Thank you. I was wondering if anyone in this thread actually knew the difference between data science and trivial analytics (e.g., back-testing.) I still have access to fairly chunky hardware from my days of working at Cloudera and Databricks, and I did a fair bit of EDA, diag, and modeling on /ES and /NQ about a year and a half back. No joy, at least with the hypotheses I came up with. This is not in any way to say that DS can't or won't produce some alpha; the limitation here is almost certainly me - I teach the stuff, but I'm more of an engineer than some brilliant researcher whose mind throws off unorthodox ideas by the minute. But I'm fairly sure that someone who simply knows the mechanics of DS tool usage and nothing more isn't going to walk away with easy millions; all that low-hanging fruit was plucked a long, long time ago. Like most other things in trading, if you're going to do well, you'd damn better be good. Just smearing a little Data Science sauce on top of your trading sandwich is not going to do it.
I was "lucky" that I have no very advanced knowledge in math, AI, algos... I tried once to use an AI software package but it was a disaster for me. So I had to work with my limited knowledge. I am strong in logical (but out of the box) thinking and have good analytical skills. So instead of number crunching, I was trying to find out WHY the market behaves like it does. It seems logical that understanding the market has more value than find out for a certain period what the ideal values are for maybe 25 different parameters. And each time you change the period you have to recalculate what the ideal numbers are then as they change all the time. So I tried to find a system that automatically adapts to changing market conditions, so that it always performs (as optimal as possible) well. So no "parameters" problems. The system should also be solid enough to overcome small anomalies without any problem. To me a good system is stable and does not suffer if the "parameters" vary within certain limits. That's necessary as the market does not always behave in a way to give perfect signals. So the system needs some flexibility.
I was "lucky" that I have no very advanced knowledge in math, AI, algos... I tried once to use an AI software package but it was a disaster for me. So I had to work with my limited knowledge. I am strong in logical (but out of the box) thinking and have good analytical skills. So instead of number crunching, I was trying to find out WHY the market behaves like it does. It seems logical that understanding the market has more value than find out for a certain period what the ideal values are for maybe 25 different parameters. And each time you change the period you have to recalculate what the ideal numbers are then as they change all the time. So I tried to find a system that automatically adapts to changing market conditions, so that it always performs (as optimal as possible) well. So no "parameters" problems. The system should also be solid enough to overcome small anomalies without any problem. To me a good system is stable and does not suffer if the "parameters" vary within certain limits. That's necessary as the market does not always behave in a way to give perfect signals. So the system needs some flexibility without losing reliability or performance.
This pretty much describes what is happening. Parameters have "meaning". What the meaning is depends upon the model. I just increased the field of vision and focus such that I hit some of the same limits as before, which indicates the system has integrity. I.e. the meta parameters are consistent although the model itself changed is evaluation intensity. Thanks for the thoughtful comment.