/atg/ - Algorithmic Trading General

machinebro edition

> previous thread
>>> Veeky Forums7581754

> Topics
Quant finance
Algorithmic trading
Systematic trading
Machine learning
Behavioral finance
Math and financial math
Financial engineering
Data scraping/feature engineering/etc

> Github (empty for now)

> Books of interest
"Trading and Exchanges" - Larry Harris
"Options, Futures, and Other Derivatives" - Hull
"Algorithmic Trading: Winning Strategies" - Chan
"Advanced Algorithmic Trading" - Halls-Moore
"Applied Predictive Modeling" - Kuhn
"Bayesian Reasoning and Machine Learning" - Barber
"The Elements of Statistical Learning" - Hastie
"Quantitative Equity Portfolio Management" - Qian
"Econometric Theory" - Stachurski
"Stochastic Calculus for Finance" - Shreve
"A Linear Algebra Primer for Financial Engineering" - Stefanica

Other urls found in this thread:


alright boys, I have a few hours time once again, waiting for convergence on my shitty gpu... Discuss, ask contribute. And I setup a github too, empty for now but I'll publish some interesting shit over the next coming days, hopefully it will give some of you some ideas.

bayesian reasoning to assess investment profitability, yes or no?
or maybe a [neuro-] fuzzy system + case-based reasoning?

That list of books is great. Looking forward to seeing what other gems come up.

personally would go for bayesian modeling over nfs or similar approaches mostly due to wider flexibility and overall simplicity (when looking at the implementation itself, the math can be fairly complex but no need to bother with that). both can work though and its mostly down to preferences, experience and problem definition itself.

This has been a great resource for learning to take what the market gives you. Futures stick to standard deviations based on IV calculated from the night before quite well.


modeling the bayesian math properly is the main reason why i'm hesitant to actually try it, i have never been able to understand it
i think you could achieve decent predictors with a cbr if you do the normalization and clustering properly
also why not gitgud instead of github?

interesting. added to my reading list. im currently compiling a list of research papers to add here, any specific topic you'd like to see mentioned?

Would love to see more about proper back-testing while avoiding biases.

I feel like the modeling math itself is understandable, now the math behind the implementation can be a brainfuck but nowadays with pymc3 or edward or similar is abstracted well enough. but again, start with what you are most familiar with. that said I think preprocessing can present some challenges there. can you link me to a paper that you are trying to implement?
> why not gitgud instead of github
ehh no particular reason really. just the most convenient option. we are unlikely to host anything controversial so it doesnt really matter where it is hosted

very interested in how this thread turns out

i see
i havent really decided on any specific paper / method to follow yet, still reading about it
i'll upload a couple of papers i think could be useful though, give me a sec

noted. I actually might write something up myself as, surprisingly, there aint really all that much out there re timeseries preprocessing, avoiding biases, etc. Chan got some good content on his blog and in one of his books but I find its all down to trial and error. For example, in a ton of academic papers they fucking standard scale (or even worse, just normalize) their features with no regards to lookahead bias and fuck yes the model just scored 80%+ accuracy...

I don't read books, but I'm a web developer and would like to get started on learning at, can you suggest a starting point?

Has anyone else tried trading with a physical modelling approach? I've gotten pretty interesting results by turning everything into dimensionless pi groups like you would find in fluid mechanics.


anonfiles .cc/file/887aad54b6cd20df073ce91c3a0950b8
heres 14 papers
analyses and methods for machine learning approaches in the stock market
(ichimoku_beginner_by_gabor_kovacs not included)

> I don't read books
well I got some bad news for you user... but seriously, are you asking for moocs/online courses? I can find something, give me some time.

constantly experimenting with that. I actually had a pretty decent state space/particle filter model working on btc/usd for a good period of time. was nothing fancy really, just defining everything in terms of classical mechanics + some concepts from gbm + adjusted drift, etc. Would just filter out noise and leave a clear signal. Id just trade on trends, unfortunately not the best approach right now. I still use the output in some of my other models. Can you describe your implementation in more detail, if you can?

Youtube videos, online courses, shit like that please. I have very poor imagination and my brain is limited to looking at what others do and doing the same ;-;

alright give me 10min. Most of the stuff on udemy and similar is kind of shit desu. There's however one very good course around I just have to remember the fucking name of the lecturer.

> experfy.com/training/courses/algorithmic-trading-strategies#curriculum

one of the best courses out there. No grand fuckery with machine learning, no overcomplicated math, strats are based on behavioral finance findings (as it should be when you are starting out). Should give you all the fundamentals you'll need to go forward with this.

Also this list might be of interst:
> quantor.co/online-courses-to-learn-algorithmic-trading-and-quantitative-finance/

nice, will look into it tonight

Thank you, my dude

Finally an actual decent thread on Veeky Forums

What's a good resource for learning stats from basics preferably with a programming language?

"Using R for Introductory Statistics"
or maybe
"R Statistical Application Development by Example Beginner’s Guide"


also if you are interested in non-frequentist stats there's this (python centric)
> github.com/CamDavidsonPilon/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers
I'd also generally recommend to just learn stats as is and then apply to a programming language of your choosing

and also, for just a plain good book on stats (and some probability) look into "All of Statistics" by wasserman

alright my algochads, I'm off for now. we'll get some actual code going next time, watch that github I'll be posting periodically and lets see what we can come up with. good luck to you all in that pursuit of alpha