Learning stock market dynamics using the kinetic ising model
In this project, I investigated an efficient approach using the kinetic Ising Model [1] to fit complex time series data. In this approach, the states in the time series data are represented by configurations of N spins, and the time evolution of these states in the time series data by an update rule...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/73031 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In this project, I investigated an efficient approach using the kinetic Ising Model [1] to fit complex time series data. In this approach, the states in the time series data are represented by configurations of N spins, and the time evolution of these states in the time series data by an update rule that depends on the spin configuration {σ_i (t)}_(i=1,…,N), and the connection weights {W_ij }_(i,j=1,…N) between the spins. Fitting a time series to the kinetic Ising model is achieved by determining the optimal set of weights {W_ij }_(i,j=1,…N), and this is done iteratively using the scheme developed by Hoang et al. [1]. We fitted the stock returns of 30 Dow Jones companies in 2014 from Yahoo finance, and also to 16 artificial data sets generated using the kinetic Ising model resulting in 32 final average weights matrices, to understand the limitations of using the fitted model to do predictions. We find that when we perform static forecasting, i.e. one-step ahead only, the prediction accuracy can as high as 77.60% for a Dow Jones stock and 98.49% for an artificial data. |
|---|