Introduction

Stock exchanges are central to the global economy, enabling firms to raise capital, diversify portfolios, and attract international investment (Majeed et al. 2021). The rapid integration of global financial markets has amplified the importance of stock exchanges and increased exposure to systemic risk and volatility (Nong 2024). Volatility, the degree to which uncertainties influence asset prices or expected returns, is a fundamental concern in finance because it directly affects investment decisions and market stability (Fang et al. 2018). Persistent volatility often manifests in clustering patterns, influencing investor expectations and shaping market behaviour (Vera-Valdés 2022; Huber et al. 2022). Financial asset volatility can arise from multiple sources, including banking and financial crises (Miyajima and Yafeh 2007; BenMim and BenSaïda 2019), natural disasters (Buigut and Kapar 2020), regulatory interventions (Grout and Zalewska 2006; Jeon et al. 2022), country risk (Kunjal et al. 2022), and geopolitical risk (GPR). GPR, encompassing wars, terrorism, trade conflicts, and political tensions, represents a key systematic risk factor that can disrupt global financial stability (Caldara and Iacoviello 2022; Fiorillo et al. 2023; Wang 2024). Recent geopolitical events such as the United States US-China trade wars, Brexit, and the Russian-Ukraine conflict have underscored the market-relevant impact of geopolitical uncertainty (Li et al. 2023; Liu et al. 2024). GPR can limit investors’ ability to value firms accurately, resulting in high stock price volatility and delayed investment decisions (Nikkinen et al. 2008; Jeribi et al. 2015; Salisu et al. 2022).

Empirical evidence indicates that elevated GPR generally depresses stock prices and increases return volatility (Zhang et al. 2023). These effects are observed at the aggregate market level (Balcilar et al. 2018; Ndako et al. 2021; Salisu et al. 2022; Zhang et al. 2023) and at the industry level. Specific industries, such as tourism, aviation, or export-oriented industries, are disproportionately affected by geopolitical or pandemic shocks (Demiralay and Kilincarslan 2019; Chesney et al. 2011; Alfaro et al. 2020; Kamal et al. 2023). Defensive industries and safe-haven assets often benefit from capital inflows during crises (Chatziantoniou et al. 2025; Bouri et al. 2019). While empirical evidence highlights industry-level heterogeneity, it mainly focuses on discrete events, which limits its applicability to real-time, evolving GPR, and the dynamic interaction between domestic and global shocks remains underexplored. This study addresses these gaps by investigating the effects of ongoing GPR on the volatility of industry returns in the Johannesburg Stock Exchange (JSE), a context not previously studied.

This study is novel theoretically as it considers the differentiated impact of GPR across industries, highlighting how industry-specific characteristics mediate risk exposure. For instance, the energy industry is highly exposed to global supply shocks, whereas basic materials are more sensitive to domestic factors such as labour strikes or electricity shortages. Empirically, this research is novel because it disentangles the effects of domestic versus global GPR on JSE industries, providing insights into which source predominantly drives volatility, unlike prior research focusing on single events or aggregate indices (see, Hoque and Zaidi 2020; Oyadeyi et al. 2024). Methodologically, this study employs the Generalised Autoregressive Conditional Heteroscedasticity-Mixed Data Sampling (GARCH-MIDAS) framework, which accommodates mixed-frequency data to distinguish short-term and long-term volatility components. This represents an improvement over traditional GARCH-type models, enabling more precise modelling of the temporal dynamics of industry volatility under geopolitical uncertainty. By applying the GARCH-MIDAS, this study quantifies how slowly moving macro-level GPR measures influence daily industry return volatility, offering richer insights for investors and policymakers than conventional approaches.

The study contributes to the literature in two key ways. First, it provides new evidence on industry-level volatility in South Africa in response to geopolitical risk. While prior studies on the JSE have primarily focused on returns (Jaffar et al. 2025), often at the aggregate market level (Hoque and Zaidi 2020) or on specific geopolitical events, such as the Russia–Ukraine conflict (Oyedeyi et al., 2024), this study examines how both domestic and global GPR influence industry-level volatility, highlighting the heterogeneous volatility responses across industries on the JSE, an analysis not previously conducted. Second, the study extends its contribution beyond in-sample analysis by conducting a rigorous out-of-sample forecast evaluation using the modified Diebold-Mariano test. This compares the predictive performance of models incorporating global versus domestic GPR in forecasting industry volatility, providing practical insights into which source of geopolitical risk contains superior information for volatility management and market forecasting.

The abovementioned contributions have important implications: investors can identify resilient industries and optimise portfolio allocation during heightened geopolitical tension. Meanwhile, policymakers can tailor domestic interventions or engage in international risk-sharing strategies depending on whether domestic or global GPR is the dominant source of risk. In the JSE context, this analysis spans nine industries; basic materials, consumer discretionary, consumer staples, energy, financials, healthcare, industrials, technology, and telecommunications, each expected to respond uniquely to geopolitical uncertainty due to structural differences.

The remainder of this study is organised as follows. Section  "Brief literature review" reviews the literature on stock market volatility and uncertainty. Section  "Data and preliminary results" details the data and preliminary results. Section  "Methodology" presents the methodology. Section  "Empirical results and analysis" presents the empirical results and analysis, and Sect.  "Conclusion" concludes the study.

Brief literature review

This study is primarily grounded in Modern Portfolio Theory (MPT), which posits that investors optimise portfolios by combining imperfectly correlated assets to maximise expected returns for a given risk level (Markowitz 1952; Balarezo 2010). Accordingly, MPT highlights how industry characteristics and exposure to macroeconomic factors, including global geopolitical risk (GPR), shape portfolio risk. Industries differ in their exposure to the specific transmission channels of GPR, such as trade restrictions, commodity price volatility, sanctions, currency fluctuations, and changes in defence spending (Caldara and Iacoviello 2022; Lim and Choi 2022; Deng and Wu 2024), which in turn leads to differences in how their returns co-move and, consequently, in the correlations observed between industries. For instance, industries like energy, with inter-regional global supply chains (Cohen Liechti et al. 2015), are particularly vulnerable, as firms indirectly connected to affected partners may experience significant disruptions along the same value chains. In contrast, aerospace and defence companies often face lower supply chain risk due to the regionalisation of operations required to meet national security standards (Baek et al. 2024).

Complementary perspectives, such as behavioural finance and the Efficient Market Hypothesis, emphasise that investor biases and information asymmetry can amplify volatility (Fama 1965; Kahneman and Tversky 1979; Thampanya et al. 2020). These frameworks suggest asset price fluctuations reflect the interplay between fundamentals, macro-level risks, and investor behaviour.

Empirically, GPR has been widely shown to influence financial market volatility. Traditional GARCH-type models, including GARCH, Exponential-GARCH, GARCH in mean, and the Threshold-GARCH, have been applied to equities, commodities, and Islamic financial markets (Bollerslev 1986; Nelson 1991; Bouras et al. 2019; Ndako et al. 2021; Salisu et al. 2022). Panel GARCH evidence suggests that international GPR shocks often exert more potent effects than domestic shocks, particularly in emerging markets (Cermeño and Suleman 2014; Bouras et al. 2019). In South Africa, domestic political dynamics significantly influence JSE volatility, underscoring the relevance of country-specific exposures (Vengesai et al. 2022). Regrettably, traditional GARCH models face limitations when dealing with mixed-frequency data, as they often fail to capture the influence of slow-moving variables like GPR on high-frequency returns (Hassan et al. 2003; Bakry 2006; Makoko and Muzindutsi 2018).

The GARCH-MIDAS framework addresses this challenge by allowing low-frequency predictors such as monthly geopolitical risk indices to shape long-run volatility while high-frequency shocks govern short-run dynamics (Engle et al. 2013; Ndako et al. 2021; Salisu et al. 2022). Empirical applications demonstrate that GPR heightens volatility in emerging and Islamic markets, though the magnitude and persistence vary by country and sectoral structure (Yang et al. 2021; Bouri et al. 2023; Nilsson 2024). Related non-GARCH approaches corroborate that GPR predicts volatility jumps, but the sensitivity differs across markets depending on institutional resilience and investor behaviour (Gkillas et al. 2018; Balcilar et al. 2018; Zhang et al. 2023).

Industry-level evidence further suggests that GPR effects are heterogeneous. Energy, financials, industrials, and basic materials industries typically exhibit higher volatility responses to geopolitical shocks, whereas the defensive industries, such as consumer staples, tend to be more resilient (Apergis et al. 2018; Bouri et al. 2023). The nature of geopolitical risk also matters: terrorism-related shocks often trigger sharper short-term volatility than broader political or trade tensions (Chatziantoniou et al. 2025). In South Africa, JSE industries have reacted differently to political shocks and the Russia–Ukraine conflict (Vengesai et al. 2022; Oyadeyi et al. 2024).

Msomi and Kunjal (2024) also find that economic policy uncertainty raises volatility in basic materials but dampens it in technology, healthcare, and financial industries, reinforcing the notion of industry-specific sensitivity to macro shocks. Despite these advances, three important gaps remain. First, most prior studies rely on discrete geopolitical events or aggregated indices rather than continuous, real-time GPR measures, limiting the generalisability of findings. Second, South African sectoral or industry analyses predominantly use traditional GARCH models, which may obscure the influence of slow-moving uncertainty factors. Third, while recent studies employ GARCH-MIDAS, few explicitly focus on geopolitical uncertainty, as most tend to examine economic policy uncertainty (EPU) or financial volatility.

This study addresses these gaps by applying the GARCH-MIDAS framework to real-time global and domestic GPR, analysing their differential effects on JSE industry volatility. Beyond in-sample estimation, the current study further advances the literature by assessing out-of-sample forecast performance using the modified Diebold–Mariano test, thereby providing evidence on the predictive value of GPR for volatility forecasting. Theoretically, this clarifies how industry characteristics condition vulnerability to global and local geopolitical shocks. Empirically, it enhances understanding of how these shocks transmit through an emerging financial market. Methodologically, it extends mixed-frequency volatility modelling to a forecasting dimension, strengthening both the explanatory and predictive relevance of GARCH-MIDAS for risk management. Collectively, these insights assist investors in identifying resilient industries and guide policymakers in designing targeted interventions depending on whether domestic or global GPR dominates market uncertainty.

Data and preliminary results

The study examines the relationship between geopolitical risk (GPR) and industry-specific volatility in the Johannesburg Stock Exchange (JSE) from 1 January 2003 to 9 October 2024, aligning with the availability of normalised global and country-specific GPR indices (Caldara and Iacoviello 2018). Daily closing prices for nine industry indices; basic materials (BM), consumer discretionary (COND), consumer staples (CONS), energy (ENERG), financials (FIN), healthcare (HLTC), industrials (IND), technology (TECH), and telecommunications (TELEC) were sourced from the Integrated Real-time Equity System (IRESS) database. The Financial Times Stock Exchanges (FTSE)/JSE All Share Index (ALSI) is a proxy for overall market trends. Real estate and utilities were excluded due to incomplete data.Footnote 1 Daily returns were calculated as:

$$\:{r}_{t}=ln\left(\frac{{P}_{t}}{{P}_{t-1}}\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:$$
(1)

where \(\:{r}_{t}\)denotes daily returns and \(\:{P}_{t}\) is the daily closing price. Monthly GPR indices at the global (GPR) and South African (GPR_ZAF) levels quantify geopolitical risk via the frequency of newspaper mentions of geopolitical events in the US, the United Kingdom (UK), and Canada, adjusted for cross-time and cross-country comparability. Compared to the Economic Policy Uncertainty (EPU) index, the GPR index captures geopolitical tensions unrelated to shocks more accurately (Bouri et al. 2023).      

Table 1 reports the descriptive statistics of daily JSE industry-specific stock returns and monthly exogenous factors. Average daily returns across industries are close to zero, as expected, with the consumer discretionary industry (COND) posting the highest mean (0.0742%) and health care (HLTC) yielding a slightly negative mean (− 0.0022%). Such small mean values are typical of daily equity returns, but the dispersion is substantial. Standard deviations range from 1.19% in the overall market (ALSI) to nearly 4% in energy (ENERG), underscoring considerable volatility. Healthcare (3.37%) also records unusually high volatility, reflecting pandemic-induced uncertainty. These results echo the stylised facts of emerging markets, where volatility is more pronounced than in developed counterparts (Bekaert and Harvey 1997).

The skewness and kurtosis measures reveal marked deviations from normality. Negative skewness dominates in most industries, suggesting asymmetric downside risk, a feature often linked to crisis-driven markets (Zaremba et al. 2022). Extreme kurtosis values, particularly for energy and healthcare, highlight the presence of fat tails and tail risk. These heavy-tailed return distributions reinforce GARCH-type models’ suitability for conditional heteroscedasticity. Turning to the exogenous monthly variables, global geopolitical risk (GPR) and its South Africa–specific counterpart (GPR_ZAF) show high skewness and kurtosis, consistent with their role as persistent, non-Gaussian processes.

Table 2 provides diagnostic tests for conditional heteroscedasticity, autocorrelation, and stationarity. The ARCH-LM tests strongly reject the null hypothesis of no ARCH effects across nearly all industries, confirming the presence of volatility clustering; a core justification for employing GARCH models. The Ljung–Box Q-statistics indicate significant autocorrelation in return series, especially in ALSI and BM, suggesting that past shocks influence present dynamics. The ADF test results uniformly reject the null of a unit root, establishing stationarity of returns and supporting their use in volatility modelling. The evidence from Tables 1 and 2 provides a strong empirical rationale for the GARCH-MIDAS framework. The high-frequency JSE return series exhibit volatility clustering, heavy tails, and persistence, warranting a GARCH structure. Meanwhile, the exogenous monthly risk variables are non-Gaussian and persistent, aligning with the MIDAS structure’s ability to integrate low-frequency predictors into volatility modelling.

Table 1 Descriptive statistics.
Full size table
Table 2 Conditional variance, autocorrelation, and unit root tests.
Full size table

Methodology

Predictability is evaluated using a GARCH-MIDAS-RV framework, in which daily stock returns is decomposed into a short-run GARCH component and a long-run MIDAS component driven by low-frequency predictors. To begin with, we specified the conditional mean equation for the daily stock return as follows:

$$\:s{r}_{i,t}=\mu\:+\sqrt{{\sigma\:}_{i,t}^{2}}\:{v}_{i,t};\forall\:i=\text{1,2},...,{N}_{t}$$
(2)

Here, \(s{r_{i,t}}\) denotes daily stock return with innovation distributed as \(\:{v}_{i,t}|{\varphi\:}_{i-1,t}\sim\:N(\text{0,1})\). The term \(\:{\varphi\:}_{i-1,t}\) represents the information set available on day i − 1 at period t. The total conditional variance (volatility), \(\sigma _{{i,t}}^{2}\) can be decomposed into short-run (\({h_{i,t}}\)) and long-run (\({g_t}\)) volatility components as follows:

$$\sigma _{{i,t}}^{2}\,=\,{g_t} * {h_{i,t}}$$
(3)
$${h_{i,t}}\,=\,(1\, - \,\alpha \, - \,\beta )\,+\,\alpha \frac{{{{(s{r_{i - 1,t}} - \mu )}^2}}}{{{g_{t - 1}}}}\,+\,\beta {h_{i - 1,t}},$$
(4)
$${g_t}\,=\,\exp \left( {{\theta _0}\,+{\theta _1}\sum\limits_{{k=1}}^{K} {{\omega _k}\left( {{X_{t - k}}} \right)} \,} \right),$$
(5)
$$\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:{X}_{t}\in\:\left\{{GPR\_ZAF}_{t},{GPR}_{t}\right\}$$

while \(\:{g}_{t}\) captures the long-run volatility component, and the MIDAS weights \(\:{w}_{k}(\cdot\:)\) follow a normalized beta polynomial.

Out-of-sample forecasts are evaluated 30 days, 60 days, and 90 days horizons, respectively. For instance, \(\:h=30,\:60,\:90\). Separate models are estimated for GARCH-MIDAS-GPR-ZAF and GARCH-MIDAS-GPR, with a benchmark GARCH-MIDAS-RV specification that excludes low-frequency predictors (i.e., \(\:{\theta\:}_{1}=0\)). The sample is divided into in-sample and out-of-sample periods, with 50% of the observations used for model estimation and the remaining 50% reserved for forecast evaluation. For the sake of robustness such as whether the forecast performance is episodic, we further structure the in-sample and out-sample periods into 75% to 25% and 90% to 10% respectively. Forecasts are generated using a rolling-window procedure in which the estimation window remains fixed while advancing sequentially through time. This approach captures evolving temporal dynamics, mitigates overfitting, and ensures that forecast errors reflect realistic information sets available at each forecast origin.

Forecast accuracy is assessed using the modified Diebold–Mariano (DM) test statistic proposed by Harvey et al. (1997), which corrects the original DM test of Diebold and Mariano (1995) for small-sample bias and multi-step-ahead forecasts. The modified statistic is given by:

$$D{M^*}=\,\left( {\sqrt {\frac{{T\,+\,1\, - \,2h\,+\,{T^{ - 1}}h(h - 1)}}{T}} } \right)DM$$
(6)

where \(\:D{M}^{*}\) denotes the modified DM statistic, \(\:T\) is the out-of-sample size, and \(\:h\) is the forecast horizon. The original DM statistic is defined as:

$$DM=\,\frac{{\bar {d}}}{{\sqrt {V(d)} \,}}\,$$
(7)

where \(\bar {d}\) is the mean loss differential, \(\:{d}_{t}=L\left({e}_{1,t}\right)-L\left({e}_{2,t}\right)\), \(\:L(\cdot\:)\) denotes the loss function of forecast errors \(\:{e}_{1,t}\) and \(\:{e}_{2,t}\) from the competing models, and \(\:\widehat{V}\left(\stackrel\;{\bar {d}}\right)\) is the unconditional variance of the loss differential. The null hypothesis of equal predictive accuracy, \(\:{H}_{0}:E\left({d}_{t}\right)=0\), is tested against the alternative \(\:{H}_{1}:E\left({d}_{t}\right)\ne\:0\). A negative and statistically significant DM statistic indicates superior predictive performance of the corresponding GARCH-MIDAS-X model (GARCH-MIDAS-GPR-ZAF or GARCH-MIDAS-GPR_GLOBAL) relative to the baseline, while a positive statistic favours GARCH-MIDAS-RV. For comparison between GARCH-MIDAS-GPR-ZAF and GARCH-MIDAS-GPR_GLOBAL, the latter is treated as the baseline model. As a result, a negative and statistically significant DM statistic in that regards indicates superior predictive performance of GARCH-MIDAS-GPR-ZAF compared to GARCH-MIDAS.

Empirical results and analysis

The GARCH-MIDAS estimation results in Table 3 reveal critical insights into how realised volatility (Panel A), global (Panel B) and domestic (Panel C) geopolitical risks shape industry volatility in the JSE. The estimates indicate that volatility in South African industry indices is associated with both persistent short-run shocks and slower-moving geopolitical variables, consistent with the dual-structure volatility dynamics documented in the literature (Engle et al. 2013; Conrad and Loch 2015). Across the models, the short-run parameters reinforce the conventional evidence of volatility clustering, as daily shocks tend to propagate over time rather than dissipate immediately. This persistence underscores that the JSE, much like other emerging markets, is characterised by prolonged episodes of heightened uncertainty, a finding that echoes Bouri et al. (2023). However, beyond these familiar short-run patterns, the long-run components reveal more economically meaningful relationships between geopolitical risk and market behaviour. The results from the benchmark GARCH-MIDAS-RV model validate the presence of intense and persistent volatility in all industries, affirming that realised volatility alone captures much of the time-varying risk structure. However, a more nuanced picture emerges once geopolitical risk is incorporated in the extended specifications, both global (GPR) and domestic (GPR_ZAF).

Table 3 GARCH-MIDAS regression on geopolitical risk and industry volatility
Full size table

The global GPR model shows that higher global geopolitical risk is associated with lower long-run volatility across several industries, particularly those more integrated into international markets such as basic materials and energy. This is consistent with existing evidence suggesting that certain industries, notably energy and consumer staples, tend to display hedging characteristics during periods of global geopolitical unrest (Caldara and Iacoviello 2018; Antonakakis et al. 2017). From a practical perspective, this suggests that investors seeking to manage exposure to global geopolitical uncertainty may consider relatively higher allocations to these industries, as they appear to act as volatility stabilisers compared to more domestically oriented industries. While the findings are consistent with hedging-type behaviour, the analysis does not formally establish the underlying mechanisms, and the observed patterns should be interpreted as indicative rather than definitive safe-haven effects.

Conversely, industries such as industrials and finance display positive sensitivities to global GPR, indicating that volatility in these industries is relatively higher during periods of elevated international geopolitical uncertainty. These findings reflect South Africa’s economy’s openness and financial integration with global capital flows, which make specific industries sensitive to global risk shocks. The elevated sensitivity of the financial industry, in particular, aligns with the notion that global uncertainty affects cross-border liquidity and capital inflows, leading to higher long-run volatility. For policymakers, this pattern emphasises that South Africa’s financial stability remains highly dependent on external conditions, implying that domestic macroprudential measures alone may not sufficiently insulate the market from global turbulence.

The domestic GPR model reveals a distinct pattern relative to the global specification. The consumer staples, financials, and industrials respond to local geopolitical tensions in the same direction as they did under global GPR, confirming that geopolitical risk, regardless of origin, is associated with South African industry volatility in a broadly consistent manner. However, a key difference emerges for basic materials and energy. Unlike their strong sensitivity to global GPR, these industries show no statistically significant response to domestic GPR, affirming that their volatility could be driven predominantly by global commodity dynamics rather than local political uncertainty. By contrast, the technology industry, which was previously unaffected by global GPR, exhibits reduced long-run volatility under domestic GPR. This finding suggests that the market may have already priced in domestic geopolitical instability, particularly recurring load shedding and governance failures at Eskom, South Africa’s state-owned electricity utility. During the sample period, these disruptions exposed the industry’s dependence on stable power infrastructure and when intensified by political uncertainty raised costs, distorted pricing and increased earnings risk. However, because such instability became a persistent structural feature of the domestic environment, investor expectations adjusted over time, resulting in lower observed volatility in response to incremental changes in local GPR. Meanwhile, the telecommunications, healthcare and consumer discretionary industries remain insensitive to both global and domestic GPR.

The broader economic implications of these results are far-reaching. The persistence of volatility across models underscores that foreign or domestic geopolitical shocks have long-lived effects on investor behaviour and market risk. For investors, this persistence means that portfolio rebalancing strategies must account for extended periods of elevated uncertainty following geopolitical events. For regulators, the decisive influence of global rather underscores the necessity of embedding global stress scenarios into prudential risk frameworks. It also signals the importance of maintaining capital flow management tools to buffer the domestic market from external contagion. Furthermore, the industry-specific responses highlight the need for tailored risk management strategies. The consumer staples industry appears to act as a hedge against both global and domestic geopolitical risk, consistent with its defensive nature. In contrast, the financial and industrial industries may require tighter regulatory oversight during periods of heightened local and global geopolitical instability to prevent excessive speculative volatility.

In line with theoretical expectations, the findings confirm that GPR effects in South Africa are heterogeneous across industries. Financials and industrials exhibit elevated volatility in response to geopolitical shocks, reflecting their exposure to global supply chains, capital flows and macroeconomic uncertainty. Conversely, defensive industries such as consumer staples remain relatively resilient, offering stability during periods of heightened risk (Apergis et al. 2018; Bouri et al. 2023). The next sub-section builds on these insights by evaluating the out-of-sample forecasting performance of the competing GARCH-MIDAS models. The analysis assesses whether incorporating geopolitical risk, global or domestic, improves the predictive accuracy of volatility forecasts relative to the baseline GARCH-MIDAS-RV model, thereby testing the practical value of these risk measures for real-time market prediction and portfolio management.

Out-of-sample forecast performance evaluation

The modified Diebold–Mariano (DM) test results reported in Table 4 provide insight into the relative forecasting performance of the competing GARCH-MIDAS specifications across industries and horizons. Overall, the evidence does not uniformly support augmenting volatility models with geopolitical risk measures. Clear improvements are observed only in the healthcare industry, where negative and statistically significant DM statistics across all forecast horizons indicate that both GPR and GPR_ZAF-augmented models outperform the baseline GARCH-MIDAS-RV specification. This suggests that geopolitical risk contains incremental information beyond realised volatility. The heightened sensitivity of healthcare volatility to geopolitical risk may reflect its strong exposure to government policy, regulatory frameworks, public health interventions, and cross-border supply chains (Caldara and Iacoviello 2022). Moreover, pharmaceutical firms operate across borders to meet international demand, exposing supply chains to geopolitical instability such as the China-US trade war, the COVID-19 pandemic, and the Russia–Ukraine conflict, which can disrupt production and trade through export restrictions and other national security measures (Kalish and Wolf 2021; Simon 2022).

Table 4 D-M test out-of-sample forecast performance results [50% sample]
Full size table

In contrast, the results for the financials industry indicate that realised volatility remains the dominant predictor relative to domestic GPR, while global geopolitical risk exerts a stronger influence than domestic geopolitical risk, as reflected in positive and significant D-M statistics. This pattern is consistent with the globally integrated nature of financial markets, where cross-border capital flows and international risk sentiment play a central role (Antonakakis et al. 2017). For the remaining industries, the absence of significant predictive gains suggests that geopolitical shocks may already be internalised in realised volatility, limiting the incremental contribution of explicit GPR measures. Alternatively, it may indicate that the transmission of geopolitical risk operates less directly through these industries’ fundamental channels.

Robustness check: is forecasting power of GPR episodic?

To assess whether the predictive power of geopolitical risk (GPR and GPR_ZAF) is stable or episodic, the out-of-sample forecasting exercise is extended from the 50% split to 75 and 90% evaluation windows. Under this framework, forecasting power is considered non-episodic (robust) if results remain consistent across alternative forecast samples, and episodic if the relative performance of the models changes materially. The results in Tables 5 and 6 reveal some heterogeneity across industries. For the healthcare industry, the predictive gains associated with GPR-based models remain consistent across the 50, 75, and 90% out-of-sample frameworks. The augmented GARCH-MIDAS specifications continue to outperform the baseline model. This stability indicates that the forecasting power of geopolitical risk in healthcare is not episodic but structurally embedded. The sensitivity of healthcare volatility to geopolitical developments therefore appears persistent and generalisable across forecast horizons and sample periods.

Table 5 D-M test out-of-sample forecast performance results [75% sample]
Full size table
Table 6 D-M test out-of-sample forecast performance results [90% sample]
Full size table

In contrast, the consumer discretionary, financials, and industrials industries display episodic behaviour. While in-sample results in Table 3 suggested that the financial and industrial industries are vulnerable to geopolitical risk, the out-of-sample analysis shows that realised volatility remains the dominant predictor. In addition, industrials industry exhibits high sensitivity to the inclusion of recent data. These findings imply that the vulnerability of these industries to geopolitical shocks is not stable over time and should not be generalised across various samples.

Conclusion

This study applied the GARCH-MIDAS framework to investigate how global and domestic geopolitical risk (GPR) influence industry-level volatility in the Johannesburg Stock Exchange (JSE). The results reveal pronounced industry asymmetries in both the magnitude and direction of the response. Globally, financials and industrials emerge as the most sensitive industries, with heightened GPR significantly increasing long-term volatility, reflecting exposure to external shocks and crisis-driven demand fluctuations. In contrast, basic materials, consumer staples, and energy, indicate partial hedging properties, with global GPR reducing long-term volatility, underscoring their resilience to external geopolitical turbulence. Domestically, volatility is amplified in the financial and industrial industries, indicating these industries’ vulnerability to local political uncertainty. Conversely, consumer staples and technology industries experience dampened volatility under domestic GPR, suggesting either reduced trading activity or flight-to-safety behaviour in response to political tensions. These findings underscore that global and domestic GPR are key determinants of volatility dynamics in the JSE, though their influence varies substantially across industries.

The economic and practical implications are twofold. Policymakers should prioritise stabilisation measures in industries most exposed to geopolitical tensions; particularly financials and industrials globally and domestically. Through improved policy communication, contingency planning, and proactive engagement in risk mitigation strategies. For investors, the heterogeneous industry responses highlight the importance of tailoring portfolio strategies to individual risk preferences: risk-averse investors may reduce exposure to volatility-amplifying industries, whereas risk-tolerant investors could strategically exploit these shocks to achieve higher returns. Methodologically, the study demonstrates the value of the GARCH-MIDAS framework in capturing both short-term shocks and the long-term effects of geopolitical uncertainty. The integration of out-of-sample forecast evaluation further enhances the practical relevance of the findings, providing investors and risk managers with guidance on the predictive power of geopolitical risk for volatility management. Specifically, GPR shows stable and consistent predictive power in the healthcare industry across all forecast windows. By contrast, for the financials and industrials industries, realised volatility dominates and the effects are sample-dependent, limiting their generalisability.

Finally, while this study advances understanding of industry-specific GPR impacts in an emerging market context, it has limitations. By examining industries individually, the analysis does not capture inter-industry spillovers or systemic volatility transmission. Future research should explore connectedness and spillover effects, identifying which industries act as transmitters and which as receivers of volatility, thereby providing a more comprehensive picture of market-wide resilience under geopolitical uncertainty.