼1

n lnðjRtjÞ þ cTt R�1t ct

o (5)

2 To maximise the log likelihood function, the Berndt et al. (1974) (BHHH) algorithm is proposed in this paper. Given that ARCH processes are highly nonlinear, maximising the log likelihood function on the assumption that the conditional density is normal renders the optimisation process invalid. Hence, the BHHH algorithm relaxes this restriction to obtain the log likelihood function by utilizing the covariance of the analytic gradients for each observation to form the conditional covariance matrix.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424406

where U represents the constant and wj4 denotes the sum of the univariate variance generated by the EGARCH model conditional of the correlation parameters. Next, we proceed to describe the data used in this study followed by a preliminary empirical analysis.

4. Data

4.1. The data

The database comprises of daily index returns on the FTSE-100, Russell 3000, CAC-40, DAX-30, IBEX- 35 and the FTSE MIB Index from January 1, 2004 to September 6, 2012 (n ¼ 2266 observations). The choice of stock markets was determined by the degree of coverage and of diversification in each index. For instance, the FTSE-100 Index lists the 100 largest blue chip companies that make up 84% of the total market capitalization in the UK. Given that the US is the origin country of the banking crisis, the US market is represented by the Russell 3000, which covers the 3000 largest public limited companies and, as such, represents 98% of the total market capitalization of US stocks. The CAC-40 lists the performance of the top 40 largest stocks on Paris Euronext based on market capitalization. In turn, the DAX-30 represents the Frankfurt Stock Exchange listing of the top 30 blue chip companies that represent 80% of the total market capitalization. To proxy southern European countries, the IBEX-35 consists of 35 of the most liquid stocks listed on the Madrid Stock Exchange General Index and the FTSE MIB Index represents 40 of the largest and most liquid stocks by capitalization that are traded on the Borsa Italiana.

For short and long term interest rates, our database includes the FTSE Global 1e3 year and 10 þ year government benchmark bond yields for the UK, US, France and Germany. The 1e3 year index is defined as the aggregate of benchmark indices of bond yields from one to three years to maturity for each country. Similarly, the 10 þ year index represents the aggregate of the indices for ten years and longer to maturity. Due to data limitations, we collected interest rate data on three year bond yields for Spain and Italy as well as ten year benchmark bond yields for both countries.3

To generate proxies for credit market liquidity conditions and investor risk aversion, we have also collected daily three month LIBOR rates, three month interest rate data and implied volatility index values (VIX) for the same countries from Datastream. However, in relation to the implied volatility index, the absence of data for both Spain and Italy means that VIX index for the Euro-zone (EU) was used to proxy investor confidence.

4.2. The T-bill Euro Dollar (TED) spread

To proxy for credit market liquidity conditions in the real economy, we computed the three month TED spread based on the London Interbank Offering Rate (LIBOR) and three month interest rates. The LIBOR, which is the expected average rate of interest over the life time of interbank loans, reflects default risk levels and wholesale market liquidity. Indeed, Taylor and Williams (2009) argue that in- creases in LIBOR are attributable to higher interest rates demanded by lenders for taking higher default risk in times of market stress. This is reflected by the spike in the TED spread shown in Fig.1 for the UK, US, France, Germany, Spain and Italy as the average rate of interest charged by banks increased as liquidity dried up during the height of the banking crisis.4,5

3 For the remainder of the paper, 1e3 year and 10 þ year benchmark yields will be referred to as short dated and long dated bond yields respectively.

4 The appealing feature of the TED spread is that it comprises the three month interest rate OIS-LIBOR spread. This represents the cost of three month liquidity from the perception of banks that is defined in terms of the risk attached to undertaking unsecured lending over the three months versus unsecured lending for one day. Further, the TED spread also incorporates the three month interest rate OIS spread, which contains a demand driven component that in a crisis causes portfolio managers to reallocate their portfolios away from high risk to low risk assets. The implication is a fall in yields and a widening of the TED spread.

5 The negative TED spread for Spain is attributable to a spike in three month interest rates that coincided with Moody's downgrade of Spanish sovereign debt in October 2011. The short term interest rate, which peaked at 5.06% in November 2011, represents investors' expectations that the country will default on its debt obligations.

Note: The country codes for the VIX index of investor confidence are: UK: UK, US: US, FR: France, BD: Germany, EU: Euro-zone.

0

10

20

30

40

50

60

70

80

90

100

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

UKVIX USEVIX

FRVIX BDVIX

EUVIX

Fig. 1. The Implied Volatility (VIX) Index of Investor Risk Appetite.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 407

Other measures of credit market liquidity risk, such as Credit Default Swap (CDS) spreads, are not considered in this study given recent evidence that they are not fully attributable to credit market liquidity risk factors (Collin-Dufresne et al., 2001; Blanco et al., 2005). Additionally, the behaviour of CDS spreads during the recent financial crisis has raised more questions regarding the usefulness of this measure of credit market liquidity risk as a leader of other markets. For instance, Forte and Pe~na (2009) and Norden and Weber (2009) both find that global stock markets tend to lead the CDS and bond markets more frequently than the opposite. By contrast, the TED spread, as a barometer of credit market liquidity, has received attention in a growing body of literature. For instance, Lashgari (2000) uses the TED spread as an independent variable on the S&P 500 index returns and reports a nega- tive and significant coefficient. The intuition is that a widening TED spread signifies deteriorating liquidity conditions thus leading to a liquidation of positions and net portfolio outflows. In later studies, Brunnermeier (2009) and Melvin and Taylor (2009) demonstrate the usefulness of the TED spread as a measure of liquidity in the wholesale market. For instance, the former study postulates that banks borrowing in the interbank market during a crisis tend to be charged a higher LIBOR and, thus, regard short term bonds a more attractive proposition given that they are considered as risk free; this leads to a spike in the TED spread as shown in Fig. 1.

4.3. Implied volatility (VIX) index

The implied volatility index (VIX) first introduced by the Chicago Board Options Exchange (CBOE) in 1993 and revised in 2003 is regarded by markets as a measure of investor fear. For the purpose of this study, we used daily VIX index values based on the revised 2003 methodology. In short, this in- corporates the S&P 500 index and averages the weighted prices of out-of-the-money puts and calls over a range of strike prices to estimate the implied volatility of stock index options. As a result, the high index readings between late October 2008 and March 2009 observed in Fig. 2 are representative of high volatility associated with large stock market downturns.

Realizing the information content of the VIX index, an increasing number of studies have used the implied volatility index as an independent variable within the GARCH framework (Blair et al., 2001; Koopman et al., 2005). Both studies reported a major improvement in the performance of daily GARCH models after adding realized volatility and the VIX implied volatility index as explanatory variables into the variance equation. However, this paper makes an important departure from previous studies in the use of the VIX index. In particular, we use the information uncovered in the VIX index to

Note: The country codes for the three month TED spreads are: UK: UK, US: US, FR: France, BD: Germany, ES: Spain and IT: Italy.

-5

-4

-3

-2

-1

0

1

2

3

4

5

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 S ep-10 Jul-11 May-12

T EDIT T EDUK

T EDUS T EDFR

T EDBD T EDES

Fig. 2. The TED Spread Measure of Credit Market Liquidity.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424408

establish whether investor risk aversion magnifies investor induced contagion sourced by the wealth effect or portfolio rebalancing hypothesis in the stock-bond relationship before and during crisis pe- riods. For instance, unlike previous studies, to capture the impact of pre-crisis, banking crisis and debt crisis periods, we include the implied volatility measure into the conditional mean as interaction dummy variables for each of the three sub-periods. Low (2004) provides intuition behind this, postulating that the VIX index contains useful information on option trader's perception of risk, and demonstrates how risk perceptions and price are related. As a result, in this paper, the VIX index proxies for investor confidence as a potential driving force behind the stock-bond relationship.

4.4. Preliminary analysis and descriptive statistics

As a starting point, Table 1 provides descriptive statistics and unconditional correlation analysis on stock and bond market returns. We show that average stock returns are generally positive for the entire sample period with the exception of the IBEX-35 and FTSE MIB indexes. The positive stock index returns coincide with negative bond yields at long and short maturities where the UK short dated bond yields reported the largest loss of 15.92%. This contrasts Spanish bond yields at long and short ma- turities which recorded an increase of 1.3% and 1.6% respectively over the same period.

Focusing on the unconditional correlation matrix, the stock-bond relationship varies considerably from relatively strong and positive readings involving UK and US bond yields to weak and negative for Spanish and Italian bonds. By discovering a relatively strong positive stock-bond correlation points to a generalized wealth effect whereas evidence of a weak and negative relationship introduces the pos- sibility of a portfolio rebalancing effect. These findings will be of particular interest when we estimate the DCC model later in the paper.

5. Empirical findings

5.1. DCC-EGARCH model estimations

A major characteristic of the recent banking crisis and European sovereign debt crisis is the un- conventional tools employed by central banks across the globe through the implementation of various quantitative easing (QE) programmes. For instance, both the US Federal Reserve and Bank of England embarked on large scale asset purchasing programs that involved long term government bonds (Gagnon et al., 2011 for the US; Joyce et al., 2011 for the UK; and Neely, 2011 for the international bond

Table 1 Descriptive statistics and unconditional correlations between stock and bond markets.

FTSE Russell CAC DAX IBEX FTSE UK US FRANCE GERMANY SPAIN ITALY

100 3000 40 30 35 MIB 1e3 10þ 1e3 10þ 1e3 10þ 1e3 10þ 3 10 1 10 Mean 0.011 0.012 �0.002 0.026 �0.001 �0.024 �0.159 �0.018 �0.081 �0.030 �0.113 0.020 �0.072 �0.035 0.013 0.016 �0.052 0.009 St.dev 1.249 1.362 1.467 1.428 1.489 1.559 4.960 0.626 4.637 1.406 4.367 1.092 16.51 1.334 4.052 1.303 4.460 1.251 Skewness 0.145** �0.357** 0.068 0.046 0.075 �0.046 0.753** 0.523** �0.734** �0.452** 1.391** �0.059** 3.485** �0.017 0.433** �1.126** 0.158** �0.895* Kurtosis 8.479** 9.765** 7.294** 6.890** 6.686** 6.132** 43.90** 230.21** 14.05** 5.038** 64.76** 3.506** 142.24** �4.744** 25.92** 13.92** 12.18** 13.56* Q(12) 56.41** 47.67** 43.14** 22.47* 27.25** 38.81** 239.98** 19.10* 66.22** 151.51** 196.43** 44.37** 230.98** 29.92** 142.74** 130.11** 587.31** 79.08*

FTSE100 1 0.570 0.902 0.842 0.819 0.828 0.213 0.261 0.360 0.368 0.168 0.259 0.095 0.324 0.017 0.107 0.065 0.073 Russell 1 0.603 0.627 0.558 0.569 0.157 0.191 0.259 0.233 0.130 0.192 0.050 0.245 0.011 0.075 0.030 0.039 CAC40 1 0.914 0.896 0.908 0.239 0.292 0.379 0.401 0.177 0.292 0.095 0.396 0.008 0.079 0.090 0.026 DAX30 1 0.825 0.841 0.219 0.280 0.356 0.378 0.165 0.274 0.091 0.359 0.010 0.105 0.067 0.053 IBEX35 1 0.893 0.244 0.274 0.356 0.378 0.168 0.261 0.118 0.398 0.063 0.018 0.086 0.061 FTSE MIB 1 0.258 0.296 0.373 0.391 0.161 0.281 0.105 0.423 �0.047 �0.024 0.076 �0.087 UK 1e3 1 0.308 0.398 0.353 0.193 0.265 0.077 0.450 0.019 0.014 0.060 0.031

10þ 1 0.337 0.378 0.201 0.379 0.060 0.438 0.039 0.225 0.070 0.191 US 1e3 1 0.548 0.323 0.404 0.073 0.449 0.063 0.219 0.096 0.182

10þ 1 0.267 0.525 0.100 0.642 0.063 0.231 0.099 0.202 FRANCE 1e3 1 0.531 0.065 0.371 0.132 0.225 0.122 0.216

10þ 1 0.098 0.775 0.100 0.433 0.144 0.429 GERMANY 1e3 1 0.125 0.019 �0.028 0.027 0.023

10þ 1 0.022 0.190 0.127 0.153 SPAIN 3 1 0.383 0.025 0.293

10 1 0.022 0.824 ITALY 1 1 0.035

10 1

Notes: The asterisks ** and * represent significance at the 1% and 5% levels respectively. The grey shaded area is the unconditional stock-bond correlation for the whole sample.

D . Petm

ezas, D . San

tam aria

/ Jou

rn al

of In tern

ation al

M on

ey an

d Fin

an ce

4 9 (2 014

) 4 01

e 4 24

4 0 9

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424410

market). On the other hand, the European Central Bank (ECB) embarked on a covered bond purchasing programme as a means of reviving the long term debt market used by banks to refine loans to the public and private sectors.

The scale of central bank intervention means that we have to adjust for the effects of QE in the EGARCH model and standardized residual returns as inputs into the DCC structure of equation (3). Hence, we begin with the conditional mean equation:

Ri;j;t ¼ g1 þ a1 X j¼1;2

Dj;t þ εt (6)

where Dj,t are dummy variables designed to capture the effects of QE on stock returns and bond yields. The term j ¼ 1,2 provides flexibility in the specified model by isolating the effects of different stages in the asset purchasing programmes on the stock and bond markets.6 However equation (6) is restrictive in that it does not model the transmitting effects of changes in liquidity conditions and the level of risk aversion under different market conditions. As a result, we expand equation (6) to enable feed through effects into the DCC model so that:

Ri;j;t ¼ g1 þ a1 X j¼1;2

Dj;t þ f1;k X

k¼1;2;3 DkDTEDt�1 þ f2;k

X k¼1;2;3

DkDVIXt�1 þ εt (7)

where k ¼ 1,2,3 represents interactive dummies that denote the pre-crisis period, the banking crisis and sovereign debt crisis, respectively. The coefficients f1,1 to f2,3 measure the impact of the TED spread and VIX index before and during the banking crisis followed by the debt crisis. The pre-crisis represents the period 1st January 2004 to 27th November 2006; the banking crisis from 28th November 2006 to 15th October 2009; and finally, the European debt crisis from 16th October 2009 to 6th September 2012.7 Hence, to generate the standardized residual returns ðεt=

ffiffiffiffiffi ht

p Þ for the DCC model,

adjusted for the effects of QE, credit market liquidity and investor risk aversion, involves estimating the following EGARCH (1,1) model on stock index returns and bond yields denoted as assets i and j respectively:

hi;j;t ¼ exp a0 þ a1 log

� hi;j;t�1

� þ a2xi;j;t�1

� (8)

where hi,j,t is the conditional variance at time t and xi,j,t�1 is the information component that contains the asymmetric term q1. Table 2 provides estimations from the DCC-EGARCH (1,1) model on all return series for the whole sample. Given the volume of results, only coefficients that are statistically sig- nificant at least at the 10% level are reported in the conditional mean equation.

A number of important observations can be made from these findings. After adjusting for the effects of QE, we find that stock market returns appear to be relatively insensitive to changes in the

6 For instance, for the UK market, we use two dummy variables; QE1 and QE2 (k ¼ 1,2) to represent two major policy ad- vances from the Bank of England. For QE1, Dj,t ¼ 1 between 5th March 2009 to 30th November 2009, and zero otherwise. This covers the initial asset purchase programme of £200bn in long term conventional gilts. The QE2 dummy variable is designed to capture the effects of three further rounds of QE from October 2011 to 6th September 2012. As such, Dj,t ¼ 1 from the 10th October 2011 onwards, and zero otherwise. Similarly, for the Federal Reserve, we employ two dummy variables to control for the impact of QE1 and QE2. For QE1, Dj,t ¼ 1 represents the period November 2008 to March 2010, and zero otherwise. This coincides with the acquisition of $1.25trillon in Mortgage Backed Securities (MBS), $200bn in Federal Agency debt followed by $300bn in long term government bonds. On the other hand, with QE2, Dj,t ¼ 1 covers the time period from November 2010 until end of June 2011, when a second wave of asset purchasing was announced which involved buying up to $600bn in long term government bonds. In relation to the ECB, we earmarked two major policy decisions to define our dummy variables. The first dummy variable Dj,t ¼ 1 covers a twelve month period between June 2009 and June 2010, to coincide with the covered bond purchasing programme. The second dummy variable Dj,t ¼ 1 is defined as the period 10th May 2010 until 6th September 2012 that coincides with the setting up of the Securities Market Programme (SMP) and the re-introduction of the Long Term Refinancing Operation (LTRO).

7 The three sub-periods used are derived from the definition of the banking crisis period provided by the Bank of International Settlements (2009) Annual Report. The sovereign debt crisis period coincides with the downgrade of Greece in December 2009 and the start of the next stage of the global financial crisis as stipulated by Reinhart and Rogoff (2009).

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 411

TED spread and VIX index. By contrast, both short and long dated bond yields are sensitive to the TED spread and VIX index, with the exception of German and to a lesser extent French and Italian benchmark bonds. The negative statistical significance of the TED spread and VIX index provide preliminary indications of a flight to (from) quality as a potential source of contagion due to tighter (relaxed) credit liquidity conditions and increased (falling) risk aversion (Chordia et al., 2005; Baur and Lucey, 2009). Although this finding is generally robust to bonds of different times to maturity regardless of market conditions, evidence in favour of the portfolio rebalancing hypothesis is coun- terbalanced by the insensitivity of stock index returns to changes in credit market liquidity and investor risk aversion.

Focusing on the conditional variance equation, the parameters (a1,a2,q1) are statistically different from zero, thus indicating the presence of EGARCH effects regardless of asset class and time to maturity. Closer inspection of the results reveals that volatility is persistent across all asset returns with the exception of UK long dated bond yields, a finding that coincides with the news innovation coefficient a2, which is dominated by the asymmetric component q1. In contrast, the estimated theta q1 is positive and statistically significant for long dated Spanish bond yields, as well as short and long dated Italian bonds, which coincides with high innovation coefficient a2 esti- mates. Nevertheless, all other model estimates show that asset returns (both stock and bond index returns) exhibit asymmetries, a finding that differs from the results reported by Cappiello et al. (2006), who find little evidence of an asymmetric effect in the conditional volatility of bond index yields.

5.2. Dynamic conditional correlation (DCC) analysis

Using the standardized residual returns generated by the EGARCH model of Table 2, we estimate DCC coefficients using equation (4) to test if the investor induced contagion hypothesis holds and whether this is caused by the wealth effect or portfolio rebalancing effect during crisis and non-crisis periods. As implied by Boyer et al. (2006), for investor induced contagion sourced by the wealth effect to hold, stock-bond correlation ri,j,t must increase in times of crisis. Conversely, for investor induced contagion to hold through the portfolio rebalancing hypothesis, the stock-bond correlation ri,j,t must decline during a crisis period.

To take into consideration that the transmission channel linking stock and bond markets may vary over time, Fig. 3 plots the time varying correlation estimations ri,j,t on stock-bond relationships for the entire sample period. The stock-bond correlation estimates exhibit a number of interesting features; for instance, the correlation estimates vary quite considerably over time, with a decline in ri,j,t observed before the banking crisis especially for stock-bond relationships involving long dated bonds. For cor- relations involving bond yields of the UK, US, France and Germany, ri,j,t increased during the banking crisis and sovereign debt crisis, thus indicating investor induced contagion is caused by a wealth effect. In all cases ri,j,t increased sharply in the aftermath of the collapse of Lehman Brothers and US House of Representatives rejection of the $700bn financial bailout in 2008, thus suggesting widespread systemic risk on the financial system (Acharya et al., 2009). However, for Spain and Italy, ri,j,t declined dramatically immediately after the banking crisis indicating that investor induced contagion dimin- ished until 2009. However, from 2010, we report investor induced contagion sourced by the portfolio rebalancing hypothesis as ri,j,t became negative during the debt crisis. Additionally, the sharp decline in stock-bond correlations that involve French bonds appears to coincide with France's AAA downgrade in early 2012. The volatile nature of DCC estimates especially during the sovereign debt crisis, suggests that the spread of the crisis is the result of a portfolio rebalancing effect caused by a “flight to quality”, which has a destabilizing effect across asset classes at a time of falling stock prices in the smaller markets.8 This is consistent with the findings of previous studies that have documented a negative impact of ratings downgrades on the stock markets (Behr and Güttler, 2008; Bannier and Hirsch, 2010).

8 For instance, whilst the FTSE 100, Russell 3000 and DAX30 all increased by 11.31%, 32.70% and 24.79%, respectively during the debt crisis period, this contrasts with the performance of stock markets in the epicentre of the crisis, such as the CAC40, IBEX35 and FTSEMIB indexes, which all declined by 11.18%, 29.04% and 34.66%, respectively.

Table 2 DCC-EGARCH (1,1) estimations e country stock and bond market returns.

Index QE adjustment Market risk factors Conditional variance Model diagnostics

g1 a1 a2 ∅1;1 ∅1;2 ∅1;3 ∅2;1 ∅2;2 ∅2;3 a0 a1 a2 q1 LB(6) LB(12)

FTSE100 0.0736** �0.0258a 0.0007a �0.0682** 0.9847** 0.0862** �0.1396** 4.9636 7.2810 (0.025) (0.013) (0.004) (0.012) (0.002) (0.015) (0.010) [0.55] [0.84]

Russell 3000 0.1280** 0.0793** 0.0007* 0.0004* �0.0652** 0.9832** 0.0845** �0.1351** 5.0176 7.6638 (0.034) (0.023) (0.0003) (0.0002) (0.013) (0.002) (0.015) (0.011) [0.54] [0.81]

CAC40 0.0910* �0.0799** 0.9775** 0.1107** �0.1570** 9.7300 10.837 (0.046) (0.012) (0.003) (0.016) (0.011) [0.14] [0.54]

DAX30 �0.0886** 0.9748** 0.1261** �0.1369** 5.7245 7.8531 (0.013) (0.003) (0.017) (0.009) [0.46] [0.80]

IBEX35 0.0048a 0.0005a �0.1055** 0.9776** 0.1470** �0.1294** 7.8395 13.196 (0.003) (0.0003) (0.012) (0.003)2 (0.016)1 (0.009) [0.25] [0.36]

FTSE MIB �0.0874** 0.9853** 0.1187** �0.1130** 3.7516 6.6119 (0.011) (0.002) (0.014) (0.008) [0.71] [0.88]

Maturity UK 1e3 �0.1697a �0.0499** 0.9995** 0.0823** �0.0483** 0.5450 2.1775

(0.099) (0.017) (0.006) (0.009) (0.006) [0.99] [0.99] 10þ �0.0585** 0.1572** �0.0392** �0.1297** �0.0165** 0.9265** 0.0825** �0.2696** 0.1871 0.2624

(0.015) (0.012) (0.002) (0.030) (0.002) (0.005) (0.008) (0.011) [0.99] [1.00] US 1e3 �0.0339* �0.0180** �0.0332* �0.0094* �0.0180** 0.9987** 0.0468** �0.0615** 8.9988 11.749

(0.010) (0.006) (0.014) (0.005) (0.006) (0.006) (0.006) (0.004) [0.17] [0.47] 10þ 0.1444** �0.0219* �0.0106a �0.0161** �0.0191** 0.9922** 0.1045** �0.0498** 5.2068 14.358

(0.041) (0.009) (0.006) (0.006) (0.005) (0.002) (0.011) (0.007) [0.52] [0.28] FRANCE 1e3 0.0487a 0.9995** 0.1144** �0.0552** 5.9047 17.121

(0.027) (0.001) (0.010) (0.006) [0.43] [0.15] 10þ 0.0004** 0.9916** 0.1099** �0.0367** 13.398 17.206

(0.0002) (0.002) (0.013) (0.007) [0.04] [0.14] GERMANY 1e3 0.0567* �0.2529a 0.9989** 0.1315** �0.0529** 7.2210 10.453

(0.026) (0.140) (0.001) (0.011) (0.007) [0.30] [0.58] 10þ 0.9950** 0.1216** �0.0350** 9.3980 11.977

(0.002) (0.012) (0.009) [0.15] [0.45] SPAIN 3 0.1179** �0.0103* �0.0123** �0.0304** �0.0014** 0.9961** 0.1444** �0.0046 9.776 11.380

(0.035) (0.0039) (0.002) (0.006) (0.0002) (0.001) (0.008) (0.007) [0.13] [0.50] 10 0.0471* �0.0909a �0.0062* �0.0109** 0.0003* 0.9941** 0.1417** 0.0491** 7.0067 14.367

(0.023) (0.051) (0.003) (0.003) (0.0002) (0.002) (0.013) (0.008) [0.32] [0.28] ITALY 3 0.0854** 0.0006a 0.9977** 0.1902** 0.0237** 1.9373 5.6783

(0.030) (0.0003) (0.002) (0.010) (0.006) [0.93] [0.93] 10 �0.0761a 0.0004* 0.9914** 0.1555** 0.0557** 16.709 23.049

(0.0451) (0.0002) (0.002) (0.014) (0.008) [0.01] [0.03]

Notes: The asterisks ** and * represent significance at the 1% and 5% levels respectively. The superscript a represents significance at the 10% level.

D . Petm

ezas, D . San

tam aria

/ Jou

rn al

of In tern

ation al

M on

ey an

d Fin

an ce

4 9 (2 014

) 4 01

e 4 24

412

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 413

A notable feature is the robustness of these results in all maturities that is not only country-specific but crosses borders.9

An issue of particular interest raised by the results is the trend in correlation estimates involving UK, US, French and German bonds and how this contrasts with stock-bond relations involving Spanish and Italian bonds in the aftermath of the banking crisis. This indicates that there are different forces at work in the transmission channels of the stock-bond relationship. For instance, the increase in ri,j,t involving bond yields irrespective of time to maturity for the UK, US, France and Germany during the sovereign debt crisis, at a time of rising stock prices in the major markets, uncovers investor induced contagion that coincides with the purchase of long term government debt under the QE programmes. An implication of the contagion model of Kyle and Xiong (2001) and a persistently high stock-bond cor- relation suggests an increase in the amount of liquidity in both the stock and bond markets due to global central bank intervention. This finding is in marked contrast to the decline in ri,j,t involving Spanish and Italian bonds regardless of maturity, which is suggestive of portfolio rebalancing by in- vestors. However, it is also worth noting that this may have been triggered by ratings downgrades of Spanish and Italian sovereign debt and the subsequent ECB intervention in the international bond markets.10

The remainder of this paper empirically investigates the feed through effects of a change in credit market liquidity conditions using the TED spread and investor confidence based on implied volatility estimates on the transmission channel of the stock-bond relationship.

5.3. DCC(Z)-EGARCH risk factor model analysis

To begin with, we re-estimated the EGARCH model using the conditional mean model of equation (6) that excludes the TED spread and VIX index to generate the standardized residual returns as inputs into the DCC model. Following this, we formulated the DCC(Z)-EGARCH risk factor model derived from equation (4):

hi;j;t ¼ rðZtÞ ffiffiffiffiffiffiffiffiffiffiffiffiffi hi;thj;t

q (9)

in which �1 � r(Zt) � þ1 is an increasing function of Zt, which in turn denotes a K � 1 vector of proxies for credit market stress and investor confidence that are used as determinants of the correlation co- efficient. As a result, equation (9) represents a useful way of identifying the transmitting effects of both factors that cause an increase (decrease) in the wealth effect or portfolio rebalancing process during crisis and non-crisis periods. To identify different market conditions, interactive dummy variables were introduced into the following risk factor model specification of the DCC function:

rðZtÞ ¼ 2 2 4 exp

� G1k;a;b

0Zt�1 �

1 þ exp � G1k;a;b

0Zt�1 � 3 5 � 1 (10)

where G1k,ij ¼ [G11,..,G12,.....,G1k] and Zt ¼ [z0,..,z2,.....,zk] in the home country a and foreign country b.11 We define G1k as the coefficients to be estimated for interactive dummy variables that are designed to capture the impact of both confidence indicators from countries a and b before and during the banking

9 Further support for the portfolio rebalancing hypothesis is reported for correlations involving Spanish and Italian bonds during the debt crisis by re-estimating the DCC model on bond yields for all countries. Additional robustness tests were carried out by estimating the DCC model on stock-bond relationships involving Euro-zone short and long dated bond yields. We find investor induced contagion caused by the wealth effect is evident during the banking crisis, whereas for the debt crisis, the portfolio rebalancing effect is the dominant source of contagion especially for EU short dated bonds. Although the results are not presented in the paper, they are available upon request. 10 Spain's credit rating was downgraded by S&P from AA þ to AA in April 2010, Moodys also downgraded Spanish debt from Aa2 to A1 in October 2011. Italian sovereign debt was downgraded from Aþ to A by S&P in September 2011. 11 represents constants and k ¼ 0,..... …,6 represents the number of parameters used to estimate the lagged correlation co- efficient estimates and proxies for credit market liquidity and investor risk appetite during the pre-crisis, banking crisis and sovereign debt crisis periods, respectively.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424414

and debt crises. Therefore, the exogenous variables used are the TED spread and implied volatility index (VIX) of Figs. 1 and 2 for each country of origin.12

The interactive dummy parameters for the TED spread are denoted as G11,.....,G13 and G14,.....,G16 for the VIX index where coefficient numbers [11]e[14], [12]e[15], and [13]e[16] capture the periods before the crisis, the banking crisis, and sovereign debt crisis, respectively.13 Tables 3 and 4 present the estimation results from the DCC(Z)-EGARCH risk factor model of equation (9) for stock-bond correla- tions involving short and long dated bond yields. Given the volume of results, the most significant findings are reported. Therefore, Table 3 reports stock-bond correlations that include the stock markets of the US, France, Spain and Italy for short dated bonds. For Table 4, stock-bond relationships involving long dated bond yields present results for France.14 Furthermore, only statistically significant co- efficients are reported up to the 5% level.

Turning to our findings, there are some notable features identified in the results; firstly, there is limited evidence that the TED spread and VIX index are the driving forces behind the stock-bond relationships before the banking crisis, in spite of earlier findings presented in Table 2 on short and long dated bond yields. Focusing on the banking crisis, we report some evidence that a widening of TED spreads reduces stock-bond correlations that include the Russell 3000 and DAX-30 indexes for long dated bond yields. This suggests that it reduces investor induced contagion sourced by the wealth effect, thus indicating evidence that funding illiquidity that leads to deteriorating liquidity conditions (Brunnermeier and Pedersen, 2009) caused some rebalancing of portfolios away from high risk assets towards quality (Chordia et al., 2005; Baur and Lucey, 2009). Nevertheless, risk aversion measured by the VIX index is a relatively more dominant driver behind the increase in stock-bond correlations involving the Russell 3000, IBEX-35 and FTSE MIB indexes, a finding that is consistent with the wealth effect.15 Interestingly, we uncover a spillover effect of risk aversion originating from foreign country b, which explains the increase in stock-bond correlations involving the CAC-40 for short dated bond yields. Spillover effects involving the TED spread and VIX index are also reported for correlations that incorporate the IBEX-35 index. Both results suggest that risk aversion levels and credit market liquidity conditions are not necessarily country-specific drivers of the asset markets, which implies that both factors may be determinants of cross-asset market volatility that crosses borders (Bianconi et al., 2013). Results on the debt crisis reveal that both the TED spread and the level of risk aversion are the main drivers behind the wealth effect and portfolio rebalancing hypotheses. On the one hand, finding a negative relationship between the TED spread and stock-bond correlations involving the Russell 3000 once again suggests that a narrowing of spreads observed in Fig. 2 mag- nifies the wealth effect with the exception of correlations involving Spanish and Italian bonds.16 On the other hand, the positive association between risk aversion and stock-bond correlations for the Russell 3000 and to a lesser extent, the CAC-40, IBEX-35 and FTSE MIB indexes involving short dated bonds implies that it contributes to the increase in the wealth effect. For correlations involving particularly Italian bonds, risk aversion contributes significantly to the increase in contagion sourced by the portfolio rebalancing effect.

12 The usefulness of the TED spread as a measure for credit market liquidity is well known, whereas for the VIX index, the information contained has been found to cause a decoupling between bond and stock prices, especially during bouts of risk aversion (Andersson et al., 2007). 13 The interactive dummies for the TED spread and VIX index are defined as G11,a,b,........,G13,a,b¼1 � DTED and G14,a,b,........ ,G16,a,b¼1 � DVIX, and zero otherwise. To maintain consistency within the analysis, the sample periods used to define all interactive dummies are the same as before. 14 In relation to stock-bond relationships that involve the FTSE-100, we find that stock-bond correlations are relatively insensitive to credit market illiquidity and high levels of risk aversion. However, for correlations involving the DAX Index with short dated bonds, we find some evidence that risk aversion explains the wealth effect. Although these findings are not pre- sented here, the results are available upon request. 15 According to the modified contagion model, this implies that risk aversion may arise from shocks to the funding constraints of investors resulting in a liquidation of positions in the asset markets (Brunnermeier and Pedersen, 2009), which, in turn, lead to higher volatility and an increase in stock-bond correlations. 16 This implies that a relaxation of credit market liquidity conditions encourages investors to participate in the stock markets at a time of more relaxed funding constraints (Brunnermeier and Pedersen, 2009) and high bond market liquidity due to the large asset purchasing programmes (Gagnon et al., 2011; Joyce et al., 2011; Neely, 2011).

FTSE100 Index and FTSE 1-3 Year Bond Returns

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

FTSE100 + UK1-3 FTSE100 + US1-3 FTSE100 + FR1-3

FTSE100 + BD1-3 FTSE100 + ES03 FTSE100 + IT03

Russell 3000 Index and FTSE 1-3 Year Bond Returns

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

RUS3000 + UK1-3 RUS3000 + US1-3 RUS3000 + FR1-3

RUS3000 + BD1-3 RUS3000 + ES03 RUS3000 + IT03

CAC40 Index and FTSE 1-3 Year Bond Returns

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

CAC40 + UK1-3 CAC40 + US1-3 CAC40 + FR1-3

CAC40 + BD1-3 CAC40 + ES03 CAC40 + IT03

DAX30 Index and FTSE 1-3 Year Bond Returns

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

DAX + UK1-3 DAX + US1-3 DAX + FR1-3

DAX + BD1-3 DAX + ES03 DAX + IT03

IBEX35 Index and FTSE 1-3 Year Bond Returns

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

IBEX35 + UK1-3 IBEX35 + US1-3 IBEX35 + FR1-3

IBEX35 + BD1-3 IBEX35 + ES03 IBEX35 + IT03

FTSE MIB Index and FTSE 1-3 Year Bond Returns

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Jan-04 Nov-04 Sep-05 Jul-06 May-07 Mar-08 Jan-09 Nov-09 Sep-10 Jul-11 May-12

MIB + UK1-3 MIB + US1-3 MIB + FR1-3

MIB + BD1-3 MIB + ES03 MIB + IT03

Fig. 3. DCC Estimates on Stock-Bond Relationships.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 415

Table 3 DCC(Z)-EGARCH risk factor model: stock-bond correlations (short dated bond yields).

Russell 3000 CAC40

UK US FRANCE GERMANY SPAIN ITALY UK US FRANCE GERMANY SPAIN ITALY

G10 0.2684 ** 0.4337** 0.3998** 0.3917** 0.1004**

(0.075) (0.105) (0.114) (0.110) (0.019) r 0.998** 0.9990** 0.9834** 0.9988** 0.9988** 0.9982** 0.9833** 0.9875** 0.9844** 0.9693** 0.9988** 0.9977**

(0.001) (0.001) (0.004) (0.001) (0.001) (0.002) (0.004) (0.003) (0.004) (0.005) (0.001) (0.002) Transmission effectsecountry a G11,i 0.0707

*

(0.034) G12,i

G13,i �0.0448** �0.048** �0.1393* �0.0842** �0.0370** (0.017) (0.017) (0.060) (0.025) (0.001)

G14,i 0.0020 *

(0.001) G15,i 0.0003

** 0.0005** 0.0010** 0.0005** 0.0005** 0.0011** 0.0005**

(8.42E-5) (8.47E-5) (0.0003) (0.0001) (0.0001) (0.0004) (0.0001) G16,i 0.0004

** 0.0012** 0.0007** 0.0005** 0.0011** 0.0004** 0.0011**

(0.0001) (0.0004) (0.0002) (7.76E-5) (0.0003) (0.0001) (0.0003) Transmission effectsecountry b G11,j e e e

e e e

G12,j e 0.0013 * e e

e (0.0006) e e G13,j e e �0.1055* e

e e (0.054) e G14,j e e e

e e e

G15,j e 0.0010 *- 0.0008** e 0.0029** 0.0005**

e (0.0004) (0.0003) e (0.0008) (0.0002) G16,j e �0.0002* �0.0006* e 0.0008* e 0.0032*

e (7.62E-5) (0.0003) e (0.0004) e (0.002) c2(6) 3.6122 5.3993 3.5376 6.0984 0.5163 0.1305 2.3983 1.9739 3.5946 2.5295 0.4219 0.1305

[0.73] [0.49] [0.74] [0.41] [0.99] [1.00] [0.88] [0.92] [0.73] [0.87] [0.99] [1.00] LB(6) 6.5759 4.8388 7.7137 56.167 2.7931 6.7748 2.9683 4.7866 2.1919 5.3473 5.1576 6.7748

[0.36] [0.57] [0.26] [0.00] [0.83] [0.34] [0.81] [0.57] [0.90] [0.50] [0.52] [0.34] LogL �1786.44 �2035.8 �4734.30 �2745.59 �668.67 �3438.10 �4689.34 �4499.03 �5401.72 �6199.26 �2844.26 �3438.10

D . Petm

ezas, D . San

tam aria

/ Jou

rn al

of In tern

ation al

M on

ey an

d Fin

an ce

4 9 (2 014

) 4 01

e 4 24

416

IBEX35 FTSE MIB

UK US FRANCE GERMANY SPAIN ITALY UK US FRANCE GERMANY SPAIN ITALY

G10 0.3350 ** 0.3048** 0.5982** 1.0852** 0.3313** 0.2889**3 0.2935** 0.1042**

(0.090) (0.096) (0.133) (0.194) (0.092) (0.096) {0.096) (0.019) r 0.9860** 0.9895** 0.9749** 0.9665** 0.9985** 0.9978** 0.9863** 0.9905** 0.2935** 0.9678** 0.9976** 0.9983**

(0.003) (0.003) (0.005) (0.005) (0.001) (0.002) (0.003) (0.003) (0.003) (0.005) (0.002) (0.001) Transmission effectsecountry a G11,i �0.1220*

(0.054) G12,i

G13,i �0.0130** �0.0251** �0.0373** (0.006) (0.007) (0.009)

G14,i 0.0009 **

(0.0003) 0.0029**

G15,i 0.0014 ** 0.0010** 0.0023** 0.0006** 0.0014* 0.0013** 0.0021** 0.0029* (0.0007)

(0.0004) (0.0003) (0.0007) (0.0001) (0.0005) (0.0003) (0.0005) (0.001) 0.0014**

G16,i 0.0035 ** 0.0004* 0.0022* (0.0003)

(0.001) (0.0002) (0.001) Transmission effectsecountry b G11,j G12,j G13,j 0.3287

*

(0.139) G14,j 0.0018

* 0.0013**

(0.0007) (0.0004) G15,j 0.0007

* 0.0006**

(0.0003) (0.0002) G16,j c2(6) 5.8873 4.0207 7.7774 2.0233 0.6494 5.1899 3.0439 3.1946 3.4875 2.2674 1.0449 1.6143

[0.44] [0.67] [0.25] [0.92] [0.99] [0.52] [0.80] [0.78] [0.75] [0.89] [0.98] [0.95] LB(6) 4.6430 4.3898 2.2651 3.5108 12.552 9.3483 3.8078 5.5880 2.7080 3.6115 3.2060 14.229

[0.59] [0.62] [0.89] [0.74] [0.05] [0.16] [0.70] [0.47] [0.85] [0.73] [0.78] [0.03] LogL �4508.26 �4460.17 �6029.26 �6289.30 �3261.56 �4876.79 �4777.25 �4381.12 �5337.94 �6291.19 �3706.726 �4937.35

Note: The asterisks ** and * represent significance at the 1% and 5% levels respectively.

D . Petm

ezas, D . San

tam aria

/ Jou

rn al

of In tern

ation al

M on

ey an

d Fin

an ce

4 9 (2 014

) 4 01

e 4 24

417

Table 4 DCC(Z)-EGARCH risk factor model: stock-bond correlations (long dated bond yields).

Russell 3000 DAX30

UK US FRANCE GERMANY SPAIN ITALY UK US FRANCE GERMANY SPAIN ITALY

G10 1.3724 ** 0.2411** 0.3317 0.6645** 0.3558** 0.3321**

(0.186) (0.073) (0.036) (0.151) (0.108) (0.110) r 0.9338** 0.9989** 0.9848** 0.9979** 0.9984** 0.9979** 0.8887 0.9786** 0.9848** 0.9890** 0.9976** 0.9975**

(0.008) (0.001) (0.004) (0.001) (0.001) (0.002) (0.010) (0.004) (0.004) (0.003) (0.022} (0.002) Transmission effectsecountry a G11,i 0.0174

* 0.0793* 0.0450*

(0.009) (0.038) (0.022) G12,i �0.0261* �0.0248* �0.0227*

(0.012) (0.012) (0.010) G13,i �0.1156** �0.0955**

(0.032) (0.033) Transmission effectsecountry b G14,i G15,i 0.0028

** 0.0003** 0.0006* 0.0004*

(0.0008) (0.0001) (0.0003) (0.0002) G16,i 0.0004

* 0.0005* 0.0017** 0.0010**

(0.0002) (0.0002) (0.0003) (0.0003) G11,j e e

e e

G12,j e 0.0267 ** e �0.0257**

e (0.009) e (0.009) G13,j e e

e e

G14,j e e

e e

G15,j e e

e e

G16,j e �0.0007** e e (0.0002) e

c2(6) 3.6672 1.0332 6.0714 11.876 1.1692 10.60547 9.0423 6.8459 18.741 13.161 2.1132 3.1585 [0.72] [0.98] [0.42] [0.06] [0.98] [0.10] [0.17] [0.34] [0.00] [0.04] [0.91] [0.79]

LB(6) 8.7621 7.0813 2.6159 6.3597 10.424 8.5272 4.9903 7.9171 4.4232 6.4405 5.7602 5.7657 [0.19] [0.31] [0.86] [0.38] [0.11] [0.20] [0.55] [0.24] [0.62] [0.38] [0.45] [0.45]

LogL �6756.81 �2994.36 �4859.72 �3263.93 �3328.86 �3366.98 �8498.47 �5820.50 �5716.98 �5494.54 �4635.70 �4762.95

D . Petm

ezas, D . San

tam aria

/ Jou

rn al

of In tern

ation al

M on

ey an

d Fin

an ce

4 9 (2 014

) 4 01

e 4 24

418

IBEX35 FTSE MIB

UK US FRANCE GERMANY SPAIN ITALY UK US FRANCE GERMANY SPAIN ITALY

G10 0.1727 ** 0.7541** 0.3814** 0.3050** 0.3367** 0.5840** 0.3011** 0.2688*

(0.024) (0.160) (0.112) (0.108) (0.037) (0.140) (0.101) (0.104) r 0.9379** 0.9742** 0.9829** 0.9901** 0.9845** 0.9987** 0.8862 0.9805** 0.9867** 0.9913** 0.9982** 0.9982**

(0.007) (0.005) (0.004) (0.003) (0.004) (0.001) (0.010) (0.004) (0.003) (0.003) (0.001) (0.002) Transmission effectsecountry a G11,i G12,i �0.1060** �0.0360** �0.0420** �0.0402** �0.0232** �0.0260**

(0.045) (0.013) (0.013) (0.012) (0.008) (0.009) G13,i 0.0198

*

(0.010) G14,i 0.0020

**

(0.0007) G15,i 0.0007

*

(0.0003) G16,i Transmission effectsecountry b G11,j e

e

G12,j e �0.0252* �0.0163* e (0.011) (0.008)

G13,j e

e

G14,j e 0.0047 ** 0.0021** 0.0018**

e (0.0017) (0.0006) (0.0006) G15,j e 0.0013

* 0.0011** 0.0006*

e (0.0006) (0.0003) (0.0002) G16,j e

e

c2(6) 4.6294 46.144 24.895 18.552 3.3865 8.6340 6.8379 4.6523 28.520 12.495 4.7612 20.912 [0.59] [0.00] [0.00] [0.01] [0.76] [0.20] [0.34] [0.59] [0.00] [0.05] [0.57] [0.00]

LB(6) 3.6761 5.2635 5.3746 4.6297 2.5697 11.780 5.2334 11.652 6.7137 6.4713 13.537 7.4972 [0.72] [0.51] [0.50] [0.59] [0.86] [0.07] [0.51] [0.07] [0.35] [0.37] [0.04] [0.28]

LogL �7248.15 �6155.21 �5934.50 �5532.16 �6861.19 �4672.50 �8554.50 �5695.32 �5660.26 �5530.91 �4678.67 �4920.74 Note: The asterisks ** and * represent significance at the 1% and 5% levels respectively.

D . Petm

ezas, D . San

tam aria

/ Jou

rn al

of In tern

ation al

M on

ey an

d Fin

an ce

4 9 (2 014

) 4 01

e 4 24

419

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424420

5.4. Credit liquidity conditions, risk aversion and the DCC estimates

So far, the results suggest that in general, the dominant force behind investor induced contagion during the crisis is the wealth effect for correlations involving the stock markets of the UK, US, France and Germany. For Spain and Italy, investor induced contagion is caused by the portfolio rebalancing hypothesis, which seems to be the main driver during the debt crisis. However, the statistical signif- icance of the TED spread and VIX index reported in Table 2 poses a question on the impact of credit market liquidity and risk aversion on the transmission channel that governs the stock-bond rela- tionship. Fig. 4 plots the difference between the DCC estimates using the standardized residual returns from the EGARCH model with and without the TED spread and VIX index in the conditional mean. A manifestation of the wealth effect due to credit market liquidity conditions, risk aversion or both during the banking and sovereign debt crises is represented by readings greater than zero. Conversely, the inducement of the portfolio rebalancing effect is identified by readings less than zero.

Noticeably, the impact of credit market illiquidity and the level of risk aversion are generally more persistent on stock-bond correlations that incorporate short dated bonds than relationships with longer dated bonds especially during the banking crisis. For instance, a manifestation of the wealth effect is reported for correlations involving the Russell 3000 index and, in particular, stock-bond re- lationships that include UK short dated bonds. Based on the Brunnermeier and Pedersen (2009) model, this suggests that a lack of liquidity and increased funding constraints imposed on investors following the arrival of shocks cause an increase in risk aversion and a further decline in liquidity in both markets. Yet despite this, the inclusion of both risk barometers has tended to reduce investor induced contagion sourced by the wealth effect for correlations involving Spanish short dated bonds. The above results are repeated for the period of the debt crisis at a time of narrow TED spreads (see Fig. 1) indicating relaxed credit liquidity conditions and low VIX index readings with the exception of two periods of risk aversion in 2010 and 2012 (see Fig. 2). However, the most notable exception here is the stock-bond relationships involving Spanish short dated bonds and the IBEX-35 with Italian bonds where a manifestation of the portfolio rebalancing effect is reported. Finally, a striking feature for stock-bond relationships involving long dated bond yields is the sensitivity of DCC estimates to the inclusion of the TED spread and VIX index, especially for correlations that include UK bonds, which is not neces- sarily restricted to crisis periods. This is hardly surprising in light of the EGARCH results of Table 2 where UK bond yields appear to be very sensitive to credit market liquidity and the level of risk aversion before the banking crisis.

6. Discussion of the results

The results presented in this paper offer important insights into how a crisis spreads. Finding ev- idence that stock-bond correlations are volatile from pre-crisis to the banking and European sovereign debt crisis indicates that the transmission mechanism that defines the correlation of assets does vary in different market conditions (Forbes and Rigobon, 2000, 2002), a finding that is also consistent with the results of Gulko (2002) and Fleming et al. (1998, 2003). Most importantly, it provides support to the investor induced contagion hypothesis.

We have also established an increase in stock-bond correlations during the banking crisis that is indicative of a wealth effect and, as such, is consistent with the implications derived from the contagion model postulated by Kyle and Xiong (2001), after relaxing the assumption of a two high risk asset markets. However, this differs from the key findings of Baur and Lucey (2009) who reported evidence of flight to quality regularly occurring during crisis periods; nevertheless, their time frame considered the pre-2007 banking crisis period. Intuitively, the increase in correlation observed during the banking crisis appears to be caused by a volatile “liquidity spiral” between both markets (Brunnermeier and Pedersen, 2009). In the modified contagion model, investors tend to substitute growth for income in a crisis; yet at the same time, shocks increase the risk that investors will hit their funding limits, which, as yields decline, leads to a liquidation of positions in bonds as lower income streams are associated with an increased risk of capital constraints (Brunnermeier and Pedersen, 2009). As a consequence, an unstable “liquidity spiral” between the two markets is the result, causing both asset classes to become more volatile and an increase in stock-bond correlations.

FTSE 100 Index and 1-3 Year Bond Returns

Russell 3000 Index and 1-3 Year Bond Returns

CAC40 Index and 1-3 Year Bond Returns

DAX30 Index and 1-3 Year Bond Returns

IBEX35 Index and 1-3 Year Bond Returns

FTSE MIB Index and 1-3 Year Bond Returns

FTSE 100 Index and 10+ Year Bond Returns

Russell 3000 Index and 10+ Year Bond Returns

CAC40 Index and 10+ Year Bond Returns

DAX30 Index and 10+ Year Bond Returns

IBEX35 Index and 10+ Year Bond Returns

FTSE MIB Index and 10+ Year Bond Returns

Fig. 4. The Impact of Credit Market Liquidity and Investor Confidence.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 421

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424422

The increase in stock-bond correlations reported is robust during the debt crisis for stock-bond relationships involving all stock indexes and bond yields of the UK, US and Germany at a time when stock markets of those countries are rising. However, according to the Brunnermeier and Pedersen (2009) model, such a scenario would coincide with relaxed funding constraints on investors to un- dertake activities that generate greater liquidity and a decline in volatility in the asset markets.17 As a consequence, the increase in stock-bond correlations observed during the debt crisis for the major countries contradict the findings implied by Baur and Lucey (2009) who postulate that a “flight from quality” and a decline in liquidity in the bond market would arise at a time of rising stock prices. Finding an increase in stock-bond correlations suggests that contagion is not necessarily restricted to negative events and falling stock prices as also implied by previous studies (King and Wadhwani,1990; Koutmos, 1996). On the other hand, for correlations involving Spanish and Italian bonds, our results also indicate that investor induced contagion does diminish after the banking crisis until the end of 2009. Negative DCC estimates from 2010 onwards suggest further evidence of investor induced contagion, but this time, caused by the portfolio rebalancing effect during the sovereign debt crisis. Whilst this is consistent with the findings of previous studies (Hartmann et al., 2004; Connolly et al., 2005; Baur and Lucey, 2009), this is indicative of a change in the transmission mechanism in different market conditions.

An added dimension of this paper is that we consider the transmitting effects of a change in financial market confidence indicators on the transmission channel of the stock-bond relationship. The inclusion of two barometers of risk is motivated by the findings of previous studies that view the TED spread (Brunnermeier, 2009; Melvin and Taylor, 2009) and VIX index (Low, 2004) as containing useful information on the state of the credit market and investor confidence; this, in turn, has a causal impact on the behaviour of different asset classes (Lashgari, 2000; Andersson et al., 2007). Our findings suggest that a major driving force of investor induced contagion is risk aversion during the banking and debt crises. Interestingly, this is counter balanced by evidence of an inverse relationship between credit market liquidity and stock-bond correlations. As a result, a widening of TED spreads, i.e., funding illiquidity conditions, observed in Fig. 1 appears to reduce investor induced contagion caused by the wealth effect during the banking crisis. This implies that some investors engage in portfolio reba- lancing and a “flight to quality”, a dynamic that is consistent with the results of Baur and Lucey (2009). A narrowing of TED spreads during the debt crisis contributed to the wealth effect for the majority of stock-bond relationships as well as the portfolio rebalancing effect for correlations involving Spanish and Italian bonds.

The intuition behind these results can generally be found in the liquidity channel postulated by Chordia et al. (2005) where the banking crisis characterized by a lack of liquidity and a high degree of uncertainty causes investors to liquidate their positions in high risk assets in exchange for low risk assets. Ultimately, this causes a shift in liquidity away from the stock market to the bond market. Conversely, the debt crisis characterized by narrow TED spreads and rising stock prices in the major markets will encourage investors to buy high risk assets, making the bond markets less liquid. When added together, this leads to a negative stock-bond relationship, which nests neatly with the corre- lation estimates reported for Spanish and Italian bonds.

Exploring the difference in DCC estimates with and without the TED spread and VIX index in the EGARCH model provides greater insight into the role of both risk barometers on the source of investor induced contagion over a period of time. The key finding reported is the manifestation of the wealth effect arising from credit market liquidity and the level of risk aversion in the pre-crisis period of 2007 and especially during the banking crisis. The results are repeated for the majority of stock-bond re- lationships during the debt crisis with the exception of correlations involving Spanish and Italian bonds where a manifestation of the portfolio rebalancing effect is reported. This is consistent with previous studies that find contagion only arises in times of a crisis (Kaminsky and Schmukler,1999; Chiang et al., 2007).

17 This period also coincides with the large scale QE programmes investigated by Gagnon et al. (2011), Joyce et al. (2011) and Neely (2011) outlined in footnote 6 that have caused bond prices (yields) to remain high (low) in a highly liquid market at a time of rising stock prices.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424 423

7. Conclusion

We provide evidence in support of the investor induced contagion hypothesis during the banking crisis and European sovereign debt crisis for stock-bond relationships. We find that the source of contagion during the banking crisis is the wealth effect that spreads across borders due to the drying up of liquidity from the stock and bond markets. However, a mixed picture is reported for the debt crisis where we report some evidence in favour of the portfolio rebalancing hypothesis as the reason behind the spread of the crisis for correlations involving Spanish and Italian bonds. Our results also indicate that risk aversion during the banking crisis contributes to the wealth effect in all instances. However, for stock-bond correlations involving Spanish and Italian bonds, we have also established that increased risk appetite magnified the portfolio rebalancing effect during the debt crisis.

Taken together, our findings imply that transmission mechanisms that define the relationship be- tween asset markets within a country and beyond change as a result of a financial crisis. Additionally, our results offer important insights to policy makers. The evidence which supports the existence of a liquidity channel in explaining how a crisis spreads, in light of credit market liquidity conditions and levels of risk aversion, implies that policy makers should consider whether liquidity levels in the financial markets are maintained through interventionist policies, particularly in periods of market stress. Finally, our results also pose questions on the unintentional consequences of central bank intervention in undertaking QE programmes on the way a crisis spreads. Such an assertion is borne out by overwhelming support for investor induced contagion sourced by the portfolio rebalancing hy- pothesis for stock-bond correlations involving Spanish and Italian bonds from 2010 onwards. Overall, our findings provide striking evidence on the time varying nature of stock-bond relationships over different market conditions suggesting that investor induced contagion explains how a crisis spreads within a country and beyond its borders.

References

Acharya, V., Pedersen, L., 2005. Asset pricing with liquidity risk. J. Financial Econ. 77, 375e410. Acharya, V., Philippon, T., Richardson, M., Roubini, N., 2009. The financial crisis of 2007-2009: causes and remedies. In:

Acharya, V., Richardson, M. (Eds.), Restoring Financial Stability: How to Repair a Failed System. John Wiley and Son Ltd. Alberg, D., Shalit, H., Yosel, R., 2008. Estimating stock market volatility using asymmetric GARCH Models. Appl. Financ. Econ. 18,

1201e1208. Andersen, T.G., Bollerslev, T., Diebold, F.X., Ebens, H., 2001. The distribution of realized stock return volatility. J. Financial Econ.

61, 43e76. Andersson, M., Krylova, E., Vahamaa, S., 2007. Why does the correlation between stock and bond yields vary over time? Appl.

Financ. Econ. 18, 139e151. Baele, L., 2005. Volatility spillover effects in European equity markets. J. Financial Quant. Anal. 40, 373e401. Bank of International Settlements (2009). 79th Annual Report. Bannier, C.E., Hirsch, C.W., 2010. The economic function of credit rating agencies - what does the watch list tell us? J. Bank.

Finance 34, 3037e3049. Baur, D.G., Lucey, B.M., 2009. Flights and contagion-an empirical analysis of stock-bond correlations. J. Financial Stab. 5 (4),

339e352. Behr, P., Güttler, A., 2008. The informational content of unsolicited ratings. J. Bank. Finance 32, 587e599. Berndt, E.K., Hall, B., Hall, R., Hausman, J.A., 1974. Estimation and inference in nonlinear structural models. Ann. Econ. Soc. Meas.

3 (4), 653e665. Bianconi, M., Yoshino, J.A., Machado De Sousa, M.O., 2013. BRIC and the US financial crisis: an empirical investigation of stock

and bond markets. Emerg. Mark. Rev. 14, 76e109. Billio, M., Pelizzon, L., 2003. Volatility and shocks spillover before and after emu in European stock markets. J. Multinatl.

Financial Manag. 13, 323e340. Blair, B.J., Poon, S.H., Taylor, S.J., 2001. Forecasting S&P 100 volatility: the incremental information content of implied volatilities

and high frequency returns. J. Econ. 105, 5e26. Blanco, R., Brennan, S., Marsh, I.W., 2005. An empirical analysis of the dynamic relation between investment-grade bonds and

credit default swaps. J. Finance 60, 2255e2281. Boyer, B.H., Kumagai, T., Yuan, K., 2006. How do crises spread? Evidence from accessible and inaccessible stock indices. J.

Finance 61, 957e1003. Brandt, M.W., Jones, C.S., 2006. Volatility forecasting with range-based EGARCH models. J. Bus. Econ. Stat. 24, 470e486. Brunnermeier, M.K., 2009. Deciphering the liquidity and credit crunch 2007e2008. J. Econ. Perspect. 23, 77e100. Brunnermeier, M.K., Pedersen, L.H., 2009. Market liquidity and funding liquidity. Rev. Financial Stud. 22 (6), 2201e2238. Campbell, J., Ammer, J., 1993. What moves the stock and bond markets? A variance decomposition for long-term asset returns.

J. Finance 48, 3e37. Caporale, G.M., Pittis, N., Spagnolo, N., 2006. Volatility transmission and financial crisis. J. Econ. 30, 376e390.

D. Petmezas, D. Santamaria / Journal of International Money and Finance 49 (2014) 401e424424

Cappiello, L., Engle, R.F., Sheppard, K., 2006. Asymmetric dynamics in the correlations of global equity and bond yields. J. Financial Econ. 4, 537e572.

Chiang, T.C., Jeon, B.N., Li, H., 2007. Dynamic correlation analysis of financial contagion: evidence from Asian markets. J. Int. Money Finance 26, 1206e1228.

Chordia, T., Sarkar, A., Subrahmanyam, A., 2005. An empirical analysis of stock and bond market liquidity. Rev. Financial Stud. 18, 85e129.

Collin-Dufresne, P., Goldstein, R.S., Martin, J.S., 2001. The determinants of credit spread changes. J. Finance 56, 2177e2207. Connolly, R., Stivers, C., Sun, L., 2005. Stock market uncertainty and the stock-bond yield relation. J. Financial Quant. Anal. 40,

161e193. Engle, R., 2002. Dynamic conditional correlation: a simple class of multivariate generalized autoregressive conditional heter-

oskedasticity models. J. Bus. Econ. Stat. 20, 339e350. Fleming, J., Kirby, C., Ostdiek, B., 1998. Information and volatility linkages in the stock, bond, and money markets. J. Financial

Econ. 49, 111e137. Fleming, J., Kirby, C., Ostdiek, B., 2003. The economic value of volatility timing using “realized” volatility. J. Financial Econ. 67,

473e509. Forbes, K., Rigobon, R., 2000. Contagion in Latin America: definitions, measurement, and policy implications. Economia 1, 1e46. Forbes, K., Rigobon, R., 2002. No Contagion, only interdependence: measuring stock market co-movements. J. Finance 5,

2223e2261. Forte, S., Pe~na, J.I., 2009. Credit spreads: an empirical analysis on the informational content of stocks, bonds, and CDS. J. Bank.

Finance 33, 2013e2025. French, K.R., Schwert, G.W., Stambaugh, R.F., 1987. Expected stock returns and volatility. J. Financial Econ. 19, 3e30. Gagnon, J., Raskin, M., Remache, J., Sack, B., 2011. The financial market effects of the Federal Reserve’s large-scale asset pur-

chases. Int. J. Central Bank. 7 (1), 3e43. Gulko, L., 2002. Decoupling. J. Portfolio Manag. 28, 59e67. Hartmann, P., Straetmans, S., de Vries, C.G., 2004. Asset market linkages in crisis periods. Rev. Econ. Stat. 86, 313e326. Hentschel, L., 1995. All in the family: nesting symmetric and asymmetric GARCH models. J. Financial Econ. 39, 71e104. Jones, C., Wilson, J., 2004. The changing nature of stock and bond volatility. Financial Anal. J. 60, 100e113. Joyce, M., Lasaosa, A., Stevens, I., Tong, M., 2011. The financial market impact of quantitative easing in the United Kingdom’. Int. J.

Central Bank. 7, 113e161. Kaminsky, G.L., Schmukler, S.L., 1999. What triggers market jitters? A chronicle of the Asian crisis. J. Int. Money Finance 18,

537e560. Kim, S.J., Moshirian, F., 2004. Dynamic stock market integration driven by the European Monetary Union: an empirical analysis.

J. Bank. Finance 29, 2475e2502. King, M., Wadhwani, S., 1990. Transmission of volatility between stock markets. Rev. Financial Stud. 3, 5e33. Koopman, S.J., Jungbacker, B., Hol, E., 2005. Forecasting daily variability of the S&P 100 stock index using historical, realized and

implied volatility measurements. J. Empir. Finance 12, 445e475. Koutmos, G., 1996. Modelling the dynamic interdependence of major European stock markets. J. Bus. Finance Account. 23,

975e988. Kyle, A.S., Xiong, W., 2001. Contagion as a wealth effect. J. Finance 56, 1401e1440. Lashgari, M., 2000. .The role of TED spread and confidence index in explaining the behaviour of stock prices. Am. Bus. Rev. 1,

1e3. Li, L., 2004. Macroeconomic factors and the correlation of stock and bond yields. In: Proceeding of the 2004 American Finance

Association Meeting. Low, C., 2004. The fear and exuberance from implied volatility of S&P 100 index options. J. Bus. 77, 527e546. Melvin, M., Taylor, M.P., 2009. The crisis in the foreign exchange market. J. Int. Money Finance 28, 1317e1330. Neely, C.J., 2011. The Large-scale Asset Purchases had Large International Effects. St. Louis Working Paper 2010-018C. Nelson, D., 1991. Conditional heteroscedasticity in asset returns: a new approach. Econometrica 59, 347e370. Norden, L., Weber, M., 2009. The co-movement of credit default swap, bond and stock markets: an empirical analysis. Eur.

Financ. Manag. 15, 529e562. Ravichandran, K., Maloain, A.M., 2010. The global financial crisis and stock market linkages: further evidence on GCC Market. J.

Money, Invest. Bank. 16, 46e56. Reinhart, C.M., Rogoff, K.S., 2009. The aftermath of financial crisis. Am. Econ. Rev. 9, 466e472. Schwert, G.W., 1990. Stock volatility and the Crash of ‘87. Rev. Financial Stud. 3, 77e102. Shiller, R., Beltratti, A., 1992. Stock prices and bond yields: can their comovements be explained in terms of present value

models? J. Monetary Econ. 30, 25e46. Taylor, J.B., Williams, J.C., 2009. A black swan in the money market. Am. Econ. J. Macroecon. 1, 58e83. Wongswan, J., 2006. Transmission of information across international equity markets. Rev. Financial Stud. 19, 1157e1189. Yang, J., Hsiao, C., Wang, Z., 2006. The emerging market linkages: further evidence. J. Appl. Econ. 21, 727e744.