When I first started out in options a decade ago, there were only basic backtesting systems, namely Option Vue (which, actually just recently went out of business due to lack of adapting and other reasons). It was not only cumbersome and slow to test but the datasets were very limited, we didn’t have the option data in a variety of market types to really test out strategies. We basically had EOD data pre 2011 and I believe 30 min data after that. That gave us what? Like 3 years of good testing ability. Further to that we had less expirations.
This is an era where trades like the Modified Iron condor were popular. These sets of strategies were borne from the limited data set I mentioned above. They had produced consistent unchallenged small gains for a handful of years only to be dinosaured when the inevitable bulldozer came while picking up these pennies (ie Aug 24, 2015). With this new data set and many more to come (Bear 2016, Low vol 2017, Feb 2018), many other trade types started to be developed, using a much more varied market data set. It was also when the first inklings of the pre-PMTT group came about when I started a skype group to talk about the Rhino trade. From the ashes of these previous trade systems, our group was born when Ron took the reigns and created PMTT.
There’s been a lot of talk in my group about curvefitting and having OOS periods now that we have access to automated tools where we can essentially backtest loads of iterations in very quick time. It’s true, when I backtested the HS3EZ, 488 and 484+2LP it took 100+ hours to properly do. I had one set of parameters and if I changed them, that’s another 100 hours :). I am certain I’ve spent 2000 hours+ backtesting in the last decade, if not more. We can now test that AND any changes in a few minutes. This creates a problem of fitting data by removing losing trades by filtering w/ new parameters etc etc. I am betwixt between the two camps of thought when speaking about the PMTT type trades only. When I am looking at the algo trading or TAA, I am firmly and obviously very focused on OOS testing and curve fitting. The edges are less and the variables much more variant. You’re searching for small edges that need a LOT of data to confirm because the edges can be or can come from something much more ambigious as is the case in algorithmic trading. It’s a definite concern for PMTT types of trades, but just not as much.
The PMTT type of trades are not the same thing as algo trading or trading futures or utilizing parameters like the TAA guys use which have less pronounced and even ambiguous edge with much more variables and variability in those variables. Our edge comes from the very robust premium inherent in the market of which acts like insurance and the pricing of this insurance is less variant and affected by less variables than other non-option trade types. The pricing of an option is via the corresponding greeks which I view as almost like a device of rubber bands which can only stretch and pull so far. We don’t need 1000 samples of bear markets and 1000 samples of low vol periods. There’s only so much that can happen in our structures. With that said and related to my betwixt comment I believe that any strategy created from a limited data set needs OOS testing before going full hog, especially if you’re only testing from 2020+. Which I am currently seeing a lot of. I firmly believe we should be using all data available to us to create these strategies (that means 2014+ at very minimum). This gives us the 2014 Oct crash and unrelenting V-Rally, the 2015 crash, the 2016 prolonged slower bear, the 2017 low vol run up, the 2018 crash, the Oct-Dec 2018 bear, the 2020 crash and subsequent huge 2021 rally and so on. I think a strategy can show returns in a full test in those markets as well as random sampling within AND it has a solid hypothesis and theory of why it should work, then it is robust enough for me to slowly add in.
In regard to risk and draw down, I also believe you can appropriately reduce overall risk with solid well thought out and well tested diversification and trade development and you can in fact limit max draw down on the portfolio of trades by doing this. By limiting draw down you increase geometric returns.
I don’t think drawdown equals risk.. It is just not that simple. You can diversify, you can mitigate, and you will have better geometric returns because of that. Risk mitigation=return. My life is just focused on this aspect, reduce risk and draw down for better geometric returns. There is volatility tax and it’s much more attractive to limit your draw down to allow for better compounding. I always say this, but its the time series of returns, the pathway we take in our bets, that is the most important.