Aug 2026· Journal of Intelligent Decision Making and Information Science· 0 citations· 22 references
TL;DR
The results show that memory-based models outperform feedforward ANN models, while the contribution of the VIX varies across models and markets.
Abstract
Volatility forecasts play an important role in financial markets. They are used for setting position limits, determining margin requirements, and guiding short-term trading decisions. Financial volatility is known to exhibit persistence and regime dependence, which makes short-horizon forecasting challenging. In addition to historical volatility measures, implied volatility indicators such as the Volatility Index (VIX) are often used as forward-looking proxies for market risk expectations. However, it is not clear whether the predictive usefulness of the VIX is consistent across different neural network architectures. This study investigates the role of the VIX in one-day-ahead realized volatility forecasting and examines whether its contribution varies across alternative neural network models. The empirical analysis focuses on major global equity markets, including the S&P 500, DAX, FTSE 100, Nikkei 225, and Hang Seng Index. Using market data from these indices, the study evaluates the forecasting performance of artificial neural networks (ANN), long short-term memory networks (LSTM), and gated recurrent unit (GRU) models. All models are estimated within a rolling out-of-sample framework and are tested with and without the VIX variable. Forecast accuracy is evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Quasi-Likelihood (QLIKE) loss functions, and Diebold–Mariano tests. The results show that memory-based models outperform feedforward ANN models, while the contribution of the VIX varies across models and markets.
Spillover of volatility shocks across borders during turbulent periods makes accurate equity market volatility forecasts especially critical for risk management, derivatives pricing, and regulatory capital. In this paper, we examine whether volatility forecasts improve when models incorporate information on how markets are connected, and whether the choice of connection measure matters. Using daily data on 29 equity indices from every major region over 2015-2025, we let each market's forecast draw on the recent volatility of the markets linked to it, with the strength of each link set either by geography, return correlation, or the Diebold-Yilmaz (DY) spillover network estimated from the data. The spillover-informed forecasting model achieved an approximately 13\% reduction in out-of-sample QLIKE loss relative to the standard Heterogeneous AutoRegressive benchmark ($p<0.001$) and attained the highest model confidence set $p$-value among the models considered. Several benchmark models that do not explicitly incorporate cross-market structure were excluded from the 90\% model confidence set. The greatest improvements were observed during periods of elevated market stress: the COVID-19 crash, the Russian invasion of Ukraine, and the $2025$ US tariff shock, spillover-informed forecasts achieved approximately 21\% lower QLIKE loss than an otherwise identical network-blind model ($p = 0.018$), with no significant performance loss during calm periods. The advantage is economically material: a volatility-targeting investor would pay $322$-$373$ basis points per year, net of transaction costs, for spillover-informed forecasts, versus an insignificant $85$-$102$ basis point for the alternatives. Finally, when the model is left to infer connections on its own, it recovers the DY network from the forecast objective alone (permutation $p=0.0005$).
Volatility is a fundamental characteristic of financial markets and plays a crucial role in investment decision-making, portfolio management, and financial risk assessment. Understanding the behaviour of stock market volatility is particularly important for frontier markets such as the Nepal Stock Exchange (NEPSE), where market fluctuations may differ from those observed in developed economies. This study investigates the volatility dynamics and forecasting performance of the NEPSE Index using daily closing prices from March 2021 to March 2026. Daily logarithmic returns were analysed following tests for stationarity and conditional heteroskedasticity. The volatility process was modelled using the symmetric GARCH(1,1) and asymmetric EGARCH(1,1) models, while model adequacy and forecasting performance were evaluated through diagnostic tests, information criteria, rolling-window out-of-sample forecasting, the News Impact Curve, volatility half-life estimation, and structural break analysis. The empirical findings confirm that NEPSE returns exhibit the stylized characteristics of financial time series, including volatility clustering, leptokurtosis, and persistent conditional volatility. The EGARCH model provides a marginally better in-sample fit and identifies significant asymmetric volatility, with negative market shocks producing stronger increases in future volatility than positive shocks of similar magnitude. However, the GARCH(1,1) model demonstrates slightly superior out-of-sample forecasting performance, indicating that the simpler specification remains more reliable for short-term volatility prediction. The analysis also identifies a structural break in July 2022, suggesting that volatility dynamics changed during the study period and that persistence estimates should be interpreted with caution. By integrating model comparison, forecasting evaluation, News Impact Curve analysis, volatility persistence, and structural break testing within a single analytical framework, this study provides a comprehensive assessment of NEPSE volatility and offers practical insights for investors, portfolio managers, policymakers, and future researchers interested in frontier equity markets.
Stock market prediction is a significant research topic in the financial sector and has been widely investigated by researchers and investors. Forecasting stock markets is challenging because of the nonlinearity, volatility, and dynamics of the data. Additionally, stock markets are affected by various internal and external factors. While most previous studies relied on historical prices and technical indicators (TIs), they ignored the influence of overall economic factors on stock markets. In contrast, this study introduces a robust framework to predict daily log returns by leveraging a combined dataset comprising historical data, TIs, and macroeconomic data, including gold and oil prices, the volatility index, the dollar index, the interest rate, the 10-year Treasury yield, the term spread, and the dividend yield. Moreover, we propose a hybrid feature selection (FS) approach that combines filter and wrapper methods, unlike in previous studies, which relied primarily on a single FS approach or neglected it entirely. The dataset represents diverse sectors and firm sizes, covering five companies: Apple (AAPL Inc.), Exxon Mobil Corporation (XOM), Goldman Sachs (GS), Pfizer Inc. (PFE), and Ford Motor Company (F). We apply a hybrid evaluation approach that incorporates 5-fold time series cross-validation (TSCV) with a holdout test set to evaluate the efficiency of the proposed models. The results revealed that despite modest improvements in statistical metrics, FS significantly enhanced the economic performance of the trading strategy, as demonstrated by better returns and Sharpe ratios. The results show that deep learning (DL) models that combine macroeconomic data and the FS process, in addition to historical data and TIs, achieved higher trading returns and produced superior risk-adjusted performance, although they demonstrated slightly higher forecast errors than traditional models. This confirms that statistical accuracy alone is insufficient for evaluating financial forecasting models. This work highlights the benefits of the proposed framework for predicting stock returns across different markets, offering significant insights for financial analysts.
Aya Nabil, S. Barakat, Ahmed Aboelfetouh et al.· Scientific Reports· 0 citations
Based on daily returns of the CSI 300 index and monthly macroeconomic variables from January 2005 to December 2025, this paper examines whether low-frequency macroeconomic information provides incremental value for stock market volatility forecasting. A GARCH-MIDAS model is employed to decompose daily return volatility into short-run and long-run components. Economic policy uncertainty, industrial value added, M2, CPI, PPI, and a composite macroeconomic state factor are incorporated into the long-run volatility component. Out-of-sample forecasting results for January 2019 to December 2025 show that industrial value added and M2 are not only significantly positive in the long-run component equation, but also perform best under the QLIKE, MSE, and MAE loss functions. The composite macroeconomic state factor has some explanatory power, although it is weaker than industrial value added and M2. Economic policy uncertainty improves forecast accuracy, but no stable and significant positive effect of EPU on the long-run volatility component is found. Overall, real economic activity and monetary liquidity are important low-frequency macroeconomic information sources for understanding and forecasting the long-run volatility of the CSI 300 index.
Yuan-Yi Xu· Advances in Economics, Manag...· 0 citations
The results show that both ML and DL models provide strong forecasting performance for the BIST100 Index, and suggest that BIST100 dynamics are shaped by the joint influence of economic fundamentals, market sentiment and liquidity conditions.
Alptekin Erdağ, M. F. Uçar, İ. Tarhan· Review of Behavioral Finance· 0 citations
This study evaluates multiple forecasting models, ranging from HAR and GARCH to Tree-based and Neural architectures, across 14 Global Equity Indices and Horizons form 1 day to 100 Trading days within a strictly chronological and capacity-controlled framework, indicating that for strongly dependent time series, nominal sample size is a misleading measure of learnability, and model capacity must be constrained relative to effective information rather than observation count.
