When working with multivariate time series with a significant number of lag components, the presence of multicollinearity among predictor lagged variables is likely. This underscores the requirement for parsimonious models in time series models that allow for parameter reduction. Diagonal Vector Autoregressive (VAR) and Multivariate Autoregressive Distributed Lag (MARDL) models are subclasses of general multivariate time series models with a significant number of lagged variables that can be identified when the coefficient matrices' parameters are restricted to the diagonal elements. The upper and lower diagonal VAR and MARDL models, as well as their variances, are derived. The prerequisites for identifying the diagonal VAR and MARDL models were found in this paper, and the models' validity was shown. To compare the performances of the new classes of multivariate lag models, data from certain macroeconomic variables such as Nigeria Gross Domestic Product (GDP), Crude Oil Petroleum (C/PET), Agriculture (AGRIC), and Telecommunication (TELECOM) are used after the first order difference of the logarithm of the series to achieve stationarity. The models were estimated, and the variances of the processes and errors were determined using the model parameters. The results show that the two models have almost the same comparative advantage. As a result, the two models complement each other when modelling multivariate lag variables.