US-to-CAD-Exchange-Rate-Forecasting

Introduction:

An attempt to forecast the US to CAD Exchange Rate.

One can view a Shiny app here, that summarizes the results of this repository.

This project is inspired by what my friend (Riley Peters) and I did in our term project at the University of Regina with a much greater emphasis on:

1. Forecasting and forecasting performance over general time series analysis.
2. Utilizing cross validation in a time series context in which models are dynamically refit per each slice of time.
3. Larger emphasis on accurate prediction intervals, including using methods proposed by Taylor and Bunn (1999) and Svetunkov (2017).
4. Shorter term forecasts; the term project we did prioritized 12 month ahead forecasts which were unlikely to be accurate given how long the time frame is.

Instructions:

Requires Docker. You may need sudo privileges in order to run the following docker commands.

To recreate this repository from scratch (and to run the Shiny app locally):

1. Clone this repository.
2. In the root directory of this repository, run:

docker run --rm -v "/$(pwd):/excaus" btang101/excaus_app make -C /excaus all

This might take about one hour to run.

3. Run, from the root directory of this repository:

docker run --rm -v "/$(pwd):/excaus" btang101/excaus_app Rscript -e "shiny::runApp('excaus_forecast')"

Then visit the address printed in the terminal using any web browser.

To reset the repository without any intermediate results, run:

docker run --rm -v "/$(pwd):/excaus" btang101/excaus_app make -C /excaus clean

Windows users, you may need to replace:

"/$(pwd):/excaus" with "${pwd}://excaus"

Dependencies:

• R (tested on 3.6):
  • alfred==0.1.1
  • caret==6.0-86
  • forecast==8.12
  • zoo==1.8-7
  • smooth==2.5.6
  • lubridate==1.7.8
  • shiny==1.4.0.2
  • shinyBS==0.61
  • shinythemes==1.1.2
  • shinyWidgets==0.5.1
  • plotly==4.9.2.1
  • DT==0.13
  • yardstick==0.0.6
  • tidyverse==1.3.0

I recommend just using the Docker container.