Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4, 2395-2404 2025 Publisher: Learning Gate DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate © 2025 by the authors; licensee Learning Gate History: Received: 6 February 2025; Revised: 14 March 2025; Accepted: 18 March 2025; Published: 24 April 2025 * Correspondence: dwifitrizalslm@telkomuniversity.ac.id Risk measurement model on top 10 cryptocurrency market capitalization Umar Al Faruq1, Dwi Fitrizal Salim2*, Farida Titik Kristanti3 1,2,3School of Economics and Business, Telkom University, Bandung-Indonesia; dwifitrizalslm@telkomuniversity.ac.id (D.F.S.). Abstract: This study conducted a large-scale analysis to evaluate the performance of traditional and Markov-Switching GARCH (MS-GARCH) models to estimate the volatility of the top 10 cryptocurrencies by market capitalization. The study compared the performance of the models using goodness-of-fit measures, specifically the Deviance Information Criterion (DIC) and the Bayesian Predictive Information Criterion (BPC). Secondly, we assess the forecasting accuracy for one-day-ahead conditional volatility and Value-at-Risk (VaR). The results obtained show that, in a manner consistent with the findings for the broader cryptocurrency market, the time-varying regime-switching model exhibits superior performance in capturing the complex volatility patterns observed in cryptocurrencies when compared to the traditional GARCH model. Keywords: Cryptocurrency, E-GARCH, GARCH, T-GARCH, MSGARCH, Volatility. 1. Introduction Blockchain technology that has developed alongside cryptocurrencies, played a crucial role in facilitating technological solutions across various fields [1]. The origin of cryptocurrencies traces back to 2009 when Bitcoin was created by an individual known as Tripathi, et al. [2]. Bitcoin was the pioneer cryptocurrency to function on blockchain, which now underpins all digital currencies. Cryptocurrencies provide a decentralized substitute to the conventional financial system, facilitating peer-to-peer transactions with low fees and enhanced security [3]. The expansion of cryptocurrencies has resulted in significant market growth with the total market capitalization is approximately at $858 billion USD in present time[4, 5]. Cryptocurrencies display varying degrees of volatility compared to traditional fiat currencies. Research by Miglietti, et al. [6] indicates that Litecoin has higher volatility compared to both Euro and Bitcoin. Striking price variations within the cryptocurrency market are instigated by factors such as market speculation, regulations, technology, and investor sentiment. Furthermore, Zhang, et al. [7] demonstrates that regulatory announcements, particularly in China, significantly heighten price volatility, liquidity, and returns, with a more pronounced effect observed during the COVID-19 pandemic. In recent years, the cryptocurrency market has encountered significant pressure due to global events, including the COVID-19 epidemic, the Russia-Ukraine conflict, monetary policy changes, and the decline of speculative markets. As a result, cryptocurrencies offer great profit opportunities but also carry significant risks, with volatility being a major challenge in risk measurement. Therefore, robust analytical models are required to effectively capture the nature of volatility [8]. This research aims to analyze and forecast cryptocurrency volatility by examining the top 10 cryptocurrencies by market capitalization over the past decade using data obtained from CoinMarketCap. The selected cryptocurrencies include Bitcoin, Ethereum, Tether, Binance Coin, Solana, XRP, USD Coin, Cardano, Dogecoin, and Avalanche. https://orcid.org/0000-0002-5026-8983 https://orcid.org/0000-0003-1672-9907 2396 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate Recent advancements in financial risk modeling and predictive analytics, such as machine learning approaches for financial distress prediction Foroutan and Lahmiri [9] and deep learning models for Bitcoin price forecasting Kristanti, et al. [10] and By assessing the volatility of cryptocurrencies through Value at Risk, as noted by Jonathan, et al. [11] the most efficient approach for estimating VaR in cryptocurrencies is emphasized, showcasing its importance in enhancing financial literacy and crafting strong trading strategies. The author will utilize GARCH and its derivatives, including SGARCH, EGARCH, TGARCH, and MSGARCH, for measurement. 2. Literature Review 2.1. GARCH Model and Its Variants in Cryptocurrency Volatility The classic GARCH model has limitations in capturing complex volatility patterns, so its extended versions such as EGARCH, TGARCH, and MSGARCH are more commonly used. EGARCH and CGARCH have proven to be more accurate in estimating volatility for short term. They are able to capture asymmetries in price fluctuations that often occur in cryptocurrencies [12]. EGARCH (1,1) also performs well in predicting volatility during crisis periods, supporting its use for emerging markets that are prone to economic instability [13]. For fiat currencies, research found that TGARCH is more suitable for currencies such as yen and ringgit, while GARCH is more suitable for yuan and US dollar [14]. This suggests that selecting the GARCH model should be tailored to the financial assets being analyzed. The TV-MSGARCH model utilizing student-t distribution yields the most effective predictions regarding Bitcoin volatility [15]. This model is designed to handle fluctuations in volatility over time. In addition, various studies indicate that MSGARCH surpasses single-regime models in forecasting Value- at-Risk (VaR) [16]. Further investigations reveal that PGARCH is better suited for student-t distributions, which tend to provide more precise representations of market movements [17]. Moreover, the GARCH (1,1) model with the student-t distribution is recognized for its robustness against changes in volatility, making it a great choice for cryptocurrency markets [18]. Simpler models can perform as well as more complex ones, provided they are able to capture return distributions with fat tail characteristics [19]. Regarding GARCH (1,1), it is the best prior for modeling volatility when combined with EGARCH to capture asymmetric effects [20]. 2.2. Effectivity of EGARCH, EGARCH, TGARCH, and MSGARCH Model in Analyzing Volatility Several studies compared EGARCH, TGARCH, and MSGARCH models to find the best approach to capture crypto market volatility. EGARCH shown to be more accurate in measuring volatility [21]. This model can capture the greater impact of bad news than good news on market volatility. However, in some cases GARCH and TGARCH provide better results [22]. MSGARCH shown more accurate VaR and Expected Shortfall (ES) estimates than single-regime models, suggesting cryptocurrency volatility is better explained by models with regime changes [23]. While TGARCH (1,1) was found more effective in capturing asymmetric shocks in the crypto market [24]. Moreover, complex GARCH models are better at explaining cryptocurrency volatility fluctuations than simpler models [25]. Lastly, the EGARCH-M model is employed to study the connection between returns and trading volume, indicating that volatility has a variable relationship with market activity [9]. 2.3. Impact of External Factors on Cryptocurrency Volatility External elements affect the volatility of cryptocurrencies. Although the volatility decreased during the COVID-19 pandemic, the correlation among cryptocurrencies has risen [26]. Regulatory actions by the Chinese government have notably influenced the price volatility and liquidity of cryptocurrencies, especially during uncertainty [7]. Events such as the COVID-19 pandemic, the Russia-Ukraine conflict, and global monetary policies exert tremendous pressure on cryptocurrency markets, creating difficulties 2397 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate in measuring and predicting volatility [27]. Among different cryptocurrencies, Ethereum (ETH) displays the greatest volatility, demonstrating higher sensitivity to market fluctuations [28]. 2.4. Effectiveness of GARCH Model in Market Analysis and Risk Management ARMA-GARCH-VaR models shown to be efficient in assessing risk within the cryptocurrency market [29]. The GARCH model effectively estimates Bitcoin's volatility, revealing clustering patterns and an inverted leverage effect through the GJR-GARCH (1,1) model, suggesting potential rise in volatility even in response to positive news [30]. The GARCH (1,1) model demonstrates volatility clustering in the crypto sector, whereas the GJR-GARCH (1,1) version identifies significant leverage effects in cryptocurrency returns [31]. However, GARCH is less effective in modeling a few cryptocurrencies, indicating that some digital assets might need different modeling approaches [32]. 3. Methods This study employs a quantitative methodology utilizing a 5-year time series data analysis to assess Value-at-Risk (VaR) for 10 different cryptos. The analytical models applied include GARCH, Exponential GARCH (EGARCH), Threshold GARCH (TGARCH), and Markov-Switching GARCH (MSGARCH). These models facilitate the examination of shifts in price volatility among crypto assets crucial for evaluating risks in unstable markets. The dataset utilized in this research comprises daily historical pricing for 10 cryptocurrencies gathered from CoinMarketCap. This study spans a decade to effectively capture the various market dynamics that unfold over time. By utilizing a wide range of historical data, this research aims to uncover significant and pertinent volatility trends that are essential for assessing the risk of crypto assets. The analysis in this study used statistical software such as R or Python, which offer specialized packages for GARCH, EGARCH, TGARCH, and MSGARCH models. The analysis method applied in this study begins with the estimation of volatility models using the GARCH Bollerslev [33] EGARCH Nelson [34] TGARCH Zakoian [35] and MSGARCH [36]. With each model as below. GARCH Bollerslev [33] ℎ𝑘,𝑡 = 𝛼0,𝑘 + 𝛼1,𝑘𝛾𝑡−1 + 𝛽𝑘ℎ𝑘,𝑡−1 (1) EGARCH Nelson [34] 𝑙𝑜𝑔ℎ𝑘,𝑡 = 𝛼0,𝑘 + 𝛼1,𝑘(|𝜂𝑘,𝑡−1| − 𝐸[|𝜂𝑘,𝑡−1|]) + 𝛼2,𝑘𝜂𝑘,𝑡−1 + 𝛽𝑘𝑙𝑜𝑔ℎ𝑘,𝑡−1, (2) TGARCH Zakoian [35] ℎ𝑘,𝑡 1/2 = 𝛼0,𝑘 + 𝛼1,𝑘𝐼(𝛾𝑡−1 > 0)𝛾𝑡−1 + 𝛼2,𝑘𝐼(𝛾𝑡−1 < 0)𝛾𝑡−1 + 𝛽𝑘ℎ𝑘,𝑡−1 1/2 (3) MSGARCH Ardia, et al. [36] ℎ𝑘,𝑡 = ℎ(𝑦𝑡−1, ℎ𝑘,𝑡−1, 𝜃𝑘) (4) Optimal model selection depends on information criteria like the Deviance Information Criterion (DIC) and the Bayesian Predictive Information Criterion (BPIC), which assess the model's efficacy in capturing the volatility of cryptocurrency asset values. Model selection is essential for statistical inference, especially in hierarchical Bayesian models and empirical Bayes models. Conventional metrics such as the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are inapplicable to Bayesian models since they depend on maximum likelihood estimation. Instead, DIC Spiegelhalter, et al. [37] and BPIC Ando [38] were created to assess model fit and complexity by taking into account the Bayesian posterior distribution. General Framework for Bayesian Model Selection: Let 𝑦 = (𝑦, … , 𝑦𝑛) be the observed data generated from the true unknown distribution 𝐺(𝑦) with density 𝑔(𝑦), and let (𝑓{𝑦 ∣ 𝜃); 𝜃 ∈ Θ ⊂ 𝑅𝑝} be the model to estimate 𝐺(𝑦) In Bayesian modeling, a prior distribution is specified for the parameters. The posterior distribution is given by: 𝜋(𝜃 ∣ 𝑦) = 𝐿( 𝑦∣∣𝜃 )𝜋(𝜃) ∫ 𝐿( 𝑦∣∣𝜃 )𝜋(𝜃)𝑑𝜃 (5) 2398 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate Whereas 𝐿(𝑦|𝜃) is the likelihood function. The prediction distribution for future observation z is as follow. 𝑞( 𝑧 ∣∣ 𝑦 ) = ∫ 𝑓 ( 𝑧 ∣ 𝜃 )𝜋( 𝜃 ∣∣ 𝑦 )𝑑𝜃 (6) With the aim of evaluating how well the predictive model approximates the true g(z) using BPIC Ando [38] and DIC [37]. DIC is defined as follow: 𝐷𝐼𝐶 = −2𝐸𝜃∣𝑦[𝑙𝑜𝑔𝐿(𝑦 ∣ 𝜃] + 𝑃𝐷 (7) BPIC is defined as follow: 𝐵𝑃𝐼𝐶 = −2𝐸𝜃∣𝑦[𝑙𝑜𝑔𝐿(𝑦 ∣ 𝜃)] + 2𝑛𝑏𝑦 (8) Furthermore, model evaluation is performed by forecasting volatility one step ahead, we use the first half of our yt data for initial model estimation and then evaluate the forecast accuracy over the remaining half (out-of-sample period). The rolling window approach is used for estimation, where the model parameters are re-estimated every 10 observations. Since the actual volatility cannot be observed, we use proxies to assess the prediction performance. We compare the predictions of GARCH models using two metrics: Mean Squared Error (MSE) and Mean Absolute Error (MAE). Table 1. Descriptive Statistics. Symbol Mean Median Std Dev Kurtosis ADA 0.0008 -0.0006 0.0467 7.0332 BNB 0.0032 0.0012 0.0467 33.7472 BTC 0.002 0.0008 0.0338 12.8558 DOGE 0.005 -0.0003 0.1002 731.0627 ETH 0.0023 0.0009 0.043 11.1073 SOL 0.0054 0.0002 0.0676 8.444 TRX 0.0023 0.0023 0.0489 70.0946 USDC -1.592E-06 9.97E-05 0.0032 51.165 USDT -5.481E-06 -1E-04 0.003 108.1218 XRP 0.0022 0.0002 0.0544 32.6233 Note: This table presents summary information for cryptocurrencies with differing levels of market capitalization. Table 2. Price Descriptive Statistics. Symbol MeanP StdDevP SkewnesP KurtosisP ADA 0.7519 0.5824 1.5795 4.8398 BNB 259.0137 208.5721 0.3033 1.9370 BTC 31501.4309 22114.3 0.7671 2.9922 DOGE 0.0909 0.10039 1.7147 6.6997 ETH 1682.4769 1244.145 0.2865 1.9842 SOL 68.6085 67.9654 0.8726 2.4722 TRX 0.0689 0.04879 1.5191 7.4769 USDC 1.0010 0.0043 3.1376 24.640 USDT 1.0008 0.0036 3.3654 30.1581 XRP 0.5332 0.3488 2.6165 12.6457 Note: This table presents summary information for cryptocurrencies with differing levels of market capitalization. P for Price. 2399 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate Table 3. Return Descriptive Statistics. Symbol Mean R Std Dev R Skewnes R Kurtosis R ADA 0.0008 0.0466 0.4932 7.0539 BNB 0.0032 0.0467 1.5818 33.8246 BTC 0.0020 0.0338 -0.3107 12.8558 DOGE 0.0050 0.0999 21.4868 735.0901 ETH 0.0023 0.0429 -0.2793 11.1223 SOL 0.0054 0.0675 0.5421 8.4644 TRX 0.0022 0.0488 3.2195 70.5145 USDC -1.33e-06 0.0029 1.1750 61.3864 USDT -4.67e-06 0.0028 0.9565 127.0100 XRP 0.0021 0.0543 2.3467 32.7287 Note: R for Return In financial risk modeling, the accuracy of Value-at-Risk (VaR) estimates is critical for risk management and regulatory compliance. Two statistical tests used to evaluate the performance of VaR models are Conditional Coverage (CC) and Dynamic Quantile (DQ) Test. The CC Test Christoffersen [39] tests whether VaR exceptions (violations) occur independently over time, ensuring correct conditional coverage. This test combines two hypotheses: Unconditional Coverage (UC), which ensures that the observed violation rate corresponds to the expected rate, and independence (IND), which ensures that violations occur independently over time. By defining the hit function as: 𝐼𝑡 = {1, 𝑖𝑓 𝑦𝑡 < 𝑉𝑎𝑅𝑡 (𝑉𝑎𝑅 𝑣𝑖𝑜𝑙𝑎𝑡𝑖𝑜𝑛) 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 The UC hypothesis states that the observed frequency of VaR violations should match the expected probability as p: �̂� = 1 𝑇 ∑ 𝐼𝑡 ≈ 𝑝 𝑇 𝑡=1 (9) Likelihood Ratio (LR) is used to evaluate this hypothesis. UC Test using: 𝐿𝑅𝑈𝐶 = −2[𝑙𝑜𝑔𝐿(�̂�) − 𝑙𝑜𝑔𝐿(𝑝)] ∼ 𝜒2 (10) IND Test using: 𝐿𝑅𝐼𝑁𝐷 = −2[𝑙𝑜𝑔𝐿(𝜋0, 𝜋1) − 𝑙𝑜𝑔𝐿(�̂�)] ∼ 𝜒2 (11) If it is statistically significant, then the VaR model fails to provide correct conditional coverage, meaning the violations are not independent and indicate volatility that is not captured by the model. Meanwhile, the DQ Test Engle and Manganelli [40] extends the CC Test by testing whether VaR violations depend on historical information, ensuring the quantile model is dynamic. The hit deviation function is defined as: 𝐻𝑖𝑡𝑡 = 𝐼𝑡 − 𝑝 then regression is performed as follows: 𝐻𝑖𝑡𝑡 = 𝛽0 + ∑ 𝛽𝑗 𝑚 𝑗=1 𝐻𝑖𝑡𝑡−𝑗 + ∑ 𝛾𝑘 𝑛 𝑘=1 𝑋𝑡−𝑘 + 𝜀𝑡 (12) where are additional explanatory variables such as past volatility and market conditions, measures serial dependence in the offense (should be zero in the correct model specification), and measures dependence on market conditions. The null hypothesis of the correct model is defined as: 𝐻0: 𝛽1 = 𝛽2 = ⋯ = 𝛽𝑚 = 𝛾1 = ⋯ = 𝛾𝑛 = 0 (13) Testing is conducted using Likelihood Ratio: 𝐷𝑄 = 𝑇𝑅2 ∼ 𝜒2(𝑚 + 𝑛) (14) where DQ is the coefficient of determination of the regression. If the DQ test result is significant, then VaR violations are influenced by historical factors and market conditions, which means that the VaR model is not appropriate. The DQ Test is more robust in evaluating dynamic risk models because it considers external factors that affect VaR violations. 2400 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate 4. Result 4.1. Top 10 Crypto Returns Figure 1. Return Cryptocurrency. Table 4. Model selection based on DIC and BPIC. Symbol DIC BPIC Both Agree ADA TGARCH-T-(3) SGARCH-T (3) - BNB SGARCH-T-(2) EGARCH-T-(3) - BTC SGARCH-G-(2) TGARCH-G-(3) - DOGE SGARCH-T-(3) TGARCH-T-(3) - ETH TGARCH-T-(3) EGARCH-T-(3) - SOL TGARCH-T-(2) SGARCH-T-(3) - TRX SGARCH-T-(3) EGARCH-T-(3) - USDC EGARCH-G-(2) EGARCH-G-(2) EGARCH-G (2) USDT EGARCH-G-(2) TGARCH-G-(3) - XRP TGARCH-T-(1) TGARCH-T-(3) - Note: Total number of cryptocurrencies is 10. Values indicate the number of time series that fit the selected model through DIC (panel A), BPIC (panel B) and both criteria (panel C). N, S and G are Normal, Student's-𝑡 and GED distributions, respectively. 𝐾 refers to the number of regimes in the model. Table 5. Effect of Inverse Leverage. Regimes Distribution EGARCH TGARCH K=1 N 0 5 0 Std 0 7 0 Ged 0 3 K=2 N 0 6 0 Std 0 8 0 Ged 0 3 K=3 N 0 6 0 Std 0 8 0 Ged 0 3 Note: Total number of cryptocurrencies is 10. Values indicate the number of time series of each model. 2401 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate 4.2. B. Out of Sample Analysis Table 6. Model selection based on VaR forecasts. Symbol CC DQ $ADA SGARCH-G-(2) SGARCH-T-(2) $BNB TGARCH-T-(2) SGARCH-T-(1) $BTC EGARCH-T-(1) EGARCH-T-(1) $DOGE SGARCH-T-(1) SGARCH-T-(2) $ETH SGARCH-N-(2) SGARCH-T-(1) $SOL SGARCH-N-(2) SGARCH-N-(2) $TRX SGARCH-T-(1) EGARCH-T-(1) $USDC EGARCH-T-(1) SGARCH-T-(1) $USDT EGARCH-T-(3) SGARCH-T-(1) $XRP EGARCH-G-(3) SGARCH-N-(2) Note: Total number of cryptocurrencies is 10. Values indicate the number of time series corresponding to the model that minimizes CC (panel A), DQ (panel B) and both criteria (panel C). N, Std and GED denote the Normal, Student's-𝑡 and GED distributions, respectively. 𝐾 refers to the number of regimes in the model. 5. Discussion Examining the ten leading cryptocurrencies, we find that generally both performance metrics (MSE and MAE) align with the most effective model. Specifically, for BTC and USDT, the two-regime TGARCH model with Student's t-distribution offers the best fit. In contrast, ETH is most accurately captured by a three-regime GARCH model with a generalized distribution. Despite detecting shifts in volatility regimes in some cryptocurrencies, the criteria of MSE and MAE occasionally indicated varying optimal models. Furthermore, DIC and BPIC sometimes presented differing opinions regarding the optimal number of regimes for a specific cryptocurrency. Table 5 indicates a reversal of the leverage effect in at least one environment. N, Std, and GED denote the Normal, Student's t, and Generalized Error Distribution, respectively. K denotes the quantity of regimes within the model. For most of these cryptocurrencies (BTC, ETH, ADA, SOL, BNB, XRP), our model shows a positive asymmetry in volatility. In particular, BTC and ETH show a clear pattern of positive asymmetry, reacting more to positive price changes. ADA and SOL also show this characteristic, but weaker. In this set of ten, only USDT shows a consistent inverse leverage effect across all model specifications. Implying positive past returns have a greater impact on volatility than negative past returns, which is an unusual finding. For DOGE and TRX, while the EGARCH model suggests that negative returns affect volatility more strongly than positive returns, the TGARCH model, at least in one of its specifications, indicates a reverse leverage effect. This suggests that the relationship between returns and volatility for these two cryptocurrencies may be more nuanced and dependent on the specific model used. USDC shows no clear evidence of positive asymmetry or inverse leverage. SOL showed the highest volatility among the ten coins, consistently across all models. ETH showed moderate volatility, while USDT and USDC showed relatively lower volatility. TRX, ADA, BNB, and XRP were somewhere in between, showing moderate levels of volatility but XRP at the end of the quarter showed its volatility. DOGE showed more fluctuating volatility, indicating greater sensitivity to market changes. Based on the behavior of the average and standard deviation across models, these cryptocurrencies could be split into some categories. BTC and ETH fall into a group that has relatively consistent averages and standard deviations across different model specifications. USDT and USDC also show stability. TRX, SOL, ADA, BNB, and XRP show some variation in mean and standard deviation depending on the model used. Generally, these ten cryptocurrencies have higher mean than the standard deviation, indicating positive average return. However, we observed instances where the standard 2402 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate deviation was higher than the mean for DOGE indicating periods of significant price fluctuations and potentially negative average returns over the observed period. For Out-of-Sample we examine the one-day forward conditional volatility and (VaR) estimates. By using MSE and MAE for volatility forecasts, we find that single-regime models generally outperform two- and three-regime models. Particularly for BTC and BNB, the TGARCH model with Student's t- distribution provides the most accurate volatility forecasts. ADA and SOL are best modeled with a two- regime GARCH model with the Common Error Distribution. DOGE, and ETH performed better with the EGARCH model, while USDT best fit the one-regime SGARCH. For forecasting VaR, Christoffersen's CC test indicates that the two-regime TGARCH model with a normal distribution is the most efficient for BNB and ADA. In contrast, SOL and ETH are best represented by a two-regime GARCH model. TRX and DOGE fit best with a single-regime GARCH model, while USDT and USDC demonstrated the highest VaR forecasting accuracy using the EGARCH model. XRP is ideally modeled with a three-regime GARCH using a General Error distribution. The DQ test yielded somewhat different results, often favoring models with fewer regimes, yet it generally aligned with the best performing model type for each cryptocurrency. Overall, while MSGARCH models tend to perform better than single-regime models, no model can be said to be the best across all ten cryptos. Asymmetric models were often preferred, especially for more volatile cryptos. Optimal models for volatility and VaR forecasting varied widely, highlighting the importance of considering individual cryptocurrency characteristics. In our Value-at-Risk (VaR) forecasting, the traditional GARCH model proved to be the best fit for ETH and USDT. However, for most other leading cryptocurrencies, the asymmetric TGARCH model is preferred (see Table 7 for a summary of in- and out-of-sample findings across the ten major cryptocurrencies). 6. Conclusions Overall, this research shows that there is no model that can universally capture the volatility patterns of all cryptos. Each asset has unique characteristics, affecting the selection. GARCH-based models, especially TGARCH and EGARCH, are often superior in capturing the complex nature of volatility, especially for cryptocurrencies with high price fluctuations such as BTC, ETH, and SOL. In addition, asymmetry effects in volatility were found for most cryptocurrencies, with some exhibiting positive asymmetry, where volatility is more responsive to price increases than price decreases. However, USDT showed an inverse leverage effect, meaning its volatility was more affected by positive than negative returns which is an unusual finding in financial assets. In the context of Value-at-Risk (VaR) forecasting, TGARCH and EGARCH models are often superior to the standard GARCH model, especially in capturing short-term volatility risk. However, the MSGARCH model with a multi regime approach shows great potential in adjusting to changes in market structure, although the optimal results still depend on the specifics of the cryptocurrencies analyzed. This study confirms that the selection of volatility models cannot be generalized for all cryptocurrencies. Instead, a more flexible approach tailored to the specific characteristics of each digital asset is required, taking into account factors such as asymmetry effects, volatility regime changes, and the corresponding return distribution. Transparency: The authors confirm that the manuscript is an honest, accurate, and transparent account of the study; that no vital features of the study have been omitted; and that any discrepancies from the study as planned have been explained. This study followed all ethical practices during writing. Copyright: © 2025 by the authors. This open-access article is distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). https://creativecommons.org/licenses/by/4.0/ 2403 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate References [1] G. Osório, "Cryptocurrencies— advantages and risks of digital money," CRC Press eBooks, 2024, pp. 160–196. [2] A. Tripathi, A. Choudhary, S. K. Arora, G. Arora, G. Shakya, and B. Rajwanshi, "Crypto bank: Cryptocurrency wallet based on blockchain," in In International Conference on Recent Trends in Image Processing and Pattern Recognition (pp. 223-236). Cham: Springer Nature Switzerland, 2023. [3] S. M. Thornton, P. K. Attaluri, E. C. Shaffrey, P. J. Wirth, A. J. Rao, and V. K. Rao, "Plastic surgery and cryptocurrency: A 2024 update," Plastic and Reconstructive Surgery–Global Open, vol. 12, no. 7, p. e6006, 2024. https://doi.org/10.1097/gox.0000000000006006 [4] S. Corbet and L. Oxley, "Investigating the academic response to cryptocurrencies: Insights from research diversification as separated by journal ranking," Review of Corporate Finance, vol. 3, no. 4, 2023. https://doi.org/10.2139/ssrn.4425965 [5] S. Kumar, S. K. Patra, A. Kumar, K. U. Singh, and S. Varshneya, "Enablers for growth of cryptocurrencies: A fuzzy– ism benchmarking," Journal of Risk and Financial Management, vol. 16, no. 3, p. 149, 2023. https://doi.org/10.3390/jrfm16030149 [6] C. Miglietti, Z. Kubosova, and N. Skulanova, "Bitcoin, litecoin, and the Euro: An annualized volatility analysis," Studies in Economics and Finance, vol. 37, no. 2, pp. 229-242, 2020. [7] P. Zhang, K. Xu, and J. Qi, "The impact of regulation on cryptocurrency market volatility in the context of the COVID-19 pandemic—evidence from China," Economic Analysis and Policy, vol. 80, pp. 222-246, 2023. https://doi.org/10.1016/j.eap.2023.08.015 [8] H. Gupta and R. Chaudhary, "An empirical study of volatility in cryptocurrency market," Journal of Risk and Financial Management, vol. 15, no. 11, p. 513, 2022. https://doi.org/10.3390/jrfm15110513 [9] P. Foroutan and S. Lahmiri, "The effect of COVID-19 pandemic on return-volume and return-volatility relationships in cryptocurrency markets," Chaos, Solitons & Fractals, vol. 162, p. 112443, 2022. [10] F. T. Kristanti, M. Y. Febrianta, D. F. Salim, H. A. Riyadh, and B. A. H. Beshr, "Predicting financial distress in indonesian companies using machine learning," Engineering, Technology & Applied Science Research, vol. 14, no. 6, pp. 17644-17649, 2024. [11] J. Jonathan, P. A. Purnama, R. Aditya Kristamtomo Putra, and D. H. Syahchari, "Evaluating value at risk in first- layer cryptocurrency token investments via monte carlo simulation," presented at the International Conference on Creative Communication and Innovative Technology (ICCIT), pp. 1–7. https://doi.org/10.1109/iccit62134.2024.10701121, 2024. [12] B. N. S. S. Kiranmai and V. Thangaraj, "Modeling volatility of cryptocurrencies: GARCH approach," presented at the In Congress on Intelligent Systems (pp. 237-251). Singapore: Springer Nature Singapore, 2022. [13] M. Mahmud and N. Mirza, "Volatility dynamics in an emerging economy: Case of Karachi stock exchange," Economic Research, vol. 24, no. 4, pp. 51-64, 2011. https://doi.org/10.1080/1331677x.2011.11517480 [14] T. Tungtrakul, N. Kingnetr, and S. Sriboonchitta, "Do we have robust garch models under different mean equations: Evidence from exchange rates of Thailand?," Robustness in Econometrics, pp. 599-613, 2017. https://doi.org/10.1007/978-3-319-50742-2_37 [15] C.-Y. Tan, Y.-B. Koh, K.-H. Ng, and K.-H. Ng, "Dynamic volatility modelling of Bitcoin using time-varying transition probability Markov-switching GARCH model," The North American Journal of Economics and Finance, vol. 56, p. 101377, 2021. https://doi.org/10.1016/j.najef.2021.101377 [16] D. Ardia, K. Bluteau, and M. Rüede, "Regime changes in Bitcoin GARCH volatility dynamics," Finance Research Letters, vol. 29, pp. 266-271, 2019. https://doi.org/10.1016/j.frl.2018.08.009 [17] S. Salamat, N. Lixia, S. Naseem, M. Mohsin, M. Zia-ur-Rehman, and S. A. Baig, "Modeling cryptocurrencies volatility using garch models: A comparison based on normal and student’s t-error distribution," Entrepreneurship and Sustainability Issues, vol. 7, no. 3, pp. 1580–1596, 2020. https://doi.org/10.9770/jesi.2020.7.3(11) [18] M. Buczyński and M. Chlebus, "Old-fashioned parametric models are still the best: a comparison of value-at-risk approaches in several volatility states," Journal of Risk Model Validation, vol. 14, no. 2, 2019. https://doi.org/10.21314/jrmv.2020.222 [19] A. Mappadang, B. A. Nugroho, S. D. Lestari, Elizabeth, and T. K. Lestari, "Measuring value-at-risk and expected shortfall of newer cryptocurrencies: new insights," Cogent Business & Management, vol. 11, no. 1, p. 2416096, 2024. https://doi.org/10.1080/23311975.2024.2416096 [20] R. G. D. S. Queiroz and S. A. David, "Does anything beat a GARCH (1, 1)? Evidence from crypto markets," presented at the In International Conference on Nonlinear Dynamics and Applications (pp. 398-408). Cham: Springer Nature Switzerland, 2024. [21] P. Chitpattanapaibul and D. Wu, "Study on stylized stock market volatility based on asymmetric GARCH model in the main area of BRI region," presented at the In 2019 4th IEEE International Conference on Cybernetics (Cybconf) (pp. 1-4). IEEE, 2019. [22] T. Panagiotidis, G. Papapanagiotou, and T. Stengos, "On the volatility of cryptocurrencies," Research in International Business and Finance, vol. 62, p. 101724, 2022. https://doi.org/10.1016/j.ribaf.2022.101724 https://doi.org/10.1097/gox.0000000000006006 https://doi.org/10.2139/ssrn.4425965 https://doi.org/10.3390/jrfm16030149 https://doi.org/10.1016/j.eap.2023.08.015 https://doi.org/10.3390/jrfm15110513 https://doi.org/10.1109/iccit62134.2024.10701121 https://doi.org/10.1080/1331677x.2011.11517480 https://doi.org/10.1007/978-3-319-50742-2_37 https://doi.org/10.1016/j.najef.2021.101377 https://doi.org/10.1016/j.frl.2018.08.009 https://doi.org/10.9770/jesi.2020.7.3(11 https://doi.org/10.21314/jrmv.2020.222 https://doi.org/10.1080/23311975.2024.2416096 https://doi.org/10.1016/j.ribaf.2022.101724 2404 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 4: 2395-2404, 2025 DOI: 10.55214/25768484.v9i4.6554 © 2025 by the authors; licensee Learning Gate [23] L. Maciel, "Cryptocurrencies value‐at‐risk and expected shortfall: Do regime‐switching volatility models improve forecasting?," International Journal of Finance & Economics, vol. 26, no. 3, pp. 4840-4855, 2021. https://doi.org/10.1002/ijfe.2043 [24] A. Ampountolas, "Cryptocurrencies intraday high-frequency volatility spillover effects using univariate and multivariate GARCH models," International Journal of Financial Studies, vol. 10, no. 3, p. 51, 2022. https://doi.org/10.3390/ijfs10030051 [25] N. A. Kyriazis, "A survey on volatility fluctuations in the decentralized cryptocurrency financial assets," Journal of Risk and Financial Management, vol. 14, no. 7, p. 293, 2021. https://doi.org/10.3390/jrfm14070293 [26] K. Yan, H. Yan, and R. Gupta, "Are GARCH and DCC values of 10 cryptocurrencies affected by COVID-19?," Journal of Risk and Financial Management, vol. 15, no. 3, p. 113, 2022. https://doi.org/10.3390/jrfm15030113 [27] D. Likitratcharoen, P. Chudasring, C. Pinmanee, and K. Wiwattanalamphong, "The efficiency of value-at-risk models during extreme market stress in cryptocurrencies," Sustainability, vol. 15, no. 5, p. 4395, 2023. https://doi.org/10.3390/su15054395 [28] S. Sarmini, C. R. A. Widiawati, D. R. Febrianti, and D. Yuliana, "Volatility analysis of cryptocurrencies using statistical approach and GARCH model a case study on daily percentage change," Journal of Applied Data Sciences, vol. 5, no. 3, pp. 838-848, 2024. [29] Y. Huang et al., "Evaluating cryptocurrency market risk on the blockchain: An empirical study using the ARMA- GARCH-VaR model," IEEE Open Journal of the Computer Society, vol. 5, pp. 83–94, 2024. https://doi.org/10.1109/ojcs.2024.3370603 [30] Y. Quan, T. Yang, C. Fei, C. Cheong, and L. Min, "Asymmetric volatility and risk analysis of bitcoin cryptocurrency market," Journal of Quality Measurement and Analysis JQMA, vol. 19, no. 2, pp. 73-79, 2023. [31] M. Khan and M. Khan, "Cryptomarket volatility in times of COVID-19 pandemic: application of GARCH models," Economic Research Guardian, vol. 11, no. 2, pp. 170-181, 2021. [32] M. Ferreira, F. J. Silva, and G. Couto, "How risky are cryptocurrencies?," Applied Economics, vol. 56, no. 58, pp. 8320- 8331, 2024. https://doi.org/10.1080/00036846.2023.2290588 [33] T. Bollerslev, "Generalized autoregressive conditional heteroskedasticity," Journal of Econometrics, vol. 31, no. 3, pp. 307-327, 1986. https://doi.org/10.1016/0304-4076(86)90063-1 [34] D. B. Nelson, "Conditional heteroskedasticity in asset returns: A new approach," Econometrica: Journal of the econometric society, pp. 347-370, 1991. https://doi.org/10.1093/oso/9780198774310.003.0005 [35] J.-M. Zakoian, "Threshold heteroskedastic models," Journal of Economic Dynamics and control, vol. 18, no. 5, pp. 931- 955, 1994. https://doi.org/10.1016/0165-1889(94)90039-6 [36] D. Ardia, K. Bluteau, K. Boudt, L. Catania, and D.-A. Trottier, "Markov-switching GARCH models in R: The MSGARCH package," Journal of Statistical Software, vol. 91, pp. 1-38, 2019. https://doi.org/10.18637/jss.v091.i04 [37] D. J. Spiegelhalter, N. G. Best, B. P. Carlin, and A. van der Linde, "Bayesian measures of model complexity and fit," Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 64, no. 4, pp. 583–639, 2002. https://doi.org/10.1111/1467-9868.00353 [38] T. Ando, "Bayesian predictive information criterion for the evaluation of hierarchical Bayesian and empirical Bayes models," Biometrika, vol. 94, no. 2, pp. 443-458, 2007. https://doi.org/10.1093/biomet/asm017 [39] P. F. Christoffersen, "Evaluating interval forecasts," International economic review, pp. 841-862, 1998. https://doi.org/10.2307/2527341 [40] R. F. Engle and S. Manganelli, "CAViaR: Conditional autoregressive value at risk by regression quantiles," Journal of Business & Economic Statistics, vol. 22, no. 4, pp. 367-381, 2004. https://doi.org/10.1002/ijfe.2043 https://doi.org/10.3390/ijfs10030051 https://doi.org/10.3390/jrfm14070293 https://doi.org/10.3390/jrfm15030113 https://doi.org/10.3390/su15054395 https://doi.org/10.1109/ojcs.2024.3370603 https://doi.org/10.1080/00036846.2023.2290588 https://doi.org/10.1016/0304-4076(86)90063-1 https://doi.org/10.1093/oso/9780198774310.003.0005 https://doi.org/10.1016/0165-1889(94)90039-6 https://doi.org/10.18637/jss.v091.i04 https://doi.org/10.1111/1467-9868.00353 https://doi.org/10.1093/biomet/asm017 https://doi.org/10.2307/2527341