Preparing data of Foreign Exchange Rate for Time Series Model
I have been working on a model that can predict the exchange rate for the Australian and US Dollar. The first thing I did was get my data from the federal reserves website.
I got monthly rates for all of the countries available before deciding on which country to work with. I put the counties into continents so it is easier to digest the information.
Based off of these I decided to go with Australia because it and New Zealand seam to be the closest to the US exchange rate. I then looked at the distribution of the exchange rate for Australia.
Seems pretty well distributed. Next I worked on making the data stationary. I use the Dickey Fuller test to determine what was the best out come. The second graph is the original data and the first graph is the difference of the natural log. Seeing the Rolling Std is closer to original and rolling mean makes the data more stationary.
I also looked at the decomposition of both the original data and difference of natural log.
From these graphs we can see that there is some seasonality in both and residual is less of a range in the first difference natural log. The seasonality for first difference natural log seems to spike more towards the end of the year but the original data seems to spike at the beginning of the year.
