It is astonishing how certain published papers in reputable journals presume to yield satisfactory results while employing the misleading LSTM forecasting method. If LSTM models were genuinely powerful, one would expect them to generate results comparable to those achieved in stock returns. However, surprisingly, they perform as poorly as ARIMA or naive forecasts when it comes to returns. This realization has led me to focus on developing deep learning models supported by textual data and NLP techniques in my current thesis, aiming to enhance predictive accuracy.
I thought I was losing my mind. I'm forecasting time series for a uni project, supervisor gave me a link to a guide, I watched some videos on the internet, even read some papers. They all did this, and I was wondering how on earth this was meant to be a forecast. Thank you for these videos, they are very useful
Thanks, great series! Something I didn’t understand from this part though: Suppose you wanted to employ a model to only predict the next day‘s/period’s returns and trade based on those predictions - which I‘d imagine to be the most common use case. Then you actually always would have the returns up to the respective current day in practice. Therefore, in that case, I don’t see anything wrong with what you are criticizing? Why would you care to predict so far ahead with all its intermediate time steps anyway? Thanks!
I am in the middle of taking your Bayesian Stat. A/B Testing on Udemy. It is hands down the best, and most enjoyable, course I have done on Udemy. I needed a good refresher for my new role in marketing as a Sr. DA, and it's helping me out a ton. I love that you include theory as well as application.
I am very glad to see discount codes because in my country your courses' prices are equals to 10% of my country's minimum wage if I don't use the codes. Discounts have been used to course. Nevertheless, prices are equals to 4% of minimum wage . lol. I have just realised that I wish I would live somewhere better :D
Finally, was so confused about what a true forecast was, and wtf these crazy accurate 'forecasts' were in the literature. Now it's clear that they are doing 1 step predictions which like you say, is cheating. Here is my question. So in order to actually achieve a true forecast with an ML model, would we keep using the newly predicted value as the inputs and so on ? Are there any other ways to truly forecast with an ML model ?
It depends on the nature of the forecast. In my time series course, we showed how to identify whether or not something can be predicted. And hence, how to avoid predicting things which are unpredictable. Furthermore, as taught in my financial engineering course, one should think about why they are trying to use past to predict future in the first place.
Please correct me if I'm wrong, ideally we can train on lets say 70% data. then we can forecast for the rest 30% data using the future_steps or sequences variable?
It is astonishing how certain published papers in reputable journals presume to yield satisfactory results while employing the misleading LSTM forecasting method. If LSTM models were genuinely powerful, one would expect them to generate results comparable to those achieved in stock returns. However, surprisingly, they perform as poorly as ARIMA or naive forecasts when it comes to returns. This realization has led me to focus on developing deep learning models supported by textual data and NLP techniques in my current thesis, aiming to enhance predictive accuracy.
Thank you for this! I thought I was going crazy that every code on earth seemed to make the same mistake.
I thought I was losing my mind. I'm forecasting time series for a uni project, supervisor gave me a link to a guide, I watched some videos on the internet, even read some papers. They all did this, and I was wondering how on earth this was meant to be a forecast. Thank you for these videos, they are very useful
Thanks, great series! Something I didn’t understand from this part though: Suppose you wanted to employ a model to only predict the next day‘s/period’s returns and trade based on those predictions - which I‘d imagine to be the most common use case. Then you actually always would have the returns up to the respective current day in practice. Therefore, in that case, I don’t see anything wrong with what you are criticizing? Why would you care to predict so far ahead with all its intermediate time steps anyway? Thanks!
I am in the middle of taking your Bayesian Stat. A/B Testing on Udemy. It is hands down the best, and most enjoyable, course I have done on Udemy. I needed a good refresher for my new role in marketing as a Sr. DA, and it's helping me out a ton. I love that you include theory as well as application.
I am very glad to see discount codes because in my country your courses' prices are equals to 10% of my country's minimum wage if I don't use the codes. Discounts have been used to course. Nevertheless, prices are equals to 4% of minimum wage . lol. I have just realised that I wish I would live somewhere better :D
Finally, was so confused about what a true forecast was, and wtf these crazy accurate 'forecasts' were in the literature. Now it's clear that they are doing 1 step predictions which like you say, is cheating.
Here is my question.
So in order to actually achieve a true forecast with an ML model, would we keep using the newly predicted value as the inputs and so on ?
Are there any other ways to truly forecast with an ML model ?
It depends on the nature of the forecast. In my time series course, we showed how to identify whether or not something can be predicted. And hence, how to avoid predicting things which are unpredictable.
Furthermore, as taught in my financial engineering course, one should think about why they are trying to use past to predict future in the first place.
I bought your new course & can’t wait to study it during the holiday break.
can u help me to reach the course cause currently i havn't enough money to buy it
thanks
@@elyacine6457 Reach out to the developer & see if they have a discount in the future within your budget.
Please correct me if I'm wrong,
ideally we can train on lets say 70% data.
then we can forecast for the rest 30% data using the future_steps or sequences variable?
It's covered in my courses ;) See either the Financial Engineering or Time Series Analysis course.
So can we not use models like Linear Regression, SVR, RFRs to predict the market ?
then for lstm case how should be split the data or not?
It's covered in my courses ;) See either the Financial Engineering or Time Series Analysis course.
@@LazyProgrammerOfficial which course of among them should we go for to learn about predicting next day trend ?
@@virusvomen6286 Both the Financial Engineering course and Time Series Analysis course are recommended!