CANVAS: a Canadian behavioral agent-based model

We develop a Canadian behavioral agent-based model (CANVAS) that utilizes Canadian micro- and macroeconomic data for forecasting and policy analysis. CANVAS represents a next-generation modelling effort, as it improves upon the previous generation of models in three dimensions: introducing household and firm heterogeneity, departing from rational expectations, and explicitly modelling the Canadian production network. This modelling capacity is achieved by harnessing large-scale Canadian micro- and macroeconomic datasets (of financial flows and national balance sheet accounts, input-output tables, government finance statistics, and the Labour Force Survey). By incorporating adaptive learning and heuristics, we equip the model to examine macroeconomic dynamics under significant uncertainty. We assess the out-of-sample forecasting performance of CANVAS against a benchmark vector auto-regressive (VAR) model and a DSGE model (Terms of Trade Economic Model, ToTEM). CANVAS advances several new frontiers of macroeconomic modelling for the Canadian economy. First, the detailed structure of the model allows for forecasting of the medium-run macroeconomic effects of the economy at the sector level. For instance, this structure allows us to assess the macroeconomic impact of the COVID-19 pandemic in Canada. Second, the realistic agent behaviour in CANVAS makes the model an ideal candidate for evaluating the effects of multiple macroeconomic policies. Third, the enriched modelling of the financial market structure allows policy-makers to conduct stress testing and assess the implication of macroprudential policies in Canada.