Advances in Financial Machine Learning (book review)

[@weekendresearcher]

Please also do a review on his next book, machine learning for asset managers, cambridge press

[@yilunlu4270]

Hello Dimitri. As a time series expert, do you agree with De Prado that applying integer differencing to make time series stationary would wipe out memory in data and hence compromise predictivity?

[@DimitriBianco]

It depends on your you build the model but my short answer is no. As an example, say you build an ARIMA model and difference the dependent variable. This model should never have an intercept. The difference dependent variable Y(t) can be written as Y(t) - Y(t-1). If you re-write you equation, you can move the -Y(t-1) to the other side which acts as a dynamic intercept.

[@yilunlu4270]

@@DimitriBianco I can see your point. Thank you :)

[@adityamukherjee1679]

Alright, I'm sold. This will be the next book I read.

[@32isaias]

I don't think you can say he doesn't understand economics or statistics. Look at the mans credentials, my gut tells me he's right and you do not know what you are talking about. Firstly this isn't a book you can just read. You have to implement and dive deep.

[@brendanlydon5272]

Hey Dimitri, not sure if you have read this textbook, but could you do a review of, “Deep Learning by Ian Goodfellow, Yoshua Bengio, and Aaron Courville”? I read Prado’s book first, but I wish I read the book I just mentioned.

[@randb9378]

Hello Dimitri, thanks for sharing your thoughts. Could you give any recommendations on where and how to start with algorithmic trading using ml?

[@DimitriBianco]

As general advice, you need a team to trade successfully. I don't work in the algo side so I can't provide much advice on it.

[@islandparadise]

Mine just arrived. Thanks Dimitri!

[@DimitriBianco]

You'll have to let me know what you think once you have read it.

[@islandparadise]

@@DimitriBianco for sure! For a 1st browse though, crap that's a lot of math for a junior developer with little math skills. But at least it's limited to equations and has some explanations. The rest should be Google-able! :)

[@sentralorigin]

would you be able to compile a list of all the texts you've reviewed alongside with your rating of them?

[@keshavkasat9465]

He already has a 30 min long video about it I think

[@sentralorigin]

@@keshavkasat9465 yea but i don't want to scroll through a 30 minute video to sort the books by rating

[@DimitriBianco]

Just let me know the topic you are interested in and I can recommend my favorite book or two. This channel will also be doing more data science and machine learning in the future.

[@RedShipsofSpainAgain]

@@DimitriBianco Thanks Dimitri. How about an introductory book on financial machine learning targeted to people who already know basic/intermediary ML but who lack experience specifically in the financial sector but would like to apply ML to the financial industry.

[@sentralorigin]

@@DimitriBianco books you've rated 5/5 for "introductory" machine learning (e.g. for a physics PhD hired as a machine learning researcher position at a hedge fund with no background in finance nor machine learning both)

[@FiorianCanuck]

Thanks Dimitri.

[@OK-dy8tr]

Thanks for the insights Dimitri !!

[@christiansong227]

Dimitri, thanks for your review on MLdP's book. I would give it 3/5. I was also disturbed by his disdain for statistics and econometric, when machine learning itself is a subset of statistics. I don't know where he is getting this idea that machine learning is a subject on its own, but if you talk to pioneers in machine learning, people like Michael Jordan at Berkeley, David Donoho (a former chief scientist at Renaissance Tech and a professor in statistics at Stanford who literally is a contemporary giant in the field of statistics), or even people like James Simon (the founder of Ren Tech) said in video that "machine learning is a statistics." So I'm also disturbed by his lack of understanding in the history of machine learning. Statisticians have been working on statistical learning theory (aka machine learning, aka pattern matching) for several centuries. The computer science department rebranded statistical learning theory as machine learning. Computer scientists weren't treated seriously by statisticians until they came up with Adaboost, but that has been the major contribution from CS people. I'll give credit to Lightboost and XGboost, but these are not new theorems... The vast majority of theorems in "machine learning" came from statistics, mathematics, and control theory. I do think economic has problems and economists should stop envying physics, and admit that they can't separate economy from politics ( it has been called for a reason "political economy") to deceive people that economy is a scientific field(no, it's not). But discarding entire field as a useless is, in my view, very wrong. Also, the issues with backtesting are not something new and I don't quite understand why he's blowing this out of proportion, when industry people are already aware of this issue...If he thinks he knows how to apply "machine learning", why did funds like AQR perform so poorly past many years? (AQR is an asset gatherer) Also, didn't people at Ren Tech tell the media outlet that they use OLS too? I don't know if you pay attention to MLdP's twitter, but he has been called out subtly by a former trader at Jane Street that MLdP seems not know what he's talking about. I just think he is like other people who want to ride machine learning hype and this kind of behavior is not something new. Have you seen Gary Marcus's crusade against deep learning? I do think deep learning is just full of shit and there has not been a single real contribution outside of recognizing static images ( of course, if you change a single pixel, then deep learning has not contributed anything to the real-world application). There have been papers calling out flaws in his papers too. You can find them online. Take care, Dimitri, and I loved your podcast !

[@DimitriBianco]

I like the insight. I actually have a video coming out soon on the field of economics which you might like. It will be titled, "The Crazy Mess of the Economic Community." I will eventually be doing more videos on machine learning concepts however they will be from a stats/math perspective. I hope to make them easier to understand than a lot of the hyped up hand waving I see a lot of with false claims as you have pointed out.

[@christiansong227]

@@DimitriBianco Hi Dimitri. That video on economy would be great ! Also thank you for your kind words. I think "If you don't learn history, you're condemned to repeat the same mistakes" is so important for knowing the history of any subject. Also, yep. I think two things that people should know are accounting and bullshit detector. I have seen people applying NLP, Bayesian methods, etc etc on predicting the movement of stock price for Tesla... and if you talk to these quants, they have zero understanding how accounting works. If you know accounting, then the equity value of Tesla is $0. It's scary that people who are managing money don't know how to read financial statement and balance sheet of a company. (I like Tesla as a product, but I'm against all the government subsidy and quite literally insane behavior of Elon Musk - that he's somehow above the security law). Bullshit detector is so valuable in real-life, not just limited to figuring out whether machine learning is really "new electricity" (to borrow words from Andrew Ng at Stanford). It's not. People with technical degrees tend to dismiss C-suite people as bunch of idiots, but these C-suite people actually can tell whether this is a bullshit or not. If you pay attention to how media portrays machine learning over the past 3 years, it went from "this will literally change the way we live" to "well, machine learning (in particular deep learning) don't work.." I would say give it another a year or two, then I expect to see many firms shutting AI/ML/DL departments. My concern is current students who don't know too much about machine learning and yet they are bought into the hype, generated by people like Yann LeCun, Andrew Ng, Yoshua Bengio, Jeff Hinton, people at OpenAI and DeepMind, etc. I hope we don't reach the point where we will have unemployed people with degree in ML, but if I do pattern matching and compare this level of ML hype to hype I saw in nanotechnology, then we will see 4th AI winter. This time though, I think this bubble in ML will explode like it did to quant / finance industry in 2007-2008 meltdown, and OpenAI will be the LTCM of IT industry.

[@janithforex4075]

@@christiansong227 Hi really appreciate your way of thinking about AI. It has been two year since you commented this. Have any of the new finding changed how you think?

[@christiansong227]

@@janithforex4075 Hi. Sorry for the late response as I don't check this account often. I think AI bubble will pop soon. People like Hinton argued that radiologists will lose their jobs (you can find this clip on YoTube.. it has been many years and radiologists losing their jobs didn't happen). I would stay away from the hype surrounding OpenAi as that organization is known for creating hype for their "research". It was back in 2018 - 2019, they were hyping solving Rubik's cube from their robotic division (now defunct lol). I think the biggest blowup of AI will be self-driving car. So many companies literally set their money on self-driving cars and I think this will be the final catalyst for popping AI bubble. I truly feel bad for the current PhD students in machine learning world.

[@OskarBienko]

Could you tell me which papers are calling out flaws in his papers? Thank you in advance!

[@nikoshazaridis2766]

Hi Dimitri, any thoughts after one year with the book? Have you used anything in your day-to-day? Any new comments?

[@DimitriBianco]

My view has stayed about the same. As more firms use data science many of the issues or common mistakes he mentions in the book I do see as common.

[@nikoshazaridis2766]

@@DimitriBianco Thank you for your reply.

[@mvgjorge]

Thanks Dimitri !

[@RoboticusMusic]

What parts are outdated for tf2?

[@DimitriBianco]

The book doesn't use deep learning, so there is no usage of tf. I will be making a video in regards to implementing this book though.

[@RoboticusMusic]

@@DimitriBianco I have the pdf but I'm no good at math. It would also be interesting to see a video explaining unresolved bugs with open source ML related libraries and fixes. I've seen people talk about unresolved memory leaks for example.

[@darkdevil905]

So how do you apply it correctly?

[@DimitriBianco]

What part of the video are you talking about?

[@Oochini]

Ever consider getting a podcast going? Even just taking the audio from this video and throwing it on stitcher or something, would be convenient for listening to reviews/advice while on the go rather than convincing myself to watch on youtube when my brain just wants to shut down and watch stupid vids haha

[@Oochini]

Thanks for the review though, it's defo one on the list to read, working through Hands-On Machine Learning by Aurelien Sol first to try get a high level overview of ML and neural nets first

[@DimitriBianco]

I just launched a podcast called, "Talking Tuesdays with Fancy Quant." It's on Stitcher, Spotify, and Apple Podcasts right now. you can also find it at the link below. What do you think of the Hands-On ML book? I bought it but haven't had time to read it yet.

[@Oochini]

I'm finding it very useful as it's quite concise and provides a good intro to the various ML models. About to move on to the second section on neural nets/deep learning. As a student trying to get into ML I'm enjoying it, not sure how a professional with real ML experience will find it

[@alex_8704]

👍👍

[@youngarmani7155]

I did not like this book.