My first SA article, A Quantitative Strategy For Trading Inverse Volatility With Impressive Backtested Results, was published October 10, 2014. A follow up SA article, Assessment Of A Quantitative Strategy For Trading Inverse Volatility 15 Months Later, was published January 11, 2016. The present Instablog article both updates the progress of the strategy following the close of Cycle 12 on November 21, 2016, and provides refined parameters for ongoing cycles.
Results Based on the Original Parameters
The original model, which included only Cycles 1-10, was purely based on the backtesting of historical and model XIV prices, dating from March 2004 through July 2014. There have been only two "live" cycles since: Cycles 11-12. Results based on the originally published parameters are collected in Table 1. For details about the original parameter values, refer to the original articles.
Superficially, the results of the two live cycles is unimpressive. Between 10/31/14 and 11/21/16, the originally published strategy would have gained only 9.75% over a timeframe of slightly longer than two years. Nothing to write home about!
Successes and Failures in Cycles 11 and 12
From the strategy's standpoint, Cycle 11 was a complete success. From a financial standpoint, not so much-but that is not my focus here. Even though the cycle resulted in a loss of principal, two of the strategy's cornerstones were confirmed. 1. The trailing stop closed the position with less of a loss, compared to the absence of a trailing stop. 2. Following the trailing stop sell, the strategy required that the cycle would not be complete until the XIV price fell further, and closed below a predesignated level. History shows this to have been completely accurate. Cycle 11 did not close in the green; nevertheless, it validated the strategy's principles and expectations for a "losing" cycle.
On the other hand, Cycle 12 can be considered somewhat of a failure because, after the XIV position was exited via the trailing stop, its market price subsequently rose, and eventually closed above the limit sell price to close Cycle 12. In other words, the strategy prematurely stopped out what might have been a successful and profitable cycle. The cause? The trailing stop was too tight. Not by much, but "almost" only scores a point in the game of horseshoes.
Adjustment of Parameters Following Cycle 12
Development of the original model identified an optimal range of values for each parameter, and parameter values that represented the midpoint of each optimal range were published. In the first two articles, the optimal ranges were not discussed. In hind sight, the omission was a probably a mistake, because it may have resulted in an overly simplistic understanding and interpretation of the single-value parameters. The newly optimized ranges and recommended values of each parameter are discussed below.
The failure of Cycle 12 forced reevaluation of the trailing stop parameter. The newly optimized range has a small detrimental effect on the outcomes of Cycles 3 and 7, relative to the original strategy parameters, but improves the model's outcomes via improvement in the outcome of Cycle 12.
Evaluation of the trailing stop trigger, i.e., the price level that must be reached before the trailing stop is set into action, revealed a small, and possibly non-statistical effect on model outcomes. Thus, the trigger was eliminated as a parameter. As a result, the trailing stop is immediately activated at the beginning of each cycle.
The reoptimized parameter value ranges are collected in Table 2. The table also identifies a recommended trigger value within the optimized range. The updated strategy has only three optimized parameters. I know of no other objective, quantitative trading strategy having fewer than three parameters that remotely comes near the success of our strategy-based on the last 12.5 years of market data (Table 3, below). The criticism that the strategy is "over parameterized," which appears in some reader comments, is without intellectual or factual merit.
Parameter 1: Limit Sell and Cycle End (Upper Limit). The optimized range is narrow (0.12%). A trigger value at the bottom of the range is recommended to ensure closing a position in case the upper range value eventually turns out to be too high.
Parameter 2: Trailing Stop. Having eliminated the trailing stop trigger, the trailing stop becomes active immediately when the position is opened. The optimized range spans 0.99%. Because a losing cycle must inevitably drop through and below the entire range, the recommended trigger value is at the bottom of the range.
Parameter 3: Cycle End (Lower Limit). The optimized range is wide (15.1%), resulting in some degree of uncertainty. If the trigger value is too low, it could result in a cycle never being properly closed. If the value is set too high, the cycle could be closed prematurely, which could subsequently result in the erroneous opening of a new cycle. For this reason, the recommended trigger value is picked to be in the middle of the range. Keep in mind, however, that the value could be off by as much as 7.5% in either direction.
Results Based on the Modified Parameters
Table 3 shows how the strategy fares when the three reoptimized parameters are applied.
Differences between Table 1 and Table 3 are boldfaced: all are the direct consequence of loosening the trailing stop. The wider trailing stop slightly degrades the outcomes of Cycles 3 and 7 (losing cycles). Cycle 12 is transformed from a losing cycle to a winning cycle.
When Will Cycle 13 Begin, and How Will It Proceed?
Three conditions must be met to trigger the beginning of Cycle 13. When these conditions have been met, a buy signal will be triggered and Cycle 13 will begin.
1. Cycle 12 must have been closed. (This condition was met on November 21, 2016.) 2. The F1/F2 futures must close (settlement price) in backwardation. As of this writing (November 28), this condition has not yet been met. 3. Subsequently, the F1/F2 futures must revert and close (settlement price) in contango.
The limit sell price and the cycle end (lower limit) values are immediately calculated based on the purchase price. The trailing stop immediately goes into effect, and its value must be updated each day that XIV closes at a new intracycle high.
The position is sold when one of two events occurs. Either 1. XIV closes above the limit sell price. This event also closes a successful cycle. Or 2. XIV closes at or below the trailing stop value. This event does not close the cycle. The losing cycle remains open, but without a position, until XIV closes at or below the cycle end (lower limit) value.
Caveats that Must Neither be Ignored nor Overlooked
The parameter ranges and recommend values can never be considered as "final." They may still require adjustment, following the completion of future cycles. This caveat was clearly stated in the original article. Its importance is emphasized by the premature exit of the XIV position in Cycle 12, based on original parameters.
Past performance neither guarantees nor implies future outcomes. This goes without saying. Nevertheless, some readers' past comments suggest that they either did not understand, or ignored the concept.
The model is based on closing prices for ETPs, and settlement prices for F1 and F2 futures. Intraday prices are not relevant to the model, and cannot be used when applying the strategy.
XIV and SVXY are volatile, and are high-risk ETPs. Nobody should invest any more in this strategy than they are willing to lose. Although a 100% loss is improbable, it could occur.
Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.