Why do we need to make our data stationary?



I have seen, before using ARIMA model, we need to convert our data into stationary model. However, the same is not true for different model like - Naive, Moving, Simple Exponential, Holt’s method. Please help me in the following questions.

  1. Why do we need to make our data stationary only in ARIMA?
  2. What are the different ways to make our data stationary?


Hi @gurjas,

Stationery data means you mean and standard deviation of data does not change in time. In trends, your mean and standard deviation is dependent on past value so you need to do the differencing of it to remove the dependency of the data.

You can think of it as removing a highly correlated variable during model reduction in linear regression. Following link explains it very good.

Hope it helps.

Vikas Jangra


Hi @gurjas,

We make the data stationary only is case of arima because the arima model looks at the past data to predict the future values. (Naive, simple exponential etc do not work that way).

The most common method would be Differencing (one level or seasonal differencing). You can also perform various transformations (square root, log, box cox).

Here are a few articles to help you understand the concept of stationarity and techniques to make the time series stationary:


Thanks… Would read and in case found some issues please help me in that…


Thanks… would read it and in case I’m still stuck somewhere kindly help me understand it better…