Transcription
Have you ever had a back test that looked like this? So certain you were going to mint money, you're already ready to quit your job. Then you start trading it and it looks like this.
This happens to all of us. And it's not just from curve fitting, but it's from a lack of assumptions and how you analyze your data. Today, I'm going to break down in a live example of a spreadsheet how you can better analyze your trades, sample data to avoid this happening to you. And the next time when you find a trade, you can robustness test it before it bites you.
A lot of viewers of the channel that are, you know, using the trade I've done. Some people are saying, "Oh, it doesn't work. The back test is tailed off." And then other people, you know, have four sharps. And this depends a lot on the path that you take to basically start trading these trades because at its core, uh, trading is a game of probability in the face of uncertain future outcomes. And back testing is when we're using historical data to basically try and better predict or make assumptions that we can in turn win.
So to kind of preface this experiment of one I've done is basically the exact same strategy which is MEIC $250 wide and 10 cent slippage. I'm going to take different paths. One would be the full year of 2022. And then what I did with that in the spreadsheet is I basically just took that yearly return and repeated it for three consecutive years. So this would basically simulate getting the um exact return you would expect from the back test.
Then I also took um MEIC from dailies forward and this time I just added 15 cents of slippage. Then I also took uh MEIC this year which has underperformed. That's just the sampling path of the trade for this year. And I repeated this same return stream for basically three consecutive years.
Then I took um only 2022 of the S&P dailies portion which is where everyone likes to trade. So you can see this six-month sample did way better than this six-month sample in 2022. And then I did MEIC with 10 cents slippage to basically say how the fixed assumptions could change.
And what I see a lot with people is they're using kind of two indicators. One would be like backwardly the gamble and ruin problem. And basically what this says is if you sit down at a card table since it has fixed outcomes and fixed inputs and the game is working against you, you're basically going to go broke um guaranteed.
Then a lot of people will also use Monte Carlo simulations which basically also uses fixed outcomes from past data and then tries to assume what the worst case is and give people a sense a false sense of safety. And something that I'd also like to put out as the other part of the Monte Carlo simulation is this was kind of developed for uh, you know, probabilities. It was actually made on the Manhattan Project and in 1913 there was the Monte Carlo fallacy at the famous casino. There's 26 blacks in a row which are basically statistically almost impossible. That's probability that's modeled a lot and the assumptions in these are fixed um moving forward, which trading is not.
And the other thing I really want to point out here too is if we look at these kind of probabilities is we have a 4% premium capture for the full year. Here we have a 2% premium capture when we just added um, you know, an extra 5 cent slippage, comparative to a 3.4% premium capture with 10 cents of slippage. All fixed inputs that could easily happen because with stop losses, we're basically going to do that. Then we have basically no PCR this year. And then the good part of 2022, the magical land of dailies, we have by far the highest PCR.
And when we're going through this, I want to very quickly show kind of what it looks like. So if we took the full year and again, this automatically scales the size with the spreadsheet, we're having a 34% return and 11% draw down and a 1.9 sharp. So we'll just kind of record this here um very quickly and then we'll go minus 11.4%. Then with the next path, we have all of a sudden 22.5% and minus 9.2%, 2%. Then with the next path, we have 1.1% and minus 4.2%. And then on the next path, we have 68.7% and minus 4%. And these are all with fixed variables effectively when we're going through this because we're going to get the same returns through time.
And what I really want to show is just the variance that we have from literally using the same exact trade mechanics and we're getting exactly what we wanted in our back test just from sampling different times in the data set. So you can see that there is a massive amount of variance between these. We can basically range from, you know, a 68% return with a 4% draw down to a negative% return. And these are all the same exact mechanics, the same exact every 15-minute time frame. And then, except for one of them, the same exact slippage and the same exact stop-loss um mechanicals. So these sample paths can drastically change.
And I want to also show you when we look at a Monte Carlo analysis here. And if you guys ever want to do this, all you need is the average return, which I've automatically queried from the inputs here, and the standard deviation, which I've automatically queried from the inputs here. And then you just do your uh norm brand for these two things. And then it'll basically say, hey, over a 20-year back test, it'll query all these. This is your max potential draw down. So this is what most people are doing to quantify their draw downs.
Now, if we just change from 10 cent slippage to uh 15 cent slippage, our Monte Carlo simulation, all of a sudden, boom, we got six more% of our draw down. And then again, same exact mechanics. If we go to the path of 2025 and essentially use this, all of a sudden now we can go bust.
So I want to really show how the assumptions of what people are using with trading of fixed results are really going to hurt them in the long term for what you're doing. Because trading is making optimal decisions in light of an uncertain future outcome. And we are doing good things by fixing our potential win size, fixing our potential loss size by credit targeting, being mechanical, and all those things. But when we only have one trade and that trade is so dependent on 5 cents of slippage or it's so dependent on a win rate going 2% up or down, that's going to be very fragile over time. And the key with all this is it's really not difficult to measure this and trade it, but it is extremely difficult to adapt to the system over time.
So what can we basically do to prevent this? So one thing we can do is we'll kind of do a walk-forward analysis to try to prevent this, which is every 6 months we'll bring forward the prior two years of data and exclude the last six months with six months out of sample, and then we can resize what we're doing and allocate to certain trades. And this is when we have a multitude of diversified trades that can kind of help us a lot.
The other thing that I see is we have to ensure that we have correct assumptions. Again, you can kind of see with the slippage here is 15 cents of slippage, 2% premium capture. 10 cent slippage, 3.4%, which makes a massive difference of, you know, the draw down effectively, you know, cutting by a significant margin uh over this period where 13%, 7%. So if you're choosing trades that need that perfect execution, it's going to be pretty fragile over time and small mistakes could compound.
The other thing with this is we need to understand how correlation really works. Another thing I'll show here is again, all these are the exact same trade mechanics, the exact same slippage except for one of them. And I have also ran a correlation matrix on these from a daily perspective. So you can see here that effectively all these trades are almost perfectly uncorrelated, except for, you know, the slippage to slippage trades, basically.
So when we're just trading condors and we're just using the same mechanics and it's so fragile on a 2% thing, your correlation is actually not good because you're sampling from only one path of data. That additional time is not a good diversification. You need to think of different diversifications like long ball without a stop. You need to think about um how do you have trend following that goes in with me. And like a thing that I don't like about this 1x condor is you're so dependent on execution and you're so dependent on basically the percentages staying right at 50-some percent.
And we can see that, you know, very routinely it breaks down. Like apart from all this stuff, when you look at this net lick line, it's not weird to go extended periods of time where, you know, you're underperforming. Like look, look at this area, like it took three months to go from here, it flattened out up here. You look at 2025 where, you know, you've basically had struggles the whole time. But if you start your path right here, you'd have a four sharp and you'd be killing it. And if you started your path here, you probably would have stopped. And this is why the sizing and everything is so important.
And another way to show this is we can look at this expectancy simulator and it'll basically randomly put in um, you know, probabilities of winning. So we can say when we win, we win 100%, when we lose, we lose 100%. And then let's say our projected win rate is 55% roughly, like MEIC, we have a 10% PCR. So let's say our win rate goes down to 53%. Now we have a 6% PCR. In the past can drastically vary now on any type of random walk moving into the future.
So this is a big problem when we're going through diversification. Diversification is not a matrix that says this, and robustness testing is not a Monte Carlo analysis that says we basically can't have trouble. But it's actually having uniquely different structures we're sampling from. So if you're sampling from a double calendar, if you're sampling from long haul, and if you're sampling from MEIC, and if you're doing that with a dozen different strategies, then the randomness of these Monte Carlo simulations and trying to fix the data with stop losses and all this stuff can really start to converge on a shorter period of time.
And if you use a long ball trade that can have an unlimited payout, that can potentially help offset weird sampling of your data when you get really bad slippage on a trade. So I hope this helped and it was just kind of an example of how we can trade things and potentially improve the performance. But we need to understand the back stops with it. And you see, you know, if we sample five different variants and we run them at a big size, it works pretty well. But if we, you know, get the best possible path, which is, you know, 2022 dailies forward and we go up 5x, we can see dramatically how much better the results are. And this is trading in the face of uncertainty and what I think a lot of people could be served at better understanding what this looks like.
So if you guys have any more questions or enjoyed this video, I'd love to hear your guys' feedback. But I wanted to show the flaws with thinking about fixed outcomes when we're looking at uncertain outcomes in futures. Thank you guys.