Introduction to Machine Learning for Finance | Hacker News Reader

Introduction to Machine Learning for Finance

algofin.substack.com

79 pointsdhruva_k27 comments

My honest opinion is that any ML related to finance that you learn outside of an investment firm is mostly a waste of time. The amount of data and nuance that successful models encompass is staggering. Unless you do something like scrape WSB and predict the next MEME stock, ML for finance makes no sense. You need credit card data, 10ks, tick data, economic forecasts, interest rates, fed moves, etc. You can carve out a niche, but if you do, forget about finding something on the internet that will tell you how to do that.

There are many games at different abstraction levels in finance, and I agree to find an edge at the level you're describing is exceedingly difficult. That said, Fama and French won a Nobel Prize with data stewardship and regressions. There's definitely a place for understanding basic financial data and ML concepts even for the retail investor.

Are u saying that finance books dont really help ?

finance book do help, 'ML for finance' books rarely do, the domain is too new and attracts many charlatans.

You're better off just learning ML from its classics like Hastie & Tibshirani , Tom Mitchell, or Bishop PRML.

And learn finance from its own classics you can find in any "financial engineering" curriculum.

There is one I liked though because of hands-on approach:

"Machine Learning for Algorithmic Trading" https://www.amazon.com/Machine-Learning-Algorithmic-Trading-...

just assume its listed "strategies" are a sort of primitive "hello world"

Anyone have any good books I can read to learn a bit more about this sort of application? I really like the finance stuff, feels like a natural marriage with DS and ML.

I don't know nearly enough to expand on what I mean.

Advances in Financial Machine Learning by Marcos Lopez de Prado is one of the best books i read last year. Especially on back testing, strategy risk are my fav.

way overrated imo, you will learn neither finance nor ML from this book, don't spend your money on this.

Given author's tendency for excessive self-promotion I would take any of his advice with a grain of a salt.

Besides he seems to have lost a ton of money for AQR, given his short tenure there. People who make money rarely publish.

if someone is really interested in finance and ML learn each on its own, in finance start with Hull Derivatives, in ML may be some coursera course.

Also quantstart blog has excellent tutorials and 'getting started' guides.

Working in the field (HFT market maker) and agree with this assessment. The book is pretty useless in practice - it's basically a vehicle of self promotion for his academic career.

>it's basically a vehicle of self promotion for his academic career

i would say that a full ~25% of technical books are this - solely for the purposes of building brand rather than communicating effectively.

What books would u recommend from ur experience? That aren't self promotion

"Trading and Exchanges: Market Microstructure for Practitioners" is a good introductory book that covers how markets work on a high level. It doesn't teach you how to build profitable systems. That you won't find anywhere anyway.

Link for the lazy of a Draft Copy available made in 2002:

http://www.acsu.buffalo.edu/~keechung/MGF743/Readings/Tradin...

Also a weird book-like slide set that explains what the book is about (and the very important concepts/market participants):

https://rkbookreviews.files.wordpress.com/2013/12/trading-an...

Thanks for the recommendations. Since, I dont work on the field as closely and mostly on a observatory capacity. Ill take a look at the recommendations.

This book is great for complete non practitioners. It was such a great read in 2015 when I graduated and went to work at a large broker-dealer.

Worked in the field, read the book. Strongly agree that the book is essentially useless. Hull is a good starting point.

Not strictly ML, but "Optimization Methods in Finance" by Cornuejols et al [1] a great reference. It introduces optimization algorithms and modeling technique as well as finance applications in alternating chapters.

[1] https://doi.org/10.1017/9781107297340

Read "Beat the market: A scientific stock market system" by Edward Thorp

What are the great successes of machine learning in finance?

Great question!

Does Renaissance Technologies count as ML? I honestly don't know where the line between quantitative/algorithmic and ML is - I suspect it's blurry.

They gather data and use it to make data driven statistical predictions -- of course they are ML!

Is there any proof of Renaissance not being just insider trading or a Ponzi scheme?

Does "Ponzi scheme" just mean "bad finance thing" now? It is impossible by definition for a fund that hasn't had inflows for 30 years to be a Ponzi scheme.

They could, for example, be buying early stage startups cheap, sell it to a friend for 3x, who sells to a VC fund for 3x, which then sells it to the public by showing huge growth.

No inflows, old investors get paid with money from new retail investors and retail investors are left with a hot potato. A Ponzi scheme.

It cannot potentially be Ponzi Scheme as their investors didn’t end up losing money.(The 0-sum was in the market participation)

The kind of returns RenTech took, it would have to be a highly elaborate insider trading and why would a bunch of scientists want to waste their time on that ?

The only thing “wonky” about RenTech was Bob Mercer’s personal political views but despite that I would still consider Bob Mercer to be a very intelligent/scientific person.

Why limit the question to finance?

Maybe because the article is titled "Introduction to Machine Learning for Finance" ?

so far no ML system created a ponzi scheme

Reply on news.ycombinator.com