Investors seeking low-cost, automated, tax-efficient and potentially alpha-generating solutions increasingly follow model portfolios of exchange-traded funds (ETF). Is there a top-down way to characterize those models? In their November 2020 paper entitled “Using Data Science to Identify ETF Model Followers”, Ananth Madhavan and Aleksander Sobczyk apply machine learning methods and cluster analysis to identify all models using at least three iShares ETFs based on monthly holdings data. Using monthly data on positions and accounts holding those positions across all iShares ETFs (370 at the end of the sample period) during January 2013 through June 2020, they find that:
- As of June 30, 2020 there are 4.1 million accounts with total assets at least $275 billion using 7,276 iShares ETF models (compared to 2,857 models in June 2015).
- In June 2015 (June 2020), accounts using models comprise 10% (17%) of total iShares ETF assets under management.
- Assets managed with ETF models is growing faster than overall ETF assets.
- Within ETF model holdings:
- U.S. equity dominates equity allocations.
- Investments in government and investment grade corporate bonds is growing slowly, but allocations to high-yield bonds and TIPS are rare.
- Allocations to Environmental, Social and Governance (ESG), fixed income and equity factor/smart beta ETFs are growing fast.
- In March 2020, the peak of the Covid-19 crash, ETF model assets flow out of fixed income, consistent with rebalancing from fixed income into equities (and enhancing performance during the subsequent equity rebound).
In summary, increasing use of automated ETF portfolio models may drive performance of certain asset classes.
Cautions regarding findings include:
- The methodology is complicated and not accessible to most investors.
- Findings are somewhat vague with respect to informing investor choices.
