Time series prediction are a massive part of finance. Instead of working with the time series themselves, I'll try and use images of them: regular plot, recurrence plot, wavelet transform, Gramian Angular Difference Field...
The tools I'd like to use are: CNN, CapNets, Transformer architecture with attention mechanism,...
The purpose will be:
- to cluster the companies from S&P500 according to different variables (stock price, stock trading volume, beta, market cap,...) and see whether the clusters are the same
- to "classify" each company: according to the past 15 days of data, will the stock price increase tomorrow?