Assessment Of A Quantitative Strategy For Trading Inverse Volatility 15 Months Later

Includes: SVXY, UVXY, VXX, XIV
by: David Easter


A quantitative strategy for trading inverse volatility was published more than one year ago.

The strategy is both quantitative and objective, and generates impressive backtested results.

Since publication, one cycle of the strategy has unfolded in real time and has resulted in a loss.

The outcome of Cycle 11 is well within reasonable probability expectations of the strategy.

Reevaluation of the strategy leads to the conclusion that changes are not warranted at this time.


On October 10, 2014, I published my first Seeking Alpha article, entitled A Quantitative Strategy for Trading Inverse Volatility with Impressive Backtested Results. The article described a strategy for trading the VelocityShares Daily Inverse VIX Short-Term ETN (NASDAQ:XIV) or the ProShares Short VIX Short-Term Futures ETF (NYSEARCA:SVXY), based on historical VIX closing data, VIX Futures settlement data, historical XIV closing prices (split adjusted), and model pre-inception XIV prices.

Since that publication, a new cycle (#11) of the strategy began on October 21, 2014, and subsequently ended on January 8, 2016. This article summarizes key elements of the strategy, and provides an update that reflects upon real-time performance over the past 14 months. Details underlying the creation of the strategy were published in the 2014 article, and will not be repeated here. Although this article will refer to XIV throughout, it is equally applicable to SVXY.

trading Objective Trading Rules for the Model

I begin by summarizing the essential trading rules of the original strategy. The rules have not changed, but the presentation has been reorganized.

  1. The buy signal, and the beginning of a new cycle occurs when three conditions are met, in chronological sequence. First, the previous cycle must have been closed, as described under Rule (6) below. Second, on a subsequent trading day, the front month (F1) and second month (F2) VIX futures must close and settle in backwardation (F1 > F2). Third, the F1/F2 futures must subsequently revert and close/settle in contango (F1 < F2). It is important to emphasize that only closing/settlement prices at 4:15 pm ET are important in the model. Neither intraday prices nor extended-hour prices are relevant, because the model was derived exclusively from verifiable historical settlement prices; reliable historical data were not available for intraday or extended-hour price action.
  2. The purchase price is established as being equal to the closing price of XIV the same day that the buy signal is triggered. Once again, only closing market prices at 4:00 ET are relevant to the model, for the reason stated above under (1).
  3. The stop loss is immediately set at 63% of the purchase price. To be activated, the closing market price must be less than or equal to the stop loss.
  4. The limit sell is immediately set at 232.8% of the purchase price. To be activated, the closing price must be greater than or equal to the limit sell price.
  5. The trailing stop trigger occurs when XIV closes above 138% of the purchase price. Once triggered, the trailing stop is set at 68% of the cycle's maximum closing price. To be activated, the closing price must be less than or equal to the trailing stop.
  6. The end of the cycle occurs when one of two events occurs: either the position is sold at the limit sell price, or the closing price falls below the stop loss value. It is important to emphasize that closing a position on the basis of the trailing stop in rule (5) does not end the cycle. Following a trailing stop sell, the cycle remains open (but without a position in XIV) until the closing price drops below the stop loss value.

How the Trading Rules Applied to Cycle 11

The previous cycle (#10) closed with a limit sell on July 1, 2014. The F1/F2 futures moved into backwardation on October 9, and subsequently reverted into contango on October 21, 2014, signaling a buy signal for Cycle 11. The purchase price (closing price of XIV) was 32.74.

The stop loss was immediately set to 20.63, the limit sell price was set to 76.22, and the trailing stop trigger was set to 45.18.

The trailing stop trigger was activated May 21, 2015, when XIV closed at 46.18. The highest subsequent closing price was 49.68 on June 23, setting the final trailing stop at 33.78.

XIV closed below the trailing stop on August 24, 2015. The XIV position was sold at the closing price of 29.58, resulting in a loss of 9.7%. Based on Rule (6), this event did not close Cycle 11.

The closing price of XIV dropped to 20.45 on January 8, 2015, which is below the stop loss price. This event signaled the close of Cycle 11.

Before Cycle 12 can begin, F1/F2 futures must close and settle in backwardation on or after January 11, 2016. When the F1/F2 futures subsequently close and settle in contango, a new buy signal will be triggered, and Cycle 12 will begin. Historically, the median interval between consecutive cycles has been about 9 weeks (see below).

Historical Performance of the Strategy Through Cycle 11

The first ten cycles are theoretical in the sense that they are the product of optimized backtesting, whereas Cycle 11 unfolded in real time. The following table summarizes relevant data for the complete sequence of eleven cycles.

The three date columns indicate the date the cycle was opened and XIV was purchased, the date XIV was sold, and the date the cycle ended. The sell and cycle-end dates are identical when the position is sold at the limit price. When the position is sold on the basis of a trailing stop, however, the sell date always precedes the end of the cycle. Three of the cycle end dates are boldfaced to indicate when this has occurred. The hiatus between cycles (difference between one cycle's end and the next cycle's start date) ranges from 1.6 to 22.8 weeks, with a median of 8.8, an average of 9.7, and a standard deviation of 6.2 weeks.

The result column indicates the outcome of each cycle. Results range from a loss of 9.65% to a gain of 153.50%. The cells containing these two values are highlighted. The arithmetic average, over eleven cycles, is a gain of 104%, with a standard deviation of 58%. Eight cycles closed with a limit sell, with cycle results that equal or exceed the limit price. Three cycles closed subsequent to a trailing stop sell: two of the three resulted in a net trading gain, and one lost money.

The cumulative balance column indicates how an initial investment of $10,000 would have theoretically grown over time. Entries represent multiples of $1000. The results do not account for trading fees, brokerage cost, or taxes, and assume that all of the proceeds from one cycle are used for the purchase of XIV in the next cycle. 11 cycles have been completed in the 11.78 years since March 30, 2014.

The maximum cycle drawdown column records the worst running (daily close) value of the position, expressed as a percentage of that cycle's purchase price. In all eleven cases, the cycle's final result is better than its maximum drawdown. The worst drawdown occurred at 31.5% in Cycle 6. Cycle 7 is unique in that the price of XIV never dropped below the original purchase price prior to being sold. The relevant cells for these two cycles are highlighted.

The number of months held is the time between buy and sell, expressed in months. Holding periods have ranged from 4.3 to 16.1 months, with an average of 8.64 and a median of 8.11 months. Eleven cycles have been completed in the 11.78-year span since March 30, 2004. The overall average is slightly less than one complete cycle per year.

The cumulative CAGR column shows the compound annual growth rate, from March 30, 2004, through the cycle's end date. CAGR represents the average annual growth rate over the specified period. The cumulative CAGR dropped below 80% following cycles 3 and 4, but has remained higher than 80% since. The current 11.78-year CAGR, following Cycle 11, stands at about 86%.

Reexamining the Model's Parameters

Following the completion of Cycle 11, I reexamined each of the model's four parameters to determine whether tweaking the original values would yield a final CAGR that exceeds 86.1%. To make a long story short, no changes to the parameter values were found that improve the historical result. The parameters stand, as originally published. That said, a few comments are in order.

The stop loss, currently 63% of the purchase price, has never yet resulted in the actual sale of a XIV position. To date, it has signaled the end of a cycle on three occasions, always subsequent to a trailing stop sell. Based on the analysis of cycles 3 and 7 only, the trailing stop could currently be set anywhere between 46.2% and 63.9% of the purchase price without affecting the results. Both of these two cycles eventually dropped to or below 46.3% before finally turning around. A value at the upper end of the range is preferable for the stop loss, to protect against the eventuality that a future position might have to be exited on the basis of the stop loss. If cycles 3 and 7 serve as any indication, it would not be surprising if the price of XIV continues dropping to somewhere near 15.13, before finally turning upward.

The limit sell price, currently 232.8% of the purchase price, identifies the best closing price (percentage) that is common to all eight of the successful cycles (i.e., cycles that close with a limit sell) without excluding any of those eight cycles. In other words, the limit sell must be close to, but cannot exceed 100% plus the lowest of the successful cycle results in the Table (132.90%). Based on current data, the limit sell can only be set within a narrow range, from 232.72% to 232.87%.

The trailing stop trigger, currently 138% of the purchase price, and the trailing stop, currently 68% of the maximum price, combine to minimize losses in unsuccessful cycles, while avoiding an unnecessary, premature sale of the XIV position when the market is weak, but likely to turn around. Based on current data, the trigger can be set within a broad range from 122.4% to 151.7%. The trailing stop must be set in a narrower range, between 67.8% and 68.0%.

Concluding Comments

The strategy is neither intended nor suitable for short-term trading. The holding period of the eleven round-trip trades (over 11.78 years) ranges from 4.3 to 16.1 months. The strategy averages slightly less than one round-trip trade per year.

Because closing/settlement prices were the basis for optimizing the model's four parameters, making use of automated intraday or extended-hours triggers will compromise the outcomes. Two historical examples will suffice to demonstrate this. (1) On August 21, 2015, the markets opened in a state of intense panic, with XIV opening at its low for the day, down 31% from the previous close. Eventually, the price recovered somewhat, closing down 18%. Using an automated intraday trailing stop would have resulted in a 24% loss in Cycle 11, more than 2.5 times worse than the actual loss. (2) Using an automated intraday limit sell would have capped gains for all eight of the successful cycles at or very near 132.8%. Cumulatively, over the eight successful cycles, doing this would have compromised total profits by approximately 24.6%!

The strategy will not be as effective for shorting the iPath S&P 500 VIX ST Futures ETN (NYSEARCA:VXX) or the ProShares Ultra VIX Short-Term Futures (NYSEARCA:UVXY) as it is for buying XIV or SVXY. This is transparently true because the maximum possible gain from shorting an ETP is 100% per trade, whereas, for each successful cycle that ends with a limit sell, the gain is 133% or more per trade.

One must not expect to make a profit on every trade! Three of the eleven round-trip trades resulted in a trailing stop sell, and one of the three lost money. I guarantee that there will be more future trades, based on the strategy, that will lose money. That is normal, and must be anticipated. The key is that over the long term, gains must exceed losses, preferably by a significant ratio.

Cycle 11 represents the first real-time test of a model that was empirically optimized from historical backtesting. The verifiable result of Cycle 11 was a loss of 9.65%. A setback? Definitely. But by itself, this occurrence does nothing to discredit the merits of the strategy. Note the following four comments.

  1. Like Cycle 11, the XIV positions in Cycles 3 and 7 were also closed on the basis of a trailing stop. The fact that neither of the two backtested cycles incurred a trading loss is a fortuitous outcome of price action earlier in the cycle that raised the trailing stop threshold above the break-even point. There is no basis for expecting that cycles like these will never lose money.
  2. Following the trailing stop sell in Cycle 11, the price of XIV eventually dropped below the stop loss threshold, closing the cycle. This action parallels the pattern observed by the model in Cycles 3 and 7, and provides additional support for the principle that a cycle is never completed and closed on the basis of a trailing stop sell.
  3. Cycle 11 is an example of "stuff happens". Post-cycle analysis of the model and its parameters confirms that the original parameters can neither be tweaked nor improved at this time to increase the 11.78-year CAGR following Cycle 11. Not even by as much as 0.0001%!
  4. I would estimate that between 25% - 30% of all future cycles will involve a trailing stop sale, and that many among these will result in trading losses. The strategy cannot prevent "stuff" from happening. What it can and does do is to provide objective exit signals that are aimed at avoiding unnecessary losses while maximizing CAGR over an extended timeframe.

Neither past performance nor theoretical modeling can provide any guarantee of future performance. The strategy discussed in this article involves a high level of investment risk. The author is not responsible for losses that are incurred by individuals, groups, or other entities who base one or more trades on any of the ideas presented.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.