Greetings Sir ,
I am very interested by the project.
I worked by the past with financial datasets using ARIMA model to predict a stock price and GARCH model to capture the volatility clustering . I have also used LSTM for the same purpose in order to put my hands on deep learning. Also , I have worked with Machine Learning more precisely SVM and Logistic Regression as Buy/Sell Signals. Lastly , I have used Reinforcement Learning using the Open gym library in order to maximize the rewards.
So , in terms of skills about the topic , I don't have any problem.
Just to make sure that I have well understood what you are waiting from me :
you want me to create a time series model that will make predictions (few seconds ahead) , to create a machine learning model that will predict either to buy/sell and finally an reinforcement learning model to help models learn.
If this is the case , the dataset should have seconds frequency because if you train the model on seconds basis and you make predictions on seconds basis this will lead to errors.
Also , do you have any specific features that you want me to work with ? because if I decide to work only with "Close" prices than I will use ARIMA . and if you have other features that you want me to incorporate to predict "Close" prices than we will use ARIMAX (and not ANIMAX).
In addition to that , if the predictions are higher than the current price : it's a buy signal otherwise a Sell Signal.
**Question : why in this case do you need a Machine Learning model to tell you when to buy and when to sell if you can deduce that from predictions of ARIMA/ARIMAX ?
--> Maybe you want to reinforce the signal by taking the intersection of ARIMA and a Machine Learning Model ? which will lead to a more effective signal and less trades (save transaction costs)
Seeking forward to get in touch
See you
Iheb