Old age mortality modelling is often associated with lack of reliable data, especially for small populations. We focus on an approach to incorporate information contained in the data from closed populations and study its impact on estimation uncertainty in an old age mortality model. We assume a two-dimensional age cohort extension of the exponential (Gompertz) model. We compare uncertainty of the parameter estimates for two models. The first is a single population model based on data solely from one country. The second is a multi-population model for a sample of populations from the central European region. Bayesian generalized linear model and a hierarchical Bayesian generalized linear model is applied. We quantify the difference in the uncertainty of the estimates of the force of mortality and whole life annuity based on root mean squared error of the predictions for different ages, cohorts and populations.