Advice for aspiring algo-traders

in fairness, i probably fall into that category.

 

That is a good motto.

A lot can be done on strategy and execution design level. I wouldn't use certain order types or entry/exit technics reliance on which would jeopardize the system too much for the period I leave it unattended for. Quite a bit of redundancy was built into execution overtime, most partial broker/data outages are handled automatically. That is one of the biggest reasons all my execution stack is custom.

I'm most vulnerable to broker's order entry going down completely and missing entries. Which is, considering volume of trades done and how often that happens is not much of a big deal for my systems.

There are few failsafes on account level where my software pretty much gives me a call. To handle things remotely/on a go I have a Slack bot. But those are even more rare than broker outages.

 

Your system's 'frequency of failure' appears low enough to not make you lose any sleep, lol.
I like the cut of your jib. What time zone are you in?

 

Regarding point 3, I wonder why one must start with a blank slate and reinvent the wheel. It seems like there could be a set of strategies or strategy templates, that are proven to have some modest edge, that one can use as a launching point.

 

Sure. Like someone is just gonna post modest edge that works, out there . You can make tons of money with modest edge. The hardest thing is to come up with one.

 

One of the better threads that I've seen come out of ET in a very long time.

Having worked with a couple hundred clients - I've come to the conclusion that the vast majority of traders have kinda sucky "discretionary" skills. In fact, client performance improves according to how much discretion I can successfully engineer out of a trading system.

 

Totally agree. We completely engineered discretion out and are fully automated.

 

I wrote AlphaPy, and yeah it's a commitment. Stopped pushing code publicly because you just get a lot of people who want shit for free.

Best,

PTR

 

did you do it by yourself???? This is very impressive.

 

Thanks RD. I did it because it was so frustrating to find any platform that could integrate machine learning into the signals. The whole idea was to integrate all possible TA features and let the model make a simple buy/sell decision with a probability at each point in time. So, you can either connect the ML signal to the systems backtesting engine, or just do traditional systems without the ML.

I just added multiple time frames with Pandas resampling, as well as Heikin-Ashi transformations. The next step is an IEX server and Streamlit dashboard.

Best,

PTR