Modeling long-term volatility memory dynamics in the Colombo Stock Exchange Open Access

Studying long-term memory in volatility dynamics is crucial in financial econometrics and asset pricing. Long-term memory refers to the persistence of past market behavior influencing future trends over extended periods, where the impact of shocks diminishes gradually rather than disappears quickly. This persistence is significant for understanding how markets process and reflect information over time. Volatility, defined as the degree of variation in asset prices, often exhibits clustering, where periods of high volatility are followed by high volatility, and stability follows periods of low volatility. These cyclical and uneven market fluctuations are critical for risk management, portfolio optimization, and derivative pricing. Understanding these dynamics enables better predictions of market behavior and optimized asset allocation for investors, reducing financial risk (Ballinari, Audrino, & Sigrist, 2022). Emerging markets, including Sri Lanka, often exhibit higher volatility and different dynamics than developed markets (Madurapperuma, 2023). Long-term dependencies on volatility challenge the notion of market efficiency, suggesting that past information may not always be fully reflected in current prices (Christensen & Nielsen, 2007). Such dependencies indicate market inefficiencies or structural breaks, warranting further examination (Kirman & Teyssière, 2001).

Although there is a vast amount of literature on volatility modeling in developed and emerging markets, applying advanced econometric techniques to frontier markets is limited. Unlike previous volatility studies in emerging markets, the Sri Lankan context offers unique characteristics that demand different modeling approaches: (1) a distinctive dual-index system that captures both broad market and blue-chip behaviors, (2) recent structural transformations through market liberalization, and (3) exposure to consecutive systemic shocks - a combination not previously examined in volatility literature. According to the Efficient Market Hypothesis (EMH), asset prices ought to fully reflect all available information (Fama, 1970), but volatility persistence in frontier markets is inconsistent with conventional models (Tripathy, 2022) and requires methodology modification. These patterns are essential in constrained markets where traditional models miss episodic volatility bursts and persistence. The Sri Lankan stock market provides excellent research material for examining frontier markets’ behavior because of its distinctive evolution and current situation. Economic reforms and foreign investment growth created disruptions that changed how stock market volatility behaved (Rathnasekara, 2022). More critically, it has faced unprecedented system shocks: a global COVID-19 pandemic followed by a severe food, energy, and sovereign debt crisis. The back-to-back crisis sequence in this emerging market enables exclusive examination of multiple system failure impacts, which has received scant attention throughout existing financial literature. Using a FIGARCH modeling approach, this paper is the first to apply it in the Sri Lankan stock market within the context of back-to-back systemic shocks, providing novel insights into how dual crises influence volatility persistence in frontier markets. Research on frontier market dynamics focuses on the Sri Lankan stock market because of its unique structure and recent market developments. This research investigates two sequential systemic shocks because earlier studies analyzed crises one at a time, thus providing valuable insights into market recovery strategies and adaptation to multiple stress events.

An analytical framework, previously absent in other markets, emerges from the current study due to its focus on the dual-index system of the Colombo Stock Exchange (CSE). The dual-index approach allows new market dynamic investigations that single-index studies fail to reveal, especially in different market segment responses to systemic shocks. The ASPI tracks general market patterns by including illiquid stocks and small-cap shares, enabling analysis of the whole financial ecosystem. The S&P SL20 Index highlights only the top 20 most liquid stocks, enabling institutional investor behavior analysis (Colombo Stock Exchange, 2021). The study demonstrates how comparing specific market indices during system-wide disturbances reveals distinctive market division patterns in frontier markets, which helps clarify market microstructural dynamics. The FIGARCH framework provides ideal tools for analyzing complex market dynamics because it offers models for short-term volatility clustering and long-memory dependencies together with fractional integration while serving markets where traditional models cannot handle the interplay between structural constraints and volatility persistence (Boubaker, Saidane, & Ben Saad Zorgati, 2022). FIGARCH models now serve developed markets with expertise, but studies of their potential during multiple systemic crises in frontier markets have not yet been conducted extensively.

This study presents a framework for studying consecutive systemic shocks in structurally constrained markets, demonstrating how FIGARCH modeling captures unique volatility dynamics that emerge under compounded crises. In addition, it shows how segmentation patterns within dual index systems can be applied to other developing financial markets. Given the global structural constraints in frontier markets (limited liquidity, restricted capital access, and a retail-dominated investor base), these findings are particularly important for quantifying volatility’s impact on market stability in such environments. The findings provide actionable insights for institutional investors to develop adaptive portfolio optimization and risk management strategies, particularly during systemic stress. This research has both theoretical and practical implications. The findings suggest refining existing models for financial theorists to better account for frontier market dynamics, particularly during consecutive systemic shocks. They provide insights for practitioners in structurally constrained risk management and portfolio optimization markets. They offer evidence-based guidance to policymakers, especially in frontier markets, on developing market stability initiatives that consider the idiosyncratic characteristics of these markets. These contributions provide an additional theoretical and practical understanding of managing frontier market volatility, laying the foundation for future research into the complex dynamics of structurally constrained markets. In addition to filling critical gaps in volatility modeling, this research establishes a path for future research to understand better the complex dynamics of frontier markets operating under systemic crisis/stress. This study establishes a novel methodological and theoretical foundation that paves the way for future research and provides avenues for understanding and managing the complexities of frontier market volatility.

Stock market volatility is one well-researched aspect, and there have been spates of empirical studies to capture the essence of stock return series based on clustering, persistence, and long memory dynamics characteristics. Niu and Wang (2013) tested long-memory and volatility clustering in financial time series and found that previous market events helped to influence future volatility significantly. The resulting outcome of the study revealed that financial time series exhibit both volatility clustering and long-term memory, indicating that market shocks have prolonged effects. Also, according to Oh, Kim, and Eom (2008), long-term memory is strong, and volatility clustering is great in high-frequency trading. This means that long-term dependencies in high-frequency trading strategies must be accounted for. Additionally, Cerqueti and Mattera (2023) utilize fuzzy clustering techniques to analyze financial time series with time-varying memory, providing a deeper understanding of how volatility evolves. The approach highlights the significance of considering the change in time memory when studying financial data. Tripathy (2022) highlighted the presence of long-memory dynamics in stock markets of BRICS countries and Barkoulas, Barilla, and Wells (2016) foreign exchange rates in the euro era, suggesting that volatility exhibits patterns persisting over time, indicating some degree of predictability. Further, Onyele and Nwadike (2021) showed the characteristics above by picking out clustering, persistence, risk-return tradeoff, and asymmetric news effects in the stock return series.

Regional studies have also contributed significantly to our understanding of stock market volatility. Tamilselvan, Palamalai, Kumar, Aswathaman, and Veerabhadrappa (2022) studied the Indian stock markets and confirmed market efficiency, persistency, and asymmetric effects as the main features of volatility. Studies by Bahadur (2009) on the Nepalese stock market and Catalina (2019) on the Bucharest Stock Exchange have also reflected time-varying volatility, clustering, and high persistence, making common underlying mechanisms ducting across markets possible. Similarly, Ali (2019) also reported the same characteristics in the Khartoum Stock Exchange, which may indicate that these volatility traits could be common characteristics in emerging markets. Das and Debnath (2022) revealed the impact of the COVID-19 pandemic on spillover for both the short and long run, considering foreign and Indian stock markets.

The long-memory effects of volatility evolution have been of important interest. Aminimehr, Raoofi, Aminimehr, and Aminimehr (2022) pointed out that the long-memory effects can critically affect market behavior and investment achievements. Comparative studies have focused on the developed and emerging stock market dynamics to reveal long-memory effects in the markets of different economies like China (Bhattacharya, Bhattacharya, & Guhathakurta, 2018). Researchers have drawn attention to the regime-dependent effects on the dynamics of stock market returns of SAARC countries, predominantly focusing on the presence of permanent volatility during bulls and the impact of currency returns on the stock market behavior (Ahmed & Mustafa, 2019). Furthermore, the studies examining the behavior of stock markets in the post-liberalized era of South Asia have provided insights into the integration patterns and volatility spillovers across the region (Gunasinghe, 2005).

While global studies have extensively examined long-memory dynamics using advanced models, several important gaps remain. First, there is limited exploration of these dynamics in smaller emerging markets, which may exhibit distinct patterns due to their unique market structures and economic conditions. Second, advanced models capable of accounting for long-term dependencies and structural breaks remain underexplored in smaller emerging markets like Sri Lanka. Third, most studies do not examine how volatility patterns evolve across distinctly different economic regimes (normal, pandemic and crisis periods). These insights from global market dynamics provide a crucial foundation for understanding how similar phenomena might manifest in emerging markets like Sri Lanka, characterized by unique structural and behavioral factors.

These studies on volatility and stock returns have evoked much interest and debate, even in the Sri Lankan stock market arena, providing new knowledge in the dynamics and real practical implications for market players and potential use by policymakers. Perera and Ediriwickrama (2020) conducted a study on the effect of idiosyncratic volatility on stock returns in Sri Lanka by using the five-factor asset pricing model, realized positive average stock returns and idiosyncratic volatility, therefore giving an insight into how volatility affects returns in the Sri Lankan stock market. They have given a theoretical extension in his discussion relating to the dynamics and behavior of the individual securities in the market. Sujenthini and Wijesinghe (2022) captured the effects of the exchange rate method on stock returns in CSE, as expressed using the GARCH approach. It derives the co-movement between the exchange rate and stock market performance, which is priceless information that could help investors and policymakers understand and mitigate the impacts of external economic factors. Further, Withanage and Jayasinghe (2017) tried to capture volatility spillovers among stock markets by investing in the BEKK-GARCH model of Sri Lanka, India, and Pakistan. This research states how the South Asian markets are interlinked and mutually influenced by each other’s volatility dynamics, which leads to an understanding of a better regional financial landscape.

Morawakage and Nimal (2016) further found the presence of volatility clustering and leverage effects in the Colombo Stock Exchange. This phenomenon aligns with findings in other emerging markets where high volatility periods are followed by more of the same, creating a feedback loop that investors must consider. Inefficiency and herding behavior have also been identified with behavioral finance factors in the Sri Lankan stock market. Herding behavior and inefficiencies, as found during the study by Hamidon and Kehelwalatenna (2020), have shown that the results are aligned with traditional finance theories and empirical results. Such pieces can be created within the context of vital influence on market volatility and investor performance. Dayaratne and Lakshman (2010) examined the impact of significant events on stock market volatility and found that CSE exhibits increased volatility during major local and international events.

Volatility spillovers between such stock markets of various Asian economies have been found, with strong evidence of the volatility transmission mechanism operating particularly between Sri Lanka, India, China, and Japan (Jebran & Iqbal, 2016). These findings have made the role of the Sri Lankan stock market one that many researchers are interested in looking at in the context of other Asian markets. Considering the implications of cointegration, correlations, and information spillovers, it implies substantial interaction with other markets, and this has been given prominent attention in the existing literature (Kuruppuarachchi, 2016). Exchange rate volatility and stock market return volatility studies concerning the emergent markets, notably the Colombo Stock Exchange, have lent validity to the relationship (Perera, 2016). Jameel and Teng (2022) find that volatility shocks in USD exchange rates and the ASPI persist over time and vary across different structural break periods. Cross-correlation analyses of stock returns in SAARC nations have assessed volatility patterns and tested the efficient market hypothesis (Singhania & Prakash, 2014). These limitations in methodology and focus discussed further in Section 2.3.4, highlight the need for more advanced approaches to understanding the dynamics of the Sri Lankan stock market.

Moreover, research on stock market development and economic growth in Sri Lanka has explored the empirical relationship between these factors, highlighting the importance of a well-functioning stock market for economic progress (Kengatharan & Vanajah, 2021). Jaleel and Samarakoon (2009) found that stock market liberalization significantly increased return volatility in CSE. Fernando (2018) finds that macroeconomic factors significantly impact stock market returns and volatility in CSE, with economic indicators such as inflation and interest rates contributing to increased market volatility. Wang, Gunasekarage, and Power (2005) find that CSE experiences significant volatility spillovers from the US stock market, highlighting the influence of global factors on the volatility of emerging markets like Sri Lanka. Recently, Riyath, Dewasiri, Siraju, Jahfer, and Sood (2024) found that stock price volatility in CSE is significantly influenced by both internal shocks and external volatility spillovers from markets such as the USA, indicating the critical role of global market dynamics in shaping local market behavior. These studies underscore the need for advanced methodologies to better capture volatility dynamics in Sri Lanka, as discussed in Section 2.3.4.

A theoretical framework of fractional integration is widely regarded as suitable for analyzing long memory in volatility dynamics (Granger & Joyeux, 1980). It asserts that long-range dependencies characterize some time series processes. That is, the current value depends on recent observations and observations in the remote past. This directly contrasts with traditional time series models, which assume the exponential decay of shocks over time (Zhu, Lee, & Hwang, 2010). Fractional integration provides for hyperbolic decay, allowing for shocks to have persistent effects, frequently witnessed in financial data (Baillie, 1996). Long–memory dynamics in financial markets frequently represent persistent volatility patterns over long periods. To tackle these problems, Baillie (1996) proposed the FIGARCH model, which uses fractional differencing in the GARCH setup. Short-term clustering and long-term dependency can be identified by the FIGARCH model, which is suitable for the Sri Lankan context where structural shifts and multiple economic regimes are present (Tripathy, 2022; Samarawickrama & Pallegedara, 2023). Using this methodological approach, we can gain useful insight into indices like ASPI and S&P SL20, which assist in better risk management and policy-making under all economic conditions (Stamos, 2022).

Perera (2022) and Hewamana, Siriwardhane, and Rathnayake (2022) refer to idiosyncratic volatility as responsible for the return volatility clustering of listed firms at the CSE, noting that firm-specific risks bring about persistent periods of high and low volatility. Similarly, Nguyen, Hoang, and Truong (2021) discuss the use of the GARCH model in capturing conditional volatility and volatility clustering, indicating that these phenomena are prevalent in financial data. According to Liu and Zhang (2015), economic policy uncertainty significantly raises market volatility and contributes to the clustering of return volatilities as markets adjust to policy changes. According to Balli, Basher, and Jean Louis (2013), volatility persists and clusters over time because of sector-specific news and events that determine these sector equity markets. Maiti (2019) observes that standard asset pricing theory overlooks idiosyncratic risk and may fail to capture this source of return clustering. Bansal, Kiku, Shaliastovich, and Yaron (2014) discuss how macroeconomic conditions, such as economic growth and inflation, influence return volatility clustering as markets respond to changes in economic variables. These studies support the hypothesis that equities on CSE significantly exhibit return volatility clustering due to various factors, including idiosyncratic volatility, sector-specific events, economic policy uncertainty, and broader macroeconomic conditions.

A growing body of evidence already supports the hypothesis that equities in the CSE exhibit significant return volatility persistence. Perera and Ediriwickrama (2020) found a positively significant link between average stock returns and idiosyncratic volatility in the CSE, indicating the persistence of volatility. Tsuji (2018) examined equity return volatility persistence in China and Japan, finding that volatility persistence was evident until structural breaks were accounted for, suggesting a complex interplay of factors influencing volatility persistence. Chkili and Nguyen (2014) investigated the movements in exchange rates and stock market returns in a regime-switching framework, and this could provide important insights for similar dynamics in return volatility within the CSE. Perera (2022) also analyzed the five-factor model and idiosyncratic volatility in Sri Lanka’s stock market, offering additional evidence for the factors influencing stock returns and their persistence. These studies support the hypothesis that equities exhibit return volatility persistence in CSE.

Different studies have shown that long-memory return volatility characterizes equities. Again, considering structural breaks, the study of Chatzikonstanti and Venetis (2015) shows that long-memory dynamics in the persistence of equity return volatility stood eliminated. Long memory in equity volatility is also illustrated by Garvey and Gallagher (2011), Quoreshi, Uddin, and Jienwatcharamongkhol (2019) Hiremath and Kumari (2015), giving reinforcements to the persistent nature of return volatility. Manap and Kassim (2011) are strong proponents that, even though information is known to arrive on a short-term memory basis, the conditional variance of stock returns stands out as a candidate for long memory characteristics, supporting the presence of long memory in equity market volatility. Bhattacharya et al. (2018) and Emmanuel, Kyei, and Gill (2016) suggest that portfolios with higher long-memory parameters tend to have higher expected returns with lower risk, highlighting a relationship between long memory and investment performance. Again, Ali, Tiwari, and Raza (2017) gave another empirical test of the long memory characteristics in return and volatility relationships in the Dhaka Stock Exchange and thus remarked on the relevance of the long memory features dynamics in the stock markets. These studies provide evidence to support the hypothesis that equities in the CSE exhibit significant long-memory return volatility, emphasizing the persistent nature of return volatility in the stock markets.

While the preceding sections highlight critical aspects of volatility clustering, persistence, and long-memory characteristics, several methodological gaps persist in the literature. Existing GARCH-family models effectively capture short-term clustering but fail to address long-term dependencies and structural breaks, particularly in emerging markets like Sri Lanka (Perera, 2022; Hewamana et al., 2022). Second, the Sri Lankan market’s frequent economic and policy changes necessitate models that account for structural breaks and regime shifts. Third, comparative analyses that compare responses across much larger indices (ASPI) versus blue-chip indices (S&P SL20) under normal and stress periods to yield conclusions on segmentation effects in the marketplace are missing. These gaps are addressed in this study by three key contributions. First, this study applies the FIGARCH model, which integrates long-memory dynamics into the GARCH framework to jointly capture both short-term clustering and long-term persistence. Second, it explores these dynamics under various economic regimes to better characterize how market stress affects volatility patterns. Third, it delivers a comparative analysis of the ASPI and S&P SL20 indices, identified segmentation effects, and their relevance in risk management and policy-making in emerging markets. The implications of these findings are practical for portfolio risk management, designing economic policies, and enhancing financial stability in countries like Sri Lanka.

Daily equity index data for this study is collected from the Colombo Stock Exchange (CSE) official website. This data pertains to daily closing prices on the All Share Price Index (ASPI) and the S&P SL 20 Index (SNP20) over the Sri Lankan stock market. The study period covers the timeframe from January 4, 2012, to April 30, 2024 (the “Full sample”). After that, the full sample is divided into three different subsamples based on various economic conditions of Sri Lanka. First, the “Normal” period reflects the normal economic conditions from January 2, 2012, to December 31, 2019, set as a reference period to give the behavior of the market under usual economic conditions. Second, the “COVID-19” period covers from January 3, 2020, to December 31, 2021, and reflects what has happened during the COVID-19 period of disruptions. This period begins shortly before Sri Lanka’s first confirmed COVID-19 case (a Chinese tourist tested positive) on January 27, 2020, with the timeline aligned to the global recognition of the pandemic in late 2019. Third, the “Economic Crisis” period runs from January 4, 2022, to April 30, 2024, showing how the market behaves when the economy is in a crisis. The starting point of this period aligns with severe economic distress, such as shortages of milk powder, gas, and fuel reported in early January 2022 due to foreign exchange shortages. The study comprehensively analyses the Sri Lankan stock market under varying economic conditions by dividing the sample into these distinct periods. Each period reflects different stress levels and external influences, offering a holistic view of market volatility dynamics. More importantly, choosing these cases is deliberate and adds methodological robustness to the paper because the checking will be based on the effects of significant economic occurrences on market dynamics. These periods were identified based on widely recognized events to ensure objective delineation, with sensitivity analyses confirming robustness across alternative date ranges. Potential selection bias was minimized by using publicly documented economic events rather than market performance to determine period boundaries, and the results remained consistent when testing adjacent months as alternative cut-off points.

Daily returns are computed using the natural logarithms of the ratio of the present-day index to the previous-day index, mathematically expressed in Equation (1):

Where R_t represents the natural log of daily stock return; Ln is the natural log operator; P_t represents the stock market index at time t; P_t-1 represents the stock market index at time _t–1. The Augmented Dickey-Fuller (ADF) test ensures that the data is stationary (Dickey & Fuller, 1979). The ADF uses equation II to test the nature of the stationary of the data series.

The ARCH family model consists of two sets of equations: The mean and variance. The autoregressive moving average (ARMA) is often used as a regressor/explanatory variable for the dependent variable in the ARCH family models’ conditional mean equation to capture the autocorrelation in the data before modeling the conditional heteroskedasticity (Engle, 1982). ARMA-GARCH models effectively capture volatility clustering, which is a common characteristic of financial time series where periods of high volatility are followed by high volatility, and periods of low volatility are followed by low volatility. This is evidenced by the significant use of ARMA(1,1)-GARCH(1,1) models in various studies to simulate and explain the volatility clustering observed in financial markets (Zeng, Shi, Tian, & Zheng, 2008; Kim, 2022). The mean equation of the ARMA-GARCH model specification is given in Equation III.

where $η0$ is constant; $η1$ is the coefficient of the autoregressive lag 1; $η2$ is the coefficient of the moving average lag 1; $εt$ is the error term. The ARCH model can capture volatility clusters in time series data (Engle, 1982). The ARCH model is given in equation - IV.

Where, ω is a constant, $δt$ is a parameter capturing the impact of past squared residuals on the current variance, $vt$ is the disturbance term in the variance equation. The ARCH model removes heteroskedasticity from the mean equation but has limitations in capturing long volatility clustering sequences. Bollerslev (1987) addresses this flaw by introducing the GARCH model. In this framework, both squared residuals and lagged conditional variances are used to describe the variance of the present time. It was developed to build upon the ARCH model to integrate past conditional variances into the equation and help capture the persistence of volatility. The estimation of the GARCH model of volatility in time series is given by Equation - V:

Where $σt2$ is the conditional variance at time t, $γj$ is a parameter that captures the influence of past conditional variances on the current variance, and p, q are the orders of ARCH and GARCH terms. To incorporate long memory in the GARCH framework, Baillie, Bollerslev, and Mikkelsen (1996) extended the GARCH model into the Fractionally Integrated GARCH (FIGARCH) model. It modifies the GARCH structure by including a term for fractional differencing $(1−B)d$⁠, in which B is the backshift operator, and d is the fractional differencing parameter that captures the long-memory feature. This fractional differencing term, $(1−B)d$ is a function involving past shocks and variances, allowing for hyperbolically decaying rates of shock impacts rather than the exponential decay seen in GARCH. Finally, the FIGARCH model captures the short-run and long-run dependence on the volatility. This is modeled using Equation VI:

Where, $ϕ(B)=1−γB$ represents the polynomial incorporating the GARCH terms, $θ(B)=1+δB$ represents the polynomial incorporating the ARCH terms.

More generally, this evolution from GARCH to FIGARCH, in particular involving fractional differencing, gives a more flexible and generalized framework within which the model can characterize the slow decay in the auto-correlation function of squared returns, which is typical in financial time series that exhibit long memory for volatility. Fractional differencing, represented by the $(1−B)d$ term in the FIGARCH model shows that the FIGARCH model gives a more flexible and parsimonious process of representing the decay of shocks to volatility than its GARCH counterpart. In general, then, fractional differencing offers the needed general framework for the FIGARCH model to relate and interrelate to almost any number of lag structures that are possibly present in the conditional strategy and, by so doing, offers a very robust model to be used by the financial econometricians in the analysis and forecasting of volatility in the financial markets. This study uses the ARMA (1,1)-FIGARCH as the best-fit model (Zeng et al., 2008; Kim, 2022; Riyath et al., 2024). Hence, the parameters of the FIGARCH model are estimated using ML under the normal distribution assumption. The Broyden-Fletcher-Goldfarb-Shanno algorithm and Marquardt steps are used to maximize the likelihood function (Chan & McAleer, 2002). The covariance of coefficients is calculated using the outer product of gradients. In order to backcast the pre-sample variance, a parameter of 0.7 is used.

The selection of ARMA-FIGARCH over alternative volatility models is guided by several methodological considerations specific to emerging and frontier markets like Sri Lanka. While other GARCH variants like EGARCH and TGARCH can capture asymmetric effects, and standard GARCH models handle short-term clustering, the FIGARCH framework uniquely accommodates both the long-memory characteristics and structural shifts that are particularly pronounced in emerging markets (Riyath & Aldabbous, 2024). The ARMA(1,1) specification was chosen based on Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values, which indicated optimal fit compared to higher-order specifications for the Sri Lankan market data. This choice also aligns with parsimony principles while capturing the essential autocorrelation structure in returns. The FIGARCH component’s fractional differencing parameter allows for more flexible modeling of persistence patterns, which is particularly relevant given the CSE’s documented history of prolonged adjustment to market shocks. This flexibility is crucial when analyzing market behavior across distinctly different economic regimes, as demonstrated by the model’s superior performance in capturing both the immediate impact and lingering effects of the COVID-19 pandemic and subsequent economic crises.

Figure 1 plots the All Share Price Index (PASPI) and the S&P SL20 Price Index (PSNP) over three subsample periods, indicating the behavior of the market with the changing economic environments. Taking the normal period from 2012 to 2019, both indices have an upward trend, meaning that the market has developed. However, the PASPI exhibited more fluctuations than PSNP, suggesting the broader market may be more sensitive to fluctuations. The COVID-19 pandemic wielded substantial influence over both indices during 2020, which both registered steep declines since the onset of the market shock. Fast forward to the end of 2021, PASPI sprung back to break past previous peaks and form new peaks beyond 12,000, while PSNP rebounded but did not recover above previous pre-pandemic highs. This shows how PSNP’s top 20 companies are weaker in the face of or responsive to the pandemic versus PASPI. The two indices declined initially and then recovered through the stressful periods. However, PASPI bounced strongly from its last peak to a high level, indicating strong resilience or a positive response to adversity. In contrast, PSNP showed a premised recovery but remained below peak levels, suggesting a more severe impact or slower recovery. Such contrasting PASPI and PSNP behaviors across economic phases require using more than one index to study financial markets from different views.

As shown in Table 1, from the full sample period, the findings indicated that the mean and median of the daily return values of RASPI were approximately zero, at 0.0243 and 0.0078, respectively. On the other hand, RSNP did not show such results, with its mean value being 0.0059 and median of 0.0085. The standard deviation of RSNP, which was higher than RASPI and RSNP, at 1.2178 compared to 0.9989, reflected a greater volatility by the top 20 companies. Figure 2 provides a visual representation of these return patterns across all three periods, clearly showing the increased volatility during the COVID-19 and crisis periods compared to the normal period. The plot effectively illustrates the more extreme fluctuations in both indices during stress periods, with RASPI showing particularly notable spikes during market transitions. In normal conditions, both indices show relative stability in the market from 2012 to 2019. RASPI shows an average return close to zero (0.0005) and a −0.0086 median. RSNP similarly shows an average return close to zero (−0.0028) and −0.0095 median. Similarly, the lower standard deviations for both indices (RASPI = 0.5825, RSNP = 0.6573) reflected a low volatility. The high mean return (0.2365) and high median (0.2898) for RASPI during the 2020 year show significant daily gains, although with substantial volatility due to the onset of the COVID-19 pandemic. RSNP also had high returns; the mean was 0.1793, and the median was 0.2385. High standard deviation is accounted for by two indices: RASPI = 1.4654, RSNP = 1.4762. During the economic crisis period of 2022–2024, the mean return for RASPI dropped to 0.0020, and the median is 0.0246, indicating low positive returns. RSNP experienced negative returns (mean = −0.0250, median = −0.0057). That indicated high volatility for RASPI, with a standard deviation of 1.5794, while RSNP’s volatility was higher, with a standard deviation of 2.0144.

Table 2 reports the ADF test concerning the stationarity of the daily returns (natural log) of the All Share Price Index (RASPI) and the S&P SL20 Price Index (RSNP). It is quite evident from the results that in the most stringent terms, across all model specifications, namely, with constant, with constant and trend, and without constant and trend, the null hypothesis of a unit root can be strictly rejected at the 1% significance level. These results suggest that the daily returns for both indices are stationary, facilitating accurate financial modeling and forecasting (Dickey & Fuller, 1979).

As presented in Table 3: Panel A, Across the full-sample period, the mean equation for both the ASPI and the S&P SL20 Index were not statistically far from zero, mainly demonstrated by the non-significance of the constant terms (ASPI: β = 0.0063, p = 0.6774 and S&P SL20: β = 0.0095, p = 0.6172). However, both indices exhibited a high return persistence, of which only a positive autoregressive (AR(1)) effect was detected clearly, with coefficients being highly significant for ASPI: (β = 0.5175, p < 0.0001) and S&P SL20: (β = 0.4367, p < 0.0001). A comparative analysis across the full sample reveals that a one-unit shock to ASPI volatility has a 51.75% carryover effect to the next period (β = 0.5175), compared to 43.67% for S&P SL20 (β = 0.4367). This quantitative difference suggests that broader market volatility takes approximately 1.2 times longer to dissipate than blue-chip stock volatility, a crucial consideration for portfolio rebalancing timing. Moving average, the MA(1) component for the two indices also differed. The ASPI had a significant negative coefficient (β = −0.2814, p < 0.0001), but the S&P SL20 had a less pronouncedly significant negative coefficient (β = −0.1902, p = 0.0123).

The hypothesis of significant return volatility clustering in equities listed on the CSE was tested using ARCH terms. For the ASPI, the significant ARCH term (β = 0.0874, p = 0.0184) indicates the presence of volatility clustering, supporting the hypothesis. This evidence is consistent with both local (Perera & Ediriwickrama, 2020), and international studies (Niu & Wang, 2013; Oh et al., 2008). The detected ARCH on the S&P SL20 was insignificant instead (β = 0.0932, p = 0.1217), thereby suggesting evidence of return volatility is less supportive of the hypothesis of return volatility clustering for this index.

We tested the conjecture, confirming that equities in the CSE have significant return volatility persistence. Both ASPI (β = 0.5478, p < 0.0001) and the S&P SL20 (β = 0.3954, p < 0.0001) have shown significant GARCH terms, thus confirming volatility persistence. This complements the hypothesis in both indices, meaning that past volatility significantly influences current volatility. These findings support the results of Perera and Ediriwickrama (2020) as well as international studies by Tsuji (2018) and Chkili and Nguyen (2014), which also showed evidence of volatility persistence in financial markets.

We tested the hypothesis that equities in the CSE exhibit significant long-memory return volatility using FIGARCH terms. We observed strong evidence supporting the hypothesis since FIGARCH terms are highly significant for both the ASPI (β = 0.6480, p < 0.0001) and the S&P SL20 (β = 0.4333, p < 0.0001), supporting substantial long-term memory in volatility. This implies that volatility shocks have a persistent and long-run effect. These results resonate with studies conducted by Chatzikonstanti and Venetis (2015), Hiremath and Kumari (2015), Zhang and Shi (2023), and Chan and McAleer (2002), which found the existence of long-memory dynamics in the volatility of stock markets.

The results from our study have extensive implications for return volatility clustering and persistence in long-term memory volatility for the Colombo Stock Exchange to investors and policymakers. The presence of substantial return volatility clustering for the ASPI makes its equities in need of firm risk management strategies to be able to adjust during periods of high volatility, and the policymakers need to come up with measures that capture these patterns in the regulatory framework to enhance the stability of the market. The lack of substantial clustering for the S&P SL20 makes it a more stable investment alternative during periods of market volatility. The observed volatility persistence in both indices underscores the importance of historical volatility in forecasting future market behavior, and guiding investors in informed decision-making and risk assessments. At the same time, policymakers should recognize the lasting impact of volatility shocks in their efforts to sustain financial stability. Additionally, the significant long-memory effects indicate that volatility shocks have enduring impacts, highlighting the value of long-term historical data in predicting future volatility. Therefore, investment strategies must incorporate these long-term volatility patterns to optimize portfolio performance, whereas, for policymakers, the focus should remain on the persistent nature of volatility in reflecting market stability through their regulatory interventions.

Table 3: Panel B indicates that the CSE exhibited distinct volatility characteristics across its indices over the normal period. The All Share Price Index displayed marginal evidence of volatility clustering, evidenced by the near-significant ARCH term (β = 0.2057, p = 0.0534). It follows that periods with high volatility are succeeded by similar periods, though this pattern does not show a strongly pronounced character. As visualized in Figure 3, both indices show moderate and comparable coefficients for clustering, persistence, and long memory during the normal period, with values generally ranging between 0.14–0.42. This balanced pattern in the coefficient values reflects the relative stability of market conditions during this period. The S&P SL20 index did not confirm this hypothesis since the ARCH term was insignificant (β = 0.1425, p = 0.2055), reflecting weaker evidence of volatility clustering. However, strong evidence of volatility persistence was found across the two indices. Significant GARCH terms for ASPI (β = 0.4164, p = 0.0019) and S&P SL20 (β = 0.3456, p = 0.0036) unmask that past volatility significantly affects current volatility. This aligns with broader financial literature, confirming that volatility shocks have enduring effects during stable economic periods. Both the indices evidenced the hypothesis of long-memory volatility, as significant FIGARCH terms were estimated for ASPI (β = 0.3758, p < 0.0001) and S&P SL20 (β = 0.3204, p < 0.0001). The influence of shocks on volatility reveals that the persistence of such an influence is spread over a long period. In the case of the ASPI, where we have marginal clustering but strong persistence and long-memory effects, investors should be prepared to face periods characterized by increased volatility. On the contrary, the S&P SL20 does not show any clustering, but the persistence and long-memory effects point to a more stable investment vehicle, although historical volatility is very important for risk management.

Table 3: Panel C indicates that the COVID-19 pandemic has dramatically altered the volatility dynamics at the CSE. The ASPI did not show clustering, as evidenced by an insignificant ARCH term (β = −0.0159, p = 0.8787). However, the S&P SL20 exhibited strong clustering, indicated by a highly significant ARCH term (β = 0.9463, p < 0.0001), suggesting more pronounced periods of volatility for the top 20 companies during the pandemic. Figure 3 clearly illustrates these dramatic shifts during the COVID-19 period, showing the highest coefficient values across all periods, with ASPI’s long memory reaching 1.145 (p < 0.0001) and SNP20 displaying strong clustering (0.9463, p < 0.0001) and persistence (0.7902, p < 0.0001). This stark contrast in coefficient patterns between ASPI and SNP20 during COVID-19 is particularly notable when compared to other periods. Both the indices supported the hypothesis of the persistence of return volatility in the pandemic period. The ASPI GARCH term (β = 0.8304, p < 0.0001) and the GARCH term for S&P SL20 (β = 0.7902, p < 0.0001) were highly significant, confirming that past volatility significantly influenced current volatility and the implication of a lasting impact of the pandemic on the market conditions. This finding is consistent with local existing studies (Adikari & Buddhika, 2023; Samarawickrama & Pallegedara, 2023). In this case, the hypothesis of long-memory volatility is supported for the ASPI but not for the S&P SL20. The rather significantly large FIGARCH term for the ASPI (β = 1.1450, p < 0.0001) shows strong long-term memory in volatility; therefore, more consistent shocks in volatility had a substantial effect on the broader market. On the contrary, the insignificant FIGARCH term for the S&P SL20 (β = 0.0099, p = 0.7760) shows weak long memory in the top 20 companies, meaning very weak persistence in volatility shocks for these companies. In the case of the ASPI, weak evidence of clustering, but strong persistence and long-memory effects were evident. While immediate volatility might not be clustered, its impact would be long-term. Hence, any investment strategies should consider this long-term impact. The S&P SL20 showed strong clustering and persistence but weak long-memory effects, which signifies that the volatility was more immediate and less persistent for that duration, making short-term volatility management very important for investors.

Table 3: Panel D indicates that neither index supported the significant return volatility clustering hypothesis during the economic crisis period. Weak evidence was also found for volatility clustering that pinged in the insignificant ASPI (β = −0.3416, p = 0.1051) and S&P SL20 (β = −0.4366, p = 0.1098) ARCH terms during the period of economic turmoil, thus suggesting that in these periods, the predictability of volatility patterns reduces. As shown in Figure 3, the crisis period exhibits a unique pattern where both indices show negative clustering coefficients while maintaining strong long-memory effects (approximately 0.5–0.56). This visual representation helps illustrate the erratic nature of volatility during the economic crisis, particularly evident in the weak persistence coefficients. The hypothesis of return volatility persistence was also not supported for either index during the economic crisis. The GARCH terms for ASPI (β = 0.0587, p = 0.8157) and S&P SL20 (β = −0.1078, p = 0.7192) were insignificant, indicating weak persistence. This indicates that weak evidence was present for this hypothesis. This result indicates that the nature of the economic crisis could have been such that these volatility shocks had no long-lasting effects, and the situation could have even become erratic. Despite the lack of support for volatility clustering and persistence, the hypothesis of long-memory volatility was supported for both indices.

At the same time, the results for the FIGARCH terms illustrated that the ASPI (β = 0.5640, p < 0.0001) and the S&P SL20 (β = 0.5013, p < 0.0001) had significant FIGARCH terms, thereby showing that the long-term effects even during the economic crisis period increased the persistence effect created by these shocks on market conditions. The long-memory effects are strong in the economic crisis period, while both the ASPI and the S&P SL20 have weak support for clustering and persistence. This implies that, in economic crises, the behavior of volatility is erratic. The evolution of volatility patterns across the three periods reveals distinct market behavior transformations. The transition from balanced ARCH coefficients in the normal period (ASPI: β = 0.2057, p = 0.0534, S&P SL20: β = 0.1425, p = 0.2055) to extreme values during COVID-19 (particularly S&P SL20’s clustering: β = 0.9463, p < 0.0001), and finally to negative ARCH coefficients in the crisis period (ASPI: β = −0.3416, p = 0.1051, S&P SL20: β = −0.4366, p = 0.1098), demonstrates how market dynamics fundamentally shift under different economic conditions. This pattern evolution suggests that traditional risk management strategies may need period-specific adjustments, particularly during transitions between economic regimes.

Table 4 presents the diagnostic tests for the full sample’s FIGARCH of ASPI and SNP20 run. In Panel A, the ARCH test has returned very high p-values with both the ASPI (F-statistic p-value = 0.8788; Obs_R-squared p-value = 0.8788) and the SNP20 (F-statistic p-value = 0.802; Obs_R-squared p-value = 0.8019). Significant p-values for the F-statistic result indicate that the residuals are homoskedastic, which means they do not have severe ARCH effects, and thus the model assumptions hold.

The Engle-Ng sign-bias test (Panel B) evaluates the presence of leverage effects, where positive and negative shocks have differential impacts on future volatility. This test indicates that the p-values in both cases in both indices are very high. Using the ASPI, the p-values were significantly high; the values were 0.9516 for sign-bias, 0.6188 for negative-bias, 0.5872 for positive-bias, and 0.8185 for joint-bias. For the SNP20, the p-values were 0.3964 for sign-bias, 0.1398 for negative-bias, 0.2546 for positive-bias, and 0.3145 for joint-bias. These high p-values suggest that no significant leverage effects are present in the standardized residuals, meaning that positive and negative shocks have similar impacts on future volatility.

The regression results for leverage effects also confirm the Engle-Ng sign bias test. Using the ASPI, the p-value for SMINUS(−1) and its interaction with lagged residuals and SPLUS(−1) and its interaction is high. The ASPI results further relate that for the SNP20, the results indicate that the p-values for SMINUS(−1), its interaction with lagged residuals, and SPLUS(−1) and its interaction are high. The results relate that there are no significant leverage impacts in the residuals. The validity tests confirm the FIGARCH models’ validity on the ASPI and SNP20 indices. The residuals show no significant ARCH or leverage effects, which support the model assumption of conditional volatility. The comprehensive diagnostic results validate the model specifications and provide quantitative support for the symmetric nature of market responses. The absence of significant ARCH effects in residuals (p > 0.80 for both indices) and the consistently high p-values in leverage effect tests (all p > 0.10) demonstrate that the market responds similarly to positive and negative shocks, suggesting a mature market structure despite the varying economic conditions.

The parameter stability test for the Nyblom White test of parameters estimated from the FIGARCH models of the ASPI and SNP20 over the full sample is undertaken, and the results are summarized in Table 5. Individual parameter stability was confirmed for the ASPI based on all the individual parameter test statistics, which were less than 10% of the critical value. However, the joint test statistic (2.3940) is greater than the 1% of the critical value (2.350) and would suggest rejecting the null hypothesis of parameter stability at the 1% significance level. The result indicates that although individual parameters might appear stable in isolation, evidence suggests a collective instability of the parameters over the sample period. On the contrary, the SNP20 was stable in individual and collective parameters. In contrast, the SNP20 exhibited stability in both individual parameters and collectively. The test statistics for all individual parameters were below the 10% critical value, and the joint test statistic (1.3072) was below the 10% critical value (1.690). This suggests that the null hypothesis of parameter stability for the SNP20 cannot be rejected at any conventional significance level, indicating that the parameters are stable over the sample period.

Confidence ellipse plots for the full sample are presented in Figure 4, formed by the ASPI and the SNP20, to assess the stability and correlation of parameters in the FIGARCH models fitted on these indices. For both indices, the confidence ellipses were predominantly circular, indicating low or no significant correlation between most parameter pairs. In the case of pairs like C(2) and C(3) and C(5) and C(6), there was a slight elongation in the ellipse, which might represent minor correlatedness. The cantering of ellipses around the origin for most parameter pairs indicates stability, suggesting that the parameters do not significantly deviate over the sample period. In the case of the ASPI model, while the individual parameter stability was confirmed, the joint test statistic indicated some collective instability. On the other hand, the SNP20 model confirmed overall stability in both individual and collective terms, thus confirming the reliability of the parameters over the sample period.

This study analyzes the performance and volatility dynamics of the All Share Price Index and S&P SL 20 Index from January 2012 to April 2024, which consists of normal economic scenarios, COVID-19 pandemics, and a subsequent economic crisis. During the full sample period, the ASPI exhibited higher mean values and volatility than the S&P SL 20, indicating that the broader market was more sensitive to economic fluctuations. This tendency was detected in all the sub-periods; however, the type and degree of volatility were of conclusive difference. The COVID-19 period was marked by increased mean returns and surged volatility for both indices, reflecting that the worst uncertainty was circulating the market. In contrast, during the period of economic crisis, both the ASPI and S&P SL 20 were detected as highly volatile, whereas for the S&P SL 20, the mean return was negative, and for the ASPI, it was positive. Advanced econometric models used in this study, such as the FIGARCH model, showed volatility clustering, persistence in volatility, and long memory in the dynamics of the two indices. The ASPI shows several features, such as clustering of returns’ volatilities during normal periods and strong persistence across periods. On the contrary, S&P SL 20 showed clustering mainly during COVID-19 but once more with a high volatility persistence. These two indices showed remarkable long-memory components. This generally means that in the market the shock to volatility lasts longer than expected.

Therefore, the empirical results greatly affect investment management and policy formulation. The findings, therefore, call for an approach in risk management for investors that combines both the short-run volatility pattern and the long-run market dynamics. More precisely, portfolio managers should apply dynamic hedging strategies, taking up the greater volatility of ASPI over that of the S&P SL20. This may include (1) position sizing using volatility forecasts, (2) stop-loss mechanisms that incorporate various volatility regimes for each index, and (3) sector diversification strategies considering the compositional differences between the indices. Strong evidence of volatility clustering and persistence implies historical volatility patterns should be part of the forecasting models of the risk management system, especially during regime transitions between normal and stress periods.

These findings also have regulatory implications for an approach to market oversight that recognizes the distinct behavior patterns of broad market indices versus blue-chip stocks. Some of the policy recommendations that policymakers could consider are: (1) the use of differentiated circuit breaker mechanisms, given the different volatility patterns between ASPI and S&P SL20, (2) improved disclosure requirements during periods of high volatility, which would add to better market transparency, and (3) intervention strategies targeted at vulnerabilities specific to each market in times of economic turmoil. These documented long-memory effects and persistence patterns could be used as early warning indicators for market stress, providing opportunities for preemptive regulatory action. Furthermore, these findings contribute to the broader understanding of frontier market dynamics, suggesting that regulatory frameworks in similar emerging markets might benefit from incorporating volatility persistence patterns into their market stability measures. The contrasting behavior of the broad market and blue-chip indices during stress periods provides valuable insights for designing market resilience mechanisms in developing economies.

Despite the comprehensive analysis and robust methodologies employed, this study has some limitations that warrant consideration. First, the study period ranges from January 2012 to April 2024. Although the COVID-19 period and the financial crisis have been captured within this ambit, it may not present long-term tendencies and patterns of market behavior. A longer term for the study would have allowed the long-run dynamics of market volatility to be better uncovered. Second, using the FIGARCH process, while advanced in nature, has limitations. These models are based on the assumption that some volatility will follow an established pattern and have inherent constraints in capturing non-linear relationships in market behavior, which is inadequate in capturing market behavior’s full reality. Additionally, the study does not detect structural breaks in the time series data, which can affect the validity of the results. Furthermore, daily data may miss important intraday volatility patterns and microstructure effects. Finally, some specific economic policies, political events, or external shocks that may influence market behavior are not substantially elaborated on in the study, and these can better explain the causes of volatility.

Given these limitations, future research may proceed in several directions to better understand market dynamics. First, an extension of the timeline to include more historical data and data for the future, where it is available, may help identify long-term trends and cyclical patterns in the market’s volatility. Second, the application of alternative advanced volatility models could better capture non-linear relationships and market dynamics. Third, incorporating techniques to detect and account for structural breaks in the time series data can provide more accurate modeling of market volatility. Structural breaks occur if there is a significant change, like the process that generates the data, and economic or political events most commonly bring these about. Their identification and allowance can boost the robustness and quality of the volatility models and allow for a more precise forecast. Future studies could also explore the relationship between trading volume and volatility patterns to understand market behavior better. Future research must also investigate the impact of incorporating macroeconomic factors, policy changes, and political events on market volatility. Factors such as interest rates, inflation, exchange rates, and political stability, among others, will be incorporated to have a clearer picture of what drives the market. Developing more sophisticated forecasting models that incorporate both short- and long-term memory effects could enhance the accuracy of market volatility prediction. This approach helps isolate the impacts of particular events or policies on market dynamics so that useful recommendations can be made for investors and policymakers.