Simple Sector ETF Momentum Strategy Robustness/Sensitivity Tests
Posted in Momentum Investing
July 7, 2010
Readers have requested sensitivity testing of the ranking interval for the simple sector momentum trading strategies described in “Simple Sector ETF Momentum Strategy Performance”. These strategies generally reallocate funds monthly, based on highest past return, to one of the nine sectors defined by the Select Sector Standard & Poor’s Depository Receipts (SPDR) exchange-traded funds (ETF):
Materials Select Sector SPDR (XLB)
Energy Select Sector SPDR (XLE)
Financial Select Sector SPDR (XLF)
Industrial Select Sector SPDR (XLI)
Technology Select Sector SPDR (XLK)
Consumer Staples Select Sector SPDR (XLP)
Utilities Select Sector SPDR (XLU)
Health Care Select Sector SPDR (XLV)
Consumer Discretionary Select SPDR (XLY)
As noted in that prior analysis, available data is so limited that sensitivity test results may well be misleading. With that reservation, here are two additional robustness/sensitivity tests: (1) additional benchmarking of prior results against an equally weighted portfolio of sector ETFs; and, (2) comparison of ranking intervals ranging from one to 12 months applied to a narrowed XLK-XLP-XLU set of sectors. Using monthly adjusted closing prices for the sector ETFs and S&P Depository Receipts (SPY) over the period 12/98-6/10 (139 months), we find that:
Calculations assume that:
- Rankings occur just before the close each month such that reallocations can take place at the monthly close.
- Return calculations all commence in January 2000 (to accommodate a 12-month ranking interval).
- The reallocation frequency is low enough that gross returns are adequate for comparisons (ignore trading frictions).
The following chart shows the average monthly gross returns by momentum from rank 1 (highest momentum) through rank 9 (lowest momentum) during January 2000 through June 2010 for a six-month ranking interval and a monthly reallocation (6-1). Results support a belief that the sector with the highest momentum outperforms those with lower momentum, but not a belief that momentum rank systematically predicts future returns. An equally weighted portfolio of the nine sector ETFs (EW), rebalanced monthly, underperforms four ranks and outperforms five ranks. The standard deviation of monthly returns is noticeably lower for EW than for the momentum portfolios.
What happens for ranking intervals other than six months?
The next chart shows the average monthly gross returns for momentum strategies based on ranking intervals ranging from one to 12 months, with monthly reallocation, applied to a narrowed XLK-XLP-XLU set of sectors (to simplify calculations). Results are disconcerting in that: (1) there is no clear pattern in returns across ranking intervals; and, (2) only four of 12 ranking intervals beat or match an equally weighted, monthly rebalanced portfolio of the three ETFs (EW). The standard deviation of monthly returns is noticeably lower for EW than for the momentum portfolios.
In this context, the six-month ranking interval (6-1) and the 10-month ranking interval (10-1) appear to be lucky. Perhaps widespread use of these ranking intervals is germane. Perhaps the three ETFs considered are not diverse enough to capture any momentum anomaly reliably.
To reiterate, the small sample size (only about 12 fully independent 12-month momentum ranking intervals) is very small for these tests. A longer sample period may produce less disconcerting results.
Also, the method/parameters derive from research on individual stocks. While arguable that the findings should carry over to ETFs, there might be confounding forces. For example, hedging practices (using a sector ETF to hedge positions in individual stocks from the sector) might affect translation of an anomaly from individual stocks to ETFs.
In summary, evidence from a rough ranking interval sensitivity test does not support a belief that a simple sector ETF momentum strategy reliably outperforms an appropriate benchmark.
A reader commented:
“Your post seems to confirm an investigation I did a while ago. While the six-month ranking period did reasonably well, if I shifted the reallocation date to a day other than the first trading day of the next month, I saw noticeably different returns. For example, I tested reallocations around option expiration dates and found that the returns were different (worse) than those for reallocations on the first of the month. Even if I shifted the reallocation date from the first to a few days prior to the end of the month, returns were worse. The overall methodology was the same – picking the top sector over the previous x months (if reallocating on option expiration, then I looked back x months on the similar option expiration day.)”
