Do exchange-traded funds (ETF) exhibit statistically reliable short-term reversal and intermediate-term momentum? In their October 2018 paper entitled “Momentum Strategies for the ETF-Based Portfolios”, Daniel Nadler and Anatoly Schmidt look for reversal and momentum in next-month performance of past winners and past losers for the following 13 universes:
- U.S. Equity ETFs: 28 US equity ETFs with returns available at the beginning of 2006.
- Multi-Asset Class ETFs: U.S. Equity ETFs plus one gold ETF, five international equity ETFs and five bond ETFs, also with returns available at the beginning of 2006.
- 11 U.S. Equity ETF Proxies: formed separately from the stock holdings as of January 2018 of each of SPDR S&P 500 (SPY), PowerShares NASDAQ 100 (QQQ) or one of the nine Select Sector SPDRs.
Every day for each universe, they reform overlapping winner (loser) portfolios consisting of the equally weighted tenth (decile) of assets with the highest (lowest) total returns over the past 21, 63, 126 or 252 trading days and hold for 21 trading days. They consider two test periods: 2007 through 2017, and 2011 through 2007. They use equal-weighted portfolios of all assets in each universe as the benchmark for that universe. They conclude that one portfolio beats another when the difference between average 21-day future returns is statistically significant with p-value less than 0.10. Using daily returns for the specified assets during 2006 through 2017, they find that:
- For the U.S. equity ETF universe:
- There is evidence of reversal for the 21-day lookback interval (losers beat winners and the benchmark) for both sample periods.
- There is no evidence of momentum (that winners beat losers or the benchmark) for any lookback interval or either sample period.
- For the Multi-Asset Class ETF universe:
- There is consistent evidence that losers beat winners for the 21-day lookback interval, but not that losers beat the benchmark.
- There is evidence that winners beat losers for 63-day and 252-day lookback intervals and both sample periods. There is little evidence that winners beat the benchmark over either lookback interval.
- For the SPY/QQQ/sector ETF proxy universes:
- For SPY stocks, there is consistent evidence that losers beat winners and the benchmark for the longer sample period, but not for the shorter sample period. There is little evidence of momentum.
- For QQQ stocks, there is fairly consistent evidence that both winners and losers beat the benchmark, but not that either beats the other.
- For sector ETF stocks, there is little evidence of reversal or momentum, except XLI stocks perform similarly to SPY stocks.
In summary, evidence overall indicates that momentum/reversal strategies are largely ineffective when restricted to U.S. equity ETFs and underlying stocks, but potentially useful when applied to a mix of asset classes.
Cautions regarding findings include:
- Testing is academic rather than practical.
- Sample periods, especially 2011-2017, are very short for evaluation of strategies using past returns up to 252 trading days (one year). There are only 11 and seven such independent lookback intervals in the sample. Use of daily overlapping portfolios offers little mitigation of this concern.
- Reported results are gross, not net. Accounting for the trading frictions from rebalancing each portfolio every 21 trading days would reduce returns. Rebalancing benchmark portfolios at the same frequency would also involve frictions. Less liquid assets may be costly to trade.
- Using large universes of ETFs or stocks likely involves some very similar assets and considerable monthly portfolio turnover due to noise. When assets are similar, momentum strategies may pick the most volatile of the group, translating to volatile portfolios.
- One motivation for momentum strategies is rotation to the best-performing assets under a variety of market conditions. Restricting a universe to equity ETFs and excluding market conditions favorable to some assets (such as for gold and U.S. Treasuries ETFs in 2008) works against this motivation.
- Testing different combinations of 13 asset universes, four lookback intervals and two sample periods introduces considerable data snooping bias, such that the best-performing combinations overstate expectations.
- Using holdings as of January 2018 for the U.S. Equity ETF Proxies impounds look-ahead bias because it ignores stocks dropped from the underlying indexes during the test periods.
See also “Momentum Strategy (SACEMS)”, various research articles on sector ETF momentum and “Doing Momentum with Style (ETFs)”.
