In this project I'm going to present the research I've done about TSA. It will cover traditional econometric techniques as well as nonlinear family of models, machine learning and eventually also deep learning techniques.
In the beginning only regression techniques will be covered, however in later on I'm about to present also classification and unsupervised methods for specifically dealing with time series data.
1. My bachelor thesis (praca licencjacka) - Nonlinear time series forecasting with selected econometric and machine learning models.
It is a comparison between econometric (AR, ARMA, SETAR) and machine learning (KNN, Decision Tree and Random Forest) models in context of specifically nonlinear time series. What I believe my thesis does differently than most of the existing research on this topic is the validation method - I have used Sliding Window Cross-Validation with retraining after every step during both training and actual forecasting. Detailed statistical properties of the residuals are covered.
This is an experimental research trying to establish whether it is possible to model the american CPI with econometric and/or machine learning algorithms. Modelling inflation is not a simple task, so this project will be continued and will extend the models used to more and more advanced ones - if you're not dealing with nTSA on your daily basis, it's likely you will see something new! I hate repetitive notebooks, so I will do my best to keep it interesting. Stay tuned!
- Basic EDA and SARIMA (CPI.ipynb)
- Structural break analysis and shortening the data (CPI_short.ipynb)
- Fitting nonlinear SETAR model to the whole dataset (CPI_SETAR.html)
This is a short comparative study that shows statistical differences between returns and log returns of NASDAQ prices for between 1980 and 1990.