ARIMA Forecasting in R Part 3 - Autocorrelations and Choosing the Model Order HD

This is the third part in the six part video series on building an accurate ARIMA forecast in R. In this video we cover auto correlations and choosing the model order. We will build and plot ACF and PACF plots and look at lags (areas outside the bounds that we need to take into account). We will eliminate the stationarity by adding a differencing step. We figure out the difference level with the diff() function on the deseasonalized data that we created in the previous video. As I tell you in the video you can plot the various differences, but as you go farther out (differences = 2+) you will have higher inaccuracy of the data. So we want to use the lowest difference level possible. In the video I use a different level of one. I also show you what the data looks like we use a different level of two or more. We then run the Dicky - Fuller augmented test again with the difference of one factored in. The results we find are similar to before except that the Dickey Fuller stat (ADF) is more negative and now is -10.81. So we can now reject the null hypothesis and that means that a different level of one is sufficient for our data here. As I state in the video the difference value becomes the D value in our Arima model. We then look at the main lag points and identify a lag at seven. These values are going to be very important and carried over into our next video where we actually build the ARIMA model. with time series forecasting. Thanks of again for watching. Please subscribe and like. God bless!