Vector autoregressive (VAR) models have become popular in marketing literature for analyzing the behavior of competitive marketing systems. One drawback of these models is that the number of parameters can become very large, potentially leading to estimation problems. Pooling data for multiple cross-sectional units (stores) can partly alleviate these problems. An important issue in such models is how heterogeneity among cross-sectional units is accounted for. We investigate the performance of several pooling approaches that accommodate different levels of cross-sectional heterogeneity in a simulation study and in an empirical application. Our results show that the random coefficients modeling approach is an overall good choice when the estimated VAR model is used for out-of-sample forecasting only. When the estimated model is used to compute Impulse Response Functions, we conclude that one should select a modeling approach that matches the level of heterogeneity in the data.