Це відео не доступне.
Перепрошуємо.
The Issue with Machine Learning in Finance
Вставка
- Опубліковано 6 сер 2024
- Students are eager to jump into the finance industry thinking they can apply the machine learning and artificial intelligence methods they have learned in school or the tech industry. The truth is, the finance industry has been making slow progress however the high analytical standard and regulations of finance make this progress much slower than the tech industry.
In this video I will cover a variety of topics including CCAR, Equal Credit Opportunity Act (ECOA), and the Consumer Finance Protection Bureau (CFPB).
SUPPORT THE CHANNEL
Quant t-shirts, mugs, and hoodies:
teespring.com/stores/fancy-quant
Connect with me:
/ dimitri-b. .
/ dimitribianco
☕ Show Your Support and Buy Me a Coffee ☕
ko-fi.com/fancyquant
Man you have described perfectly my day to day job. You literally have demystified all of the fuzz around the Machine Learning hype in banking.
The Issue with Machine Learning in Banking*
Well covered and pretty comprehensive take on the topic!
Thanks!
Such a useful video, thanks mate
Love what you do man keep it up I love the stuff you put out GO COUGS !
Thanks for the feedback. GO COUGS!!!
Dimitri, in the video you talk a lot about regulations as a barrier for ML models to be used, do you thing the current regulations that don't allow for these models to be applied are justified in the sense that the amount of money that is spent on them have a very good reason that justifies them or is it more a resistance to change (here it would make sense as you said in the video as a competitive advantage from established banks over fintechs and new market players). If there are good reasons could you give an example? Also, very nice video, keep it up.
great video man!
Thanks!
SAS License costs close to $ 100,000. But relevant to Banking as it assures your data confidentiality unlike the open sourced one.
Great video from someone who just started practicing machine learning, thanks.
had the same issue at the bank i work at, impossible to get IT security to allow installing packages in R - had to purchase the premium version.
Hey Dimitri, can you do a video on P quants vs. Q quants?
Yeah, I'll add it to the list.
Hey Dimitri, sorry if this is a bit off topic for this video, I was wondering if you would know any good learning materials/books etc. for basic quantitative portfolio management? I'm starting up a team in my university!
If you are looking for a machine learning book specifically on finance I would take a look at "Advances in Machine Learning" by Prado. The book is less than $35 on Amazon (linked below). This is on my reading list and is highly recommended by others however I have not had time to read it yet.
amzn.to/2ydx6o7
The universal logic applies here. START WITH SOLVING THE BUSINESS PROBLEM, NOT WITH THE TECHNOLOGY TO SOLVE THE PROBLEM.
This is an interesting atmosphere. You mentioned hedge funds at the end. Does less regulation mean more machine learning?
Hedge funds and investing are a whole other realm. Banks (which I talk about here) make money from selling products such as mortgage and auto loans. The data banks have on these customers is very detailed and data sets can be 200mm+ records and thousands of variables (100gb+ in size). Hedge funds will use market data and scrape the web for other non traditional data sources (2 Sigmas and Citadel do this). There isn't regulation on a lot of these data sources however the issue of ethics comes into play. For example, if I can scrape data from Facebook and it has personal information available what should we use and what shouldn't we use. Facebook recently got hammered for their data issues and Google just announced last week that they hid a data breach. I think the tech industry will have to catch up to banking in standards but for now hedge fund data is wide open.
some of your viewers are market guys, I thought IBs have internal prop trading desks that would have their own quants using ML. Not even necessarily for evaluation but as a market research and sentiment analysis tool. Id say that is much more accessible to apply without proprietary methods.
Prop trading no longer exists at IBs due to regulation. If you are interested in trading you should look at my interview with a buddy of mine who was an FX derivatives trader at UBS.
ua-cam.com/video/z0E2ug1rfgg/v-deo.html
Hello Dimitri, with regards to what James Simons does at Renaissance Funds: What is it that they do?
Renaissance uses statistical arbitrage which is common in the industry. This is different than machine learning but linked below is a nice write up on how Renaissance makes money.
www.quora.com/What-are-the-investment-strategies-of-James-Simons-Renaissance-Technologies-I-understand-he-employs-complex-mathematical-models-along-with-statistical-analyses-to-predict-non-equilibrium-changes
He uses a lot of signal processing and machine learning
Is learning econometrics still useful? Where wud it help ? What is it that can still be done very well by econometrics and cannot be done by big data models ?
It is still useful and I wish more people took it. Most financial models are done using standard statistics (econometrics). Credit risk, market risk, operational risk, and ppnr risk. ML is mainly used for fraud models however more banks are using it for marketing credit models.
100% correct. P.S I work in Machine Learning in Finance and come from the tech sector + academia. Completely spot on here with your points. Key point: You really have to work for an innovative firm. I deploy models on a daily basis. Impact of automation is serious (6-7 figures) but if you do it wrong you will lose your firm more money. In finance you can only really work on a small subset of ML/AI use cases and with limited datasets but you can make a huge impact.
Yes, the details make a big difference. Thanks for your feedback and insight.
It's the same in other fields too
Dimitri .. I'm taking a class an online FInancial Engineering class and was wondering if you could help me. Perhaps a private lesson or two. Please let me know. Thanks
In general, no I don't provide one on one tutoring due to time constraints. If you have specific topics however I might be able to make videos about them.
Can you make a video using stochastic calculus and how you use it in financial models?
Baking?
hmmm I don't really get what pricing errors you can make. For loans you'll know that there is something wrong if your rate is out of market. Some loans are even repriceable so you can increase rates as needed.
For other instruments, its based on market price and you just earn on spread or fee.
For OTC derivatives, it maybe possible.
Loans don't have a market price. Every firm will price them differently. Try buying a car and see what rates everyone gives you.
@@DimitriBianco If survey is done on various banks offering, it should not be too far from each other. I just used the term "out of market" if the rate is too far.
I don't see a real distinction between statistical learning and machine learning. Machine learning is an extension to classic statistical modeling, perhaps with more focus on approximate numerical methods, performance, and computational efficiency+scale. Of course ML has kind of become synonymous with lack of interpretability and playing fast-and-loose with the assumptions (based on business use-cases that may only care about performance), but this doesn't define what ML is. Interpretable ML is a big research area (side note: ML is a lot bigger than neural networks). There are also attempts to reintroduce or emphasize formal guarantees into ML-like methods (e.g. probabalistic circuits). I pretty much agree on your forecasts. Banking changes slowly and I don't think the way ML is often done outside of finance can just be moved to finance without changing things a lot. I think "ML in finance" is continuing to become its own domain, particularly in banking, for the reasons you gave. Like "ML in medicine", it will not be something that happens overnight.
Strong regulations are a good thing, but I also think a lot of the way banking functions is suboptimal. I am speaking more about their attitude to infrastructure technology, not so much ML. I think seeing open-source as a security risk (btw open-source and the cloud are different things) is a bit of a joke. Many of the industry standard security tools are open-source. I don't see how proprietary software is more secure when you don't actually know the code being run or can be certain that vulnerabilities are being fixed. Aside from that, if you really care about security and correctness, you should be investing into formal methods and other correctness proving (like the aerospace industry) - not merely assuming that paying a company for proprietary software (so that some party can be held liable) or doing what other banks do will save you.
The big secret is machine learning is just a sub topic inside of statistics.
Does the rise of data science and machine learning makes learning econometrics redundant/ useless ?
Nope. Data science is really just an extension of statistics. It has areas where it is better and areas where it is worse.
hi dimitri, i want to know, how is the CFI certification on quants?
is it a good one to try?
I haven't seen anyone in the industry with it, so I can't comment on it.
@@DimitriBianco actually it was a typing error, I wanted to ask, CQF certification, certification for quantitative finance
I don't think you understand neural networks.
What doesn't he get, exactly?
I've talked to others in the industry, and they more or less agree with his sentiment.