Unpacking Alpha in Innovation-Driven ETFs: A Comparative Study of Artificial Intelligence and Blockchain Funds

Abstract

This paper evaluates the performance and portfolio role of Artificial Intelligence (AI) and Blockchain exchange-traded funds (ETFs) based on monthly returns from 2010 to 2025. The findings show that both AI and Blockchain ETFs generate positive alpha and high standalone returns but also display considerable drawdown risk. Their weak correlations with each other and with broad indices highlight diversification benefits, particularly when combined with U.S. benchmarks. Portfolio optimization reveals that Global Minimum Variance (GMV) and Tangency portfolios ascribe lower weights to these ETFs, while Risk Parity portfolios have a more balanced exposure, helping to diversify risks. Efficient frontier analysis highlights that the inclusion of AI and Blockchain ETFs improves the attainable risk–return profiles, even if they are not a dominant allocation. The findings stress that AI and Blockchain ETFs are suitable as satellite holdings. When applied judiciously, they offer the potential to improve diversification and risk-adjusted performance; however, concentrated bets subject investors to undue downside risks. Positioning portfolios around broad-based indices and overlaying modest thematic tilts emerges as a prudent approach to capturing innovation-driven upsides without compromising long-term portfolio resilience.

1. Introduction

Thematic investing strategies have changed the way assets are managed by giving investors specific exposure to technology via specialist ETFs. Investors are especially interested in artificial intelligence (AI) and Blockchain (BC) exchange-traded funds (ETFs) because they might change the game and increase in value over time (Bhumichai et al., 2024).1 These funds allow investors to gain exposure to AI applications across sectors and blockchain uses in finance and supply chains (Adel, 2024; Taherdoost, 2022). Studies indicate that combining AI with blockchain is a potential way to improve data security, transparency, and operational efficiency (Shinde et al., 2021; Wang et al., 2021).

The growth of thematic ETFs focused on AI and Blockchain creates new opportunities and risks. These funds are getting attention from investors because they provide them with access to new and developing technology. However, investors should consider their high volatility, the possibility of big losses, and the fact that their long-term effects on portfolios are still unknown. Investors need to look at the risk–return profile to make smart choices on how to focus on one area or spread their money out over several sectors.

While previous studies have thoroughly investigated broad-based and sector ETFs, far less focus has been directed into theme funds especially associated with AI and Blockchain. Current research on specialized ETFs, such as cannabis (Fabozzi & Malhotra, 2025) and healthcare (Malhotra & Marino, 2024), elucidates performance trends but fails to assess innovation-driven ETFs through comprehensive frameworks, including multi-factor models, downside risk metrics, and portfolio optimization. Consequently, a major deficiency exists in comprehending the performance of AI and Blockchain ETFs in comparison to benchmarks and their potential to provide unique value for diverse portfolios. Assessing the risk–return characteristics of AI and blockchain-focused ETFs requires empirical evaluation (Karoui et al., 2024).

This paper investigates whether AI and Blockchain ETFs outperform traditional benchmarks on a risk-adjusted basis and their potential role in diversified portfolios. The study analyses data from January 2010 to March 2025 using Sharpe, Sortino, Omega ratios, multifactor regression for alpha, Treynor–Mazuy models for market timing, and VaR/CVaR for downside risk. It also explores how incorporating these ETFs with indices like the Russell 3000 impacts portfolio efficiency.

This study explains the dynamics of AI and Blockchain ETFs as both stand-alone assets and as part of a portfolio construction. The study finds that AI and Blockchain ETFs produce positive alpha but exhibit high volatility and tail risk. This work provides a nuanced methodology for investors who might want to think about these assets as a satellite rather than core holdings. The paper’s findings have practical implications for portfolio managers and investors in balancing innovative opportunities with prudent risk control and diversification principles.

The paper is structured as follows: Section 2 reviews prior research on ETF performance, including thematic ETFs. Section 3 details the dataset characteristics. Section 4 explains the methodology for evaluating the risk-adjusted performance of AI and blockchain ETFs. Section 5 presents empirical findings. Section 6 concludes with key insights and implications.

2. Literature Review

Extant research on ETFs has continued to grow in quantity and diversity. Among the early contributions to the literature were investigations of the performance and efficiency of broad-market ETFs (Poterba & Shoven, 2002). The results of these studies supported prior evidence on the positive impacts of ETF structure on performance and tax efficiency and found that the expense ratios of sector ETFs were slightly lower than those of sector mutual funds, on average, and that ETFs were better at tracking their benchmarks during periods of market stress. Overall, the work on sector funds by Harper et al. (2006) revealed the structural advantages ETFs could offer as compared to more conventional investment vehicles.

Building on this concept, Ang et al. (2006) went on to investigate the features and risks surrounding ETFs in general. They argue that, in terms of ETFs’ three most fundamental characteristics—liquidity, transparency, and cost-effectiveness—exchange-traded funds are powerful tools for gaining market exposure. However, they also emphasize the risks associated with ETFs and the need to consider factors such as tracking error, leverage, and tax treatment. Their research suggests that while ETFs are efficient, they may not be devoid of certain practical performance issues. Svetina (2010) conducted a study comparing ETFs to retail and institutional index funds, concluding that ETFs, on average, outperform retail funds and are on par with institutional vehicles, thus reinforcing their place in both retail and institutional contexts.

Attention then shifted to the implications of ETF proliferation for active management. Blitz and Huij (2012) documented how passive, index-tracking strategies reduced reliance on actively managed funds, reshaping asset allocation practices. Yet questions regarding product design and efficacy soon arose. Blitz and Vidojevic (2021) reported that many factor-based ETFs underperformed their benchmarks, raising concerns about whether innovative structures consistently added value or merely repackaged traditional exposures.

Other studies have considered effects at the market level, and on behavior. Chen (2015) found that commodity ETF performance variation is attributed to investor sentiment, leading to tracking errors in times of increased market volatility. Corbet and Twomey (2014) argued that trading of ETFs can increase volatility, especially in illiquid markets, and how ETFs impact market stability.

Sector-focused research has been especially influential in demonstrating how ETF performance varies across industries. Sharifzadeh and Hojat (2012) compared the performance of exchange-traded funds to index funds. They found that while ETFs may offer better returns in some instances, these differences are not statistically significant, implying that other factors likely play a larger role in investors’ choices between ETFs and passive index mutual funds. Blitz and Huij (2012) noted that sector ETFs heightened both systemic and idiosyncratic risks relative to more diversified products. Madhavan and Sobczyk (2016) emphasized liquidity constraints as drivers of tracking error during stress periods, while Broman and Shum (2018) linked ETF liquidity spillovers to increased stock-level volatility in high-volatility states. Elton et al. (2019) found no evidence of persistent alpha among passive mutual funds and ETFs tracking the same indices once differences in systematic risk and fund structure were controlled for, emphasizing that structural and cost-related factors rather than active management explain most return differences, complementing Dolvin (2014) who examined actively managed ETFs and compared their performance to passive funds. He found that active ETFs were more volatile and offered no absolute return advantage, confirming that active management does not systematically outperform passive benchmarks. However, Dolvin noted that when evaluated on relative risk metrics—such as the Information and Treynor ratios—some active ETFs provided diversification benefits, particularly those with higher trading volumes.

More recently, the literature has grown to include themed ETFs, which show that investors want more specific exposures. Fabozzi and Malhotra (2025) studied cannabis ETFs and found that they consistently did worse than standard stocks and had higher chances of losing money. Many observers attributed the strong performance of certain funds during the pandemic to market dislocations rather than to superior security selection or market-timing ability. Malhotra and Marino (2024) examined healthcare ETFs and found that U.S. investors would get some diversification advantages but overall, the risk-adjusted returns would be better. Malhotra and Napoleon (2024) examined Diversified Emerging Market ETFs (DEMETFs), demonstrating that they surpassed international benchmarks and improved global diversity. Collectively, these studies show that specialized ETFs may be useful, but they also warn that success depends on the situation and is typically linked to market cycles and structural design.

Research on technology sector-specific themed ETFs remains limited. Malhotra (2023) examined the performance of technology-oriented mutual funds and exchange-traded funds, finding that technology mutual funds surpassed both technology ETFs and market benchmarks like the NASDAQ-100 in terms of both absolute and risk-adjusted returns. Few studies have evaluated the performance of ETFs centered specifically on artificial intelligence and blockchain using multifactor models, advanced risk-adjusted metrics such as the Sharpe, Sortino, and Omega ratios, or downside risk frameworks like Value at Risk (VaR) and Conditional Value at Risk (CVaR). This lack of research provides motivation for the present study. By comparing AI and blockchain ETFs to benchmarks including the S&P 1500 Information Technology, the NASDAQ 100 Technology, and the Russell 3000 Indices, this research seeks to determine if these funds deliver superior risk-adjusted returns and alpha beyond standard market factors.

This study also looks at extreme risks and how these ETFs impact entire portfolios, evaluating whether they are most effective as central positions or as supporting elements alongside diversified approaches. Building on earlier research, this study examines innovation-focused ETFs to shed light on how thematic investing in the technology sector works for both investors and academics.

3. Data

This study uses monthly returns for Artificial Intelligence (AI) and Blockchain exchange-traded funds (ETFs) from January 2010 to March 2025. The data is from Morningstar Direct. AI ETFs are defined as funds investing in companies that develop AI products and services, enhance AI technologies, or allocate at least 25 percent of their portfolios to firms heavily engaged in AI-related research and development. Some of these ETFs also employ AI methodologies to guide their security selection. Blockchain ETFs, by contrast, invest in firms applying blockchain technologies across industries or in cryptocurrency-linked instruments such as futures, options, and related products.

The growth of these funds over the sample period has been substantial. The number of AI ETFs expanded from 28 in 2010 to 79 by March 2025, collectively managing $75 billion in assets with an average turnover ratio of 87.09 percent and a net expense ratio of 0.54 percent. Blockchain ETFs rose from 14 to 81 over the same period, overseeing $152 billion with a turnover ratio of 50.27 percent and an annual expense ratio of 0.52 percent. These statistics highlight both investor demand and the rapid institutionalization of innovation-driven ETFs.

To benchmark the performance of ETFs, we utilized monthly returns for three popular indices: S&P 1500 Technology Index, NASDAQ 100 Technology Index, and Russell 3000 Index. Each benchmark serves as a distinct reference point tailored to the composition and investment goals of AI and blockchain ETFs. The S&P 1500 Technology Index measures the performance of the broad U.S. technology sector. The NASDAQ 100 Technology Index includes high-growth companies that make it a particularly relevant benchmark for ETFs focused on disruptive industries. The Russell 3000 Index, on the other hand, captures the performance of the entire U.S. equity market, allowing us to contextualize the performance of thematic ETFs within the broader market.

This multi-benchmark method offers a balanced view of AI and blockchain ETF performance across sectors and innovation, showing if they deliver value beyond traditional investments.

Finally, to assess their role in portfolio construction, we examined combinations of AI and Blockchain ETFs with broader equities. These equally weighted portfolios simulate practical asset allocation strategies, testing whether high-growth, high-volatility thematic ETFs can enhance the overall risk–return tradeoff when paired with more stable exposures such as the Russell 3000 Index. While AI and Blockchain ETFs have demonstrated strong standalone returns, their volatility raises doubts about their suitability as core holdings. Evaluating them within diversified portfolios allows us to determine whether they improve efficiency, reduce downside risk, and deliver more consistent long-term performance.

Table 1. Summary statistics of monthly rates of returns for Artificial Intelligence ETFs, Blockchain ETFs and benchmark indices for the period January 2010 to March 2025.

	Monthly Mean Return (%)	Standard Deviation of Monthly Returns (%)	Return to Risk Ratio (%)
Artificial Intelligence Exchange Traded Funds	4.23	27.39	0.15
Blockchain Exchange Traded Funds	5.39	36.55	0.15
S&P1500 Information Technology Index	1.54	5.31	0.29
NASDAQ-100 Information Technology Index	1.50	5.04	0.30
Russell 3000 Index	0.95	5.85	0.16
Equally weighted portfolio of AI ETFs and Russell 3000 Index	2.59	14.64	0.18
Equally weighted portfolio of Blockchain ETFs and Russell 3000 Index	3.23	19.12	0.17
Equally weighted portfolio of AI and Blockchain ETFs	4.88	23.26	0.21
Equally weighted portfolio of AI ETFs, Blockchain ETFs, and Russell 3000 Index	10.55	48.55	0.22

presents the monthly return characteristics of Artificial Intelligence (AI) ETFs, Blockchain ETFs, and multiple benchmark indices from January 2010 to March 2025. These summary statistics include average monthly returns, standard deviations, and risk-adjusted return measures, helping assess both the potential rewards and inherent risks across investment options.

Blockchain ETFs had the highest average monthly return at 5.39%, but they were also the most volatile with a 36.55% standard deviation. AI and blockchain-themed ETFs delivered an average monthly return of 4.23%, albeit with high volatility. Both had a modest return to risk ratio (0.15), indicating high potential but significant risk.

Broader equity indices like NASDAQ-100 Technology and S&P 1500 Information Technology Indices had lower average returns of 1.54% and 1.50%, but better risk-adjusted performance with returns per unit of standard deviation at 0.29 and 0.30. Large-cap technology stocks showed stability and efficiency.

Pairing thematic ETFs with broader market indices improved return-risk profile. For example, combining blockchain-focused ETFs with the Russell 3000 Index reduced volatility to 19.12% while maintaining a strong average return of 3.23%.

Diversified portfolios that mix blockchain and innovation-focused ETFs with U.S. market indices showed strong performance, achieving average monthly returns of over 10% and better risk–return efficiency. Despite high volatility, these combinations offered improved compensation for risk.

Table 1 indicates that concentrated thematic ETFs can yield high returns but are risky alone. When integrated into diversified portfolios, they enhance return potential without significantly increasing risk.

4. Methodology

We build on the descriptive analysis of the previous section and employ a multi-stage methodological framework to evaluate the risk–return performance of Artificial Intelligence (AI) and Blockchain exchange-traded funds (ETFs) relative to benchmarks. The benchmarks include the S&P 1500 Information Technology Index, the NASDAQ 100 Technology Index, and the Russell 3000 Index. These indices collectively capture sectoral (technology-specific), market-wide (broad U.S. equities), and diversified exposures, thereby providing a comprehensive comparative context for the thematic ETFs.

4.1. Risk-Adjusted Performance

To assess risk-adjusted performance, we use Sharpe ratio (Sharpe, 1966), the Sortino ratio (Sortino & van der Meer, 1991), and the Omega ratio (Keating & Shadwick, 2002). The Sharpe ratio evaluates excess return per unit of total volatility, while the Sortino ratio refines this by penalizing only downside deviations, thereby emphasizing risks most relevant to investors. The Omega ratio offers a broader perspective by comparing the probability-weighted gains and losses across different thresholds, capturing return asymmetries that conventional measures may overlook. Collectively, these measures provide a robust framework for evaluating whether the elevated returns of thematic ETFs adequately compensate for their volatility and downside risk.

Beyond ratio-based analysis, we implement a multifactor asset pricing framework to decompose ETF returns and identify their underlying drivers. Factor models are central to investment research as they distinguish between returns attributable to systematic risk exposures and those representing abnormal performance. This study applies an extended Fama–French framework, enabling us to test whether AI and Blockchain ETFs generate alpha—returns unexplained by exposure to established factors such as market, size, and value.

4.2. Multi-Factor Asset Pricing Models

The concept of alpha is at the core of multi-factor asset pricing models. A statistically significant alpha indicates that these ETFs create value beyond standard risk factors, potentially reflecting unique sector exposures, structural design advantages, or favorable positioning within innovation cycles. As a result, the methodology not only measures relative performance but also provides insights into whether AI and Blockchain ETFs represent genuine sources of differentiated returns or simply repackage existing market risks.

To estimate alpha, we apply the Fama-French five-factor model (Fama & French, 2015), extended to include the momentum factor, which captures return persistence driven by recent price trends. The inclusion of momentum is especially relevant for AI and blockchain ETFs, as the companies underlying these funds often experience sharp valuation swings influenced by technological breakthroughs, innovation cycles, and investor sentiment. Ignoring the momentum effect in this context may distort alpha estimates and overlook a well-documented return anomaly.

Equation (1) presents the six-factor model used in this study.

$$R_{i,t}-R_{f,t}={\alpha }_i+{\beta }_1\times \left(R_{m,t}-R_{f,t}\right)+{\beta }_2\times {SMB}_t+{\beta }_3\times {HML}_t+{\beta }_4\times {RMW}_t+{\beta }_5\times {CMA}_t+{\beta }_6\times {MOM}_t+ {\epsilon }_{i,t}$$

(1)

where $R_{i,t}$ = the percentage return for fund i in month t.

$R_{f,t}$ = the yield on US Treasury bill month t.

$R_{m,t}$ = the return on CRSP value-weighted index for month t.

$\left(R_{m,t}-R_{f,t}\right)$ = The variable is the market risk factor and represents the excess return of the overall market and accounts for the general risk associated with investing in the stock market

${SMB}_t$ (Small minus Big) = The SMB factor measures the historical performance difference between small-cap stocks and large-cap stocks. It’s calculated as the return of a portfolio of small-cap stocks minus the return of a portfolio of large-cap stocks. A positive SMB suggests that small-cap stocks have outperformed large-cap stocks.

${HML}_t$ (High minus Low) = The HML factor represents the historical performance difference between value stocks and growth stocks. It’s calculated as the return of a portfolio of value stocks (those with a low price-to-book ratio) minus the return of a portfolio of growth stocks (those with a high price-to-book ratio). A positive HML indicates that value stocks have outperformed growth stocks.

${RMW}_t$ (Robust minus Weak) = This factor measures the historical performance difference between profitable and unprofitable companies. It’s calculated as the return of a portfolio of profitable companies minus the return of a portfolio of unprofitable companies. A positive RMW suggests that profitable companies have outperformed unprofitable companies.

${CMA}_t$ (Conservative minus Aggressive) = The CMA factor represents the historical performance difference between conservative (low investment) and aggressive (high investment) companies. It’s calculated as the return of a portfolio of conservative companies minus the return of a portfolio of aggressive companies. A positive CMA indicates that conservative companies have outperformed aggressive companies.

${MOM}_t$ (Momentum) refers to the tendency of an asset’s relative performance to persist over time. In other words, assets that have exhibited strong performance in the recent past are likely to continue outperforming in the near future, while those with poor recent performance tend to underperform.

ε_i,t = an error term.

4.3. Robustness Checks

Literature on thematic ETFs has often relied on static models and relatively short time horizons, which may obscure important dynamics in innovation-driven and volatile sectors such as Artificial Intelligence (AI) and Blockchain. To ensure that our findings are not artifacts of model specification or sample period, we conduct several robustness checks that extend beyond the unconditional multi-factor framework.

First, we assess robustness by incorporating conditional alpha evaluations of AI and Blockchain ETFs. We also compute Value at Risk (VaR) and Conditional Value at Risk (CVaR), or Expected Shortfall (ES), to reinforce our conclusions and test their stability across alternative methodological frameworks.

The conditional models acknowledge the dynamic nature of factor exposures in innovation-driven ETFs, while the incorporation of downside risk measures extends the empirical framework to better capture fat-tailed losses and extreme volatility that are frequently underestimated by static models. This dual approach allows us to verify whether the risk-adjusted outperformance of AI and Blockchain ETFs is robust to alternative specifications and to provide investors with a fuller understanding of the risks and opportunities these funds present.

4.4. Conditional Factor Models

Traditional performance measures such as Jensen’s (1968) alpha implicitly assume constant risk exposures. Despite their popularity, these measures can be misleading if applied without adjustments in dynamic financial environments, such as rapidly changing sectors like artificial intelligence (AI) and blockchain technology, where the risk premiums and factor sensitivities can shift significantly over time.

To address these shortcomings, Ferson and Schadt (1996) introduced a conditional performance evaluation framework that integrates lagged public information variables into the estimation of alpha. Their approach replaces unconditional betas with conditional, time-dependent betas and expected returns, offering a more dynamic view of fund performance. By including lagged variables such as the three-month Treasury bill rate (TR3M), the slope of the term structure (SLOPE, measured as the yield differential between long- and short-term Treasuries), the corporate bond quality spread (QS, defined as the BAA–AAA spread), and the dividend yield of the S&P 500 Index, the conditional model better captures temporal shifts in risk–return dynamics. Applying these variables with a one-month lag enhances predictive accuracy, thereby improving robustness in performance evaluation.

The following equation presents the resulting conditional models, where Z_j,t−1 represents the demeaned value of the unconditional elements:

$$R_{i,t}-R_{f,t}={\alpha }_i+{\beta }_i\left(R_{m,t}-R_{f,t}\right)+\mathsf{\delta }\times \left\{z_{t-1}\times \left(R_{m,t}-R_{f,t}\right)\right\}+{\beta }_2\times {SMB}_t+{\beta }_3\times {HML}_t+{\beta }_4\times {RMW}_t+{\beta }_5\times {CMA}_t+{\beta }_6\times {MOM}_t+{\epsilon }_{i,t}$$

(2)

where $z_{t-1}$ represents one or more lagged information variables (e.g., dividend yield, term spread, lagged market return). The product $z_{t-1}\times (R_{m,t}-R_{f,t})$ creates an interaction term (or terms) between each conditioning variable and the market excess return.

Therefore, $\delta$ represents a vector of coefficients corresponding to those interaction terms.

This framework allows us to determine whether the alpha observed for AI and Blockchain ETFs persists after adjusting for time-varying factor exposures.

4.5. Market Timing and Security Selection

A fund manager’s ability to select assets that will generate superior returns in the future is known as selectivity, whereas market timing refers to the ability to adjust portfolio holdings in anticipation of changes in market conditions or asset price movements. Classic studies by Treynor and Mazuy (1966), Kon and Jen (1978), Henriksson and Merton (1981), and Lee and Rahman (1990) examined these dimensions in mutual funds, consistently finding that managers tend to underperform in both market timing and selectivity.

Treynor and Mazuy (1966) introduced a quadratic extension to the Capital Asset Pricing Model (CAPM) to assess these skills. By adding a squared market return term, their model tested whether portfolio returns exhibit a convex relationship with market performance, which would indicate successful timing ability. These early studies found that while fund managers sometimes displayed modest skill in security selection, evidence of consistent market timing ability was weak.

This study analyzes the market timing and selectivity of AI and Blockchain ETFs. Although these funds are passive, their focus on specific innovative sectors can introduce implicit timing and selection characteristics. We use the Treynor and Mazuy (1966) model, which adds a quadratic term to CAPM, to evaluate these factors.

$$R_{i,t}-R_{f,t}={\alpha }_i+{\beta }_1\left(R_{m,t}-R_{f,t}\right)+{\beta }_2\times {\left(R_{m,t}-R_{f,t}\right)}^2+{\epsilon }_{i,t}$$

(3)

The coefficient β₂ provides insight into the manager’s ability to accurately predict market performance by examining the relationship between the portfolio return and the market return in a non-linear manner. A positive and statistically significant β₂ indicates superior market timing skills, suggesting that the manager can effectively anticipate market movements. Conversely, a negative and statistically significant β₂ suggests poor market timing abilities. If β₂ is not greater than 0, it indicates that the manager lacks market timing abilities. On the other hand, α_i represents selectivity, which refers to the manager’s skill in selecting individual securities that outperform the market.

4.6. Conditional Market Timing and Selectivity

To provide a more robust assessment, we also estimate conditional models of market timing and selectivity, following the framework of Ferson and Schadt (1996). This approach incorporates lagged public information variables such as interest rates, the slope of the yield curve, credit quality spreads, and dividend yields into the model to capture time-varying factor sensitivities. By doing so, we test whether the apparent performance of AI and Blockchain ETFs persists once macroeconomic conditions and evolving risk premiums are considered.

$$R_{i,t}- R_{f,t}={\alpha }_i+{\beta }_i\left(R_{m,t}-R_{f,t}\right)+{\delta \times \left\{z_{t-1}\times \left(R_{m,t}-R_{f,t}\right)\right\}+\beta }_2\times {\left(R_{m,t}-R_{f,t}\right)}^2+ {\epsilon }_{i,t}$$

(4)

where $Z_{t-1}$ represents a vector of demeaned lagged public information variables. The interaction term allows market betas to vary with economic conditions, while the quadratic term retains the test for non-linear market timing ability.

By applying both unconditional and conditional frameworks, our analysis evaluates whether AI and Blockchain ETFs display any evidence of selectivity or market timing—either through their structural design or implicit exposure to dynamic innovation-driven sectors. This approach also connects our study to broader literature, while testing whether these thematic ETFs’ performance is better attributed to systematic sector exposure rather than any form of timing or selection skill.

4.7. Analysis of the Downside Risk of Monthly Returns

Value at Risk (VaR) and Expected Shortfall (ES) are key tools for assessing extreme downside risk in portfolio returns. For volatile AI and Blockchain ETFs, these measures reveal resilience in tough market conditions.

VaR estimates the maximum expected loss over a specified horizon at a given confidence level, quantifying potential downside exposure. For thematic ETFs, a higher VaR indicates greater vulnerability under market stress, while a lower VaR signals relatively stronger downside protection. By comparing VaR across AI ETFs, Blockchain ETFs, and benchmark indices, we evaluate how investors might calibrate exposure to balance growth opportunities with loss protection.

Expected Shortfall (ES), also known as Conditional VaR (CVaR), refines this analysis by estimating the average loss conditional on exceeding the VaR threshold. ES captures the severity of losses in tail-risk scenarios and provides a more comprehensive picture of downside vulnerability. An ETF with a lower ES not only avoids frequent extreme losses but also tends to limit the magnitude of those losses when they occur, highlighting differences in resilience across strategies.

Together, VaR and ES provide a richer assessment of the risks embedded in AI and Blockchain ETFs. These measures complement the conditional alpha models by addressing tail risk, thereby ensuring that our conclusions regarding performance and risk-adjusted returns are robust to both dynamic exposures and extreme outcomes.

5. Empirical Results

We begin our empirical analysis of the monthly returns of ETFs relative to various benchmark indices by examining correlation among the monthly returns of ETFs and benchmark indices.

5.1. Correlation Analysis

Table 2. Correlation among monthly returns of Artificial Intelligence ETFs, Blockchain ETFs, S&P1500 Information Technology Index, NASDAQ-100 Technology Index, and Russell 3000 Equally Weighted Index based on monthly returns from January 2010 to March 2025.

	Artificial Intelligence ETFs	Blockchain ETFs	S&P1500 Information Technology Index	NASDAQ 100 Technology Index	Russell 3000 Equally Index
Artificial Intelligence ETFs	1.00
Blockchain ETFs	0.21	1.00
S&P 1500 Information Technology Index	0.22	0.22	1.00
NASDAQ 100 Technology Index	0.23	0.22	0.97	1.00
Russell 3000 Index	0.27	0.22	0.74	0.77	1.00

reveals how the monthly returns of AI ETFs, Blockchain ETFs, and major market indices move in relation to one another from January 2010 to March 2025.

Table 2 shows that ETFs focused on AI and Blockchain are largely uncorrelated with one another. With a correlation of 0.21, the movements of one have little impact on the other. This finding is not surprising as AI ETFs tend to react to the developments in automation and machine learning, while Blockchain ETFs are more responsive to innovations in decentralized finance and the digital-asset market. Since the two themes capture distinct innovation cycles, investors who include both ETFs in a portfolio can meaningfully diversify their holdings and dampen portfolio-level volatility.

AI and Blockchain ETFs show low correlation (0.21–0.27) with major indices like the S&P 1500 Information Technology, NASDAQ-100 Technology, and Russell 3000, indicating they move independently from the broader market. Adding these ETFs to traditional portfolios allows investors to access technology growth without depending on overall market trends.

The benchmark indices, in contrast, are almost perfectly correlated. The S&P 1500 Technology and NASDAQ-100 Technology indices have a correlation of 0.97, and both exhibit strong relationships with the Russell 3000 (0.76–0.79). These high correlations imply that traditional indices provide little diversification benefits when held in combination with one another.

5.2. Analysis of Risk-Adjusted Performance

Table 3. Risk-adjusted performance (Sharpe, Sortino, and Omega ratios) for AI and Blockchain ETFs, S&P1500 Information Technology Index, NASDAQ-100 Technology Index, and Russell 3000 Index based on monthly returns from January 2010 to March 2025.

	Sharpe Ratio	Sortino Ratio	Omega Ratio
Artificial Intelligence ETFs	0.11	0.76	2.41
Blockchain ETFs	0.12	0.72	2.39
S&P 1500 Information Technology Index	0.04	0.06	1.11
NASDAQ 100 Index	0.04	0.05	1.10
Russell 3000 Index	−0.06	−0.07	0.86
Equally weighted portfolio of AI ETFs and Russell 3000 Index	0.09	0.31	1.59
Equally weighted portfolio of Blockchain ETFs and Russell 3000 Index	0.10	0.40	1.78
Equally weighted portfolio of AI And Blockchain ETFs	0.15	0.81	2.60
Equally weighted portfolio of AI ETFs, Blockchain ETFs, and Russell 3000 Index	0.19	0.85	2.77

evaluates the performance of Artificial Intelligence (AI) ETFs, Blockchain ETFs, and benchmark indices by adjusting returns for risk using three measures: the Sharpe ratio, Sortino ratio, and Omega ratio. These metrics help investors understand how effectively each investment compensates for the risk it carries.

AI and blockchain ETFs outperformed traditional benchmarks across these measures. Blockchain ETFs had a Sharpe ratio of 0.12 and an Omega ratio of 2.39, which means that the returns were good for each unit of risk. Their peers, AI focused ETFs, had comparable outcomes, which suggests that both groups did better than traditional indices when taking into consideration volatility and downside deviation.

On the other hand, sector-specific benchmarks such as the S&P 1500 Information Technology Index and the NASDAQ-100 Technology index showed much lower Sharpe and Sortino ratios that range from 0.04 to 0.06 with Omega ratios close to 1.10. These values imply that while these indices are more stable, they offered limited excess returns relative to their risk levels over the sample period. The Russell 3000 Index, a measure of broader U.S. equity markets, fared even worse, with negative Sharpe and Sortino ratios and an Omega ratio below 1.0, indicating poor risk-adjusted performance overall.

Mixed portfolios of thematic ETFs with broad U.S. indices had more favorable outcomes. For example, the Sharpe ratio increased to 0.10 and the Omega ratio increased to 1.78 for portfolios combining the blockchain ETFs with the Russell 3000 Index. Portfolios, including both blockchain and innovation ETFs had the highest Sharpe ratios (up to 0.19) and Omega ratios approaching or above 2.8. The results suggest that portfolios combining thematic ETFs with diversified equities improve returns and stability.

Omega ratios over 2 for both AI and Blockchain ETFs show that returns are skewed toward strong gains, with high volatility. Investors in these areas have better odds of large positive returns. Adding these ETFs to market indices further increases Omega ratios, indicating more efficient return distributions and greater chances of surpassing target returns.

Table 3 points to a clear benefit in using thematic ETFs as complementary holdings rather than stand-alone bets. While individually they carry considerable volatility, their inclusion in diversified strategies meaningfully improves overall portfolio efficiency and offers better protection against downside outcomes.

5.3. Portfolio Optimization

To better understand how volatility and cross-market comovement influence the role of Artificial Intelligence (AI) and Blockchain exchange-traded funds (ETFs) in diversified portfolios, we extend our analysis beyond simple equal-weighted portfolios to a systematic optimization approach. Using monthly data from January 2010 to March 2025, we estimate long-only, mean–variance efficient portfolios and compare them with the Russell 3000 Index, which represents broad U.S. equity exposure.

The optimization results in

Table 4. Optimal Portfolio Allocations and Risk–Return Characteristics: AI and Blockchain ETFs with the Russell 3000 Index.

Portfolio Type	AI ETFs	Blockchain ETFs	Russell 3000 Index	Mean Return (%)	Volatility (%)	Sharpe Ratio
Two-Asset (AI + Russell 3000)	0.18	–	0.82	1.53	6.97	0.21
Two-Asset (Blockchain + Russell 3000)	–	0.14	0.86	1.57	7.84	0.19
Two-Asset (AI + Blockchain)	0.58	0.42	–	4.78	22.54	0.21
Global Minimum Variance (GMV)	0.00	0.00	1.00	0.85	5.86	0.14
Three-Asset (Sharpe-Maximizing) (Tangency Optimization)	0.19	0.14	0.67	2.27	9.70	0.224
Risk Parity (Equal Risk Contribution)	0.17	0.12	0.71	1.96	8.81	0.223

provide important insights into the role of AI and Blockchain ETFs in diversified portfolios.

The analysis of two-asset portfolios demonstrates limited diversification advantages. When combined with Russell 3000, allocations to AI and Blockchain ETFs were relatively modest at 18 percent and 14 percent, respectively, due to their higher volatility compared to the benchmark. The mean monthly returns for these portfolios (1.53 percent and 1.57 percent) slightly exceeded the standalone return implied by the Russell 3000, while corresponding Sharpe ratios of 0.21 and 0.19 indicate only incremental improvements in risk-adjusted performance. Notably, the portfolio comprising solely AI and Blockchain ETFs recorded a significantly higher mean return (4.78 percent), accompanied by pronounced volatility (22.54 percent), which highlights the speculative and cyclical characteristics inherent in these sectors.

Among alternative strategies, the Global Minimum Variance portfolio minimizes risk by investing entirely in the Russell 3000, producing minimal return and the lowest Sharpe ratio. The three-asset tangency portfolio allocated 19% to AI ETFs, 14% to Blockchain ETFs, and 67% to the Russell 3000, achieving the highest Sharpe ratio of 0.224. Even minimal tech ETF exposure improved portfolio efficiency. The risk-parity portfolio assigned 17% to AI, 12% to Blockchain, and 71% to Russell 3000, with a similar Sharpe ratio of 0.223.

The results suggest that AI and Blockchain ETFs can enhance portfolio efficiency when used judiciously. The three-asset Sharpe-maximizing portfolio demonstrates that small allocations to these thematic ETFs improve risk-adjusted performance when anchored by a broad market index such as Russell 3000. Risk-parity optimization further confirms their contribution by balancing their higher volatility against stable assets. Pure AI–Blockchain exposures or unconstrained tangency portfolios carry excessive downside risk. Small, disciplined allocations to innovation-focused ETFs help investors pursue growth without sacrificing portfolio stability.

5.4. Empirical Analysis of Multi-Factor Model

Table 5. Net monthly alpha for Artificial Intelligence and Blockchain ETFs. Alpha is based on monthly returns from January 2010 to March 2025.

	Artificial Intelligence ETFs	Blockchain ETFs
Adjusted R2	0.04	0.04
alpha	1.13	3.18
Mkt-RF	1.66 ***	1.65 **
SMB	0.54	−0.06
HML	−0.89	−0.53
RMW	−0.11	−1.73
CMA	1.90	−0.90
MOM	0.29	−0.84

*** statistically significant at 1% significance level, ** statistically significant at 10% significance level.

(based on Equation (1)) reports the results of a multifactor regression analysis estimating net monthly alphas for Artificial Intelligence (AI) ETFs and Blockchain ETFs over the period January 2010 to March 2025. The model includes exposures to the Fama-French five factors (Mkt-RF, SMB, HML, RMW, CMA) and the Carhart (1997) momentum factor (MOM). The reported alpha represents the risk-adjusted excess return unexplained by these common risk factors, while the adjusted R² indicates the proportion of return variability explained by the model.

Blockchain ETFs produced a monthly alpha of 3.18 percent, suggesting that they delivered substantial abnormal returns above what would be expected based on traditional risk exposures. In comparison, AI ETFs achieved a lower alpha of 1.13 percent. Both alphas were estimated with modest explanatory power (adjusted R² = 0.04), indicating that these factor models explain only a small portion of the variability in ETF returns, unsurprising given the speculative and innovation-driven nature of the sectors.

Factor loadings offer additional insight. Both AI and Blockchain ETFs show strong and statistically significant exposure to the market excess return (Mkt-RF), with coefficients of 1.66 and 1.65, respectively. This affirms their high market sensitivity, consistent with their elevated volatility observed in earlier tables.

The results indicate that other factors exhibit variation. The size factor (SMB) is positively associated with AI ETFs (0.54) but close to zero for Blockchain ETFs (−0.06), indicating a slight preference for smaller-cap companies by AI ETFs. Both ETFs have negative loadings on value (HML), profitability (RMW), and investment (CMA) factors, suggesting a growth-oriented focus common in innovation-focused sectors. Blockchain ETFs show significant negative loading on RMW (−1.73), indicating an inclination toward firms with lower profitability characteristics. Additionally, both ETFs exhibit negative exposure to the momentum factor (MOM), with Blockchain ETFs at −0.84 and AI ETFs at 0.29.

Table 5 highlights that while traditional factor models (Equation (1)) explain only a small share of returns, Blockchain and AI ETFs consistently generate positive and meaningful alpha. These results reinforce their role as high-growth, high-risk instruments potentially offering returns beyond what is captured by conventional asset pricing factors.

Table 6. Net monthly alphas for Artificial Intelligence ETFs, Blockchain ETFs and benchmark indexes for five-factor plus momentum model from January 2010 to March 2025.

	Artificial Intelligence ETFs	Blockchain Exchange Traded Funds	S&P 1500 Information Technology Index	NASDAQ 100 Technology Index	Russell 3000 Index
January 2010 to March 2025	1.13	3.18	−1.02 ***	−0.98 ***	−1.35 ***

*** statistically significant at 1% significance level.

compares the alpha based on expanded asset pricing model that includes the Fama-French five factors and the momentum factor estimates for Artificial Intelligence (AI) ETFs, Blockchain ETFs, and three major benchmark indices that include S&P 1500 Information Technology Index, NASDAQ 100 Technology Index, and Russell 3000 Index over the period from January 2010 to March 2025.

According to Table 5, Blockchain ETFs demonstrate the highest net alpha at 3.18 percent per month, while AI ETFs follow with a net alpha of 1.13 percent. Although these figures indicate potential for positive risk-adjusted performance exceeding what standard risk factors account for, the alphas are not statistically significant. Consequently, the findings do not provide sufficient evidence to conclude that either AI or Blockchain ETFs achieved abnormal returns during the sample period.

In sharp contrast, all three benchmark indices reported negative and statistically significant alphas at the 1 percent level. The S&P 1500 Information Technology Index (−1.02 percent), NASDAQ 100 Technology Index (−0.98 percent), and Russell 3000 Index (−1.35 percent), underperformed relative to what would be predicted by their exposures to market, size, value, profitability, investment, and momentum factors.

These negative alphas imply that, even after adjusting for common sources of risk and return, the benchmark indices failed to deliver excess performance over the study period. This underperformance could reflect the broad-based market challenges during the post-2010 period, such as volatility during the COVID-19 pandemic or global macroeconomic headwinds, which the more focused and agile thematic ETFs were better positioned to navigate.

Table 7. Net monthly alphas for Artificial Intelligence ETFs and Blockchain ETFs based on conditional five-factor plus momentum model from January 2010 to March 2025.

	Alpha
Artificial Intelligence Exchange Traded Funds	1.25
Blockchain Exchange Traded Funds	3.38

presents the conditional alphas (Equation (2)) for Artificial Intelligence (AI) and Blockchain ETFs based on a conditional five-factor model augmented with momentum. Conditional alpha measures the average excess return an investment generates after accounting for time-varying exposures to key risk factors. Unlike traditional alpha, this approach adjusts for changing market conditions, making it particularly relevant for assessing strategies that respond dynamically to evolving economic environments.

Based on monthly returns from January 2010 to March 2025, AI ETFs have a modes positive alpha of 1.25 percent, but Blockchain ETFs have a greater alpha of 3.38 percent. However, these alphas are not statistically significant, so there’s insufficient evidence that their outperformance is real or repeatable.

Positive alphas may appear promising, but they lack statistical support for sustained outperformance. These ETFs can benefit from tech enthusiasm, yet such gains are typically short-lived, indicating their speculative and cyclical nature rather than long-term structural strength.

5.5. Empirical Analysis of Market Timing and Security Selection

Table 8. A summary of results from unconditional and conditional Treynor and Mazuy (1966) models. For the Treynor and Mazuy (1966) models, α_i measures selectivity whereas β₂ measures market-timing. T-stats are in parentheses.

	αi	β2
Treynor and Mazuy Model (Equation (3))
Artificial Intelligence ETFs	0.74 (0.31)	0.03 (0.47)
Blockchain ETFs	1.38 (0.43)	0.04 (0.48)
Conditional Treynor and Mazuy Model (Equation (4))
Artificial Intelligence ETFs	2.87 (1.17)	−0.08 (−1.06)
Blockchain ETFs	0.87 (0.26)	0.08 (0.78)

presents the results of the unconditional and conditional forms of the Treynor and Mazuy (1966) model (Equations (3) and (4)), which are used to examine the market-timing and security-selection properties of AI and Blockchain ETFs. In this framework, the intercept term (α_i) captures the manager’s ability to select outperforming assets, while the coefficient on the squared market excess return (β₂) represents the manager’s ability to time the market.

In the unconditional Treynor–Mazuy model (Equation (3)), both AI and Blockchain ETFs show modestly positive selectivity (0.74 and 1.38) and market-timing coefficients (0.03 and 0.04), but all have low t-statistics, indicating no statistically significant skill. The excess returns are likely due to structural or thematic exposures rather than active selection or timing.

The conditional Treynor–Mazuy model (Equation (4)) incorporates lagged macroeconomic variables including interest rate spreads, credit quality differentials, and dividend yields—as conditioning factors to account for time-varying sensitivities to market conditions. The inclusion of these variables alters the results. AI ETFs exhibit a higher conditional alpha of 2.87 (t = 1.17), which may indicate enhanced selectivity performance after adjusting for economic conditions; however, this coefficient does not reach statistical significance at conventional thresholds. In contrast, the associated timing coefficient, β₂, is negative (β₂ = −0.08, t = −1.06), implying that AI ETFs have exhibited some degree of underperformance during rising market environments. This finding is consistent with the underperformance of the innovation-related sectors in the early stage of the economic recoveries.

In contrast, blockchain ETFs exhibit a positive, though modest, conditional alpha (0.87, t = 0.26) and a positive timing coefficient (β₂ = 0.08, t = 0.78). While these parameters are not statistically significant, the reversal in sign relative to AI ETFs may suggest that blockchain-focused funds demonstrate a more favorable response to improving market conditions. This could be attributed to their sensitivity to speculative momentum and the adoption of digital assets.

Both models show no statistically significant evidence of selectivity or market-timing ability for AI and Blockchain ETFs. The small positive alpha likely reflects the benefit of their themes, not managerial skills, which aligns with existing research showing market timing ability is uncommon even among active fund managers. As a result, investors’ actual performance from these ETFs likely stems from sectoral concentration and innovation-driven momentum, not from dynamic reallocation or predictive trading strategies.

5.6. Empirical Analysis of Downside Risk

Table 9. Value-at-Risk and Expected Shortfall for Artificial Intelligence ETFs, Blockchain ETFs, S&P 1500 Information Technology Index, NASDAQ 100 Technology Index, and Russell 3000 Index.

	Value Risk (VaR) at 95% Confidence Level	Conditional Value at Risk (CvaR) at 95% Confidence Level
Artificial Intelligence ETFs	−7.34	−9.31
Blockchain ETFs	−11.64	−15.64
S&P 1500 Information Technology Index	−8.15	−9.25
NASDAQ 100 Information Technology Index	−7.54	−9.12
Russell 3000 Index	−8.68	−11.21
Equally weighted portfolio of AI ETFs and Russell 3000 Index	−6.5	−9.74
Equally weighted portfolio of Blockchain ETFs and Russell 3000 Index	−7.34	−9.31
Equally weighted portfolio of AI And Blockchain ETFs	−8.71	−11.12
Equally weighted portfolio of AI ETFs, Blockchain ETFs, and Russell 3000 Index	−24.60	−32.10

summarizes the downside risk profile of Artificial Intelligence (AI) ETFs, Blockchain ETFs, and several benchmark indices using two widely adopted risk measures: Value at Risk (VaR) and Conditional Value at Risk (CVaR), both calculated at the 95 percent confidence level. These metrics estimate the potential for extreme losses in adverse market conditions over the monthly return horizon from January 2010 to March 2025.

Standalone blockchain ETFs have the most downside risk, with a monthly VaR of −11.64 percent and a CVaR of −15.64 percent. These numbers show how much risk investors take when they hold these ETFs on their own. They have a lot of potential for high returns, but they are also very sensitive to big drops, which makes them especially vulnerable when the market is stressed.

The S&P 1500 Information Technology, NASDAQ-100, and Russell 3000 are examples of broad market indices that have lesser downside risk. Their VaR numbers range from −7.54 to −8.68 percent, and their CVaRs are usually between −9.12 and −11.21 percent. This shows that they are reliable, basic parts of a portfolio, but even these indices can lose money when the markets are volatile.

Blending blockchain ETFs with traditional equity indices reduces downside exposure meaningfully. For example, a 50-50 combination of blockchain ETFs and the Russell 3000 narrows VaR to −7.34 percent and CVaR to −9.31 percent and it is significantly better than holding blockchain ETFs alone.

The portfolios that combine both blockchain and emerging tech ETFs with a broader index exhibit higher overall downside risk due to the elevated volatility of their components. However, their performance should not be viewed purely through the lens of short-term risk. As discussed earlier in Table 1 and Table 3, these diversified portfolios also deliver some of the strongest risk-adjusted returns. The higher VaR and CVaR figures—such as −24.60 percent and −32.10 percent for the portfolio including both thematic ETFs and the Russell 3000, should therefore be weighed against their superior efficiency in capturing upside when markets perform well.

VaR and CVaR results show that thematic ETFs are sensitive to downside shocks, with standalone Blockchain ETFs experiencing greater tail losses. Diversifying with broad indices reduces these risks and improves the Omega profile, illustrating how thematic ETFs can both capture innovation gains and enhance portfolio asymmetry when combined with stable benchmarks.

6. Summary and Conclusions

The study evaluated the performance of blockchain-themed ETFs, as well as other AI ETFs, over an extended period, specifically from January 2010 to March 2025. To analyze their performance, we employed various techniques, including raw return comparison, risk-adjusted return ratios, multifactor regression analysis, and downside risk assessment. The objective of this research is to determine whether the inclusion of these thematic ETFs provides unique value in a portfolio when compared to traditional investment benchmarks, such as the S&P 1500 Information Technology Index, NASDAQ 100, and Russell 3000 Indices.

The raw return analysis indicates that the thematic ETFs in question consistently outperformed the selected benchmarks in terms of average monthly returns throughout the sample period. However, they also exhibited a substantially higher level of volatility, which suggests a more turbulent risk profile. In contrast, the ratios including Sharpe, Sortino, and Omega revealed more favorable risk–return characteristics for these ETFs, indicating that investors were adequately compensated for their risk-taking—particularly when those risks were mitigated through diversification.

Multifactor and conditional alpha analyses show that AI and Blockchain ETFs have positive, but statistically insignificant, alphas. These gains seem mainly due to sector concentration and investor interest in innovation, rather than lasting outperformance or unique managerial skills.

There is no strong evidence of market-timing or stock-selection skill. Small positive alphas and weak timing coefficients show these ETFs mainly provide exposure to cyclical, high-growth sectors rather than generating active returns. Downside analysis reveals sensitivity to market corrections, especially for Blockchain ETFs. However, including these ETFs with broader market holdings can reduce risk and enhance return profiles.

As expected, the downside risk analysis confirmed that these ETFs, especially the Blockchain theme, remain highly sensitive to a market downturn. Held in isolation, the Blockchain ETF was the most prone to tail risk. By blending it with a broad-based benchmark, such as the Russell 3000, investors can partially reduce their downside risk while keeping an eye on the positive return potential. This finding is an additional argument in favor of diversification when allocating capital to such high-volatility industries.

Portfolio optimization shows that AI and Blockchain ETFs receive limited allocation in global minimum variance and tangency portfolios, indicating their role in diversification rather than as core holdings. Risk-parity portfolios also favor a modest inclusion of these ETFs, which help broaden the efficient frontier without taking a leading position.

AI and Blockchain ETFs provide thematic exposure and periodic innovation-driven gains, but their excess returns lack statistical strength. They are best used as tactical or satellite holdings to boost diversification, not as main alpha generators. Investors interested in innovation should use measured and diversified allocations to manage volatility and sector risk.