Predictability of Stock Market

Soumyajit Roy

Background

We all know about the unpredictability of Stock Market. Try to model this data using some mathematical formula is not a new approach. The main motivation for doing this has come from Renaissance Technology. The founder of this company is James H. Simmons, a renowned mathematical genius who started to work on this being bored from Mathematics. His company is one of the most successful Hedge Fund and investment firm of all time with return more than 60%.

J.H.Simmons

Motivation

A team of researcher from MIT published a paper where they rejected the hypothesis that Stock market is totally unpredictable. So there is a scope that we can predict Stock Market Movement. There may be a very complicated pattern,so we can model using Machine Learning and Deep Learning methods to capture the complex relationship of the movement.

Background Dataset

There are huge amount of data is generating in every second in Stock Market regular days. There are some packages in Python which provides us with NSE live data as well as historical data for desired time. Ex:- yfinance,nselib,jugaad_data etc.

Previous Work

It’s been a “Hot Topic” for years as everyone wants to be rich. There’s been many approaches for this modelling. Some of them are,

1. Markov Chain with estimated TPM .

2. Some advanced Stochastic Approach

Time series modelling approach

4. Machine Learning Approach

5. Using ANN

6. Using LLM and Generative AI(StockGPT) (https://arxiv.org/search/?query=stock+market+modelling&searchtype=all&source=header)

Approach to the problem

It has been seen that Transformer based architecture is not very successful. Also all other approaches have there own accuracy and justification. I am thinking if I could create a DNN architecture which can capture the very long term stochastic dependence efficiently(RNN and LSTM but with a longer dependency ). I want to mix Time series approach with DNN based approach to see if model efficiency increases.

Referance

https://medium.com/@TejasEkawade/getting-indian-stock-prices-using-python-19f8c83d2015
https://en.wikipedia.org/wiki/Jim_Simons