Objective: This study provides an integrated empirical assessment of how geopolitical shocks and military-conflict news are associated with returns, conditional volatility, liquidity conditions, and investor behavior in Iranian financial markets. The empirical design combines stock, foreign-exchange, gold, liquidity, retail-flow, and news measures and distinguishes military threats, realized actions, and de-escalation signals. It also examines whether specifications containing news information show better out-of-sample predictive performance than conventional market-only baselines. The direct conflict of June 2025 and the prolonged episode beginning on 28 February 2026 are treated as distinct empirical environments because their intensity, persistence, and potential structural instability differ. Accordingly, the analysis is framed around event-time associations, conditional responses, and predictive relationships, not unrestricted causal effects or a new risk-management theory.



Method: The research is quantitative, longitudinal, and ex post facto, with the market day/news day as the unit of analysis. Daily observations cover 3 January 2020 to 19 July 2026. Domestic equity variables, including TEDPIX, the equal-weighted index, selected industry indices, trading value, market value, retail buying and selling, and market breadth, are obtained from TSETMC. Free-market US-dollar, 18-carat gold, and benchmark-coin quotations are obtained from TGJU. International controls include Brent oil, global gold, VIX, MSCI Emerging Markets, and the Caldara-Iacoviello geopolitical-risk index from FRED/EIA, FRED/CBOE, FRED/LBMA, MSCI, and the Federal Reserve dataset. The Persian news corpus is drawn from the archived sources in the research dataset; each item retains its source, publication time, and archival identifier. Reposts are deduplicated and news is aligned with Tehran market hours. At least 2,500 items are manually labeled using stratified sampling, and a fine-tuned ParsBERT classifier is evaluated against TF-IDF+SVM and XLM-R alternatives. Market returns are measured as log changes; the dollar-adjusted equity return is the log change in the stock-index-to-exchange-rate ratio; retail net flow is scaled by trading value; and turnover, Amihud illiquidity, and advance-decline breadth measure activity and liquidity. Unexpected threat, realized-action, and de-escalation news shocks are obtained after removing predictable components of probabilistic news indices. Event studies estimate short-horizon abnormal responses, EGARCH-X models conditional mean and variance, behavioral regressions examine retail-flow and liquidity outcomes, and time-ordered out-of-sample models evaluate predictive performance while controlling for global, calendar, sanction/negotiation, and exchange-rate-regime variables.



Findings: In the [0,+1] event window, realized military-action shocks are associated with a 3.42% decline in the dollar-denominated stock index and a 2.67% decline in the equal-weighted index, whereas the free-market dollar and domestic gold show positive abnormal returns of 2.91% and 3.36%, respectively. Retail net money flow declines by about 0.41 standard deviations over [0,+3]. In the EGARCH-X variance equation for the equal-weighted index, the realized-action coefficient is 0.34, compared with 0.16 for the threat coefficient; a Wald test rejects equality at p=0.021. The de-escalation coefficient is negative and smaller in absolute magnitude than the realized-action coefficient, which is consistent with an asymmetric conditional-volatility response. The ParsBERT classifier records a Macro-F1 of 0.86 versus 0.74 for TF-IDF+SVM. In out-of-sample forecasting, the market-plus-news XGBoost specification attains an AUC of 0.68 for stock-return direction versus 0.58 for the market-only logistic baseline, while the exchange-rate-volatility specification reports a QLIKE of 0.176 versus 0.214 for the market-only GARCH baseline. Because both the feature set and model class differ in these comparisons, the contrasts are treated as comparative predictive evidence rather than an isolated estimate of the marginal contribution of news.



Conclusion: The findings support a multi-market empirical assessment of geopolitical financial risk in Iran. The larger decline in dollar-adjusted and equal-weighted equity measures, together with positive abnormal returns in the dollar and domestic gold, is consistent with a shift toward currency and inflation hedges around realized military shocks. The smaller decline in the nominal aggregate index should not by itself be interpreted as preservation of real investor wealth, because exchange-rate depreciation and the weight of export-oriented firms may partly affect nominal resilience. The reported estimates are event-time associations, conditional responses, or predictive comparisons according to the method used; the design does not establish structural causality. For portfolio and fund managers, separating threats from realized actions can inform stress scenarios, liquidity buffers, and conditional currency/gold hedging after independent out-of-sample validation. For regulators, the evidence may inform discussion of transparency, reopening rules, and market depth, but it does not establish a unique optimal policy response. Data quality, media endogeneity, market closures, price limits, and regime changes remain important limitations.