Granger`s (1969) concept and definitions of causality, feedback, and instantaneous causality and Akaike final prediction error criterion and extended by Chan et al (1982) to fit a multivariate autoregressive model are used. The objective of the paper is to distinguish between simple, feedback and instantaneous causality. The notion of feedback between endogenous and exogenous variables in the bivariate AR model and its extension to the tri-variate AR models is presented. As an application to these Causality notions, the paper aimed to reach the optimal lag structure in forecasting some monetary variables in the Egyptian economy from 2005 to 2018. Variables selected were “current deposits of local currency", “loan totals" and “quasi-money". The three variables were correlated, and each variable was used as an endogenous function of itself lagged and the other two variables as exogeneous. The study also aimed to test if prediction is improved if current values and previous values are used in the prediction equation.
.. The analysis showed that a one-way simple causal model exists from “loans total" to “current deposits of local currency", and from “quasi-money" to “loans total". Instantaneous causality and feedback occur between the three variables.