The Impact of the Ukrainian Crisis and Global Shocks on Stock Market Connectedness in Sub-Saharan Africa: Evidence from Diebold and Yilmaz (2012) Spillover Analysis

The study of financial connectedness and cross-market linkages has emerged as an important topic in financial lexicon. It has been widely demonstrated in empirical work that financial crises are not the only channels through which exogenous shocks are transmitted to financial systems in the globe, however the political, economic, and health catastrophes also create transmission mechanisms and threatening economies and financial systems.

The examination of financial connectedness among financial markets through return and risk spillovers provides a significant overview of information transmission between stock markets, where a shock or an abrupt change in one stock market might affect the price and risk levels in another markets [1]. Over the last two decades, academics, research scholars, and investors have had heated discussions over the financial connectedness among stock markets across the globe. Indeed, this issue has provided an impetus for much empirical analysis that aimed at investigating the degree of interdependence between regional and international stock markets.

The financial theory suggests that the interdependence and connectedness among financial markets can enhance efficiency of capital allocation and increase the potential benefits from portfolio diversification [2]. Nonetheless, it has been argued that the increased level of co-movements among stock markets due to technological innovation and improvement posed risks and shocks spillover across emerging and developed stock markets alike. Consequently, the issue of financial contagion and volatility spillover has become a ubiquitous topic in the financial lexicon especially during crises that befell due to financial and geopolitical instability. It is imperative for policy makers in developing economies nowadays to maintain a minimum level of traction of their financial systems during the crises periods, and this will most likely necessitate a constant vigil on the degree of financial connectedness of and expected volatility spillovers to domestic stock markets.

The presence of a dominant role of certain markets in creating and transmitting spillover effects diachronically has gained increased importance during the crises periods since stock markets contribute to crises transmission through financial contagion and volatility spillovers. The ongoing Ukrainian crisis is blanketing the world economic outlook and financial forecasts since February 2022. The repercussions and consequences of this crisis are expected to have a perilous aggregate impact on financial markets co-movements across the world.

In the last fifty years, the developing economies in the African continent have received tremendous amounts of foreign development aids (amounted to almost 1 trillion dollar) [3]. Although foreign aid flows to Africa have failed to achieve full poverty alleviation and high level of economic prosperity. However, it has been noticed that the development aid utilization accompanied with full-fledged reform policies in some African countries has exhibited significant positive effect on the financial landscape and FDI flows in African region, where the adoption of new technology and automated trading systems enabled for noticeable improvement in the stock market platforms and investment toolkits [4]. Moreover, a sharp increase in the number of operating stock markets can be witnessed in Africa as the number of operating stock exchanges rose from eight in 1989 to 23 in 2007, and then to 38 stock markets by the end of 2020. The stock markets in the majority of African countries have been characterized by weak integration with international counterparts. The partial segmentation of the African stock exchanges from major international markets has instigated further investigation on potential role of African stocks in enhancing the risk/return tradeoffs of international investment portfolios especially during the periods of financial turmoil and political unrests. Henceforth, the research in this paper aims to answer the following research questions: First: what are the main alterations and changes, if any, in dynamic integration among main stock markets in the African region and developed markets in the wake of the Ukrainian crisis? Second: did this crisis lead to more significant effects on connectedness and spillover among these markets as compared to the crises of COVID-19 and global financial crisis in 2008? Third: do political, economic, and health crises lead to structural change in stock markets behavior in the African continent?

This paper aims to investigate the interconnectivity between African stock markets and major stock markets in both emerging and developed countries. Specifically, it focuses on three key aspects. Firstly, the study examines the relationships between African stock markets and global markets during the ongoing Ukrainian crisis. Secondly, it compares the impact of significant financial, health, and political crises in the new millennium, namely the Global Financial Crisis in 2008, the COVID-19 pandemic in 2020, and the Ukrainian crisis, on the transmission of returns and volatility among African and developed stock market indices. Lastly, the research conducts a comparative analysis to explore the role of domestic and international events in initiating structural shocks and sudden changes in stock markets across the African continent.

For these aims, research in this paper covers time windows of three main crises, 2008 financial crisis (GFC), the outbreak of pandemic of COVID-19 in early 2020, and the ongoing crisis in Ukraine. The present study employs Bai-Perron [5] techniques to identify the dates and times at which major changes in the stock market performance of African nations have occurred during times of crisis. This paper proposes the multiple structural breaks test that has been applied to major stock markets in Africa for the last fifteen years.

Diebold and Yilmaz [6] spillover index is also utilized to measure the financial connectedness and spillover of return and risk from developed market indices to their counterparts in African continent. This paper also utilizes the impulse response analysis to capture the dynamic response patterns of African stock indices to shocks in global markets.

The contribution of this study is mainly derived from its prompt response to the need for comprehensive investigation for safe and resilient investment conduits through which investors could diversify their investment portfolios in the wake of the ongoing Ukrainian crisis. In addition, this paper is one of the earliest, if not the first, attempts that assess the potential role of African stock markets in international portfolio diversification during the ongoing Ukrainian crisis. The remainder of the paper is organized as follows: part two demonstrates literature review. Part 3 encompasses data and method of study. Part 4 illustrates the outcomes of this study, and part 6 concludes the paper.

This study analysis the daily stock prices for a sample of six stock markets in Africa, in addition to a group of developed countries. The sample of African markets including Egypt, South Africa, Namibia, Morocco, Nigeria, Cote D’ Ivoire. The stock markets in Egypt, Morocco, and South Africa, respectively, are the oldest stock exchanges in the region, while the markets in Namibia, Nigeria, and Cote D’ Ivoire are in top ten list of performers in last few years. The Asian Composite is selected to proxy for developing markets, whereas the Euronext 100 and S&P indices represent advanced markets.

This study assesses the implications of the Global Financial Crisis of 2008, the COVID-19 pandemic of 2020, and the Ukrainian crisis of 2022 on African markets using the MSCI indices from the DataStream. We apply daily data to gain an understanding of the short-term alterations in the markets. The Bai-Perron [5] test for multiple structural breaks is employed in our data range from July 2008 to July 2022. The Diebold and Yilmaz [6] spillover index is utilized to measure the interconnectedness of the markets for the three sub-periods, and an Impulse Response Analysis is administered to assess the response of African markets to shocks in other markets.

3.1 Testing for multiple structural breaks: Bai-Perron (2003)

In this paper, we employ Bai-Perron [5] to investigate the existence of multiple structural breaks in African and developed stock market indices. Despite the fact that this test does not allow for a clear depiction of the most significant structural break in a time series, which can be argued as one of its main limitations, nevertheless this model is recognized as more desirable than other tests for structural changes as it allows for determination of number structural break points as well as the simultaneous estimation of multiple break dates [47]. In this paper, the utilization of the Bai-Perron test is expected depict if political, economic, and health crises can equally lead to structural change in stock markets behavior in the African continent?

In accordance with Pai-Perron, the multi-linear regression model with m breaks and (m+1) regimes is presented as follows:

$x_t=y_t^{\prime} \beta+z_t^{\prime} \delta_j+u_t\left(t=T_{j-1}+1, \ldots ., T_j\right)$

where, $j=1, \ldots, m+1$, $T_0=0$ and $T_{m+1}=T$. x_t is dependent variable at time t. $y_t(p \times 1)$ and $z_t(q \times 1)$ are vectors of covariance. $\beta$ and $\delta_j(j=1, \ldots, m+1)$ represent the corresponding vectors of coefficients. $\left(T_1, \ldots, T_m\right)$ are unknown break points, and u_t is disturbance at time t. Using the matrix form of multiple linear regression described above, we can obtain the following:

$X=Y \beta+\bar{Z} \delta_j+U$

where, $X=\left(x_1, \ldots \ldots, x_T\right)^{\prime}$, $Y=\left(y_1, \ldots . ., y\right)^{\prime}$, $U=\left(u_1, \ldots \ldots, u_T\right)^{\prime}$, $\delta=\left(\delta_1^{\prime}, \delta_2^{\prime}, \ldots, \delta_{m+1}^{\prime}\right)^{\prime}$, and $\bar{Z}$ is matrix which diagonally partition Z at $\left(T_1, \ldots, T_m\right)$, i.e., $\bar{Z}$ = diag $\left(Z_1, \ldots, Z_{m+1}\right)$. The $\delta^0=\left(\delta_1^{0 \prime}, \delta_2^{0 \prime}, \ldots, \delta_{m+1}^{0 \prime}\right)^{\prime}$ and $\left(T_1, \ldots, T_m\right)$ used to denote the true values of $\delta$ and the true break points.

A structural stability test against a fixed number of breaks (examining whether or not there will be no structural breaks), a structural stability test against an unknown number of breaks, and a structural stability test against a set of breaks (which is test of $\ell$ versus $(\ell+1)$ breaks. The test amounts to the application of $(\ell+1)$ tests of the null hypothesis of $\ell$ break against the alternative hypothesis of $(\ell+1)$ breaks).

3.2 Testing for return and volatility spillovers: Diebold and Yilmaz (2012)

To test the conditional heteroskedasticity of time series variables in this study, Engle [48] also suggests using Dynamic Conditional Correlation (DCC) of the GARCH model. We then use the generalized VAR method and variance decomposition theorem to construct the connectedness and spillover index introduced by Diebold and Yilmaz (2012) to show how returns and volatility are spilled over between stock market indices. The covariance stationary VAR(p) is presented by Diebold and Yilmaz (2012) as follows:

$K_t=\sum_{i=1}^p \Psi_i K_{t-1}+\varepsilon_t$

where, K_t represents the n×1 vector of the endogenous variables, $\Psi_i$ are the n×n autoregressive coefficient matrices, $\varepsilon_t$ represents a vector of the serially uncorrelated errors. The moving average representation is written as $K_t=\sum_{j=0}^{\infty} \mathrm{B}_j \varepsilon_t$, $\mathrm{A}_j$ satisfies the recursion of the form $\mathrm{B}_j=\Psi_1 \mathrm{~B}_{j-1}+\Psi_2 \mathrm{~B}_{j-2}+\cdots .+\Psi_p \mathrm{~B}_{j-p}$ with B₀ is the identity matrix of n×n, and $\mathrm{B}_j$ for $j<0$. The H-step ahead of generalized error forecast variance decomposition is:

$\varphi_{i j}(H)=\frac{\sigma_{j j}^{-1} \sum_{h=0}^{H-1}\left(e_i^{\prime} B_h \Sigma_{e_j}\right)^2}{\sum_{h=0}^{H-1}\left(e_i^{\prime} B_h \Sigma B_h^{\prime} e_i\right)}$

The $\Sigma$ denotes the variance matrix of errors vector, and the $\sigma_{i j}$ represents the standard deviation of the error terms of the jth equation. The $e_i$ denotes n×1 vector on the ith component and zero otherwise. The connectedness index encompasses n×n matrix c$\varphi(H)$ = $\left[\varphi_{i j}(H)\right]_{i, j=1,2}$, where where entries provide the contribution of the variable j to the forecast error variance of variable i. In a generalized decomposition of the variance, each entry of the variance decomposition matrix is normalized based on its row sum since, in the generalized decomposition, the contributions of both own and cross variables do not sum to one.

$\widetilde{\varphi_{\imath \jmath}}(H)=\frac{\varphi_{i j}(H)}{\sum_{j=1}^n \varphi_{i j}(H)}$

The $\sum_{j=1}^n \varphi_{i j}(H)$ = 1 and $\sum_{j=1}^n \varphi_{i j}(H)$ = n by construction. The $\widetilde{\varphi_{\imath \jmath}}(H)$ allows testing of the pairwise directional connectedness from j to i at H horizon. The transmission of the effect from j to i can be represented by $C_{i \leftarrow j}(H)$, while he opposite direction causality from i to j is presented by $C_{j \leftarrow i}(H)$. The net pairwise directional connectedness is presented as:

$C_{i j}=C_{i \leftarrow j}(H)-C_{i \leftarrow j}(H)$

Based on the above, the total directional spillover index from all variables to i is indicated for by C_(i←.) (H), is calculated as:

$\begin{aligned} C_{i \leftarrow .}(H)=\frac{\sum_{j=1, j \neq i}^N \widetilde{\varphi_{\imath \jmath}}(H)}{\sum_{i, j=1}^N \widetilde{\varphi_{\imath \jmath}}(H)} \times 100 =\frac{\sum_{j=1, j \neq i}^N \widetilde{\varphi_{\imath \jmath}}(H)}{N} \times 100\end{aligned}$

Hence, the net total connectedness is defined as:

$C_i(H)=C._{\leftarrow i}(H)-C_{i \leftarrow .}(H)$

Lastly, the aggregation of variance decomposition across variables in the system (stock market indices in our example) indicate for the total connectedness index that can be found by:

$\begin{aligned} C(H)=\frac{\sum_{i, j=1, i \neq j}^N \widetilde{\varphi_{\imath \jmath}}(H)}{\sum_{i, j=1}^N \widetilde{\varphi_{\imath \jmath}}(H)} \times 100 =\frac{\sum_{i, j=1, i \neq j}^N \widetilde{\varphi_{\imath \jmath}}(H)}{N} \times 100\end{aligned}$

According to Diebold and Yilmaz (2012), the technique can show the immediate impact of the current political situation on the transmission of financial shocks between African and developed economies.

3.3 Impulse response analysis

The utilization of impulse response function aims to depict the dynamic patterns and trajectory of series variables over the period of time under examination. In other words, this test enables for investigating the response of variables in the system to one impulse (shock) in other variable. The impulse response analysis is conducted using the Vector Autoregressive (VAR) theorem to allow for interactions among variables in the system. The simple bivariate VAR model is:

$\begin{gathered}\mathrm{a} t=110-112 \mathrm{~b} t+\gamma 11 \mathrm{a} t-1+\gamma 12 \mathrm{~b} t-1+\varepsilon \mathrm{a} t \\ \mathrm{~b} t=120-121 \mathrm{a} t+\gamma 21 \mathrm{~b} t-1+\gamma 22 \mathrm{a} t-1+\varepsilon \mathrm{b} t\end{gathered}$

where, l₁₀ and l₂₀ represent intercept terms while b and a denote the time series. As a_t and b_t affect each other, -l₁₂ and -l₂₁ are the simultaneous influence of unit change of b_t on a_t, and γ₁₂ and unit change in b_t-1 on a_t, respectively. The ε_yt and ε_zt represent white-noise disturbances to highlight innovations or shocks in a_t and b_t respectively in response to one standard deviation shock in other variables. γ₁₁ reflect the influence of b_t-1 on a_t and γ₂₂ the effect of b_t-1 on b_t.

The moving average representation system in VAR can be represented by:

$\left[\begin{array}{l}a_t \\ b_t\end{array}\right]=\left[\begin{array}{l}\bar{a} \\ \bar{b}\end{array}\right]+\sum_{i=0}^{\infty}\left[\begin{array}{ll}\mho_{11}(i) & \mho_{12}(i) \\ \mho_{21}(i) & \mho_{22}(i)\end{array}\right]\left[\begin{array}{l}\varepsilon_{a t-i} \\ \varepsilon_{b t-i}\end{array}\right]$

This moving average enables for interaction among sequences of variables. Coefficient Ʊ_i demonstrates the shock effect of error terms on the time path of dependent variables in the equation above. The functions Ʊ₁₁(i), Ʊ₁₂(i), Ʊ₂₁(i) and Ʊ₂₂(i) show impulse response functions that represent the behaviour variable in response to various impulses and shocks in other variables in the system.

This paper attempts to measure the main spillover channels among major stock markets in the African continent and a selected group of main international stock market indices. Using a group of research methodologies, the outcomes of the paper highlight the main return and volatility spillovers among the aforementioned indices during three major crises in the last two decades. However, we would like hereby to highlight the main limitations our paper might have witnessed. For instance, the use of the Diebold and Yilmaz (2012) model for return and volatility spillover lacks the required flexibility to capture the asymmetric behavior of time series data, where the effects and spillovers among studied variables are expected to exhibit nonlinear trajectory. In this regard, we believe that future research on this topic needs to improve the instrumental econometrical techniques to avoid such weakness such as the nonlinear time series models. Moreover, we urge future research to consider a time series model for structural breaks that can allow for concurrent determination of multiple as well as single most significant break in the time series. Finally, future research and attempts can improve the validity and comprehensiveness of the outcomes by enlarging the sample of studied markets and include more time observations.