COVID-19 related media sentiment and the yield curve of G-7 economies

Abstract

We explore the connectedness of the components of the sovereign yield curve (slope, level and curvature) across G-7 countries and media sentiment about COVID-19. The recent pandemic is a unique opportunity to identifying the transmitters and receivers of risk. Our results indicate that media sentiment along with the US yield curve components are main transmitter of spillovers, whereas Japan is the leading recipient of spillover. Among the European countries, we notice France as a major transmit, whereas Germany and UK switch role as transmitter and receiver alternatively. The results are important for mapping the interconnectedness between countries. In addition, policy makers can use them when devising disaster plans to prepare for future market crises.

1. Introduction

One of the basic principles in finance theory dictates the importance of the diversification of investments to reduce risk in one’s portfolio. Effective diversification, at least theoretically, involves asset allocation to different sectors, stocks and countries that are weakly correlated. Extant literature shows that market crises such as the late 1990s’ dot.com bubble, the global financial crisis of 2008, the subsequent European sovereign debt crisis as well as the evolution of international capital markets have made countries more interconnected (Zaremba et al., 2019, Spierdijk and Umar, 2014, Naeem et al., 2020, Kenourgios et al., 2014). Consequently, this interrelationship may violate the premise regarding the benefits associated with asset allocation in different capital markets. In fact, such events have proven that the correlation between the various markets strengthens and becomes more linked in times of market turbulence. This phenomenon has inspired researchers to devote much of their attention to the interconnectedness between countries, and their different types of assets.

A rapidly growing stream of studies has explored risk spillovers and uncertainty across economies various financial markets and across various assets such as equities, fixed income and commodity markets (e.g., Colombo, 2013, Conefrey and Cronin, 2015, Gupta et al., 2018, Fernández-Rodríguez et al., 2016, Umar, 2017, Antonakakis et al., 2018, Cronin and Dunne, 2019, Cipollini and Mikaliunaite, 2020, Liow and Song, 2020, Riaz et al., 2020, Zaremba et al., 2021, Zhao et al., 2021). In this respect, many of the existing studies focus their examination on the connectedness and risk spillovers between European Union countries (Antonakakis and Vergos, 2013, Claeys and Vasicek, 2014, Fernández-Rodríguez et al., 2016, Ehrmann and Fratzscher, 2017, Greenwood-Nimmo et al., 2017, BenSaïda, 2018, Chatziantoniou and Gabauer, 2021, Umar et al., 2019). In addition, a plethora of studies have explored connectedness between commodities (e.g., Antonakakis and Kizys, 2015, Yoon et al., 2019, Umar et al., 2019a, Umar et al., 2019b, Balcilar et al., 2021), cryptocurrency markets (e.g., Koutmos, 2018, Antonakakis et al., 2019, Ji et al., 2019, Zieba et al., 2019, Umar et al., 2021, Aharon et al., 2021), and asymmetric connectedness (e.g., Baruník et al., 2016, Baruník et al., 2017, BenSaïda, 2019). We seek to contribute to these studies by examining connectedness during market crises (e.g., Alter and Beyer, 2014, Greenwood-Nimmo et al., 2017), and more specifically, to several studies examining connectedness and contagious effects during COVID-19 (e.g., Adekoya and Oliyide, 2020, Gubareva and Umar, 2020, Adekoya et al., 2020, Bissoondoyal-Bheenick et al., 2020, Corbet et al., 2020, Umar and Gubareva, 2020, Umar and Gubareva, 2021, Umar and Gubareva, 2021a).

The COVID-19 pandemic is an example of the destructive and adverse contagious effects following a crisis in the global capital markets, which also highlights the effects of the increasing connectedness between countries. Unlike former crises, COVID-19 is unique in being a global crisis that began outside the financial markets but has affected every country directly or indirectly. Indeed, the perceptions about the spillover of risk from country to country resemble the spread of the virus itself. Thus, exploring the possible spillover effect during this period is a real opportunity to observe the dynamics of risk transmission and its movement between and within markets.

Our work joins studies dealing with risk spillovers during market crises, particularly the literature dealing with contagious effects within debt markets. The yield curve has attracted the attention of both academics and practitioners. Several studies have shown that it can predict economic activity (e.g., Ang et al., 2006, Fernández-Rodríguez et al., 2015, Harvey, 1988, Wheelock and Wohar, 2009). This demonstration, in turn, has inspired a vast literature focusing on yield curve modeling over decades, with a strong emphasis on government bond yields and their term structures.

However, as opposed to previous studies that concentrated on the overall yield curve, our study explores three components of the yield curve. Adopting this approach provides us with more in-depth information for different time frames that might not emerge when considering the overall yield curve. Specifically, our goal is to investigate the connectedness of the slope (short-term), curvature (medium-term) and level (long-term) of the sovereign yield curve across the G-7 economies during the COVID-19 pandemic. To accomplish this goal, we will map each of the G-7 countries’ yield curves to identify net transmitter or net receiver of risk spillover. Given that the G-7 group contains the most advanced economies in the world, our results should be relevant for practitioners such as investors and fund managers as well as for analysts in terms of asset pricing, portfolio choice, and capital investment allocation for the short and long terms. The connectedness of the yield curve components of Asia pacific and Eurozone countries have been studied by Gabauer et al., 2020, Umar et al., 2021a, respectively. We add to this strand of literature by incorporating the impact of media sentiment on the connectedness of the components of the yield curve during a systemic crisis induced by the covid-19 pandemic. Thus, we add to the literature on the relationship between the yield curve and market sentiment by investigating a unique time period marked by great uncertainty in the global financial markets. Our sample period covers the beginning of 2020 until August 2021. Therefore, we can explore the initial emergence of the pandemic, its peak and the lockdowns, followed by the gradual opening of world economies. The findings may be useful for policy makers and leaders of the G-7 countries by identifying their respective country's role as a transmitter or receiver of global shocks. Such a determination may help them prepare better for possible future shocks.

To achieve our objective, we employ a two-step procedure. The first step uses the Nelson–Siegel model of Diebold and Li (2006) to estimate the level, slope, and curvature of each component of the yield curve for each G-7 economy. The second step uses the TVP-VAR methodology to measure the dynamic connectedness of these three latent variables during the COVID-19 period. We also use data from Ravenpack’s media coverage on COVID-19 (namely, the MCI, the Media Coverage Index).

Our main results can be summarized as follows. First, the Media Coverage Index (MCI) and USA are the leading transmitters of spillover across all the components of the yield curve for the G-7 countries. Second, Japan is a consistent receiver of risk from the G-7 countries with regard to all components of the sovereign term structure. Among the European countries, France seems to be a consistent transmitter, Germany and UK exhibit switching roles of transmitter and receiver on alternative basis. Italy is also a receiver, but only with regard to the level and slope of the yield curve. Third, the relationship is dynamic, as we document several switches in the roles of transmitter and recipient. Our results have important implications for investors, policy makers and future studies on the topic of connectedness and spillover.

The remainder of this research is structured as follows. Section 2 describes the data sources and outlines the methodology. Section 3 discusses the main results, and the final section summarizes and concludes.

2. Data and methodology

The sample covers the entire COVID-19 crisis, from January 1, 2020, to August 31, 2021.1 Thus, our sample period allows us to analyze the dynamics of yield curve connectedness during all the different phases of the pandemic. We use two steps to quantify the interconnectedness of the elements of the yield curve in the G-7 countries. In the first step, we compute the slope, level and curvature using the Diebold and Li (2006) model based on the Nelson–Siegel framework (1987). While there are other alternative estimation approaches, the Nelson and Siegel approach is considered one of the best due to various benefits it provides. The model generates parsimonious estimates for the three components of the yield curve, which improves its predictive power (Diebold and Rudebusch, 2013). Moreover, the model is suited to any kind of yield curve and is closely connected to economic theories.

To measure the three components with the Diebold and Li (2006) framework, we retrieved data from Bloomberg on zero coupon bond yields. We use monthly maturities of 3, 6, 12, 24, 36, 48, 60, 72, 84, 96, 108, 110, 180, 240, and 360 months for each of the seven countries under investigation.

2.1. Estimation of yield curve components

To analyze the dynamism of the yield curve, we follow the Diebold and Li (2006) modification of the (Nelson and Siegel, 1987) model. Diebold and Li (2006) as well as Diebold et al. (2006) claim that the component of the yield curve depend on an VAR(1) structure that enable us to develop a state-space model for model the yield curve components. Diebold and Li’s (2006) state-space equation is as follows:

$$y_t\left(\tau \right)={\left(\begin{array}{ccc}1& \left(\frac{1-e^{-\lambda {\tau }_1}}{\lambda {\tau }_1}\right)& \left(\frac{1-e^{-\lambda {\tau }_1}}{\lambda {\tau }_1}-e^{-\lambda {\tau }_1}\right)\\ 1& \left(\frac{1-e^{-\lambda {\tau }_2}}{\lambda {\tau }_2}\right)& \left(\frac{1-e^{-\lambda {\tau }_2}}{\lambda {\tau }_2}-e^{-\lambda {\tau }_2}\right)\\ \begin{array}{c}⋮\\ 1\end{array}& \begin{array}{c}⋮\\ \left(\frac{1-e^{-\lambda {\tau }_M}}{\lambda {\tau }_M}\right)\end{array}& \begin{array}{c}⋮\\ \left(\frac{1-e^{-\lambda {\tau }_M}}{\lambda {\tau }_M}-e^{-\lambda {\tau }_M}\right)\end{array}\end{array}\right)}^{,}{x}_t+u_t,u_{t}N\left(0,R\right)$$ (1)

$$\stackrel{\sim }{x_t}=\mathrm{\Gamma }{\stackrel{\sim }{x}}_{t-1}+{\eta }_t{\eta }_{t}N\left(0,G\right)$$ (2)

where $y_t\left(\tau \right)$ denotes an M × 1 dimensional vector for yield rates and $u_t$ denotes an M × 1 vector of error terms. $x_t=[L_t,S_t,C_t]$ is a 3x1 dimensional vector containing the latent factors of the yield curve with L_t denoting the level factor, C_t denoting the curvature factor, and S_t denoting the slope factor. In the subsequent transition equation, $\stackrel{\sim }{x_t}=x_t-{\overline{x}}_t$ describes the matrix of demeaned time-varying form parameters, $\mathrm{\Gamma }$ denotes the dynamic interaction between form parameters, and ${\eta }_t$ denotes the error vector with dimension 3 × 1. The assumption is that ${\eta }_t$ and $u_t$ are independent. $G$ is a diagonal matric with dimension M × M. Lastly, $R$ denotes a 3 × 3-dimensional variance–covariance matrix.

2.2. The TVP-VAR connectedness approach

In our second step, we use Antonakakis, Chatziantoniou & Gabauer (2020) TVP-VAR frameework to calculate the dynamic connectedness between the COVID-19 media coverage index and the components of the yield curve. While a more common thread in the connectedness-based literature is Diebold and Yılmaz’s (2012) VAR approach, Antonakakis, Chatziantoniou & Gabauer (2020) suggest an extended connectedness approach to address several potential drawbacks of the VAR approach. More precisely, they allow the variances to be distinguished from Koop and Korobilis’ (2014) forgetting factors as opposed to the stochastic volatility of the Kalman Filter calculation. As a result, in contrast to the random option of a rolling window VAR, they deploy a time-varying parameter. Furthermore, they maintain that their revised method is also important for analyzing dynamic connectedness at lower frequencies and small time-series data. Moreover, their approach avoids the potential loss of observations and is more robust in the case of outliers. Given the short period of time series available for our analysis, this approach is particularly suitable for our analyzing the dynamic connectedness.

To study the spillover mechanism in a dynamic manner, we follow Antonakakis, Chatziantoniou & Gabauer (2020). Based on the Bayesian Information Criterion (BIC), we use a TVP-VAR(1) with dynamic time-varying volatility, as described by the following equations:

$$Y_t={{\beta }_{t}Y}_{t-1}+{\epsilon }_t,{\epsilon }_{t}N[0,S_t]$$ (3)

$${\beta }_t={\beta }_{t-1}+v_t,v_{t}N[0,R_t]$$ (4)

$$Y_t={A_t\epsilon }_{t-1}+{\epsilon }_t$$ (5)

where the N × 1 vectors are $Y_t$,${\epsilon }_t$ and $v_t$, and the N × N matrices are $A_t$, $S_t$, ${\beta }_t$ and $R_t$. Eq. (3) is the description of the scheme by Wold. The time-varying vector moving average’s (VMA) coefficients are the center of the connectedness index developed by Diebold and Yilmaz (2012). It incorporates the generalized impulse response function (GIRF) and the generalized forecast error variance decomposition (GFEVD), both of which were established by Koop et al., 1996, Pesaran and Shin, 1998.

In forecasting variable i, which is attributed to shocks on variable j, our attention is on the h-step error variance. The algorithm’s equation is:

$${\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g(\mathrm{h})=\frac{{\sum }_{t-1}^{h-1}{\mathrm{\psi }}_{\mathit{ij},t}^{2,g}}{{\sum }_{i=1}^{\mathrm{N}}{\sum }_{t-1}^{h-1}{\mathrm{\psi }}_{\mathit{ij},t}^{2,g}}$$ (6)

where ${\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g(\mathrm{h})$ is the h-step ahead of GFEVD, ${\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g\left(\mathrm{h}\right)={\mathrm{S}}_{\mathit{ij},t}^{-\frac{1}{2}}{A}_{h,t}{\sum }_t{\epsilon }_{\mathit{ij},t}$, ${\sum }_t-$ the covariance matrix for the error ${\epsilon }_{\mathit{ij},t}$ and ${\sum }_{j=1}^N{\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g\left(\mathrm{h}\right)=1,$ ${\sum }_{i,j=1}^N{\stackrel{\sim }{\phi }}_{\mathit{ij},t}^N\left(\mathrm{h}\right)=\mathrm{N}.$ We construct the total connectedness index (TCI) reflecting the network’s interconnectedness, computed as:

$$C_t^g\left(\mathrm{h}\right)=\frac{{\sum }_{i,j=1,i\ne j}^N{\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g\left(h\right)}{{\sum }_{j=1}^N{\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g\left(\mathrm{h}\right)}\times 100.$$ (7)

First, we try to figure out variable i’s spillovers to all other j, reflecting the complete directional relationship to others, and identified as:

$$C_{i\to j,t}^g\left(\mathrm{h}\right)=\frac{{\sum }_{j=1,i\ne j}^N{\stackrel{\sim }{\phi }}_{\mathit{ji},t}^g\left(h\right)}{{\sum }_{j=1}^N{\stackrel{\sim }{\phi }}_{\mathit{ji},t}^g\left(\mathrm{h}\right)}\times 100.$$ (8)

Second, the spillovers of all variables j are calculated to variable i, referred to as the total directional connectedness  from  others and described as:

$$C_{i\leftarrow j,t}^g\left(\mathrm{h}\right)=\frac{{\sum }_{j=1,i\ne j}^N{\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g\left(h\right)}{{\sum }_{i=1}^N{\stackrel{\sim }{\phi }}_{\mathit{ij},t}^g\left(\mathrm{h}\right)}\times 100.$$ (9)

Third, to derive the net total directional connectedness, we quantify the differences between the total directional connectedness  to  others and the total directional connectedness  from  others:

$$C_{i,t}^g\left(\mathrm{h}\right)=C_{i\to j,t}^g\left(\mathrm{h}\right)-C_{i\leftarrow j,t}^g\left(\mathrm{h}\right)$$ (10)

Subjectively, the net total directional connectedness demonstrates whether vector i propels the network ($C_{i,t}^g\left(\mathrm{h}\right)$ > 0) or is instead propelled by the network ($C_{i,t}^g\left(\mathrm{h}\right)$ < 0).

3. Empirical findings

The descriptive statistics of the term structure components of the yield curve --level, slope, and curvature--for the G-7 countries are reported in Table 1 . Italy and Canada have the highest absolute mean values for level and slope component, whereas Italy and France have the highest absolute mean value for the Curvature components. In contrast, Japan has the lowest mean value for all the yield curve components. We notice excess kurtosis, suggesting that these series distributions are leptokurtic relative to normal distributions. The Jarque-Bera normality test rejects all series normality at 1% statistical significance. We report the statistics and p-values for Augmented Dickey Fuller (ADF test) unit root test and the Elliott, Rothenberg & Stock unit root test and confirm the stationary of the first difference of all the yield curve components. Fig. 1 shows the graphical depiction of the series employed for this study.

Table 1.

Descriptive Statistics of Term Structure Components.

Panel A: Level
	Canada	France	Germany	Italy	Japan	UK	USA	MCI
Mean	2.257	1.293	0.581	2.977	0.306	2.030	1.848	67.271
Median	2.121	1.321	0.521	2.944	0.425	1.982	1.761	72.410
Maximum	3.233	2.083	1.438	3.733	2.199	3.032	3.190	82.950
Minimum	1.185	0.663	−0.044	2.467	−3.725	0.888	0.754	0.090
Std. Dev.	0.376	0.361	0.365	0.303	0.891	0.414	0.665	15.541
Skewness	0.814	0.113	0.568	0.448	−0.826	0.234	0.448	−2.811
Kurtosis	2.941	1.919	2.295	2.286	4.671	2.569	1.984	11.053
ADF test	−7.791	−8.210	−9.039	−7.741	−9.887	−9.342	−9.279	−7.791
ERS test	−9.985	−8.966	−9.204	−7.517	−9.609	−9.294	−9.314	−9.985
ADF sig	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
ERS sig	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
Jarque-Bera	48.018	22.033	32.350	23.759	7.330	99.835	33.195	48.018
Probability	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
#Obs	434	434	434	434	434	434	434	434

Panel B: Slope
	Canada	France	Germany	Italy	Japan	UK	USA
Mean	−1.935	−1.905	−1.228	−3.359	−0.471	−2.016	−1.477
Median	−1.924	−1.899	−1.121	−3.312	−0.598	−1.968	−1.278
Maximum	−0.226	−1.346	−0.740	−2.072	3.461	−0.638	0.042
Minimum	−3.190	−2.729	−2.061	−4.098	−2.507	−3.207	−3.000
Std. Dev.	0.712	0.364	0.348	0.312	0.882	0.513	0.824
Skewness	0.748	−0.393	−0.746	0.089	0.786	−0.296	−0.284
Kurtosis	3.426	2.088	2.375	4.387	4.562	2.428	1.969
ADF test	−6.945	−7.559	−8.901	−7.965	−9.779	−9.515	−8.781
ERS test	−9.451	−8.014	−9.433	−7.324	−9.584	−9.591	−9.244
ADF sig	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
ERS sig	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
Jarque-Bera	43.737	26.235	47.297	35.342	88.840	12.266	25.049
Probability	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
#Obs	434	434	434	434	434	434	434
Panel C: Curvature
	Canada	France	Germany	Italy	Japan	UK	USA
Mean	−1.170	−2.361	−1.918	−2.275	−0.868	−2.025	−2.393
Median	−1.601	−2.310	−1.846	−2.423	−0.964	−2.122	−2.094
Maximum	0.901	−1.079	−0.856	0.645	1.836	0.053	−0.494
Minimum	−2.670	−3.035	−2.772	−3.308	−2.096	−4.191	−4.390
Std. Dev.	0.860	0.321	0.428	0.743	0.576	0.794	0.988
Skewness	0.749	0.383	−0.282	1.233	0.887	0.634	−0.489
Kurtosis	2.298	3.544	2.232	4.585	5.219	2.945	2.071
ADF test	−6.803	−7.654	−7.716	−9.814	−9.716	−9.354	−7.881
ERS test	−9.476	−7.906	−8.201	−8.984	−9.533	−9.807	−10.341
ADF sig	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
ERS sig	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
Jarque-Bera	49.590	15.933	16.450	155.742	146.286	29.243	32.985
Probability	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***	0.000***
#Obs	434	434	434	434	434	434	434

Notes: The table presents a summary of the statistics for the key variables of the study, namely, the level (Panel A), slope (Panel B) and curvature (Panel C) of the sovereign term structure. The descriptive statistics reported here are as follows: Mean, Median, Maximum, Minimum, Skewness, kurtosis (Kurt). The p-values of Augmented Dickey Fuller test (ADF test) and Elliott, Rothenberg & Stock Unit Root Test (ERS test) are reported. We report The Jarque-Bera test, the Augmented Dickey Fuller test (ADF test) and Elliott, Rothenberg & Stock Unit Root test (ERS test) statistics and their corresponding p-values. The ***, **, and * denotes statistical significance at 1%, 5%, and 10% levels, respectively. and finally, the total number of observations for each variable (#obs).

Fig. 1.

Time series graphs of the components of yield curve and MCI.

We fit a TVP-VAR (1) model and compute the time-varying connectedness of the system.2 Fig. 2 illustrates the total connectedness of the G-7 system for the sovereign yield curves’ components (slope, level, and curvature) during the COVID-19 period from January 2020 to August 2021. As the figure indicates, there is an obvious dynamic connectedness in each of the yield curves’ components across the G-7 countries, supporting our approach of regarding the relationship as dynamic rather than static. Fig. 2 shows that the slope and yield components of the yield curve exhibit the highest level of connectedness. Furthermore, we see that the connectedness of all the components peaks at the end of first quarter of 2020, i.e., March 2020 that coincides with the period of global lockdowns and peak of the pandemic crisis. Subsequently we notice that at the end of the second quarter of 2020, when the global effect of pandemic started to subside and there was gradual opening of the world economies, the connectedness tends to decrease in value. Subsequently with the second and third waves, we see another peak in connectedness. Notably, the connectedness reaches its highest levels during the first three months, when market uncertainty was at its highest level, along with the rapid escalation of the disease. This result is consistent with Adekoya and Oliyide (2020) who find similar results regarding the connectedness between commodities and capital markets during COVID-19. They argue that this relationship is evident in the initial months when the pandemic was spreading aggressively throughout the world and when newspapers and media coverage gathered momentum. As we will discuss later, all our results imply that the attention paid to COVID-19, reflected in the media coverage index (MCI), is a driver and main transmitter of shocks to the economic system.

Fig. 2.

Total Connectedness Measures of the Yield Curves’ Components.

Next, we turn to the mapping of the G-7 countries in terms of being a transmitter or receiver of such shocks. Fig. 3 illustrates the net directional connectedness (top panel) between the slope of the yield curve and the MCI the spillover of each variable to all other variables (2nd panel) and the spillover from all the variables to a specific variable (3rd panel). We focus mainly on the top panel depicting the net directional connectedness as it identifies the net transmitter and net receiver of shocks and as such is obtained by taking a difference of the 2nd and the 3rd panel. As the figure indicates, the MCI is a net transmitter of shock. In particular, we notice that during the first wave (first quarter of 2020), the MCI is a major transmitter of spillover. Similarly, USA and France are predominantly net transmitters, except for short periods when around the end of first quarter when they switched to being net receivers. As mentioned above, this is the peak of first wave of crisis induced by pandemic and almost the entire world came to a halt at that point in time. Germany and UK switch between transmitter and receivers of spillover during different periods of the pandemic. Interestingly, UK and Germany seem to exhibit opposite role as transmitter and receiver during most part of the crisis. Italy, Japan and Canada are predominantly net receivers throughout sample period.

Fig. 3.

Spillovers between the slope of the yield curve and the MCI.

A similar picture emerges when we consider the level (Figs. 4 and 5 ) component of the yield curve. We notice that MCI, France, USA are predominantly net transmitter of spillovers. In addition, we notice that Germany is a predominant transmitter, whereas UK along with Italy, Canada and Japan is a predominant receiver for spillover for the level component. As before, we notice that the connectedness level changes with the phases of the pandemic. Lastly, we discuss the connectedness of the curvature component of the yield curve and MCI depicted in Fig. 6 . Like the previous cases, we notice that MCI is a consistent transmitter of shocks with a distinct peak at the end of first quarter of 2020. USA seem to be a predominant transmitter of spillover during the entire sample period. However, we notice that other countries switch between transmitter and receiver of spillover during the sample period except for Japan which is predominantly a net receiver throughout the period.

Fig. 5.

Spillovers between the curvature of the yield curve and the MCI.

Fig. 4.

Spillovers between the level of the yield curve and the MCI.

Fig. 6.

Portfolio Weights and Hedge Ratios.

The above results show the evolution of the connectedness over the sample period. In Table 2 we report the average connectedness results for the three components of the yield curve and MCI. The top, middle and bottom panels of Table 1 report the results for level, slope and curvature component of yield curve, respectively. We notice that on average MCI, USA and France are net transmitters of spillovers as evident by positive values in the last row of each panel for each of these countries. On the contrary, Japan is a consistent recipient of spillover for all the three components. Germany is net transmitter for the level and slope component, while a net receiver for the curvature component. On the contrary, Italy and Canada are net recipients of spillovers for the level and slope component but net transmitters of spillovers for the curvature. Lastly, UK is a net transmitter of shocks spillover for slope component but is a net receiver for the other two components.

Table 2.

Average connectedness of yield curve components and MCI.

Panel A: Average connectedness of Level Component and MCI
	USA	Germany	Japan	France	Italy	Canada	UK	MCI	FROM
USA	55.90	13.10	2.20	12.40	3.30	17.20	7.60	2.50	58.40
Germany	12.30	47.20	3.10	32.30	4.80	4.50	8.80	1.40	67.10
Japan	7.90	8.20	77.60	8.50	3.50	1.30	5.20	2.10	36.70
France	8.50	32.00	3.20	48.30	12.10	3.40	6.10	0.70	66.00
Italy	4.00	7.00	2.20	19.80	75.00	1.30	3.40	1.50	39.30
Canada	24.20	3.90	0.70	4.30	1.90	74.30	3.40	1.50	39.90
UK	9.40	13.20	3.30	10.40	2.90	3.90	69.50	1.60	44.80
MCI	3.30	0.50	0.80	0.80	0.40	0.50	1.10	106.80	7.50
Contribution TO others	69.70	77.90	15.60	88.40	29.00	32.00	35.70	11.30	TCI
NET directional connectedness	11.30	10.80	−21.00	22.40	−10.30	−7.90	−9.10	3.90	38.32
Panel B: Average connectedness of Slope Component and MCI
	USA	Germany	Japan	France	Italy	Canada	UK	MCI	FROM
USA	65.40	11.30	2.20	10.10	4.60	10.30	9.80	0.60	48.90
Germany	10.30	51.50	2.30	29.30	7.70	3.30	9.20	0.60	62.80
Japan	8.20	8.00	77.20	13.10	1.80	1.30	4.50	0.30	37.10
France	7.10	29.70	3.70	55.60	6.30	3.10	6.80	1.90	58.70
Italy	8.90	6.30	1.10	6.20	76.50	1.60	12.60	1.10	37.80
Canada	15.20	3.90	0.70	3.60	2.30	85.10	3.30	0.20	29.20
UK	10.40	11.20	2.90	8.70	6.80	3.80	70.10	0.30	44.10
MCI	0.10	0.20	0.20	0.70	0.20	0.00	0.10	112.90	1.40
Contribution TO others	60.20	70.60	13.20	71.80	29.70	23.40	46.20	4.90	TCI
NET directional connectedness	11.30	7.80	−24.00	13.10	−8.10	−5.70	2.00	3.50	34.99
Panel C: Average connectedness of Curvature Component and MCI
	USA	Germany	Japan	France	Italy	Canada	UK	MCI	FROM
USA	95.70	2.70	1.00	2.90	3.60	7.10	0.90	0.40	18.60
Germany	3.80	92.00	1.20	7.70	3.80	3.20	1.80	0.70	22.20
Japan	2.10	1.00	86.90	12.90	2.40	7.70	1.10	0.30	27.40
France	2.50	4.70	4.60	83.10	8.70	6.00	2.00	2.70	31.20
Italy	3.70	2.50	4.20	6.60	94.20	2.00	0.90	0.30	20.10
Canada	8.00	0.50	2.30	8.40	3.60	89.50	1.00	1.00	24.80
UK	2.10	1.90	1.20	1.80	1.60	2.90	102.60	0.20	11.70
MCI	0.20	0.10	0.20	0.50	0.00	0.20	0.20	113.00	1.20
Contribution TO others	22.30	13.20	14.70	40.80	23.70	29.00	7.90	5.70	TCI
NET directional connectedness	3.70	−9.00	−12.80	9.70	3.60	4.20	−3.80	4.40	17.21

Notes: This table reports the average connectedness of the each of the yield curve component for each country and media coverage Index (MCI). The second last row of each panel shows the contribution of each variable to all other variables in the system, the last row shows the net directional connectedness depicting whether a variable is a net transmitter (positive values) or net receiver (negative values). The last column shows the contribution from all other variable to a given variable in the system. Lastly, TCI denotes total connectedness index.

4. Discussion and policy implications

Our findings that connectedness peaks with phases of the pandemic underscores the lack of diversification opportunities during periods of heightened uncertainty. The role of MCI as a net transmitter adds another dimension to this analysis. Our findings are consistent with those of Youssef et al. (2021), who examine the spillover effects of the EPU index on indices of eight major stock markets. While the existing literature attributed to the heightened and interconnected nature of global stock markets during times of market distress, as opined by Zhang et al., 2020, Zhang et al., 2020 and Cepoi (2020), our findings extend this literature in the fixed income markets domain by specifically analyzing the impact of the pandemic across the entire yield curve. Thus, we are able to isolate the impact of the pandemic for various investment horizons ranging from short-term, medium term to long term.

Apart from MCI, US appears to be the main transmitter of spillover to all other markets. Our finding again underscores the central role of US in the global financial architecture and thus have a central role as a transmitter of risk spillovers (Bissoondoyal-Bheenick et al., 2020, Zhang et al., 2020, Zhang et al., 2020). Although the Pandemic is different from the global financial crisis (GFC) which originated from USA, yet the role of US as a main transmitter of spillover is consistent. The US government’s3 initial slow response in managing COVID-19 may have created uncertainty, and it failed to take a leading role when the crisis erupted. The role of Canada as main recipient of shocks is consistent with Su (2020) who suggests that the geo-politico-economic proximity of Canada to the US benefits Canada which derives its market decisions from the US. Therefore, Canada appears to be a main recipient of spillover for short and long term.

Among the European countries, we notice that France and Germany are predominant transmitters, whereas Italy is the predominant receiver of spillover for short and long term. This maybe attributed to the central role of these two economics in eurozone (Umar et al., 2019). However, Italy is a transmitter of spillover over the medium term. The heightened level of connectedness also supports Riaz et al. (2019) finding that the impact of news announcement on European financial markets is more pronounced during periods of heightened uncertainty. Among the European countries an interesting pattern is noticed between Germany and UK, which appear to switch between transmitter and receiver role on alternative basis. These results can be useful for devising hedging strategies. Lastly, the role of Japan is a net recipient of spillover maybe attributed to the diminishing role Japan's in global financial markets relative to other emerging economies and the economic stagnation, dubbed “Japan's lost decades,” which has been exacerbated by a series of unfortunate events, such as the earthquake (Su, 2020).

An important insight for discussion is how our results may, for example, yield practical guidelines for the benefit of investors or alternatively for policy makers at the state level. In practical terms, fixed income investors which are involved in international financial investments across different countries (such as the G-7), may wish to keep their eyes on the dynamic connectedness between the public sentiment as reflected by Media Coverage Index (MCI). Tracking the behavior of connectedness across time may be a signal tool for preparing and balancing their exposure to market risk. For example, under times of increased connectedness or alternatively by observing higher values of the MCI index, investors may wish to shift funds or decrease their investments in sovereign debt investments in countries which are documented as main absorbers of risk transmission. Alternatively, under such periods, investors can consider buying more fixed income options and derivatives to reduce their potential exposure to market risk, which seems to affect not only the equity market, but the fixed income as well. These guidelines are valid for investors holding fixed income investments under different horizons explored here, that is, the short (slope), medium (curvature) and long term (level). Therefore, investors can better shape their hedging strategies by a preliminary use of the information about the connectedness across time.

Similarly, policy makers at the state level, may wish to also track their country sovereign debt connectedness with the MCI, to better manage their own disaster plans. In their efforts of stabilizing the upcoming fluctuations in the fixed income market, they can for example monitor the exposure level to contagious risk in advance, and accordingly initiate in preliminary steps to enhance market stability such as providing more information. Such step is an important measure under elevated times of uncertainty but is mostly important in the timing of providing such transparency. Our study suggest that state policy makers can better tune up their decisions during times of crisis and around periods characterized by high volatility.”.

4.1. Hedging and portfolio implications

We explore the hedging and portfolio effects for an investor who is interested in G-7 government bonds. We employ the FTSE government bond index4 for each of the G-7 countries and quantify optimal hedge ratios for an investor who takes a long position in one fixed income index (b1) and a short position in the other fixed income index (b2). Following, (Kroner and Sultan, 1993, Kroner and Ng, 1998), we construct hedge ratios as presented in Eq. (11) below:

$${\beta }_{b1b2,t}=h_{b1b2,t}/h_{b2b2,t}$$ (11)

where $h_{b1b2,t}$ denote the conditional covariance of the two fixed income bond indices and $h_{b2b2,t}$ is the conditional variance of the shorted fixed income bond index. We employ a DCC model to compute the conditional variance and covariance.

Similarly, we can quantify the portfolio weights as

$${\omega }_{b1b2,t}=\frac{h_{b2b2,t}-h_{b1j,t}}{h_{b1b1,t}-2h_{b1b2,t}+h_{b2b2,t}}$$ (12)

Such that

$${\omega }_{b1b2,t}=\left\{\begin{array}{c}0,if{\omega }_{b1b2,t}<1\\ {\omega }_{b1b2,t},if{0\le \omega }_{b1b2,t}\le 1\\ 1,if{\omega }_{b1b2,t}>1\end{array}\right)$$ (13)

${\omega }_{b1b2,t}$ is the weight of the long bond in one dollar portfolio of the two bonds indices (for each pair). Thus, ${1-\omega }_{b1b2,t}$ is the weight of the shorted bond index.

Table 3 reports the results of portfolio weights (first panel) and hedge ratios (second panel) for each pair of the G-7 government bond indices (country1/ country2 means a long position in first country’s bonds and a short position in second country’s bond). We report the average, standard deviation and the 5th and 95th quantiles of the portfolio weights and hedge ratios. Overall, our results corroborate our connectedness analysis. We notice that US bonds exhibit higher hedging attributes for other G-7 economies. Similarly, we notice that with the European economies, Germany and France exhibit higher hedging attributes.

Table 3.

Portfolio Weights and Hedging ratios.

	Portfolio weights				Hedge ratios
	Mean	SD	5%	95%	Mean	SD	5%	95%
Canada/France	0.49	0.15	0.25	0.73	0.26	0.12	0.09	0.47
Canada/Germany	0.53	0.15	0.27	0.75	0.47	0.13	0.26	0.66
Canada/Italy	0.58	0.16	0.28	0.82	0.47	0.12	0.27	0.68
Canada/Japan	0.68	0.19	0.31	0.96	0.46	0.11	0.32	0.66
Canada/UK	0.86	0.12	0.62	1.00	0.30	0.08	0.20	0.46
France/Canada	0.51	0.15	0.27	0.75	0.25	0.10	0.08	0.39
France/Germany	0.54	0.16	0.24	0.77	0.47	0.15	0.22	0.70
France/Italy	0.58	0.17	0.28	0.85	0.43	0.14	0.19	0.64
France/Japan	0.63	0.15	0.36	0.87	0.29	0.12	0.09	0.47
France/UK	0.85	0.13	0.62	1.00	0.28	0.08	0.13	0.41
France/USA	0.17	0.10	0.03	0.36	0.59	0.20	0.31	0.93
Germany/Canada	0.47	0.15	0.25	0.73	0.50	0.11	0.30	0.65
Germany/France	0.46	0.16	0.23	0.76	0.50	0.13	0.26	0.68
Germany/Italy	0.84	0.27	0.22	1.00	0.93	0.06	0.78	0.99
Germany/Japan	0.77	0.17	0.52	1.00	0.67	0.13	0.41	0.82
Germany/UK	0.93	0.09	0.75	1.00	0.40	0.07	0.28	0.52
Germany/USA	0.18	0.10	0.05	0.36	0.49	0.18	0.17	0.74
Italy/Canada	0.42	0.16	0.18	0.72	0.54	0.12	0.33	0.75
Italy/France	0.42	0.17	0.15	0.72	0.51	0.13	0.29	0.72
Italy/Germany	0.16	0.27	0.00	0.78	1.01	0.05	0.94	1.11
Italy/Japan	0.81	0.18	0.52	1.00	0.78	0.09	0.62	0.88
Italy/UK	0.89	0.11	0.67	1.00	0.42	0.08	0.28	0.58
Italy/USA	0.17	0.10	0.06	0.37	0.48	0.18	0.17	0.74
Japan/Canada	0.32	0.19	0.04	0.69	0.66	0.15	0.43	0.93
Japan/France	0.37	0.15	0.13	0.64	0.42	0.13	0.17	0.59
Japan/Germany	0.23	0.17	0.00	0.48	0.89	0.11	0.66	1.06
Japan/Italy	0.19	0.18	0.00	0.48	0.96	0.09	0.83	1.10
Japan/UK	0.74	0.16	0.44	0.95	0.37	0.10	0.24	0.56
Japan/USA	0.18	0.09	0.08	0.35	0.23	0.20	−0.12	0.55
UK/Canada	0.14	0.12	0.00	0.38	0.77	0.2	0.48	1.12
UK/France	0.15	0.13	0.00	0.38	0.71	0.2	0.36	1.03
UK/Germany	0.07	0.09	0.00	0.25	0.95	0.18	0.67	1.27
UK/Italy	0.11	0.11	0.00	0.33	0.90	0.17	0.65	1.20
UK/Japan	0.26	0.16	0.05	0.56	0.65	0.15	0.42	0.89
UK/USA	0.03	0.04	0.00	0.12	0.91	0.29	0.42	1.38
USA/Canada	0.78	0.10	0.62	0.92	0.13	0.11	−0.01	0.32
USA/France	0.83	0.10	0.64	0.97	0.22	0.09	0.12	0.36
USA/Germany	0.82	0.10	0.64	0.95	0.17	0.09	0.06	0.29
USA/Italy	0.83	0.10	0.63	0.94	0.15	0.09	0.06	0.28
USA/Japan	0.82	0.09	0.65	0.92	0.06	0.06	−0.03	0.16
USA/UK	0.97	0.04	0.88	1.00	0.13	0.05	0.06	0.22

Notes: This table shows the portfolio weights and the hedge ratios for an investor who takes a long position in the first country's government bond and a short position in the second countries government bonds. (First country/second country).

5. Conclusions

Our investigation into the dynamics of the connectedness between the components of the G-7 countries’ sovereign yield curve and the impact of the coverage of the COVID-19 pandemic in the media, expressed in Ravenpack’s MCI, reveals that the latter is a significant transmitter of shocks. In addition, the shocks reflected in the index affect both the G-7 countries and the elements of their sovereign yield curves differently. Based on our analysis of the changes, we conclude that the level of connectedness reached its peak during the initial period of the pandemic’s outbreak (January - March 2020). From this point forward, the connectedness remained strong, supporting the contention that the benefits of diversification decline during times of market turmoil and crises.

Our conclusions have major consequences for academics and practitioners. From the academic point of view, despite the plethora of studies on yield curves (e.g., Harvey, 1988, Haubrich and Dombrosky, 1996, Ang et al., 2006, Erdogan et al., 2015), we provide a more nuanced look at the components of the yield curve, not just its overall movement, as an indicator of economic growth. It is helpful to document how we can use changes in the components of the yield curve to indicate changes in economic growth over time.

From the more pragmatic and practical point of view, our findings may help policy makers in several ways. First, our results highlight the importance of monitoring the overall connectedness of various factors as well as the net effects they have on their countries. By doing so, they can be better prepared for upcoming shocks, especially in extreme market conditions, and take measures to support the economy. Experience has shown that the timing of such measures is no less important than the types of measures taken. Thus, for example, policy makers can minimize panic and uncertainty by launching public information campaigns about the measures they are taking. Such transparency will hopefully lead to more stable markets. The results can also help individual and institutional investors in terms of asset allocation and risk management. They can use our findings to rebalance or hedge their investment strategies. A potential extension of this work can be by employing an alternative VAR-X model, with media sentiment X as an exogenous variable5 . Similarly, an empirical extension encompassing emerging economies would present interesting comparative results.

CRediT authorship contribution statement

David Y. Aharon: Conceptualization, Investigation, Data curation, Methodology, Resources, Formal analysis, Writing – original draft, Writing – review & editing. Zaghum Umar: Conceptualization, Investigation, Data curation, Methodology, Resources, Formal analysis, Writing – original draft, Writing – review & editing. Mukhriz Izraf Azman Aziz: Conceptualization, Investigation, Data curation, Methodology, Resources, Formal analysis, Writing – original draft, Writing – review & editing. Xuan vinh Vo: Investigation, Formal analysis, Methodology, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

Data will be made available on request.