We apply a model-based approach to forecasting vote shares in Canadian federal elections using individual-level data from the Canadian Election Study (CES). Drawing on the voter-level forecasting framework proposed by Camatarri (2024), we estimate weighted logistic regression models that incorporate party identification, ideological self-placement, retrospective economic evaluations, government trust, and sociodemographic controls to predict vote choice, then aggregate predictions to estimate aggregate electoral outcomes. We examine the 2019, 2021, and 2025 federal elections, employing stratified bootstrap inference and Bayesian validation via Markov chain Monte Carlo to assess the stability and uncertainty of our estimates. The paper contributes to debates about the applicability of voter-level modelling to multi-party parliamentary systems and the temporal dynamics of voting behaviour in the Canadian context.