It is highly unusual for the VIX Volatility Model to completely reverse a gain of +54.3% which was made from 11-8-16 through 3-17-17. That said, such a reversal, though rare, can occur. In this case, it was due primarily to the CBOE VIX Volatility Index having one of its worst gap downs in its multi-decade history. This black swan event occurred on April 24, 2017.
Fortunately, VVM is designed to withstand such black swan events. Complete backtests that include the key change discussed below have been conducted from Jan-2009 to present, or more than 8 years. They show that time to recovery from the three drawdowns greater than -20% has been on the order of a few weeks or less:
Drawdown date range (number of days): % down
1) 8-5-13 to 9-26-13 (38 trading days): -27.0%
Time back to high watermark: 14 trading days
2) 10-28-15 to 11-11-15 (10 trading days): -26.8%
Time back to high watermark: 27 trading days
3) 1-12-16 to 2-3-16 (10 trading days): -38.4%
Time back to high watermark: 13 trading days
4) 3-30-17 to 5-18-17 (35 trading days): -40.9%
Time back to high watermark: <in progress>
These results include a key adjustment having been made which is discussed below.
Steep Performance Trajectory
Note that in the first three drawdowns shown above, recovery was fast because volatility was highly elevated. VVM tends to perform best during such periods even when accounting for drawdowns. For example, if you take the drawdown of -38.4%, it set new performance highs 13 trading days later, completely reversing the -38.4% loss, then went on to achieve another +30.5% gain over the next 25 trading days.
Singularly, it's performance trajectory, while volatile at times, is quite steep. Keep in mind this is based largely on backtests though extreme caution has always been taken to avoid critical errors to which backtests are often prone. One big comfort factor is that the current VVM which includes the adjustment made starting with the 4-3-17 sell signal was put to the test starting September of 2016 as I wished to see how it traded in real-time before making it live. It managed to continue its steep performance trajectory matching what it achieved in backtests up until the 4-3-17 sell signal. That said, I wrestled with whether to let members aware of these tests as they may come across as highly coincidental: great performance in real-time up until it goes live on VOSI with the 4-3-17 sell signal. The coincidence is haunting, and there are no good excuses. Nevertheless, the path to achievement is sometimes paved with unfortunate timing. But when it comes to timing the markets, all that matters is profits. The rest is meaningless.
Also note that the current recovery to high watermark is taking place during one of the lowest periods of volatility thus may take longer than in the prior three drawdowns. As mentioned, VVM tends to make its strongest gains during volatile periods. Otherwise, it tends to sit on sell signals, riding an uptrending XIV.
A Key Change
Certain adjustments were made prior to 11-8-16 as noted in the VIX Volatility Model (VVM) Results Table. Note how the growing pains "debugging" process from 12-30-15 to 11-8-16 resulted in a real-time low in the results at $59,931 on 11-8-16. The model then shot up +54.3% through 3-17-17 to $92,488. Another adjustment was then made starting with the 4-3-17 sell signal. As you can see, since that adjustment was made on 4-3-17, losses reversed the 54.3% gains.
While the timing of the adjustment was a bit unfortunate in that a black swan event occurred shortly after the change was made, I will be closely monitoring how VVM performs with and without the recent adjustment made that started with the sell signal on 4-3-17.
That said, the reason behind the key change that allows for a greater fail-safe loss is to substantially reduce the number of whipsaws. In running tests back to early 2009, or more than 8 years, as well as spot checks conducted in prior periods, the risk/reward was materially improved as a result of this adjustment.
More specifically, profits were boosted even beyond the backtested triple digit percentages shown in the small 2009-2016 performance table in the VVM portion of the results section. This came as a result of :
1) suffering far fewer whipsaw trades thus fewer losses
2) staying in position longer to realize substantial profits in some cases that otherwise would have been reduced by the whipsaws along the way
The downside is that with the fail-safe adjustment, the strategy's worst total drawdown (losses from peak to trough) increased to -40.9%. Further, the size of a loss for an individual signal can be greater (see below for specifics). That said, the time to recover losses is much faster than it was prior to the key change as the performance trajectory has steepened.
1) Keep in mind UVXY is highly volatile so after further thorough testing, a fail-safe of 12% has been set. Note, in rare cases, the fail-safe may exceed 12% by a substantial amount.
2) For XIV used in sell signals, a fail-safe of 6% has been set. Note, in rare cases, the fail-safe may exceed 6% by a substantial amount.
In addition to the above, greater losses could also occur due to rare, sizable gap-downs. Brexit and the recent "Frexit" on April 24 are two of the only cases in backtests going back to Jan 2009 where the model was on the wrong side. That's not to say the model has not experienced overnight gaps against its position, but such gaps resulted in smaller one-day losses and often, the whole signal ended up being profitable. IMPORTANT: Fail-safes may trigger at a smaller loss depending on the situation. Based on backtests and real-time trading, the maximum fail-safes are triggered roughly 11% of the time, thus profits shown in backtests have far outweighed such infrequent events. In highly unusual cases, such maximum fail-safes can cluster such that the same buy or sell signal may trigger the maximum failsafe more than once. This would be frustrating were it not for the VVM's ability to make strong gains throughout its entire testing period of more than 8 years.
Please Consider This Carefully
Success especially in volatile strategies such as this one depends on one's ability to avoid getting shaken out by fear during the inevitable drawdowns. VVM walks the tightrope at times as it often avoids getting pushed prematurely into cash while protecting the downside by switching out of its signal if necessary.
Gaps higher or lower can occur especially during one-off events. Note that while gap downs in stocks should be sold, this does not apply to this strategy which deals in volatility. That said, one may wish to manage their stop losses to fit their own personal risk tolerance levels. While some prefer to have their own maximum stop loss levels in place, others prefer to position size smaller so they are more likely to stay in position by following the model's signals.
Remember to expect some drawdowns along the way (though usually contained to less than 20%) as this strategy can be volatile, and nothing goes up in a straight line. That said, time to recover the high watermark is usually fairly rapid as shown above. Note that the model did achieve a +54.3% return from 11-8-16 to 3-17-17 in real-time trading conducted by an actual member, and has done this a number of times over this sort of timeframe in backtests as well as in real-time trading that began in September of 2016. Keep in mind that just because it has been able to do this in backtests and in real-time trading does not mean it must automatically do it again for the following cautionary reasons:
1) Though very rare events, markets can materially change. While the strategy is designed to change with material changes in markets as it is not a static black box strategy, profitability will be affected if the strategy cannot adequately adjust to the changes. There is no guarantee that the strategy will be able to adequately adjust to all market changes in the coming years as some market changes may be less adaptable than others.
2) Backtests, while necessary, are still based on past data, thus no backtest can 100% guarantee that results will continue into the future. This was the motivation in using real-time testing starting in September of 2016 which did prove out up until the 4-3-17 sell signal. That said, the longer the backtest, the better. While it is noteworthy that the strategy was able to achieve a +54.3% return from 11-8-16 to 3-17-17 in real-time trading, as well as strong real-time (but non-live VOSI) results from September to end of 2016, it is a brief timeframe.
3) Should the adjustment made starting on the 4-3-17 signal prove to create too much risk in the coming months even though backtests and real-time currently show otherwise, the strategy will revert back to before the adjustment was implemented.