Short Premium Research Dissection (Part 23)

Today I continue with thoughts about our author’s comparison between high-risk and limited-risk strategies (Part 21 table).

I wrote last time about normalizing for DDs and made mention of the top 3 included in her tables. Per my conservative approach, I think it’s important to heed the worst DD and possibly position size based on [150% of] that [since your worst DD is always ahead of you as discussed in second and third paragraph here].

Just looking at worst losses can be misleading, however, which is why I am interested in seeing average* and distribution. One or two outliers [even out of many] can make a bad average look much worse. Given this possibility, I am interested in seeing (or quantifying) the whole [histogram] distribution. Average [win and average] loss along with percentiles (distribution) are part of my standard battery (see second paragraph here).

If the average loss is skewed by an outlier, then maybe I calculate the average without the extreme and attempt to add a hedge to cushion me from greater-than-average losses. I would never re-run the backtest and admire the much-improved performance, though (hello curve fitting). This is something I would do as proactive insurance if it can be done at a low enough cost to preserve profitability for the whole system.

If the average loss for high-risk is significantly larger and deemed to be trustworthy, then an increased position size for the limited-risk strategy may be justified. The standard battery would tell us these things.

Getting back to my Part 22 DD analysis, I am surprised the MDD difference between high and unlimited risk is only a few percentage points despite a horrific crash of 2008’s magnitude. This is only 5% of the total account, however, so perhaps the 34.2%—a percentage of a percentage (often misleading as described in the paragraph below first excerpt here)—better reflects the difference.

This got me looking more closely at the equity curves, which is where something clearly seems amiss. The high-risk equity curve(s) in Part 15 goes horizontal for roughly six months between 2008-9. This makes sense because of the VIX < 30 filter. The defined-risk curve(s) in Part 20, however, is V-shaped across that time interval. It makes sense that these defined-risk positions are hedged and therefore would lose less (or even profit?) during volatility crush but---VIX < 30! That is applicable to both. The defined-risk curve should be horizontal for the same 2008-9 period. If this throws off the remainder of the curve(s) then is her reported CAGR erroneously high? Are her reported DDs erroneously low?

Aside from all of the critical observations I have made throughout this mini-series, perhaps nothing is worse than obvious data flaws or inconsistencies. Any and all conclusions are based off the data and calculations made from it.

This is really bad news. I hope the mistake is mine.

* Arithmetic mean. Use of median is another potential way to correct for outlier distortions.