Machine Learning and Data Science Blueprints for Finance
Vložit
- čas přidán 19. 03. 2021
- I review the book, "Machine Learning & Data Science Blueprints for Finance" by Tatsat, Puri, and Lookabaugh. As machine learning and data science have become the hot topic in finance these days it is becoming more important to really understand the basics. Most of the basics come in the form of traditional statistics and the scientific method. Besides fitting lines to data, a full range of tests need to be conducted to really understand your data, the model structure, the output, and its usage.
Machine Learning and Data Science Blueprints for Finance (my affiliate link)
amzn.to/3tFe3gz
Rating: 2/5 STARS
Quant t-shirts, mugs, and hoodies:
teespring.com/stores/fancy-quant
Connect with me:
/ dimitri-bianco
/ dimitribianco
☕ Show Your Support and Buy Me a Coffee ☕
ko-fi.com/fancyquant
On the bright side, I learned a ton about the right way to perform rigorous data science modeling from this video
Thank you Dimitri, we are learning a lot through you of what is happening in the industry
Thank you very much for sharing your insights and giving us more than just a book review. Subscribed!
Thank you for the video and review Dimitri!
Which books would you recommend for the finance industry (quantitative) used in investment firms, hedge funds? Many thanks!
Expected a book review but got more than that. I like, subsribed.
Thanks!
the point you made about professionalism made me curious, might not be a bad idea to make a video about the process you mentioned with a real example. I think it would be educational and definitely worth it.
He made some from time to time.
Really glad I found this channel. I couldn't agree more with your perspective. If you can't holistically explain the why behind your model selection based on a hypothesis, then you're misrepresenting the data... I am more impressed by the specification of the model, not a simple performance metric... Thanks.
Sir love your passion for your work. I admire that you have a vision for how the new technology should shape to integrate with finance.
Thanks!
I wasn’t interested in the book, but I really loved your comments here! You still need to use your brain while doing ML. Thank you!
Reading these kinds of books make me really appreciate AFML ;)
Thanks for the review.
Dimitri, your thoughts and comments are very precise. It would be awesome if you write a book! Give it a chance, success for sure💪🏼😆
Hello Dimitri,
Sorry if I missed it. I recently bought "Artificial intelligence in finance" and "Python for algorithmic trading" by Yves Hilpisch. I am going through them, but I am not enjoying them much, they seem to have many of the problems of the book in the video. What do you think of those, maybe compared to the one of the video?
And using these books together with a book that explains the data science process from model choice to statistical testing in practice? What would you suggest for such a book?
Hey, it's been two years since you made this comment so I guess you ended up finding some good resources. If so, would you be willing to give some suggestions?
@jimjohnson357 such a long time! I read several books in the meantime, but the best resource was going on Arxiv, printing 40+ papers on ML/AI and reading them
Thanks for the review Dimitri, and
please do review on analysis of financial time series by Ruey Tsay
I love this stationarity pet peeve.
You should review more 2-star books like this. Really entertaining, but more importantly, very insightful.
I try to avoid low ranking books. I make money by recommending books and providing my Amazon affiliate link where I make a small amount.
I do see the value in pointing out issues in books though. I was surprised so many people liked the video and mentioned they learned more about ML models.
@@DimitriBianco That's a fair point.
Speaking just for myself (and hoping this reply is congruent with the video since it's been several days since I watched it): since this is not my main field of study, and since I have a strong interest in learning about quant finance, all of these ML and statistical topics are part of kind of a nebulous cloud in my head. I think I have a basic understanding of many of them on their own, but not necessarily the proper knowledge & experience to tie them all together.
If I'd have read this book on my own, it's possible I might have seen the issues you pointed out, but the odds are I wouldn't have. You tied everything together nicely. By emphasizing the point that ML and stats are not necessarily separate things (ML is built on top of stats), and the importance of understanding the structure of the data before anything else, I think you bring clarity to a poorly understood topic.
It's also entertaining to see someone give their unfiltered opinion on a topic they are passionate about...especially when it has to do with a product that was sent by the creator of that product specifically for a review.
@@DimitriBianco Old comment but shout out for the transparency. It's obvious affiliate links make you money but nowadays not many people would be that transparent.
Can you recommend some books the parts that this books misses ( model validation & statistical analysis of the data) for Machine Learning in general?
There really aren't very many book on ML and finance. The best one I have seen is "Advances in Financial Machine Learning." The author has similar complaints in the book as the ones I mentioned in this book review. I'll link my affiliate link to book below and a book review I did on that book.
The book on Amazon:
amzn.to/3f8KSPe
My review:
czcams.com/video/gqE8KxA6DM8/video.html
Love this video! Pls keep making more
Extremely valuable and insightful
Which book would you suggest for time series modelling/analysis ?
I would like to know too
"Time Series Analysis and its Applications" by Shumway and Stoffer is a nice hands on book to start, "Statistics and Data Analysis for Financial Engineering" by Ruppert is another cool book to start (it also contains other important topics) and i personally used it. If you need some in-depth book i really suggest the bible which is Time Series Analysis by Hamilton
@@andresrossi9 Thank you !! 😁
@@mohitchandnani2559 you're welcome!
Does time series models really work?
Hello Sir, glad you reviewed this book :)
Thanks
I saw too many so called data science gurus that have soooo much limited knowledge about math and stats.
Would this be a book for a person who wants to familiarize themselves with this area of finance?
That's what I think.
They can't put every aspect of testing in the book, otherwise they would end up with a 1200 page book.
Who's going to buy that?
No. Starts from stats and math books first.
Brother please write a book series.
is there a complimentary recourse that can help with the issues you raised? like some book or course
do you know any good books for time series with python?
Check it out: Advances in Financial Machine Learning. Dimitri also had a book review video on that.
This is the best book I have seen for finance and ML.
For time series with Python....I don't know of any. The main packages for time series in Python are lacking by a long shot. This is why firms either design their own proprietary packages or use SAS.
I guess this book is a good cookbook for Kaggle competitions that use financial datasets.
This book is for kindergarten players.
Hi Dimitri, I agree that testing the robustness of the model is essential for an effective deployment. However, when optimizing a model, I usually use a suitable metric (RMSE, ROC, …, F1) in order to tune the hyperparameters, and then test the various things you have brought up during the video for the generated model. Is there something wrong with this approach or you are just criticizing those who only tune the hyperparameters to maximize a single metric without performing data analysis/feature engineering?
I'm criticizing those who seek to optimize based on a single metric. The real world is far too complex to be optimized with a single metric. There is always a trade off and understanding the trade offs and model risks are important especially in finance. The 2008 crisis is a good example of this story of failure. We'll see another sort of model failure soon in finance.
@@DimitriBianco Understood, thank you
@@DimitriBianco it won’t if the wall streets is run by mathematicians.
@@DimitriBianco Hey Dimitri, what do you mean by another model failure in finance (if you can talk about that! ?
Great video as usual. Its been a time since you've done a books recommendation video. Can you please do one on recommending resources about modeling ?(specially things you mentioned here)
A read list also would be sufficient more than enough :D
hi can you make another quant reading list for 2021? Thanks!
I'll consider it but not a lot would change.
hey cousin, do you have a good machine learning for finance book to recommend?
There really aren't very many book on ML and finance. The best one I have seen is "Advances in Financial Machine Learning." The author has similar complaints in the book as the ones I mentioned in this book review. I'll link my affiliate link to book below and a book review I did on that book.
The book on Amazon:
amzn.to/3f8KSPe
My review:
czcams.com/video/gqE8KxA6DM8/video.html
Would this book be good for learning the basic principles?
Not at all.
I wouldn't recommend it. You are better off finding textbooks on specific topics like statistics and ML and then just applying it to finance. I'm not sure why so many people think applying it to finance is drastically different than other fields.
Will reading this book guide me to starting the next medallion fund?
Nope
@@DimitriBianco very sad
All time-series modelling requires stationary, not all models. I don’t even understand why I would need stationarity in cross sectional data.
Stationarity is the same concept as central limit theorem which is required in all forms of statistics and probability. It is just a more complex version due to the ordering of time-series.
@@DimitriBianco Yes, stationary is indeed the CLT. Not all models require the CLT, even things as fundamental as your poison distribution for small values of the mean (lambda), CLT is not assumed. I’ll go a step further and add the negative binomial distribution in there as well. My point is that saying all models require “Stationarity/CLT” is very crude.