Forecasting a country's exchange rate is a cimbersome exercise requiring a handy luggage of economics knowledge. Furthemore, this isn't merely limited to pure economics and is a subject of significant effect of politics, news, social moods, neighbors, and many other factors.
This repository is an experiment to model one exchange rate applying a number of classical econometric models as well as machine learning and deep learning techniques.
Some of the techniques used for this project include the following:
- simple time series models
- classical econometric models such as SARIMA and the like
- boosting methods
- neural networks (Recurrent Neural Networks)
Additionally, I provide the exploratory data analysis with visualizations and a number of stats employed.
The project aims to analyze and forecast Kyrgyz Republic's national currency (som or KGS) to US dollar (USD) exchange rate.
Experiments include the Facebook's one stop shop for time series analysis:
kats==0.1.0
Other dependecies include traditional DS packages:
pandas==1.1.5