Nooby McNoob becomes a quant

[sle]

capacity constrained strat? What is that?

That means that a strategy has small possible absolute dollar output. For example, imagine that there is an ETF that frequently trades at a significant discount/premium to NAV. If this ETF has very little daily volume and you can only make a small amount in dollar terms, you'd say that it's a capacity constrained strategy.

 

[sle]

Why can't folks talk in colloquial terms???

Mainly because you get used to a specific jargon and assume that other people know it. For example, a lot of Americans would say "X is built like a linebacker" which colloquially would be "X is tall and weighs a lot". If you work in finance you might as well learn the jargon.

 

[digitalnomad]

I am also technically inclined both when it comes to mathematics and statistics as well as programming. I learn quickly and have no reaction to stress. What should I lose or add here?

New member here, but been through the grinder already.

Here is my advice for intraday auto-trading:

1) Start with capacity constrained markets (as sle metioned)

2) Focus on event-driven trading/testing (not easy to test)

3) Use a quality tick data provider and software that can implement it

 

[quant1]

Correct, I do plan to develop my own but only after getting experience doing the simple things so I can shake out bugs in my understanding of the platforms I will be using.

Frequencies: I want to stay away from sub-minute trading. I do not have enough confidence to go for long term trades (multiple weeks/months), but that is the eventual goal. I would likely start out creating a simple intraday momentum-based scalping-type algorithm. I know you hear "scalp" and think "omg, bots, can't compete" but I am not competing at the sub-minute level.

Funds: I will start out with $25K and boost it if I can show returns that are somewhat uncorrelated with the market. I'm not sure how to handle losses though. Should I recapitalize to $25K if I can explain the cause of my losses correctly? In any case, if I show uncorrelated streams, I have... uh... much richer people than me who would probably invest with me.

Bouncing ideas: I have people in the theoretical side of things, but not successful traders. This is why I am here. I know people talk down about stocktwits, but I learned a lot from those peeps the last time around.

Languages: C++, Python, Typescript. I've almost literally done it all except for LOB apps (thank fucking god) Hilariously, I used to consult for a well-known industry de-facto standard in derivatives pricing and I'm very proud of the fact that they still use my software in the field. Pride, but no royalties, unfortunately.

Sounds like a very reasonable way to start out. You definitely have the right skill set. The way to work around the unexplained losses situation is by decoupling alpha models from execution models. If you have an alpha model that back tests well but fails live, then you can test the model over the live period to confirm signal strength. If it is still strong, then your execution/fees are the problem. If not, the signal must be refined.

 

[nooby_mcnoob]

Sounds like a very reasonable way to start out. You definitely have the right skill set. The way to work around the unexplained losses situation is by decoupling alpha models from execution models. If you have an alpha model that back tests well but fails live, then you can test the model over the live period to confirm signal strength. If it is still strong, then your execution/fees are the problem. If not, the signal must be refined.

Glad to hear you say that.

I was of the impression that most backtest frameworks allow you to model commissions and slippage as well. Are you saying that even when you model these, the differences from reality are large enough to trash your model in live trading?

 

[quant1]

Glad to hear you say that.

I was of the impression that most backtest frameworks allow you to model commissions and slippage as well. Are you saying that even when you model these, the differences from reality are large enough to trash your model in live trading?

I see the confusion. I'm saying that you should have separation between the models for alpha and execution. And then, you backtest using both at the same time. Basically, you want to be able break apart the effects of the models to the greatest extent possible.

 

[algofy]

Some quant talk and methods make things more complicated than it needs to be. I intentially stay away from all quant education material and I just develop and test in a common sense approach that greatly reduces curve fitting and then I just throw a ton of mud at the wall and every once in a while something sticks. Like allot of things in life, keeping it basic and simple works better, at least for me.

 

[nooby_mcnoob]

Some quant talk and methods make things more complicated than it needs to be. I intentially stay away from all quant education material and I just develop and test in a common sense approach that greatly reduces curve fitting and then I just throw a ton of mud at the wall and every once in a while something sticks. Like allot of things in life, keeping it basic and simple works better, at least for me.

To each their own, for sure, but the guys at Renaissance Technologies would like to have a word with you. Speaking of which... Anyone here know anything about the actual models they are using?

 

[quant1]

No claim to specify methods, but it seems signal processing is of great interest there.

 

[sle]

To each their own, for sure, but the guys at Renaissance Technologies would like to have a word with you. Speaking of which... Anyone here know anything about the actual models they are using?

Yes.
No (to the next question).