Stationarity and Differencing - Applied Time Series Analysis in Python and TensorFlow

👉 Get the course for 87% off: https://www.udemy.com/course/applied-time-series-analysis-in-python/?couponCode=TSPYTHON2021 Email me for a coupon: peixmarco@gmail.com ------------------------------------------------ Let’s introduce the concepts of stationarity and differencing. They are both key concepts in time series analysis as they will come up very often as we progress through the course and start modelling. A time series is said to be stationary when its properties do not change with time. Therefore, it has constant mean and variance, and covariance is independent of time. For example, if a time series has some kind of trend or seasonality, then its mean is not constant, so it not stationary. However, white noise is indeed stationary. Here, we see the ACF plot for a time series. We saw this plot in the previous lesson and we know that this time series with a trend will have a similar ACF plot. Therefore, we can also say from the correlogram that the time series is not stationary. Now, differencing is what is called a transformation. It is a way to make a time series stationary, by removing trend and stabilizing the mean. Taking the algorithm is another popular transformation used to reduce variance. Differencing is simply the difference between consecutive observations. We can difference a time series more than once if necessary, until we make it stationary. However, we never difference more than twice. Here, we see again the ACF of white noise, and so we know that it is a stationary process. So now, you know how to recognize stationarity by looking at the ACF plot. In the next lesson, we will also see how we differencing will impact the stationarity of a time series, and then, we will code each step in Python.