Background My trading is 100% fully automated and run against a live (real money account). Primarily focused on micro future contracts MES, MNQ, MYM, and M2K. However, I am now starting to venture out into individual stocks as well. More on that in a moment as its partially the reason for this journal. All of my strategies are generated via machine learning (self developed platform - not using off the shelf software). My execution engine is also self developed. I currently have 448 independently trained strategies focused on the 4 micro contracts listed above running live. Half of those strategies can "swing" meaning hold the contract for up to 21 days, while the other half are pure day trading strategies and are flat by market close. The strategies are evenly divided between long and short. Futures Swing Live Results: 2022-03-01 $30,036.00 2022-02-01 $46,064.75 2022-01-01 $-26,675.25 2021-12-01 $17,146.50 2021-11-01 $7,451.00 Futures Daytrade Live Results: 2022-03-01 $22,113.50 2022-02-01 $18,462.50 I have less live data on the daytrade strategies as I just started training/running those starting in late January. Strategy Generation I have 7 r5.8Xlarge boxes running at AWS constantly training new strategies. Generally speaking I can build about 28 new strategies per week. Which is what I normally target for a new contract. 14 long and 14 short specific strategies. This week I am training a daytrade strategy (flat by EOD) for AAPL. Each week I plan to train a new stock Goals/Journal Discussion I have 3 primary goals for starting this journal: #1 I am bored... everything is automated, even the strategy generation for the most part. Takes only a few minutes of manual work each day to sanity check new strategies. I am looking for this journal to provide something to do. #2 To show what is possible with a fully automated system and hopefully inspire others. #3 To get ideas on which stocks/markets I should train strategies on next - also to discuss best ways to trade those markets. For example, I am now training AAPL. However, I am debating the best way to trade. Should I trade the shares directly or use options in liu of shares? Historically, I found even with winning strategies where the option contract is held less than a day, a noticeable drag exists. My concern with trading the shares directly is if it becomes hard to borrow during extreme moves. I need to be able to reliably count on my strategy executing on both the long and short side. For those who frequently trade AAPL on Interactive Brokers, have you ever seen it have an issue executing a short? I already have the ground work in place to train SBUX as my next stock. After SBUX, which stock would you recommend next?
Looks interesting! I ran some ML generated algos on ES and NQ back in 2019. They made a bunch of money then gave a good chunk of it back...I retired them. Honestly, there was way too much risk involved. How much data do you use for training?
I actually train the strategies on SPY, IWM, QQQ, and DIA - but execute them against the futures. My data goes back to 2002.
Leveraged ETFs provide better volatility than individual companies. UVXY traded ~52M shares friday and rose ~2.8%. Good volume & volatility. UVXY movement is often a mirror image of SPY movement.
What are the inputs to your ML system. What kind of pre-processing are you applying to them? What ML algorithms are you using?
Unfortunately, this is an area I am not willing to discuss. The "secret" sauce to what makes my strategies work is in the data and algorithms. Garbage in = Garbage out.... I believe solving this issue was the most significant discovery I made in my 20 year trading career. Now I can train new strategies essentially at will.
Curious... In your 20 year trading career - since when did you become fully automated? Were you initially trading discretionary? What's the capital base you're utilizing to generate those returns?
Then what's the point of doing a journal other than to say, "Look at how much money I am making?" We already know that it is possible to generate highly profitable systems. You don't have to disclose the pre-processing part, but if you're not even willing to share what inputs your system considers (does it look at only the underlying it is trading, does it look at multiple instruments, does it look at derivatives, does it look at market breadth indicators, VIX, etc...) then what will others learn from reading this journal that they don't already know?
I started out discretionary, however even from the earliest days of my trading career I was attempting to automate. I never was a really successful discretionary trader. Would have been better off keeping my money in a low cost index fund. I started this year with roughly $800k.
I am willing to discuss overall infrastructure... stock selection... lessons learned.. The strategy generation and data that goes into the strategy are off limits. The inputs that I feed into the ML algorithms are critical to the outcomes. About as much as I am willing to say is that it does not just look at the underlying.