When you test a billion combinations, the probability that at least 1 of them pass out of sample is huge. This is still overfitting
What platform do you use to run multiple intraday strategies?
- Thread starter fan27
- Start date
When you test a billion combinations, the probability that at least 1 of them pass out of sample is huge. This is still overfitting
The attached package provides an overview of my current prototype. I am currently re-writing it in C#. The idea is that given a theory of the market, you can easily test it.
For example, in the attached example, I tested the theory that when ETFs have experienced significant selling over a given period of time, at some point a bounce is likely. My goal is to be able to help myself and others find profitable trading strategies. I will offer my technology as an SDK, GUI application and/or provide a consulting service where I will assist users build strategies for their specific platform based on their needs or ideas.
I hope to have the SDK and consulting services ready sometime in Q1 2018.
For example, in the attached example, I tested the theory that when ETFs have experienced significant selling over a given period of time, at some point a bounce is likely. My goal is to be able to help myself and others find profitable trading strategies. I will offer my technology as an SDK, GUI application and/or provide a consulting service where I will assist users build strategies for their specific platform based on their needs or ideas.
I hope to have the SDK and consulting services ready sometime in Q1 2018.
You should look at genetic algos that find strategies. They do the same thing as what you suggest and are more unconstrained. What you are doing is just constraining the search space. It still suffers from the same problem of data mining and overfitting
That is the whole idea. I want to constrain the search space so I can get the results I need very quickly.
How do YOU test to avoid data mining and overfitting?
That's the billion dollar question. Imagine someone has achieved that, why wouldn't he start a hedge fund and get billions of funding? His only cost is machine which is dirt cheap.
To search quickly, you can use other optimization methods like particle swarm or genetic algos
Why would I look to use something like particle swarm or genetic algos when I already have something that solves my problem? Look, I am not looking to develop some software that solves all the mysteries of the market and produces some billion dollar magical algorithm. That is not reality. If my software can efficiently find profitable strategies that have a reasonable chance of performing well in the future through MINIMIZING curve fitting, I will be happy and busy trading those strategies.
i'm asking you imagine a scenario where someone has achieved what you said. Finding profitable strategies without expensive researcher quants .
He will be able to scale big time and run a hedge fund managing billions. How many guys do you know who are doing that?
Hmmm....maybe I will be that guy
If you can do that, why would you need to hook it up to consumer products and sell mass market?