The obvious risk in shorting UVXY and other vol ETNs is the ever-present threat of a large spike that can wipe out a trader’s account. In studying the historical price gyrations, one can assemble broad probabilities for the frequency and degree of these spikes.

## Assumptions

In conducting this study I had to define the parameters to measure against, the most important of these being where to set the starting point. The most obvious to me was to use an All Time Low as the baseline. All spikes were measured from the most recent ATL to the highest print during the regular session (trough to peak.) This of course leaves out other instances such as spikes from a large pull-back that did not reach a new ATL first, as can be seen by studying the many ‘after-shocks’ from the larger spikes. It also leaves out premarket and after hours prints that typically aren’t included in most OHLC data. Spike ranges include percentages right below the next level, so a 49.7% spike is still logged as a 25% spike. Only after it surpasses 50% is it counted as a 50% spike occurrence.

This study was initially conducted in 2019 so the data set included about 7 years’ worth of data from inception in 2011 when UVXY was 2x levered, and about 2 years at 1.5x levering which began in 2018. While the levering change should certainly effect the results, the impact was minimal and actionably no different.

## Smoothing

Recognizing that trading is inherently messy and unpredictable, precision is less valuable than the larger picture to trade against. Attempts to redefine the assumptions to acquire finer precision will always result in a smaller data set which greatly reduces reliability. What I sought in an end product was a balance between precision, confidence and usability. Additionally, choosing significant levels for categories, conformed to the same idea. I also let the results inform where to set measurement levels and time frames.

## Vol Respects No Clock

Timing markets is elusive if not impossible, so the output of this study can only give ranges as to when and how large the next spike might be. Spikes can be terribly early or late, but the overall cadence has held true long after I initially ran this study.

## Frequency

A spike of 10% or more happens approximately once per month (12 times per year) inclusive of larger spikes. This was confirmed by my End Of Week metrics which showed a median spike of 9.2% each week

Spikes of 25% or more happen every other month or 6 times per year

Spikes of 50% or more happen once every 6 months or twice per year

And spikes of 100% or more happen once every 18 months or every year and a half.

## Recovery

Equally important in how often spikes occur, is how long it takes for the peak of a spike to crush and decay to a new ATL.

10%+ spikes recover in 2 days

25%+ spikes recover in 2 weeks

50%+ spikes recover in 2 months

100%+ spikes recover in 6 months

## Implementation

How do you use this information? This cadence can inform almost every imaginable strategy for volatility trading.

- Costs to carry a long position can be prohibitive. Waiting until a certain spike is ‘due’ or overdue is a way to tip the odds of profiting going long in your favor, even if only slightly.
- Shorts can open hedges or lighten up when a spike is due.
- Playing the crush with long or short Puts is informed by the approximate time to print a new ATL
- Knowing the next spike level that’s due can inform which strikes to use to fulfill a conservative or aggressive stance
- You can extrapolate the probability for the current spike level to reach the next threshold

Understanding that these are merely probabilities and not predictions is key. 100%+ spikes have happened with as few as 10 months between the (Feb 2018 then Dec 2018) to over 2 years between. Aftershocks from the March 2020 spike produced almost a half dozen 50%+ spikes within 6 months and didn’t produce a new ATL for almost 12 months. No single instance can or should inform a trade, but larger perspective demonstrates that these probabilities have and continue to hold true.