Abstract
The diversification and low-cost benefits of Exchange Traded Funds (ETF) have increased their popularity among retail and institutional investors. Fund managers are investigating incorporating Artificial Intelligence (AI) technologies into ETF portfolio construction for efficiency and optimal performance. The gap in the literature is that a comprehensive risk-adjusted performance analysis has not been performed focusing on actively managed traditional ETF and AI ETFs tracking the S&P 500. It is important to note that a comparison between active and passive ETFs has been performed and papers on how AI influences the returns of portfolios of ETFs, mutual funds, etc., have been conducted. However, no comparison has yet been done on the performance of active traditional ETFs and AI ETFs tracking the S&P 500 Index.
The S&P 500 Index is the unit of analysis for this study because it accounts for a significant share of the U.S. overall market capitalisation, which is why it is often regarded as the greatest measure of how U.S. equities are performing overall. Moreover, machine learning techniques may be employed to keep trading within the acceptable risk coefficient throughout portfolio creation and optimization (Min-Yuh & Lin, 2019). Trading strategies that maximize Sharpe ratios may be created using AI algorithms (Baek, Lee, Uctum, & Oh, 2020), which will improve the risk-adjusted performance of ETFs relative to the benchmark.
This quantitative study based on secondary numerical data compared the performance of AI ETFs and actively managed traditional ETFs tracking the S&P 500, from 2019 to 2021, using five risk-adjusted ratios: the Sharpe ratio, Treynor ratio, Sortino ratio, Information ratio, and Jensen’s alpha. Furthermore, the tracking error of AI ETFs vs traditional ETFs tracking the S&P 500 for the period 2019 to 2021 was analysed and due to the unequal sample size, the statistical significance of tracking errors of the AI ETFs and actively managed traditional ETFs tracking the S&P 500 was compared using Welch’s t-test. In addition, a regression analysis was performed to examine the factors contributing to performance as measured by alpha. The factors include fund expenses, tracking error, and average daily trading volume.
The study found that actively managed traditional ETFs outperformed AI ETFs across all the risk-adjusted performance measures except the Jensen Alpha. The annualised three-year Jensen’s alpha
v
was zero for AI ETFs, whilst negative for actively managed traditional ETFs. Contrary to the risk-adjusted performance, the tracking error achieved by the top five AI ETFs was substantially lower than the tracking error values achieved by the top five actively managed traditional ETFs. Leading to the conclusion that AI ETFs were less aggressive than traditional active fund managers in attempting to achieve active returns. Moreover, with AI ETFs achieving a low tracking error, the majority of the funds had a positive alpha whereas actively managed traditional ETFs mostly achieved a negative alpha throughout the period. The Welch’s t-test proved that there was no statistical significance in the tracking error of actively managed traditional ETFs and AI ETFs at a 5% significance level. The regression model showed that based on long-run coefficients, tracking error is the main factor that best explains the movement or change in alpha. This might imply that insofar as valuating active ETFs, the investor might be better placed looking at active ETFs with a high tracking error because they are more likely to achieve alpha. However, correlation does not mean causation meaning that the strong correlation between a high positive tracking error and alpha does not mean that a high tracking error translates to the fund being able to achieve a positive Jensen’s alpha.
This study contributes to addressing the gap in theory by conducting a comprehensive risk-adjusted performance analysis focusing on actively managed traditional ETFs and AI ETFs tracking the S&P 500. The study further contributes to the practice of investment management by demonstrating that AI could contribute to returns on a risk-adjusted basis.
Keywords: Alpha; Artificial Intelligence; Exchange Traded Funds; Risk-adjusted Performance; S&P 500 index; Tracking Error.