In the first stage, the data were retrieved from the Web of Science database using the search queries shown in Table 1. Once the search was performed, 120 documents were obtained.

Table 1

Search queries
"GARCH MIDAS" + "VOLATILITY" "GARCH MIDAS" + "VOLATILITY FORECASTING" "GARCH MIDAS" + "ENERGY" "GARCH MIDAS" + "EMERGING MARKETS"

Font: Own elaboration.

To ensure quality, the database was filtered as suggested by Ferrari (2015) to include exclusively:

Articles written in English.

Articles were excluded if they met the following conditions:

Duplicated records.

Subsequently, the remaining articles were reviewed manually to discard those unrelated to the topic examined. After this process, 30 articles were obtained and considered eligible. Each study was screened in two steps:

Title and abstract screening to ensure thematic relevance.

From the retrieved articles, a narrative literature review was conducted. The reviewed manuscripts were summarized and categorized to identify specific patterns in previous research. Based on the above, a classification of the analyzed manuscripts was proposed, which is detailed in the results section.

In the second stage, a bibliometric analysis was conducted on the database of selected documents. Therefore, the number of publications per year, citations trends, the number of articles and countries by author, and the collaboration networks among authors' countries were identified and analyzed.

In addition, a conceptual structure analysis of the topic studied was conducted using machine learning techniques, particularly the k-means clustering algorithm. This method is commonly used to "address the machine learning task of clustering, which involves finding natural groupings of data" (Lantz, 2019, p. 271). The main advantage of this approach is that it does not require a specific result associated with the data, which allows groups to be determined from the data without any preconceived criteria:

In our analysis, the k-means exercise was based on the co-currencies of words found in the abstracts, titles, and keywords of the articles included in the database. To accomplish this task, a natural language processing routine was applied to form the conceptual clusters. The results obtained from this step are presented in the following section.

Risk of Bias and Limitations

Although the PRISMA framework was applied, it is important to acknowledge some limitations. First, the exclusive use of the Web of Science database may have left out relevant studies indexed in other sources. However, this decision allowed us to focus on high-impact, peer-reviewed journals in economics and finance, which strengthens the reliability of the results. Second, the restriction to articles in English may introduce a language bias. However, it also ensured consistency in the analysis and comparability between studies. Finally, although manual review is not completely free of subjectivity, this process allowed for a more careful assessment of the relevance and quality of the included studies. Overall, these considerations do not weaken the study, but rather reinforce the robustness and credibility of the results.

Following the previously described methodology, the selected articles were analyzed to gain insights into the literature related to the topic studied. This exercise is presented in two sections related to the (i) evolution of econometric models and (ii) the relationship between oil shocks and volatility in emerging markets.

Recognizing the pivotal role of modeling and forecasting volatility, researchers and financial practitioners stress its importance in designing effective hedging and investment strategies. Regulatory bodies also heavily depend on accurate volatility forecasts to monitor turbulence in financial markets (Salisu et al., 2022). To address the challenges posed by volatile market conditions, researchers have diligently proposed methodologies for understanding and forecasting volatility. Among these approaches, the Autoregressive Conditional Heteroskedastic (ARCH) and Generalized Autoregressive Conditional Heteroskedastic (GARCH) models stand out.

The arch model, introduced by Robert Engle in 1982, aims to model volatility during both turbulent and calm periods. It forecasts future changes in variance over time based on past errors, providing valuable insights into macroeconomic and financial phenomena (Engle, 1982; Bollerslev et al., 1994). However, challenges arise due to a non-negativity restriction in the conditional variance formula (Bollerslev, 2023). Building on the arch model, Bollerslev (1986) developed the Generalized Autoregressive Conditional Heteroskedastic (GARCH) model, introducing a more flexible lag point and incorporating features from the arch model and regression analysis. Despite its advantages, the classic GARCH model faces limitations in estimating many coefficients and capturing asymmetric patterns in the second conditional moments, which affect the accuracy of the results (Cappiello et al., 2003).

In response to these limitations, the GARCH-MIDAS model (Engle et al., 2013) emerges as an advanced and less error-prone alternative for volatility modeling and forecasting. By combining GARCH models with Mixed Data Sampling (MIDAS) approaches, this model allows the incorporation of time series data with different frequencies, proving valuable for modeling economic, financial, and environmental phenomena.

While GARCH-MIDAS models have found success in analyzing variables like interest rates, exchange rates, and asset price volatility in developed countries, their application in emerging markets remains relatively unexplored. The unique challenges faced by emerging markets, including data quality issues, structural instability, and nonlinear relationships among variables, require specialized attention (Alper et al., 2012a, 2012b; X. Yu et al., 2021; Salisu et al., 2022; Wang & Li, 2023; Hong et al., 2023).

The GARCH-MIDAS model has been researched and used because it has become a widely accepted tool for forecasting volatility in financial markets, especially in developed economies and in certain emerging markets such as China and the BRICS nations. However, its application in Latin American contexts remains significantly underexplored, despite the region's marked exposure to oil price shocks and macroeconomic uncertainty (X. Yu et al. 2021). In the literature, there is a notable and scarcely addressed gap between developed and emerging countries, particularly visible in Latin America. This difference is explained by the structural dependence on commodity exports, fiscal sensitivity to oil revenues and a marked vulnerability to shocks in international crude oil prices. Economies such as Colombia, Mexico, and Brazil show volatility patterns that differ significantly from those observed in the BRICS countries or in Asian markets, due to factors specific to the region such as political instability, exchange rate fragility, and the asymmetric transmission of global energy shocks (Ratti & Vespignani, 2016).

Although classified as emerging economies, Latin American nations have been systematically underrepresented in empirical applications of mixed-frequency models such as GARCH-MIDAS. This methodological exclusion limits the generalizability of the findings from other regions and restricts the understanding of volatility dynamics within the Latin American context. Several studies have highlighted that incorporating economic, political, and environmental uncertainty variables into forecasting models can significantly improve their accuracy and relevance (Hong et al., 2023; Salisu & Gupta, 2021a, 2021b, 2021c). Consequently, broadening the empirical scope to include Latin America would not only close a gap in the literature, but also strengthen the design of financial policies adapted to the structural characteristics of the region. Likewise, the integration of macroeconomic uncertainty variables such as political risk, climate shocks, and institutional fragility in the volatility forecast remains underdeveloped in Latin American contexts. Expanding the empirical scope to include these dimensions could enhance the explanatory power of GARCH-MIDAS models and support stronger financial decision-making across the region (Galindo & Nuguer, 2023; Blanchard & Galí, 2007; Ayres et al., 2025).

In response to this identified gap, this review underscores the need for future research to systematically apply mixed-frequency models to Latin American financial markets, not only to fill a methodological gap, but also to inform public policy in economies where volatility has direct social and fiscal consequences.

Oil shocks are unexpected events in which the oil price experiences strong growth. This event can be caused by factors such as speculation and uncertainty in the financial market, problems with oil production and demand, armed conflicts, political disagreements, among others (Tumala et al., 2023). Consequently, they have a major impact on the global economy, as this raw material accounts for about 2.5% of the world's gross domestic product (GDP). Oil shocks, therefore, generate fluctuations in economic activity and cause significant disruptions in production systems-such as energy and transportation-as well as in macroeconomic variables, including excessive increases in inflation (World Bank, 2023). Therefore, the investigation of these sharp increases in oil prices has had a significant impact on financial markets.

Now, considering that volatility allows to understand in advance the movement of individual stocks and the stock market (Baum et al., 2010; Bloom, 2009), volatility forecasting has been tested in markets where oil shocks occur, as well as developed countries or some emerging countries such as the BRICS, i.e. Brazil, Russia, India, China and South Africa (Salisu et al., 2022; Salisu & Gupta, 2021a, 2021b).

When analyzing the conceptual structure of the reviewed manuscripts, two clusters were identified (Figure 1). The blue cluster, which groups most of the analyzed documents, refers to concepts such as stock markets, prediction, volatility, exchange rates, energy assets and references to econometric models which include words such as midas and GARCH suggesting that this cluster exposes volatility forecasting as its main focus. The second cluster, identified by the red color, is related to other financial applications different from volatility forecasting including concepts such as variables, portfolio management, macro-economic, among others. In consequence, this cluster is classified as "other applications". This result is expected, considering that the main application of the GARCH-MIDAS model is related to volatility forecasting. However, it is worth noting that there are not specific countries within the clusters.

Figure 1 / Font: Own elaboration.

The bibliometric analysis was performed to gain insights into the trends and concentration of academic production, as well as the authors, collaborations, and countries related to the reviewed documents. Figure 2 shows the number of articles per author within the analyzed period. It can be observed that "F. Ma" is the most productive author, with a total of 12 articles published over a five-year period (2019-2023), placing him first in terms of productivity in this category. In second place, we find the author "A. A. Salisu" with 11 published articles, which demonstrates a high level of productivity. In third place, both authors "R. Gupta" and "L. Wang" are founded with 9 published articles. "C. Liang" has contributed a total of 7 published articles, making him the fourth most productive author on the list. On the other hand, authors "Y. Wei" and "X. Wu" share fifth place in terms of productivity, both with 6 articles published during the analyzed period.

Figure 2 / Font: Own elaboration.

Production per country is shown in Figure 3, which provides an interesting view of the scientific production of articles in several countries. As can be observed, China leads academic production in terms of quantity, with 65 articles and a considerable proportion of them produced in collaboration with other countries. It is worth noting that the following country in terms of articles production is Nigeria, with only 7 articles, which are developed in collaboration with other countries in most of the cases. Turkey and the United States have similar results.

Figure 3 / SCP: Single Country Publications; MCP: Multiple Font: Own elaboration.

While the number of articles produced is valuable information, so is the period in which the articles were produced (Figure 4). As can be seen, the most productive author has been C. Liang with 7 articles, one in 2021, five in the following year, and one in 2023. The second most productive is X. Wu, with 6 articles produced in the same number of years as the previous one. In third place is L. Zhang, with 5 articles.

Figure 4 / Font: Own elaboration.

Regarding the annual scientific production, it can be observed in Figure 5 that it has evolved significantly over the years. From a modest beginning in 2007 with a single article, the number of publications remained at low levels until 2014. However, since 2015 there has been a marked and constant increase in scientific production. In 2023, 31 articles were reached, which reflects a substantial increase compared to previous years.

Figure 5 / Font: Own elaboration.

These results are in line with the average number of citations, which shows an exponential growth with recurrent fluctuations in the years where significant changes were made in the production of annual articles (Figure 6). This continued increase in scientific output suggests sustained interest in the topic and may be related to additional investments in research, technological advances, or a growing community of researchers. These data offer an encouraging view of academic activity and commitment to research in this specific area, with constant and promising growth in scientific production in recent years.

Figure 6 / Font: Own elaboration.

Collaboration between countries to produce articles has become more intense during the last few decades, as this relationship allows the development and research of articles to be more efficient and functional, enriching the knowledge of different cultures and opinions. In this regard, the literature review developed evidences a high level of collaboration between countries, as shown in Figure 7. In the obtained results, it is worth noting the high degree of cooperation between China and various regions, including France, Lebanon, the United Kingdom, United States, Nigeria, and South Africa and Asian countries.

Figure 7 Font: Own elaboration.

Building on the previous discussion, it is essential to highlight the theoretical value of the GARCH-MIDAS volatility forecasting model. Beyond its analytical depth, this approach can be translated into practical tools for a wide range of stakeholders in emerging markets, particularly in Latin American economies. Below are concrete examples, supported by empirical studies, that show how mixed-frequency models enhance real-world decision-making.

In Nigeria and South Africa, Tumala et al. (2023) applied a GARCH-MIDAS framework to separate oil shocks from idiosyncratic market movements. They showed that hedge funds that used these mixed-frequency volatility signals were able to anticipate days of extreme volatility and adjust their futures hedges accordingly, leading to a measurable reduction in the Value-at-Risk of their portfolios during the 2022 oil price rally.

Hong et al. (2023a) demonstrated that integrating a financial uncertainty index into a GARCH-MIDAS model significantly improved early warning signals about systemic stress in China's A-share market. By incorporating daily volatility forecasts into its macroprudential dashboard, the People's Bank of China was able to preemptively adjust liquidity requirements, thereby mitigating near-term market turbulence.

Tumala et al. (2023) demonstrate how disentangled oil shocks can be used to assess stock market volatility in emerging economies such as Nigeria and South Africa. Using the GARCH-MIDAS framework, they integrate low-frequency macroeconomic indicators with high-frequency financial data to produce more accurate volatility forecasts. This approach offers valuable insights for designing countercyclical fiscal frameworks in oil-dependent economies, where volatility thresholds can guide sovereign stabilization fund transfers and improve debt service performance during periods of oil price stress.

Perilla (2023) expanded the GARCH-MIDAS approach by incorporating monthly series of co₂ emissions along with daily oil prices to assess the links between climate and finance in Latin America. This hybrid model provided development agencies and central banks with prospective risk assessments on green energy investments under volatile oil markets.

These examples highlight the versatility of mixed-frequency volatility models. Whether calibrating dynamic hedges, strengthening macroprudential supervision, designing countercyclical policies, or assessing environmental-financial risks, GARCH-MIDAS offers a data-driven compass for navigating uncertainty in emerging economies.

The three authors contributed equally to: conceptualization, methodology, investigation, data curation, formal analysis, project administration, visualization, writing - original draft, writing - review & editing.