Hello there:
I have taken many courses on quantra… and Learned a lot in the process…Neverteless I have a traing Idea that I want to try out and use machine learning for that.
My idea is the following… I want to spot bubbles in the US stock market, and as stated in a course in quantra… I want to use machine learning to spot the "Low of the day" and the "High of the day" of those bubbles so that I can simply use orders to buy or sell near those prices and make a profit…
I am spotting the following bubbles for that I think, will continue to run on Monday 15th, April 2024… (today's date is Saturday 13th april 2024 )
Ticker "HUSA", Houston American Energy Corp.
Ticker "AGX", Argan Inc.
Ticker "WINT", Windtree therapeutics Inc.
If you have checked those stocks… they have made big moves… So Is there anything you could recommend me to tackle the problem of choosing features in order to forecast the "LOW of the day" and the "High of the day" of those three stocks stated above for Monday 15th , 2024??
How can I begin to tackle this problem ?? how would you do it ??
Hi Ghery,
This is an interesting idea. Obviously, you will have to backtest it before you can say it is good (or not).
Let's try a different way to tackle this idea.
As a starting point, you must have gathered the OHLC data of the stocks in question. If you are considering ML, you already know that you need a target variable.
Consider a classification model. Maybe you can answer the question, "Is the stock in a bubble?" The answer is yes, or no.
Depending on the strategy, you would buy or sell depending on the answer.
Once the target variable is established, what features can you use to predict the bubble.
The easiest line of thinking is a rapid increase in the price might suggest a bubble. So, you could check the percentage increase in the close price (or low, high, open) as a features.
Apart from that you can have a feature which is related to fundamentals. If the fundamentals are the same (or worse) and still price is increasing, then it could be a bubble.
Can you have sentiment indicators as well? You can try sentiment analysis for this exercise.
Of course these are just certain examples and you can add your own features to it. You can check the feature importance as well to make sure the features have "some" predictive power over the target variable.
Hope this helps.