The 7 Reasons Most Machine Learning Funds Fail Marcos Lopez de Prado from QuantCon 2018
Vložit
- čas přidán 30. 06. 2024
- This talk, titled The 7 Reasons Most Machine Learning Funds Fail, looks at the particularly high rate of failure in financial machine learning. The few managers who succeed amass a large number of assets, deliver consistently exceptional performance to their investors. However, that is a rare outcome. This presentation will go over the 7 critical mistakes underlying most financial machine learning failures based off of Marcos López de Prado’s experiences and observations.
To learn more about Quantopian, visit bit.ly/mlqc2018.
The slides for this presentation can be found at bit.ly/2DyUNdc.
Bio of the Speaker:
Dr. Marcos López de Prado is the chief executive officer at True Positive Technologies LP. He founded Guggenheim Partners’ Quantitative Investment Strategies (QIS) business, where he applied cutting-edge machine learning to the development of high-capacity strategies that delivered superior risk-adjusted returns. After managing up to $13 billion in assets, López de Prado acquired QIS and successfully spun out that business in 2018.
López de Prado is a research fellow at Lawrence Berkeley National Laboratory (U.S. Department of Energy, Office of Science). A top 10-most-read author in finance based on SSRN's rankings, he has published dozens of scientific articles on machine learning and supercomputing and holds multiple international patent applications on algorithmic trading.
Marcos earned a Ph.D. in Financial Economics (2003), a Ph.D. in Mathematical Finance (2011) from Universidad Complutense de Madrid, and is a recipient of Spain's National Award for Academic Excellence (1999). He completed his post-doctoral research at Harvard University and Cornell University.
Disclaimer
Quantopian provides this presentation to help people write trading algorithms - it is not intended to provide investment advice.
More specifically, the material is provided for informational purposes only and does not constitute an offer to sell, a solicitation to buy, or a recommendation or endorsement for any security or strategy, nor does it constitute an offer to provide investment advisory or other services by Quantopian.
In addition, the content neither constitutes investment advice nor offers any opinion with respect to the suitability of any security or any specific investment. Quantopian makes no guarantees as to accuracy or completeness of the views expressed in the website. The views are subject to change, and may have become unreliable for various reasons, including changes in market conditions or economic circumstances. - Věda a technologie
1. Sisyphean Quants 6:53
2. Integer Differentiation 15:26
3. Inefficient Sampling 24:32
4. Wrong Labeling 30:32
5. Weighting of non-IID samples 38:29
6. Cross-Validation leakage 45:33
7. Backtest Overfitting 48:31
8. QnA 1:00:30
Great lecture. Very Insightful. Thank you for sharing :)
I'm looking forward to reading the book.
Outstanding presentation, the best I've seen in quant finance.
Metalabeling can also be used to differentiate between exit strategies and entry strategies - the first one has to be greedy and have an opinion for any given moment when you are in the market, whereas the second one needs to balance the degree of certainty with the expected outcome (given the exit strategy) minus commissions / slippage.
Someone is thinking, finally.
There's a nugget or two in this talk. Thanks for sharing.
This talk is gold!
Amazing!it's just surprised to me,every time i only learn how to use a ML model on some datasets, however overlooking that which conditions we use it in a wrong way
i wished i read his books or papers two years ago...i read lots of financial research papers and replicated them. But not a single paper works in real life. but I now know the exact reason.
You took the words right out of my mouth! :). I've been replicating some research papers recently and its really amazing that none of them actually work :) . It seems people are just interested in publishing rather than producing knowledge. But this talk was really helpful in giving me a direction to move forward. cheers and good luck on your research
can relate ! this lecture should be the first thing you watch before you read any other researches
@@ahsabour488 so true. academia is plagued with impracticality for the sake to appear fancy
How do you reproduce something that doesn't work? Technically then it has failed replication.
Why does no one in the hedge fund space seem to like this guy? I think he's got a lot of great ideas that make sense from a theoretical point of view, especially when you consider the low signal to noise ratio of the statistical modeling problem in finance. Everyone seems to obsess over details but never think about the big picture like MLDP does.
Who exactly doesn't like this guy?
Is not that they do not like him. He is one of the smartest guys on the street and usually likes to patent or safe credit for his ideas. Hedge Funds do not like to patent or publish anything because the minute they published the stretegy does not work anymore. So there is a conflict of interest between MLP and any hedge fund. If it were not the case he will be an unknown HF billionaire. Now he is not a HF billionaire but he is known worldwide
Hes got some good ideas, but most of his ideas don't really do much in practice. Like fractional diffwrencing is pretty much useless - regular differencing works fine, and if it doesn't, fractional differencing is usually not going to help. He also has a huge ego - it's really exhausting reading his books because he comes across as so dogmatic and egotistical. And ultimately he states a bunch of obvious truths like they're new ideas giants haven't been aware of, and he doesn't have a track record to show for it. Just a bunch of degrees.
@@alejandrorodriguezdomingue174 I didn’t know much about his work but I liked his talk, even though I think most problem he presents here are quite basic to the research community. But the single fact that he patents his idea is against my values n turns me off.
@@dl662 why?
Marcos Lopez de Prado, you should get a Nobel Prize for your contributions to Finance!
Not meant to be disrespectful to his work, but really most he’s talked about is ml101, the kind of problems ppl in control n medical research comm have been dealing with on a daily basis. A Nobel prize feels a bit humorous to me.
@@dl662, do you understand what he is saying? Are you experienced with Finance ML?
@@dl662 You are right. But sometimes Nobel prizes have been really humorous too 🥰
Thanks for sharing, excellent!
Thank you Marcos!
24:33 Pitfall #3: Inefficient Sampling
非常に素晴らしい。ロペスさんのもとで働きたいわ。。
なんちゃってデータサイエンティストが学べることがとても詰まっている。
The link for the presentation is not valid anymore:( please helpppppp
Could someone help me out with understanding the Integer Differentiation section? If I'm understanding correctly, when d=0, the time series is simply the original prices themselves and when d=1, the series is the returns from one price to the next.
What would happen to the series if you had d=.5 for example? Let's say that the original price series was [32.45, 31.25, 33.61, 33.15, 33.40, 32.97].
Thanks in advance!
Check the python libraries for fractional differencing bruv
liked the three barrier method!!
Is Numeraire the final solution here?
Lol I love the jab at economists being an example of a “high recall algorithm”
@@anshupandey6260 I think it is high R2
@@anshupandey6260 some measure of true positives, false negatives and so on
higher recall = higher true positives (0.5-1.0)
buenas tardes sr marcos mentor financiero una pregunta el machine learning es usado también en las acciones meme como game stop y amc? gracias por la respuesta
I had difficulty understanding:
- what's the thing about memory .. even technical indicators are looking at history prices at n periods. Neural networks are using time series to not only look at historical prices but also the sequence.
- small groups are also ideal in developing trading algorithms. Max is 3 per group: 1 with the trading/model idea, 1 programmer, 1 tester. In some cases, 1 person can do all.
- all available data should be given to all groups and its up to the groups to select what to use.
iI am not 100% sure if my idea is right but I think with memory the problem is related to the differentiation.
The time series data of stocks resembles "non-stationary-data", meaning its mean, variance and autocorrelation changes over time.
To make the data more digestible for the predictive models this data is transformed into stationary data by differentiating the whole data, effectively stabilizing the mean. This differentiation leads to a loss in memory.
So even if you feed the model with data of the last 100 days there is NO (or low) memory within this data due to the differentiation and therefore the model can not leverage its memory supportive architecture
so frustrating when no one listening when I say how important data cleaning is
i'm wading through the first couple chapters of his book and skimming/previewing latter chapters. Feels like the book is telling us to machine learn on autocorrelated features. I don't think this actually works in practice and the book focuses a lot on "management" (building a pipeline of "discovery") rather than the actual hard results that he says matters more than back testing or whatever else he says in his book. He even mentions there are no concrete strategies revealed. He seems more like an academic who doesn't actually trade anything, but writes books. The ultimate ultimate of a financial guru fraud with fancy degrees in the end.
A lot of things are made up out of thin air. It's just like when people make up patterns and have funny names for them ("double monkey ass bottom top flag"). He does this in his book a lot; "triple-barrier-method" for example. Does he present any sort of study that shows this method is effective for entering or closing trades? no. It's just thrown out there like he pulled something out of his a**.
yeah the ideas are abundant but i didnt see any live or forward test or proof. The way he discredits efficient market theory and flat out saying markets are predictable is insane.
38:00 false breakout vol filtering
Massive respect for Marcos
I have had no luck with fractional differentiation, that is, I haven't got any useful results from this.
Anyone got any different results?
Most of that is waste of time
This dude is good in finance and good at yapping people too.
Good presentation although it's basically just a synopsis of his book.
Whats the name of his book?
@@divyeshpoddar2972 "Never gonna give you up"
@@willrocksBR oooooohhh that's dirty
What does it mean fractionally differentiated?
Taking integer number of differences to stationarize a time series
Edward Norton is very smart!
this man is my neighbor
48:37
so basically when the price is moving up - I press the green button?
Is green good?
2023, ChatGTP enters the ..... Chat.
Exactly! Who would have though so?
59:00 so he run many trials and got good results? Maybe he overfitted the results!
존나어렵네
4 years later MLLF or MILF as i call it, are doing fine.
Whole presentation is a load of rubbish. Each of his points can be countered with an opposite example. I doubt this man has made even one profitable trade in his entire life.
Could you elaborate?
@@OskarBienkoI doubt that he is capable of elaborating his assertions.
i love how you made such a loaded comment without elaborating especially when the topic is complex. Brilliant. You added value to the discussion