In this study, I focus on several types of uncertainty indexes: TEU, GEPU, REPU, GGPR, and RGPR. Yuan et al. (2022) used multivariate QVAR to disclose that the oil and stock markets are more susceptible to Russia and China’s EPU. Apostolakis et al. (2021) employed TVP-VAR and IRF approaches, revealing heightened connectedness among FSI, EPU, and oil prices during COVID-19. Wu et al. (2019) found that the tendency to invest in US Treasury bonds drops during heightened GEPU, and Demir et al. (2018) noted that GEPU plunged the yield of Bitcoin toward the extreme quantiles. Stolbov et al. (2018) demonstrated that, in the longer term, EPU significantly causes systemic risk in nine European countries, including Russia. Furthermore, Liow et al. (2018) found that shocks related to EPU instigate financial market stability.

Del Gaudio (2023) discovered that GPR significantly intensified FSI in the advanced economy, while NguyenHuu and Örsal (2024) established that GPR substantially exacerbates financial instability only at higher quantiles of financial sector indices. Umar et al. (2022) showed that the returns of European assets related to finance reacted heterogeneously due to GPRs, and Aysan et al. (2019) found a significant negative impact of GPR on Bitcoin returns. Bossman et al. (2023) revealed the potential hedging characteristics of the euro and Swiss franc against GPR.

On the other hand, Gök et al. (2022) observed varying levels of causality from TEU to Bitcoin, gold, and US 10-year bonds in terms of volatility and returns. Wu et al. (2021) found that cryptocurrency return and volatility are Granger-caused by TEU and GEPU. During the COVID-19 period, French (2021) noted that Bitcoin returns were significantly affected by TEU, and Aharon et al. (2022) found a similar causal relationship from TEU to cryptocurrency returns. Additionally, Hung et al. (2023) suggested that crypto prices are affected by heightened TEU and GEPU across varying time frequencies.

Several studies have discussed the nexus between energy, stock, and commodity markets with the dynamics of financial stress-related spillovers. Recently, Sohag et al. (2023a) identified a detrimental response of Russian financial stability due to both demand and supply shocks in the oil price in the short term. Günay et al. (2023) revealed that emerging markets are influenced by Russian and Brazilian stock markets. Long and Li (2023) uncovered heightened spillovers among the ­financial stress indices of six countries during extreme market conditions. Hoque et al. (2023) investi­gated the dynamic spillovers among Bitcoin, gold, and sectoral as well as ­regional financial stress indices, finding heightened connectedness during major events. Meanwhile, Elsayed et al. (2022) found that financial stress was the net receiver of risk transmission from the oil market during the mid to longer time horizon, and vice versa in the short term. Long et al. (2021) established a strong connectedness of financial stress indices with GPR. Ozcelebi (2020) revealed that the financial stress indices of developed economies lead to EMPI. Additionally, Elsayed and Yarovaya (2019) suggested that due to the Arab Spring, there is evidence of shock transmission to the financial stress indices of MENA countries.

From the first section, it is evident that there is no study regarding the connected­ness of uncertainty measures to solely Russian financial stress. Moreover, from the second section, we have identified that only Sohag et al. (2023a) work solely on the Russian Financial Stress Index, unravelling the effects of disaggregated oil market shocks on it. Therefore, we want to fill this research gap and investigate the dynamic connectedness and spillovers of uncertainty to Russian financial stress using CQ, R-CQ, and TVP-VAR approaches.

The Financial Stress Index of Russia was extracted from the Analytical Credit Rating Agency (ACRA). The GGPR and RGPR indices from Caldara and Iacoviello (2022), the GEPU index from Davis (2016), the REPU index from Baker et al. (2016), and the TEU index from Baker et al. (2021) were collected. Table 1 provides a detailed data description.

Counting newspaper articles about global uncertainties, Caldara and Iacoviello (2022) developed a measure of adverse geopolitical expansions and associated dangers since 1900, along with their advancement and economic ramifications. The automated text-search outcomes provided the basis for the geopolitical risks (GPRs) indicator from the digitized archives of ten newspapers. By tallying the amount of news regarding hazardous issues printed within each publication per month, they produced the index as a portion of the entire amount of news articles. They computed both global and country-specific GPRs. Following the methodology­ of Baker et al. (2016), Davis (2016) utilized 21 nations’ GDP‑weighted EPU indices to generate the global EPU (GEPU) index. The EPU index of each country measures how often three terms — “economy,” “policy,” and “uncertainty” — appear in newspapers within a given country. Baker et al. (2021) used a Twitter database to count English tweets regarding “economy” and “uncertainty” from 2011 to generate the TEU index.

In the context of the bivariate model, the quantile co-movement among the variables is investigated using the cross-quantilogram (CQ), recursive cross-quantilogram (R-CQ). Finally, the TVP-VAR is used to look into the ­dynamic connectedness.

The CQ technique, proposed by Han et al. (2016), is used to assess dual connecting relationships amid two sets of corresponding time series. It has a number of notable characteristics that provide rationale for using the method. First, even when there is an irregular distribution and very erratic observations, the approach may still be used to estimate bivariate volatility spillover across two markets. Second, using the CQ approach, we may determine how much a market will shock an observer under various quantiles of the data. Third, the approach enables­ simultaneous evaluation of the effectiveness of each variable, their length, and their overall trajectory (Sohag et al., 2022).

CQ between two stationary time series {y_1t ≤ q_1t(τ₁)} and {y_2t–k ≤ q_2t–k(τ₂)} is generated by the following equation (1), where lag order (k = ±1, ±2) for a group of τ₁ and τ₂ is denoted by k.

$p_{\tau }(k)=\frac{E\left[{\psi }_{{\tau }_1}(y_{1t}\le q_{1t}\left({\tau }_1\right)){\psi }_{{\tau }_2}(y_{2t-k}\le q_{2t-k}\left({\tau }_2\right))\right]}{\sqrt{E\left[{\psi }_{{\tau }_1}^2(y_{1t}\le q_{1t}\left({\tau }_1\right))\right]}]\sqrt{E\left[{\psi }_{{\tau }_2}^2(y_{2t-k}\le q_{2t-k}\left({\tau }_2\right))\right)}}$ (1)

where stationary time series is denoted by y_i,t, i = 1, …, 6 represents the RFSI, GGPR, RGPR, GEPU, REPU and TEU respectively and t = 1, 2, …, T. The ­cumulative distribution and corresponding probability density function are denoted by F_i (.) and f_i (.) for y_i,t, i = 1, 2, 3. Corresponding quantile function is, q_it (τ_i) = inf{υ: F_i (υ) ≥ τ_i} for τ_i ∈ (0, 1) and ψ_a (u) = 1[u < 0], where a is the process of quantile-hit. The CQ technique enables the detection of uniform transition in both series as well as serial dependency between variables at ­distinct quantiles.

During analyzing cross sectional dependence between two stationary time series events {y_1t ≤ q_1t(τ₁)} and {y_2t–k ≤ q_2t–k(τ₂)}, ρ_τ (k) = 0 indicates no cross-sectional dependence from event {y_2t–k ≤ q_2t–k(τ₂)} to event {y_1t ≤ q_1t(τ₁)}. We can detect how the cross-quantile dependency between the chosen variables varies across various spans of time by predicting how ρ_τ (k) varies with the lag length k. This allows us to quantify the degree and duration of reliance. In our case we consider taking lags as k = 1, 3, 6, 12.

After that, using a Ljung–Box kind test with the test statistic obtained as equation (2), we determine the statistical significance of ρ_τ (k).

$Q_{\tau }^{\ast }(p)=\frac{T(T+2)\sum _{k=1}^{p^{\mathrm{\#}}}{\hat{\rho }}_{\tau }^2(k)\ast }{(T-k)\mathrm{\#}}$ (2)

where the cross-quantilogram, denoted by ρ̂_τ (k), was computed as follows:

${\hat{\rho }}_{\tau }(k)=\frac{\sum _{t-k+1}^T{\psi }_{{\tau }_1}(y_{1t}-{\hat{q}}_{1t}\left({\tau }_1\right)){\psi }_{{\tau }_2}(y_{2t-k}-{\hat{q}}_{2t-k}\left({\tau }_2\right))}{\sqrt{\sum _{t-k+1}^T{\psi }_{{\tau }_1}^2(y_{1t}-{\hat{q}}_{1t}\left({\tau }_1\right)}\sqrt{\sum _{t-k+1}^T{\psi }_{{\tau }_2}^2(y_{2t-k}-{\hat{q}}_{2t-k}\left({\tau }_2\right))}},$ (3)

where the estimated quantile function is calculated by q̂_it (τ_i)(i = 1, 2, 3). Stationary bootstrap is utilized for the estimation of the null distribution of the CQ by equation (3) and the Q-statistic by equation (2).

To investigate time series data, the recursive cross-quantilogram (R-CQ) method employs a rolling window, where the window size stands for a certain time interval. The reliance between systemic risk and market circumstances is often measured using CQ, which are computed inside each window to evaluate­ the link between two series. The CQ calculates the likelihood of a variable exceeding a quantile based on another variable’s value at various quantiles, such as the lower (τ = 0.05), middle (τ = 0.50), and higher (τ = 0.95). The R-CQ technique recursively creates CQs for each rolling window to analyze evolving relationships over time. Overall, the R-CQ technique is useful for evaluating systemic risk and market circumstances, detecting and analyzing market bubbles.

We have applied the updated version of the time-varying parameter vector auto­regressive (TVP-VAR) method, as modified by Antonakakis et al. (2020), to the earlier versions of the connectedness approach by Diebold and Yilmaz (2012) and Diebold and Yılmaz (2014). The approach is superior to its earlier versions since it applies both forgetting factors and the “Kalman filter” in the model ­approximation process through the utilization of the variance-covariance matrix, allowing it to capture most of the information contained in the data.

With p lag length the TVP-VAR model can be defined as:

y_t = φ_t x_{t –1} + ε_t ε_t | I_{t –1} ~ N (0, Σ_t),

vec (φ_t) = vec (φ_{t –1}) + w_t w_t | I_{t –1} ~ N (0, W_t), (4)

where endogenous temporal series of order (m × 1) is represented by y_t, while lagged vector of order (pm × 1) regarding y_t ranging from (t – p) to (t – 1) is denoted by x_{t –1}. Vectors of error terms are denoted by ε_t and w_t. All recognized facts are represented by I_{t –1} till t – 1. Time-varying variance-covariance matrices are denoted by Σ_t and W_t.

During the estimation phase of “Generalized forecast error variance decomposition” (GFEVD) both time-varying “variance-covariance” matrices as well as coefficients are provided. The Z-step ahead GFEV denoted by φ_ij (Z) is decomposed initially by the generalized VAR model and then the row sum will be used for normalizing it. Using the “Wold representation theorem”, the TVP-“vector moving­ average” (VMA) is obtained from the TVP-VAR model as the penultimate step of the decomposition process. This transformation process is as follows:

$y_t=\sum _{i=1}^p{\phi }_{it}{y}_{t-1}+{\epsilon }_t=\sum _{j=0}^{\mathrm{\infty }}{A}_{jt}{\epsilon }_{t-j}$

${\phi }_{ij}(Z)=\frac{{\sigma }_{jj}^{-1}\sum _{z=0}^{Z-1}{\left(w_i^{\mathrm{\prime }}{A}_z\mathrm{\Sigma }{w}_j\right)}^2}{\sum _{z=0}^{Z-1}{\left(w_i^{\mathrm{\prime }}{A}_z\mathrm{\Sigma }{A}_z^{\mathrm{\prime }}{w}_i\right)}^2}$

${\tilde{\phi }}_{ij}(Z)=\frac{{\phi }_{ij}(Z)}{\sum _{j=1}^N{\phi }_{ij}(Z)}$ (5)

where the approximated SD for the error of variable j is denoted by the σ_jj, for the error term vector ε, Σ is the matrix of variance and the identification vector is represented by w_i taking 1 as the i^th component and zero otherwise.

Using the “Minnesota prior” the “Kalman filter” is utilized according to Antonakakis et al. (2020), with (0.99, 0.99) as the popular standard falloff factors during the approximation process.

All the connectedness components including total connectedness (TC), directional spillovers received by element i from j (DC_i←j), and transmit from i to j (DC_i→j). Consequently, net directional spillovers (NET) and net pairwise directional connected­ness (NPDC) are computed as follows:

$TC(Z)=\frac{\sum _{i,j=1,ijj}^N{\tilde{\phi }}_{ij}(Z)}{\sum _{i,j=1}^N{\tilde{\phi }}_{ij}(Z)}\times 100=\frac{\sum _{i,j=1,ijj}^N{\tilde{\phi }}_{ij}(Z)}{N}\times 100,$ (6)

$D{C}_{i\leftarrow j}(Z)=\frac{\sum _{j=1,j\ne i}^N{\tilde{\phi }}_{ij}(Z)}{\sum _{i,j=1}^N{\tilde{\phi }}_{ij}(Z)}\times 100=\frac{\sum _{j=1,j\ne i}^N{\tilde{\phi }}_{ij}(Z)}{N}\times 100$ (7)

$D{C}_{i\to j}(Z)=\frac{\sum _{j=1,j\ne i}^N{\tilde{\phi }}_{ij}(Z)}{\sum _{i,j=1}^N{\tilde{\phi }}_{ij}(Z)}\times 100=\frac{\sum _{j=1,j\ne i}^N{\tilde{\phi }}_{ij}(Z)}{N}\times 100$ (8)

$NE{T}_i(Z)=D{C}_{i\to j}(Z)-D{C}_{i\leftarrow j}(Z)$ (9)

$NPD{C}_{ij}(Z)=\frac{{\tilde{\phi }}_{ji}(Z)-{\tilde{\phi }}_{ij}(Z)}{N}\times 100$ (10)

The generated TC using formula mentioned in equation (6) does not remain within the range [0, 1]. Hence, the adjusted connectedness regarding total is calculated using the formula:

$TC(\text{adjusted})=\frac{\sum _{i,j=1,i\ne j}^N{\tilde{\phi }}_{ij}(Z)}{N-1}\times 100.$ (11)

Analysis begins with a descriptive analysis and then progresses into the empirical findings and discussion. Here we provide tabular data (Table 2) to comment on the findings of several normality tests, such as skewness, kurtosis, and Jarque–Bera (JB), which all consistently demonstrate that the variables under examination are not normal. Moreover, the ERS unit root test verifies that the variables in this study are stable over time at I(0) level.

As a result, econometric techniques based on quantiles are a good fit for this fat tailed data. Quantile-based data analysis approaches, such as CQ, recursive R-CQ and TVP-VAR are useful for exami­ning the quantile dependency and dynamic connectedness of the variables of interest when the data lacks normality.

We have applied this bivariate method of identifying quantile dependence of several types of uncertainty indexes with Russian Financial Stress Index. There are four distinct memory scenarios shown on the cross-quantilogram heatmap, one for each of the four possible latency patterns (monthly, quarterly, bi-annual, and annual). Color gradients may also be used to show the direction of a dependency between two variables. The greater the cell’s redness, the greater its quantile dependence. The less dependent a cell is, the bluer it is. There is no association between the variables, as seen by the light green cells. To roughly estimate the statistical significance of quantile dependence, a Ljung–Box test is utilized. The (*) icon on the heatmap denotes statistical significance at the 10% level for the quantile dependence of the respective cells.

Fig. 1 reveals the spillover effects of TEU to RFSI in terms of quantiles across different memory structures. It is found that there is strong positive dependence at the bearish states (towards the lower quantiles of both variables) of the market in the initial memory, and the strength of this positive spillover effect gradually diminishes towards longer (quarterly, bi-annual, and annual) memory structures. Specifically, in the quarterly memory, there exists strong negative dependence across the middle range (q0.25–q0.75) quantiles of both series. In the bi-annual memory, there is a slightly positive association at the middle (q0.50–q0.70) quantiles of both series.

This spillover effect tends to be negative at the bearish as well as mid-range (q0.30–q0.70) quantiles of both variables. These findings are well supported by Hung et al. (2023); Gök et al. (2022) and Aharon et al. (2022). Use of cryptocurrency in Russia is a common matter and this sector is very much vulnerable to TEU fluctuations. So Russian financial stability can be disturbed by TEU.

Figure 1.

Cross-quantile dependence between Tweeter-based Economic Uncertainty (TEU) and Russian Financial Stress Index (RFSI). Note: The horizontal axes show the quantile distribution of RFSI. The bars on the graph are color-coded, with blue representing a negative association and red representing a positive association. The intensity of the colors corresponds to the strength of the association between the two variables. Source: Author’s calculations.

Fig. 2 depicted the quantile dependence of GEPU to RFSI across different lag structures to see whether the spillover effect persists over time or not. It is evident that there is strong positive spillover effect at the bullish (towards the higher quantiles of both series) as well as mid-range market conditions in case of initial memory though there was some negative dependence towards the higher quantiles of RFSI and lower quantiles of GEPU. While in the quarterly memory, there is a strong positive spillover effect at the bearish states of the market, and negative dependence is also found towards the lower quantiles of RFSI and higher quantiles of GEPU. Interestingly, the quantile dependence vanishes in the direction of longer memory (bi-annual and annual).

That means the Global Economic Policy Uncertainty has significant positive dependence on Russian Financial Stress Index at the initial memory but the spillover effect dissolved gradually towards longer memory. These findings are in line with Antonakakis et al., (2014); Arouri et al., (2016) and Brogaard and Detzel, (2015).

Figure 2.

Cross-quantile dependence between Global Economic Policy Uncertainty (GEPU) and Russian Financial Stress Index (RFSI). Note: The horizontal axes show the quantile distribution of RFSI. The bars on the graph are color-coded, with blue representing a negative association and red representing a positive association. The intensity of the colors corresponds to the strength of the association between the two variables. Source: Author’s calculations.

Quantile dependence scenario between REPU and RFSI is represented in Fig. 3. A strong positive spillover effect is evident at the bullish market states in case of initial memory length while the dependence turns to a strong negative at the bearish market condition for quarterly memory structure. In addition, there exists strong positive dependence towards higher quantiles of RFSI and lower quantiles of REPU.

Bi-annually, there is no significant dependence. In the case of annual memory length, significant heterogeneous dependence is found. Unlike the GEPU, REPU has long-lasting heterogeneous spillover effect on RFSI. Yuan et al. (2022); Apostolakis et al. (2021) provide support in favor of this findings.

Figure 3.

Cross-quantile dependence between Russian Economic Policy Uncertainty (REPU) and Russian Financial Stress Index (RFSI). Note: The horizontal axes show the quantile distribution of RFSI. The bars on the graph are color-coded, with blue representing a negative association and red representing a positive association. The intensity of the colors corresponds to the strength of the association between the two variables. Source: Author’s calculations.

Fig. 4 depicts the quantile dependence between GGPR and RFSI. Significant strong positive dependence is observed at the middle quantiles of both series, while strong negative dependency is found in the bearish states of the market with the initial memory. For quarterly memory, strong positive quantile dependence is observed in the bearish states. Additionally, there is strong negative dependence towards the lower quantiles of RFSI and higher quantiles of GGPR. Although there are significant positive and negative spillover effects of GGPR on RFSI in the initial memory, all these heterogeneous effects wash out across the longer memory structures. Similar findings are also evident in the research of Aysan et al. (2019); Bossman et al. (2023) and Del Gaudio (2023).

Figure 4.

Cross-quantile dependence between Global Geopolitical Risk (GGPR) and Russian Financial Stress Index (RFSI). Note: The horizontal axes show the quantile distribution of RFSI. The bars on the graph are color-coded, with blue representing a negative association and red representing a positive association. The intensity of the colors corresponds to the strength of the association between the two variables. Source: Author’s calculations.

The quantile dependence between RGPR and RFSI is presented in Fig. 5. It is evident that there is strong positive spillover effect at both bearish and bullish­ market conditions for initial memory. Gradually, the significance of positive quantile dependence diminishes towards longer memory structures. For the quarterly memory, there is positive dependence at the bullish states and some negative spillovers during middle quantiles. Towards the middle quantiles, there is positive and slightly negative dependence for bi-annual and annual memory respectively which is supported by Del Gaudio, (2023) and NguyenHuu and Örsal (2024). Therefore, RGPR has long lasting spillover effects on RFSI than GGPR. Similar findings are also found for the spillover effects of REPU and GEPU on RFSI.

Figure 5.

Cross-quantile dependence between Russian Geopolitical Risk (RGPR) and Russian Financial Stress Index (RFSI). Note: The horizontal axes show the quantile distribution of RFSI. The bars on the graph are color-coded, with blue representing a negative association and red representing a positive association. The intensity of the colors corresponds to the strength of the association between the two variables. Source: Author’s calculations.

Rolling window based quantile approach is used to crosscheck the spillover effects of the variables under scrutiny over time. Around 20% of the available observations (30) is taken as window size and 100 bootstrap is set for generating efficient recursive outputs. The rolling sample window approximations are represented by the threads from top-to-bottom, when both markets experience lower quantiles (5%), median quantiles (50%) and upper quantiles (95%). Time variation in spillover measurements can be shown using this method. We use cross-quantilogram to analyze time-varying dependency under normal and extreme market situations, using time series at the lower (τ = 0.05), intermediate­ (τ = 0.50), and higher (τ = 0.95) quantiles. The blue line is the time-varying cross-quantilogram, and the red lines are the 95% confidence interval that was calculated using 100 replicates of the bootstrap method.

Fig. 6 intensely depicts the spillover effect between TEU and RFSI from both side across various quantile combinations including 5, 50 and 95 quantiles. Under the lower market trajectory (τ = 0.05), there exists a positive spillover effect from each side of TEU and RFSI around 2014 (Crimea crisis) and 2018–2023 during the COVID-19 outbreak and recent Russia–Ukraine conflict. Moving towards mid quantiles (τ = 0.5) there is a positive spillover from both sides around 2011–2012 (European debt crisis), 2014 (Crimea crisis) and 2020–2023 recent pandemic and Russia–Ukraine conflict. While in the higher quantiles (τ = 0. 95) strong positive risk spillovers are found during 2018–2023 from both sides due to impactful recent events. These are consistent with earlier findings.

Figure 6.

Recursive cross-quantilogram-based volatility spillover effect of Tweeter-based Economic Uncertainty (TEU) with Russian Financial Stress Index (RFSI). Note: Graphs are generated based on three specific quantiles, i.e., 0.05, 0.5 and 0.95. Source: Author’s calculations.

Fig. 7 apparently exposes the quantile dependence between GEPU and RFSI from both sides across various quantile combinations including 5, 50 and 95 quantiles­. Under the lower market trajectory (τ = 0.05), during 2011–2012 (European debt ­crisis) we found positive spillovers of GEPU to RFSI, and towards 2018–2023 (­during recent pandemic and Russia–Ukraine conflict) slightly weak positive dependence is evident. From RFSI to GEPU there is no such significant spillover on GEPU. In the middle quantile (τ = 0.5), there is evidence of positive quantile dependence throughout the sample period from both ends. Moving towards the highest (τ = 0.95) quantiles, RFSI receive and also transmit risk spillovers to GEPU in the time range 2018–2023. This also resembles earlier findings.

Figure 7.

Recursive cross-quantilogram-based volatility spillover effect of Global Economic Policy Uncertainty (GEPU) with Russian Financial Stress Index (RFSI). Note: Graphs are generated based on three specific quantiles, i.e., 0.05, 0.5 and 0.95. Source: Author’s calculations.

Fig. 8 vividly portrays the spillover effect between REPU and RFSI from both sides across various quantile combinations. Under the lower market trajectory (τ = 0.05), a positive spillover effect is observed from both sides around 2018–2023, encompassing the China–U.S. trade war, COVID-19, and the ­conflict between Russia and Ukraine. Moving towards the mid quantiles (τ = 0.5), there is an upward trend of risk spillovers from both sides around 2020–2023. In the case of the higher quantiles (τ = 0.95), an upward trend from both ends around 2019–2023 is observed, although it is not statistically significant.

Figure 8.

Recursive cross-quantilogram-based volatility spillover effect of Russian Economic Policy Uncertainty (REPU) with Russian Financial Stress Index (RFSI). Note: Graphs are generated based on three specific quantiles, i.e., 0.05, 0.5 and 0.95. Source: Author’s calculations.

Fig. 9 vividly discloses the quantile dependence between GGPR and RFSI from both ends across various quantile combinations. Under the lower market conditions (τ = 0.05), positive risk spillover effects are evident from both ends around 2011–2014 (European debt and Crimea crises) and 2020–2023 (COVID-19 and Russia–Ukraine conflict). In the middle quantiles (τ = 0.5), positive quantile dependence is observed around 2014 (Crimea issue), 2018 (China–U.S. trade war), and 2020–2023 (pandemic and Russia–Ukraine conflict) from both sides. Moving ­towards the highest (τ = 0.95) quantiles, positive risk spillovers are found throughout­ the range, especially around 2011–2012 (European debt crisis) and 2020–2023 (pandemic and Russia–Ukraine conflict). These findings support earlier outcomes.

Figure 9.

Recursive cross-quantilogram-based volatility spillover effect of Global Geopolitical Risk (GGPR) with Russian Financial Stress Index (RFSI). Note: Graphs are generated based on three specific quantiles, i.e., 0.05, 0.5 and 0.95. Source: Author’s calculations.

Fig. 10 reveals the quantile dependence between RGPR and RFSI from both sides across various quantile combinations. Under the lower market trajectory (τ = 0.05), evidence of positive spillover effects from both ends is observed during­ the years 2011–2012, 2014, and 2022–2023. In the middle quantiles (τ = 0.5), positive quantile dependence is evident around 2011–2012, 2015, and 2020–2023 from both sides. In the case of the highest (τ = 0.95) quantiles, positive spillovers are found throughout the study period, especially around 2011–2012 (European debt crisis) and 2020–2023 (pandemic and Russia–Ukraine conflict). These findings are consistent with earlier outcomes.

Figure 10.

Recursive cross-quantilogram-based volatility spillover effect of Russian Geopolitical Risk (RGPR) with Russian Financial Stress Index (RFSI). Note: Graphs are generated based on three specific quantiles, i.e., 0.05, 0.5 and 0.95. Source: Author’s calculations.

The estimated parameters of the TVP-VAR model are presented in Table 3. The results demonstrate that the 95% confidence intervals include all forecasted posterior means. The diagnostic tests show that the time-varying parameters converge to the posterior distribution, in accordance with the convergence diagnostics of the Geweke statistics. The null hypothesis for Geweke statistics is that the samples are drawn from the stationary distribution of the chain, and the test is based on the equality of means of the first and last parts of the Markov chain. The number of iterations is also enough for posterior estimation because all of the factors regarding inefficiency are quite negligible.

Therefore the estimated TVP-VAR model is robust and will provide reliable results. The dynamic total directional connectedness measures are presented in Table 4 based on a TVP-VAR using lag of order “1” (BIC) and 10-step ahead-generalized forecast error variance decomposition with a rolling window of size 50. This estimation is carried out according to the updated version of Gabauer (2021) who modified the static TVP-VAR model proposed by Diebold and Yilmaz (2012).

The diagonal elements of the matrix represent the individual contributions of each element to volatility spillover. The off-diagonal elements indicate the contributions from or to other elements. In addition, the table’s columns tie one variable to all the others independently, while the rows link the contribution of each variable to the prediction error variance of that variable in the system. The average TCI for the analyzed variables is 32.30%, indicating that its dynamic network may determine the system’s internal connectivity. The RGPR is transmitting volatility shocks to other variables with highest forecast error variance value of 41.82% in the system. TEU is receiving the highest volatility shocks from all other variables of the system with forecast error variance value of 38.64%. The average total connectedness index (TCI) of the system is 32.30% and the dynamic TCI of the system is depicted in Fig. 11.

It is observed that the TCI varies over time among the variables under conside­ration. Fig. 11 clearly shows that the TCI increases to almost 50% around 2011 (European debt crisis) and fluctuates between 39% and 36% during 2014 (Crimea issue). The TCI increased from 26% to 29% just after the outbreak of COVID-19 in 2020, a trend supported by Long et al. (2021), and from the start of the Russia–Ukraine conflict in 2022, the TCI increases to 32-33%.

The dynamic net total directional connectedness is presented in Fig. 12. Positive numbers represent the system’s net transmitting role, while negative values represent its net receiving role. RFSI receives volatility shocks during COVID-19, as outlined by Long et al. (2021), but interestingly, it acts as a transmitter just after the start of the Russia–Ukraine conflict. REPU acts as a receiver of volatility shocks during the Crimea crisis, COVID-19, as well as the Russia–Ukraine conflict.

GEPU acts as a transmitter from the beginning of the Crimea conflict in 2014 and during the conflict with Ukraine. TEU is the receiver of such volatility spillovers from others throughout the entire sample period. GGPR acts as a risk transmitter during the Crimea (2014–2015), COVID-19 (2020–2021), and the recent conflict with Ukraine (2022–2023), while RGPR was the receiver of volatility shocks during­ the Crimea (2014–2015) and the recent conflict with Ukraine (2022–2023).

The dynamic net pairwise total directional connectedness is revealed in Fig. 13. These pairwise dynamic connectedness graphs provide evidence of the pairwise time frequency connectedness. From the graph, it is apparent that TEU and GEPU were the net pairwise receivers of volatility shocks from RFSI during the conflict between Russia and Ukraine, while REPU, GGPR, as well as RGPR, were the net pairwise volatility risk transmitters during the Russia–Ukraine conflict. Overall, we found that all other variables are highly interconnected with each other.

The dynamic net pairwise total directional connectedness can be visualized by a network plot, depicted in Fig. 14. This plot reveals the overall net pairwise transmission and reception of volatility shock scenarios visually. Blue nodes represent the net transmitters of volatility risks, while yellow vertices represent the receivers of those volatility shocks. The magnitude of (TO) spillover effects is represented by node size, while the direction of the spillovers is shown by arrows, and the strength of connectedness is indicated by the thickness of the connections.

In this case, GGPR, GEPU, and RGPR act as net transmitters, while REPU, TEU, and RFSI act as net receivers of volatility shocks. RFSI directly receives volatility spillover shocks from RGPR, while it indirectly receives risks from GGPR, GEPU, and REPU via RGPR. Additionally, REPU directly receives shocks from RFSI, which aligns with the findings of Long and Li (2023) and Ozcelebi (2020) regarding how FSI can transmit shocks to others.

Finally, TEU directly receives shocks from all variables except RFSI, while it indirectly receives shocks from RFSI via REPU, as supported by the research of Ozcelebi (2020).

Figure 11.

Dynamic TCI of the TVP-VAR approach with lag length of order 1 (BIC) and 10-step ahead-“generalized forecast error variance decomposition” (GFEVD). Source: Author’s calculations.

Figure 12.

Dynamic net total directional connectedness of the TVP-VAR approach with lag length of order 1 (BIC) and 10-step ahead GFEVD. Source: Author’s calculations.

Figure 13.

Dynamic net pairwise total directional connectedness of the TVP-VAR approach with lag length of order 1 (BIC) and 10-step ahead GFEVD. Source: Author’s calculations.

Figure 14.

Dynamic net pairwise connectedness network plot of the TVP-VAR approach with lag length of order 1 (BIC) and 10-step ahead GFEVD. Source: Complied by the author.

We have found that Russian financial market stress is very much interconnected with different sources of uncertainties. Uncertainty regarding geopolitical tension can directly transmit to financial sector of any country through the channel of trade and capital flow and the investor’s sentiment regarding investment. This is the core understanding of the black swan theory established by Taleb (2007). There is another well-known concept, the risk aversion theory of Arrow (1971) and Pratt (1976). This theory describes the investor’s tendency of lingering during any unpleasant situation due to uncertainty regarding probable return from any potential investment into any financial market, therefore exacerbating the financial stress of that country. In addition, the Minsky–Kindleberger theory of financial ­instability provides a compelling conceptual link between financial unpredictability­ and uncertainty regarding economic policy (EPU) (Kindleberger et al., 2005; Minsky, 1982). This concept relays that EPU can transmit risk to the financial stability of a country. That’s why according to these well-established concepts risk can be transmitted to financial stress of a country from different sources of uncertainties. We have proved these concepts empirically for Russian financial market stress. However, based on the market structure and institutional quality of a country, the severity and persistence of its financial market stress can vary.