Have you ever wondered if you can automate the strategy creation process?
It is a process that still requires time and effort to learn to go through all the steps and then get to the final product.
But if the steps are somehow the same all the time, can’t I use an automated strategy development software?
I tried it and I want to tell you what happened in this post …
Enjoy the reading
Hi guys, hi from Andrea Unger.
The machine to deliver strategies!
Well, you know what I do, I’m a systematic trader and I develop my own trading systems.
In the Unger Academy, you can find all the information on how I develop and I teach how to develop trading systems, so people are taken from scratch throughout the process of development to get a final product, which is a strategy.
On top of it, you know, I am a fan of diversification, I would say that you need a portfolio of strategies or one strategy will never be enough and you have to produce many strategies to get your own portfolio.
This diversification has to be done in terms of markets and approaches, trend-following, counter-trend, bias, or whatever.
So this is easy.
So obviously it takes a while to get all the principles, it’s easy from some point of view, but it takes time to get through the process in a decent way.
Strategy development software: where does the idea come from?
I’ve been asked and I wonder if there were some software to do it for you.
I mean all this process, in which you start from an idea and you go to the end, the strategy, it’s done step by step, there is a number of tasks to do, all these are explained in the Unger Academy, but if these tasks are so clear, let’s say, to go through, why shouldn’t we have a piece of software that does this for us?
So as I wondered about this myself, I looked about the offers out there, and there is actually a lot of software that claims to do what I was looking for and what many of you ask me about.
So I went out there and I tested it, I took it and not only, I even cooperated with a person, a very skilled person, a software developer, with a precise idea in mind to developed this kind of software.
We’ve worked through it and we run through tests.
Normally the tests of these machines are the standard ones.
You take a portion of data, the “in sample”, on this portion of data you test and you develop something that works and then you take a portion of data which are unknown during the development phase, the “out-of-sample”, there you test how the model you produced works or keeps on working.
If it’s fine, okay, if not you take another direction.
Normally, the bottom phase, you run further tests on an “out out-of-sample”, I mean, another hidden portion of data that is a sort of a live test of this.
So actually I tried to cheat the machine and I really kept data from the feed, I didn’t feed the machine with all the data I just gave only a portion of data to the machine, keeping the remaining portion as a real live market, so to be sure that somehow the software was not having a peek into the future data.
I was able to cheat the machine, I mean, I don’t believe the software was tricky, of course, I believe that the work brought that way doesn’t, unfortunately, produce the expected results, in fact, as a matter of fact, the basket of strategies that can be produced through the process that we put in place, failed in the final verification, the so claimed “live trading”.
Fortunately, the live trading was with a hidden portion of data and not with the real money phase because that would have been very very expensive.
So what happened? Why did this happen?
The developers’ main pitfall
The danger we have as developers is to develop with… is the over-fitting.
I mean to look too in-depth into the data and to catch moves which are, actually part of the noise of the market, the moves of the market are real moves triggered or driven by financials, driven by the identity of a market and noise.
It’s like a drunk guy going home from a pub.
This guy knows where his house is, so the main direction is something he knows, so he knows where to go, but he probably goes zigzagging.
So if you want to know the identity of this guy, the address, the living address, we can find his residential address, seeing the main move, the straight move home, we don’t have to care about all the zigzags because that would be obviously misleading us.
So we have to detach the noise from the main move if you understand the main move we have something to go on.
Sometimes this guy is even too drunk and he even forgets where to go to and he goes to the park, that is an extreme condition when sometimes markets miss completely the identity they have and they get completely crazy.
It doesn’t happen too often, but it happens, and this is obviously the extreme case.
If we look at the normal cases or the normally drunk guys on Friday evening, that’s what we have to work on.
So is the Machine able?
Man VS Machine: who’s the winner?
We are not always, but is the machine which is smarter normally than human beings, not really smarter but more efficient of course… isn’t it able to find out where the noise is and to keep it away from the normal behavior?
I fear not!
I fear that in the end, we are still running through a data mining process where we look into the noise to find good results.
You know, I mentioned this in other videos, I don’t use out-of-sample.
Because I’m lazy?
Not because I don’t want to cheat myself!
Normally what we do is: we test the “in sample”, we verify the “out-of-sample”, if out of sample is good we are happy, if it’s bad we jump back into the “in sample”, we re-play all the parameters then again we look into the out-of-sample and so we do this until we find a satisfactory out-of-sample.
This is cheating because we run through the process so many times until we stumble across something that works, this is a pure coincidence and this is exactly the same that the machine does because the machine even though is able to start from scratch, not really from the work already done, by resampling parameters, but even though it were able to start from scratch again, it still has such a strong computing capacity that scrambling data it will lead to something that works.
So it will find among tons of solutions, those solutions which for pure coincidence work also in out-of-sample.
Obviously not always, I don’t mean the 100% of the strategies will fail, but more or less most of them which is what is important for us, are not what we desire and even in the “in sample” “out-of-sample” when we add an another out out-of-sample phase, a second out-of-sample, we will somehow find something that works well.
Strategy development software: The AI limit
So the cleanup process, the robustness test process, however you call it, in reality, it is just something that is looking into the number of things we need to find those which work.
This is not a robustness test, it’s just picking and picking is not what we want because the picking is done by choosing the pick that works, but not because that one is good, that’s the one that for pure coincidence is working.
The problem, if it’s a problem on this software, is that normally they come with big claims and big names as genetic, out-of-sample, walk forward analysis, Monte Carlo simulation, all things that have big scientific noise in our head and they are good I mean they are all good and the software themselves are good, but unfortunately, I believe, I strongly believe, not for the strategies’ production.
So actually you could use such software to test your own stuff, which means that once you develop strategies, with your knowledge and with your decision-making process, but the steps you went through are your steps, then you could use all the robustness features of this software to scramble your strategies to a point when you verify if it really is as good as you wish.
I mean, I have something that we use in the Unger Academy as well, it’s something that probably we will shape a bit better in the future just to have something really good to go through this, let’s call it, evaluation process, but the point is that the construction process, so far, I don’t believe it can be made by a machine.
Strategy development software: why it can’t be real?
If you think for one moment about this… we are retail traders, this software is normally for retail traders, so if it were so simple, I mean if you could buy a product for 1000 bucks which just pushes out tons of strategies, if not tons, dozens… Wouldn’t guys like Goldman Sachs be able to produce an infinitely more efficient product to really deliver the best strategies in the world?
Are they doing it?
No, they are not!
At least, not in that way.
“You know that Renaissance began a fund doing this…” Not really.
They’re working on a number of things, they are scrambling all the possible data but that’s all human-driven…
I mean, the investigation is done with powerful tools, but the direction of the investigation comes from above.
There are niches of investigation done through powerful machines, but the direction of each niche comes from a human being and so it’s human-driven.
Then the output has such a large application because the fund is very big so that Mr. Simons can obviously play all the small bricks into this enormous pool where his guys are trading, but it’s still cleverly driven by a human being.
It’s not an automated process with big names in it, unfortunately not.
So the reality is that we still have to work, to work hard.
This software is nice, I will probably do something similar but if you believe that I came here today to announce the launch of a new product for Unger Academy the ultimate machine where you click on the mouse and you get your strategy well finished, no, unfortunately not!
The solution according to Unger Academy
I believe something like this will be hardly coming even in the next future, probably some software where you can, as said, evaluate your strategies that’s a possibility, but not something that is doing the job for you.
In the end, come on!
Developing, analyzing markets, understanding markets is not that bad.
I would have loved such software just to gain time, to make more things, to take them to a different level, but I am still compelled to sit here and develop myself my ideas and my strategies and this is the truth, which is not that terrible, as said, but it’s worth to know that the miracle machine, in my opinion, is not out there.
If you purchase such a product, you believe it will make what you desire, I fear you will come back after a while with some losing strategies in your mind and say “Hey guy, you were right!”.
So now tell me what you mean.
I hope not for you, but I fear that it will be the outcome of such a decision.
You have to work, to study and to understand what you are doing.
If you want to do it with Unger Academy, here below you find a link you get more information or keep on following us, every time we post a new video here you will get the information and maybe there is something useful for you too.
That’s it guys, so see you next time.
Ciao from Andrea Unger.
Ciao from Italy.