Statistics Glossary → Exogenous variables (VARX, ARIMAX)
Exogenous variables (VARX, ARIMAX)concept
External variables in a time-series model that the model itself does not explain but that are assumed to influence the series being explained.
Endogenous variables are those inside the model that influence one another and are forecast; exogenous variables enter the model from outside and are assumed not to be affected by the endogenous ones. In an ARIMAX model (regression with ARIMA errors), the dependent series is explained by a regression on exogenous variables and the error of that regression is modelled as an ARIMA process. In a VARX model, exogenous variables such as weather, holidays or policy changes are added to a VAR system of several interacting endogenous series.
To forecast with an exogenous variable, its values for the forecast periods must be known or forecast separately; the uncertainty in those values is usually not reflected in the forecast intervals, which can therefore look too narrow. Treating a variable as exogenous is a strong assumption: if it is affected in return by the endogenous series (e.g. interest rates and inflation), treating it as exogenous gives biased results, and it is better added to the VAR system as an endogenous variable. In ARIMAX, because of the ARIMA error structure, the coefficients may not be interpretable as directly as ordinary regression coefficients.
An everyday example
You are forecasting an ice-cream seller's daily sales. Sales depend on their own past but are also strongly affected by air temperature. Temperature is not affected by sales, so it is an exogenous variable. To forecast next week's sales, you also need to give the model next week's temperature forecast.
In MerQur
It is used in ARIMA Forecast through 'Independent variables (ARIMAX)', with future regressor values taken from rows whose dependent variable is empty, and in VAR through 'Exogenous variables (VARX)'.
Analyses where this concept appears
ARIMA ForecastVAR (Vector Autoregression)