Short Premium Research Dissection (Part 19)

I continue the research review today with the graph and table shown in Part 18.

Anytime we get one of these “hypothetical portfolio growth” equity curves, I also want to see the standard battery. Going back to Part 15, this includes things like: number of trades, number of wins (losses), distribution of (winning/losing/all) DIT, distribution of losses including max/min/average [percentiles], average trade [ROI percentiles], average win, PF, number of trades, CAGR, max DD %, CAGR/max DD %, standard deviation (SD) winners, SD losers, SD returns, total return, PnL per day, BPR, CAGR/SD returns, etc. I have not been absolutely consistent with the battery, which is why I write “things like” and “etc.” when describing it. The gist is to include enough statistics to provide deep context for performance.

As mentioned in the second paragraph below the graph shown in Part 14, I don’t understand why daily trades were not studied. This would give us a much larger sample size. We wouldn’t get an equity curve, but the equity curve itself does not tell us certain essential details anyway (hence the standard battery).

Our author concludes that sizing approach makes a huge (small) difference with the 10- (16-) delta put. She points out max loss potential is 181% for the 10-delta put, which is why the contract difference is so great between the two approaches. I would like to know number of winners/losers to the upside/downside because the 10-delta put would only underperform the 16 in the face of downside losers. She told us most of the losers actually occur to the upside (see this table).

Either way, I think max loss potential is less important than profitability. The green curve appears to increase ~120% in 11 years, which is 7-8% CAGR. That seems mediocre compared to the SPX average annual return (which would be nice to see plotted as a control comparison). I have struggled throughout over how to understand “hypothetical portfolio growth” (last questioned in Part 16). CAGR could be multiplied by “median margin percentage” to get a better idea of profit potential, but lacking the standard battery we have no drawdown or volatility (SD) information to complete the performance picture.

Our author next addresses why a stop-loss (SL) should be implemented with risk already defined:

     > As the results… demonstrate, using a SL can allow for more contracts,
     > which can amplify returns over time.

This is consistent with the green curve soaring far above the blue. However, the black curve slightly outperforms the red until the very end. Without inferential statistics, I would guess the SL helps in one of two cases (100% or none). Other SL levels could have been tested but were not. Is one of two sufficient to conclude it works and therefore include it as part of the strategy? If not then doing so may constitute curve-fitting.

I want to see a decent number of trades stopped out to know our author is not curve-fitting by sniping the worst (see second paragraph Part 14). Different SL levels will result in different numbers of trades being stopped out. I am skeptical because she did not explore this.

Another mitigating factor is because both total return and drawdown (or SD) are integral components of performance, differences due to leverage or contract size are not necessarily differences at all. Total return-to-drawdown (or SD) ratio is constant regardless of contract size until the extremes are reached (see paragraph below Part 15 graph).

Lumping together SL and position sizing also creates confusion. She position sizes based on a 100% SL, which is also used as the SL. Position sizing could be based on any SL level since it’s just the maximum acceptable loss divided by trade loss at the SL point. As an independent variable, SL level is just another parameter that increases total number of permutations in multiplicative style as shown in the second paragraph of Part 13. Muddling the picture even further is her treatment of allocation as an independent parameter for optimization when allocation is really just another facet of position sizing.