Common Mistakes in Stock Price Prediction: Prices As Targets (Episode 21)
Vložit
- čas přidán 16. 07. 2024
- VIP COUPONS 75% OFF BELOW (check lazyprogrammer.me for the latest coupons if after Nov 17, 2021)
Time Series Analysis: deeplearningcourses.com/c/tim...
Financial Engineering: deeplearningcourses.com/c/ai-...
PyTorch: Deep Learning and Artificial Intelligence: deeplearningcourses.com/c/pyt...
Reinforcement Learning (with BONUS extra content): deeplearningcourses.com/c/art...
Transformers + LLMs: deeplearningcourses.com/c/dat... - Věda a technologie
For the astute viewer:
I forgot to mention one thing. In the intro to this video, I mention "appeal to authority", and you might immediately think that my original statement about only bloggers and marketers covering this topic is also a kind of "appeal to authority". That's true!
At least, it would be true if that were my main argument. But if you watch the last video closely, you'll recognize that this is just an offhand remark. It's merely presented as an interesting observation, not as the focal point of the discussion.
Of course, I am not saying this whole niche is a waste of time (otherwise, I wouldn't be here talking about it or making courses on it, would I?). I mention in this lecture that it simply requires "extra caution and vigilance", even in my own courses.
In my courses and videos, you can indepdently verify any claim I've made. I follow my own rules. All claims are demonstrated via code so as to be indisputable. In fact, I very much enjoy presenting this code as it is usually to debunk what the bloggers / marketers have done.
If I had a model with very high predictive power that always lost money, I’d just do the opposite and retire someplace nice. Great video. I just found your channel and really appreciate the time/effort you’ve invested in making such clear and articulate explanations of fairly complex concepts. Do any of these models ever use fundamental analysis techniques, such as looking at the quarterly income and balance statements? Metrics like percent change in revenue, cash on hand, debt, operating income/loss, etc.? I mostly see people talking about day trading bots.
stumbled upon you while making all of the above mistakes using chatgtp and decided to learn abit to see how i can fix some of the problems occurring. Then behold the light behind the tunnel shown on me and now i got alot of courses to complete and books to buy and digest the information.
I’ve subbed to your channel using my personal account.
But I did want to say THANK YOU!
I wish I’d discovered your channel in 2020, when I started my project.
Thank you for being the only CZcams channel that realises you can’t use price and should use return.
And your explanation of statistical techniques is very accessible, especially for someone of my limited intellect. I spent months watching MIT’s CZcams videos on finance. I could have saved hours watching this.
I’d like to say I hope you gain many subscribers. But I’d rather people continue to make mistakes for purely selfish reasons. Sorry, not sorry.
thanks this new view to predictions understanding , you are the best bro
Hey I wanted to ask if there will be any upcoming coupons for the financial engineering course. I also was wondering what's about the VIP content? will it still be available again or when is it available? I don't get the description on this part since the course is already release for a while now. Thanks!
Thanks for your inquiry. Actually at the time you posted your question there was already a coupon on my website. It expired a few hours ago but I just created new ones, again on my website. Go to lazyprogrammer.me.
Read the posts if you want to know what the VIP content is. The course contains both "normal" and "VIP" contents. The post explicitly describes which part is "normal" and which part is "VIP".
You can read more about how the VIP courses work here: lazyprogrammer.me/how-do-vip-courses-work/
hi i had a few questions.
im currently predicting percentage change (return) of a given stock.
If i run a model, can i include percentage change and other technical indicators? can i also include High, low, close, open?
Or should I make the TA indicators using high, low, close, open prices and then drop them before feeding it into the model? I assume this should be ok because stock return uses stock prices and if that's ok surely the technical indicators would be as well?
Sorry, i should probably just do classification of up or down. and also include TA indicators, percentage change, high, low, close?
I just really want to do this properly. And ofc i compare this with a naive model
Only way to know if it'll work is to try!
Thank you
Great video!
I was wondering if predicting prices (electricity not stocks) is a bad idea even if the targets are stationary according to the ADF statistics test, my intuition is that differencing already stationary data means that we unnecessarily wipe the memory that is present in the data . What do you think about forecasting "stationary" prices, is it still a bad idea?
Only one way to find out!
I have a question: so if I want to predict sales and I am using a LSTM I am required to use sequences as inputs. While doing that I teach the models how to form the targets given a certain sequence. When I want to predict into the unknown future, is it wrong to assume that I am only able to predict one step ahead due to the requirement of sequences? If I am wrong is there a way to work around it?
All discussed in the Time Series course my friend!
I will purchase it but I need a short term answer: can we only predict one step into the future with LSTMs without feeding back our results?
@@kington2311 The course teaches both one-step and multi-step forecasts. It shows 2 methods of making multi-step forecasts: 1 being the method you mentioned, where predictions are used to make predictions, and the 2nd method does not.
@@LazyProgrammerOfficial Thank you very much!
so far from watching these videos over and over to take notes, i have created a program that has 9/15 accuracy from a model that only has one 850 epoch training session. i am now working on implementing a retraining period after every 5 predictions that will include the new input data in the model and also save the best weights.
i am too broke to buy a course but if i make this work i will definitely purchase
that was a terrible program that i had before upon further testing. i now am working on combining an svm, gradient boost, randomforest, GARCH, and ARIMA for an ensemble prediction. if that goes well i will see if i can also effectively add an lstm or VARMA.
You rock 🔥
There is a whole bunch of papers about price prediction with ML, they touch random walk hypothesis in their introduction and then propose methods which finally have nothing to do with it, I tried to understand what their results really mean but failed hopelessly.
It seems to me that these papers are either useless or sth is wrong about them.
For example the 2020 work of Mehtab, Sen from NSHM Knowledge Campus, with all their referenced related work which also claims to detect some significance but when I check these papers it's a mess again.
How should I interpret them all?
Yep, generally these papers are not great. The metrics they show aren't useful because they never compare to the naive forecast / random walk forecast, nor do they typically provide the code so that the reader can compare themselves.
@@LazyProgrammerOfficial there is no scam included they sell along with the paper, so I don't get why such work exists.
Now I got what they did in the aforementioned report, they define time slots and predict the return when buying at the opening of the current time slot and selling at the opening of the next time slot.
BUT for predicting the % change between opening prices, they feed in the current time slot's high, low, and close price - which were of course not known at its opening. Since the close price is known, the opening price of the next slot will be similar, so they just looked ahead. They were so stupid to post a figure where you can see that the close price has almost all the weight in their network. 47 pages and lots of code for that? WHY?
@@pona5s It helps to think of who writes such papers. For example, many students who simply want to get their project done and get a good grade while working on a "cool" subject. Professors who want to publish papers for the sake of publishing papers, which must be done if they wish to stay employed. They apply ML because there is a "hype" around it, which makes it easy to gain interest and attraction to their works.
GREETINGS AND RESPECT FROM INDIA !!!!
The question is: why aren't YOU predicting stock prices and be millionaire and instead you make youtube videos?
You're making a mistake in assuming what I am claiming. I am not claiming to be teaching an algorithm that can predict stock prices. Therefore, all that is necessary is to listen more carefully. ;)