Shorting Leveraged ETF Pairs
Posted in Volatility Effects
September 16, 2009
Studies of leveraged exchange-traded funds (ETF), such as those summarized in “The Unintended Characteristics of Leveraged and Inverse ETFs” and “The Performance of Leveraged ETFs over Extended Holding Periods”, find that the frequent rebalancing actions necessary to maintain targeted leverage substantially affect long-term performance. A reader observed:
“I’ve read so many articles about how the leveraged ETFs are screwy, and they chew up both sides of the market due to their rebalancing, etc. So I’ve been shorting equal amounts of the long and short double ETFs. I’m short the QID and the QLD, short the TWM and UWM, short the UGL and the GLL, and short the DIG and DUG. I figure, if they are bad longs, they must be good shorts. My thinking is that in a STRONGLY trending market, the position may lose some ground, at least temporarily. But in a weakly trending market, or sideways, both will decay nicely. When I look back on the ones that are a few years old, they just melt away (one side more than the other).”
Does this reverse thinking work? To check, we examine the inception-to-date performance of paired short positions for Ultra S&P500 ProShares (SSO) / UltraShort S&P500 ProShares (SDS) and Ultra QQQ ProShares (QLD) / UltraShort QQQ ProShares (QID). Using daily adjusted closes for these 2X and -2X ETFs for the period 7/13/06 (the first date prices for all four are available) through 9/11/09, we find that:
First, we focus on the SSO/SDS pair. Over the entire sample period, while the S&P 500 Index declines by 16%, SSO (2X) and SDS (-2X) both underperform their leverage targets over the long term by falling 46% and 23%. The long-run performance of SDS seems especially out of kilter. The following chart compares the evolutions of cumulative returns for the index, SSO and SDS over the entire sample period.
How does shorting equal dollar amounts of both ETFs at the outset perform?
The next chart shows the evolution of the cumulative return for equal initial short positions in SSO and SDS established at the close on 7/13/06 and held through 9/11/09. We assume that: (1) there is no rebalancing of the two short positions; and, (2) the returns on the proceeds from the short sales cancel the costs of maintaining the short positions. The cumulative return is mostly negative during the first half of the sample period and mostly positive during the second half, with a large surge since March 2009. The terminal value of an initial $20,000 investment on 7/13/06 ($10,000 in each of SSO and SDS) is $27,362. Again, this evolution assumes zero net cost for shorting.
The evolution for equal initial short positions in QLD and QID is generally similar, but it reaches lower values and is more volatile during the first half of the sample period. It is then less volatile during the second half of the sample period. The terminal value of an initial $20,000 investment on 7/13/06 ($10,000 in each of QLD and QID) is $27,366.
Under what conditions does shorting of a leveraged ETF pair work best? Is volatility of the underlying, as an indicator of level of difficulty in rebalancing the ETFs to the targeted leverages, key?
The following scatter plot relates the net daily return for the shorted SSO-SDS pair to same-day CBOE Volatility Index (VIX) over the entire sample period. The Pearson correlation for these two series is 0.00, and the R-squared statistic is therefore also 0.00, indicating no relationship.
The comparable scatter of net daily return for the shorted QLD-QID pair versus same-day CBOE NASDAQ Volatility Index (VXN) has an R-squared of 0.01 and is much more dispersed vertically for low values of volatility.
In case there is a non-linear relationship, we try a ranking.
The final chart shows the average net daily return for the shorted SSO-SDS pair by quintile of same-day VIX over the entire sample period. Results suggest that the shorted pair tends to produce positive returns when volatility is high. The standard deviation of daily returns is extremely high for the top quintile.
Results for the shorted QLD-QID pair are similar, but the average daily net return for quintile 2 is markedly more negative and the standard deviation of daily net returns is much less pronounced for the highest quintile.
We tested relationships between daily changes in VIX and VXN and net daily returns for the associated shorted ETF pairs, but got inconsistent results.
There may be liquidity effects not fully evident from VIX and VXN relationships that materially affect fund rebalancing costs.
We also tested relationships between net daily return for the shorted pairs and several measures of trend in the underlying (recent slope and above/below short-term simple moving averages), but found nothing notable.
Overall impressions are: (1) the sample period is not long enough for reliable inferences that might guide a strategy of shorting leveraged ETF pairs; and, (2) the interval from late 2008 through early 2009 is very disruptive to inferences about such a strategy.
In summary, shorting pairs of 2X and -2X leveraged ETFs may pay off over long periods as rebalancing effects grind on fund values, but data for guiding inference is skimpy and somewhat “wild.”
