Global commodity shocks are associated with volatility in frontier financial markets, affecting exchange rates and equity indices. This study examined volatility transmission from global commodity shocks to Kenya’s USD/KES exchange rate and the NSE 20 Share Index using daily data from November 1997 to December 2024. While GARCH specifications capture clustering, they are sensitive to structural breaks and regime changes, which distort persistence and weaken risk measures. Machine learning approaches provide alternatives capable of capturing nonlinear dependencies, abrupt volatility bursts, and regime-independent dynamics. Empirical evidence demonstrates that the 2008 Global Financial Crisis and COVID-19 induced permanent volatility regime changes. This study examined volatility transmission from global commodity shocks to a frontier financial market, focusing on the USD/KES and the NSE 20 Share Index. Structural break-detection was integrated through the Iterative Cumulative Sum of Squares algorithm, alongside APARCH, FIGARCH models and ML architectures (XGBoost, LSTM). In Kenya volatility is characterized by strong persistence and long-memory dynamics, with limited evidence of leverage effects. Break-adjusted models improve inference by correcting spurious persistence, while machine learning approaches demonstrate superior tracking of volatility during stress regimes. Volatility transmission is nonlinear, break-sensitive, and state-dependent; hybrid ML–econometric methods enhance crisis forecasting and regime-sensitive financial stability analysis.

Global commodity shocks are associated with volatility in frontier financial markets, affecting exchange rates and equity indices. While GARCH specifications capture clustering, they are sensitive to structural breaks and regime changes, which distort persistence and weaken risk measures. Machine learning approaches provide alternatives capable of capturing nonlinear dependencies, abrupt volatility bursts, and regimeindependent dynamics. Empirical evidence demonstrates that 2008 Global Financial Crisis and COVID19 induced permanent volatility regime changes. This study examined volatility transmission from global commodity shocks to a frontier financial market, focusing on the USD/KES exchange rate and the NSE 20 Share Index. Structural break detection was integrated through the Iterative Cumulative Sum of Squares algorithm, alongside APARCH, FIGARCH models and ML architectures (XGBoost, LSTM). In Kenya volatility is characterized by strong persistence and longmemory dynamics, with limited evidence of leverage effects. Breakadjusted models improve inference by correcting spurious persistence, while machine learning approaches demonstrate superior tracking of volatility during stress regimes. We show that volatility transmission from global commodity shocks to a frontier market is nonlinear, break-sensitive, and state-dependent, and that hybrid ML-econometric methods improve forecasting during crisis-period. Findings highlight persistence distortion, horizondependent performance, and relevance of regimesensitive modelling frameworks for financial stability in structurally evolving economies.

This study examined the comparative performance of GARCH family models including GARCH, EGARCH, GJR GARCH, APARCH, and FIGARCH within a horizon and regime aware framework to assess forecasting accuracy. Using daily prices of equity and foreign exchange markets in Kenya covering 1997–2024, volatility was modelled and validated through Value at Risk and Expected Shortfall back tests to establish economic relevance. The results reveal strong horizon and regime dependence: EGARCH performs well in capturing short run volatility in the equity market and turbulent foreign exchange re-gimes, while FIGARCH dominates in calm equity markets and at medium term horizons. Risk validation confirms that FIGARCH delivers reliable tail risk forecasts for equities, whereas EGARCH excels in turbulent foreign exchange markets. Unlike prior comparative studies that focus on efficient markets with stable volatility structures, this study applied GARCH family models to a frontier market, comparing forecasting accuracy across vary-ing horizons and regimes. The study advanced beyond best fit evaluations by linking forecasting performance to horizon length, asset type, and regime shifts, thereby contrib-uting new evidence on modelling volatility in African frontier markets and offering in-sights relevant for regulators, institutional investors, and policymakers concerned with financial stability.

Frontier financial markets face a diagnostic gap in forecasting volatility: linear and single-regime GARCH fails to capture breaks, spillovers, and regime transitions. Despite the importance of these markets, there is a gap in the literature: lack of a Kenya-specific, regime-sensitive Early Warning System (EWS) that can integrate long-memory filtering of volatility with nonlinear classification models. Therefore, policymakers lack signals to anticipate systemic stress. This study constructed a multi-stage pipeline using 6703 daily observations of the NSE 20 Share Index, USD/KES exchange rate, and Brent spot prices (1997–2024). K-Means clustering, FIGARCH filtering, LSTM–XGBoost ensemble classification, a logistic threshold model, and VaR/ES back-testing were used to generate alarm signals and validate risk-management performance. Across model estimation and evaluation phases, time horizons were assessed, showing that one-day signals were reactive and ten-day forecasts diluted precision, while five-day predictions achieved the strongest balance. Across model iterations, accuracy improved from 0.86 in the initial FIGARCH–LSTM pipeline, to 0.92 in the unbalanced classification model, and 0.94 in the weighted baseline, culminating in 0.98 with the final hybrid LSTM–XGBoost ensemble. The ensemble model delivered robust detection across Calm (F1 = 0.99), Moderate (F1 = 0.82), and Stress (F1 = 0.69) regimes. Stress thresholds were developed to translate regime signals into actionable alarms. This study provides an empirical application of a hybrid framework, combining long memory modelling of volatility with the adaptability of deep learning models to deliver Basel-compliant tail-risk alarms and a state-dependent policy matrix for regulators.

In frontier markets, financial volatility exhibits long-memory properties and regime-dependent asymmetries that standard linear models do not capture. This leads to inaccuracies in forecasting risk when a single model is applied across regimes. This study investigates the horizon- and regime-dependent performance of volatility models within a horizon- and regime-sensitive evaluation framework that applies single-regime Generalized Autoregressive Conditional Heteroscedasticity (GARCH) variants alongside a Hidden Markov Model (HMM). We evaluate the predictive accuracy of GARCH, Exponential GARCH (EGARCH), Glosten-Jagannathan-Runkle GARCH (GJR-GARCH), Asymmetric Power ARCH (APARCH), Fractionally Integrated GARCH (FIGARCH), and an HMM. Diebold–Mariano test statistics reveal that predictive superiority is sensitive to the chosen benchmark. When EGARCH is the benchmark, results highlight the importance of leverage effects, whereas a FIGARCH benchmark demonstrates that short-memory models are rejected as horizons increase. While short-memory models capture immediate clustering, FIGARCH maintains stable performance via hyperbolic decay. HMM provides a superior in-sample fit by capturing transitions between calm and turbulent regimes. Economic validation through Value-at-Risk (VaR) and Expected Shortfall (ES) backtesting indicates that FIGARCH and APARCH offer more reliable coverage for early warning systems during market stress. The findings emphasize that forecasting in a frontier market requires asset-specific approaches where benchmark selection dictates the interpretation of model superiority.

Frontier financial markets face a persistent diagnostic gap in forecasting volatility: traditional linear and single regime GARCH models fail to capture structural breaks, nonlinear spillovers, and regime transitions that define fragile market dynamics. In particular, ARMA–FIGARCH specifications could not exhaust residual linear persistence, motivating the integration of Long Short Term Memory (LSTM) networks to close this gap. This limitation leaves policymakers without timely, actionable signals to anticipate systemic stress. To address this gap, this study developed a hybrid Long Short Term Memory (LSTM) neural network combined with Fractionally Integrated GARCH (FIGARCH) econometric filtering to design a regime sensitive early warning system (EWS) in a frontier financial market. Using daily data from the NSE 20 Share Index, USD/KES exchange rate, and Brent crude volatility spanning 2000–2025 forecasting horizons were first assessed, showing that one day signals were overly reactive and ten day forecasts diluted precision, while five day predictions achieved the strongest balance with 98% accuracy and robust detection across Calm (F1 = 0.99), Moderate (F1 = 0.82), and Stress (F1 = 0.69) regimes. Econometric validation confirmed that conditional heteroskedasticity was adequately captured by the FIGARCH specification, though residual autocorrelation required AR refinements to improve model fit. Building on these diagnostics, probabilistic stress thresholds were developed to translate regime signals into actionable alarms, distinguishing transitional phases from full stress episodes and providing regulators with a practical buffer for intervention. The hybrid model thus advances methodological innovation by combining long memory volatility modelling with deep learning adaptability, while delivering Basel-compliant tail-risk alarms and a state-dependent policy matrix for regulators, offering policy relevant tools for monitoring volatility and crisis preparedness. The findings underscore the importance of regime sensitive approaches in frontier markets, offering credible and timely signals to anticipate and mitigate systemic stress, and contributing to both academic literature and practical financial stability oversight.