Let's get acquainted with some of the most common models for forecasting time series:

• Simple moving average. This model is often used for datasets that do not have seasonality or pronounced trends (such type of data is called stationary data). In this model, we create predictions based on the mean value of previous data points.

• Autoregressive model. The model is based on the principle of linear regression with the addition of a dependent variable, a linear function of past values of the dependent variable. This model can be used only for stationary data;
• Moving average model - this model also works like linear regression, but instead of using previous values as predictors, this model makes predictions using a linear combination of samples generated from the white noise process. It's important to admit that Moving average and Simple moving average are two different models. This model also can be used only for stationary data;
• Autoregressive integrated moving average model. As the name implies, this model combines the previous two with an additional differencing step. What advantage can this give? For example, the ability to represent a time series with non-stationary properties;
• Vector autoregressive model. The main feature of this model is the ability to predict multivariate time series.

We briefly covered 5 types of models that we will use in the future. In the following sections, we will closely examine the mathematical "mechanism" and implementation in Python of some of these models.

Note

At the moment, the concept of stationarity may seem unfamiliar and incomprehensible to you but don't worry, it will be discussed in more detail later.