Short Premium Research Dissection (Part 33)

I continue today with the second-to-last paragraph on allocation.

All graphs from previous sections assume allocation. Some graphs study allocation explicitly (e.g. Part 20). Others incorporate a set allocation to study different variables (e.g. 5% in Part 18). Return and drawdown (DD) percentages may be calculated from any of these allocation-based graphs.

I remain a bit uneasy about the fact that so many of the [estimated] CAGRs seen throughout this report seem mediocre (see fifth paragraph here). I am familiar with CAGR as it relates to long stock, which is why I have mentioned inclusion of a control group at times (e.g. paragraph below graph here and third paragraph following table here).

While [estimated] CAGR has me concerned, CAGR/MDD would be a more comprehensive measure (see third-to-last paragraph here). Unfortunately, I am not familiar with comparative (control) ranges for CAGR/MDD on underlying indices, stocks, or other trading systems. Unlike Sharpe ratio and profit factor—metrics with which I am familiar regardless of market or time frame—I rarely see CAGR/MDD discussed.

The larger takeaway may be as a prerequisite to do or review system development. I would be more qualified to evaluate this research report were I to have the intuitive feel for CAGR/MDD that I have for Sharpe ratio and profit factor.*

With regard to the Part 32 graph, our author writes:

     > The two management approaches were profitable… Holding trades
     > to expiration was an extremely volatile approach, while closing
     > trades after 10 days… resulted in a much smoother ride.

That [green] curve looks smoother, but volatility of returns cannot be precisely determined especially when four curves are plotted on the same graph. This is a reason I promote the standard battery (see second paragraph of Part 19): standard deviation of returns and CAGR/MDD are the numbers I seek. Inferential statistics would also be useful to determine whether what appears different in the graph is actually different [based on sample size, average, and variance].

Now back to the teaser that closed out Part 32: did you notice something different between that graph and previous ones?

For some unknown reason, we lost three years of data: 2007-2018 in previous graphs versus 2010-2018 in the last.

This “lost data” is problematic for a few different reasons. First, 2008-9 included the largest market correction in many years. Any potential strategy should be run through the period many people consider as bad as it gets especially when the data is so easily available. Second, inclusion of system guidelines thus far has been made based on largest DDs and/or highest volatility levels: both of which included 2008 [this isn’t WFA]. Finally, when something changes without explanation, the critical analyst should ask why. Omitting 2008 from the data set could be a great way to make backtested performance look better. This would be a curve-fitting no-no, which is why it raises red flags.

* This is a “self-induced shortcoming,” though. Any mention of CAGR/MDD in this mini-series comes
   from my own calculation (e.g. second paragraph below table in Part 20). Our author makes no
   mention of these while omitting many others as well.