We develop an agent-based model (ABM) for the Austrian economy using data from national accounts, input-output tables (IOTs), government statistics, census data and business surveys. The model incorporates all economic activities (producing and distributive transactions) as classified by the European system of national accounts (ESA); and all economic entities, i.e. all juridical and natural persons, are represented by agents (at a scale of 1:10). This large amount of data and high level of detail enables us to set all our parameter values according to the data of our benchmark year, i.e. our model is not subject to the parameter identification problem. We show that this model is able to improve on vector autoregressive (VAR) and autoregressive moving average (ARMA) models in out-of-sample prediction. In a comparison over forecasts horizons up to 12 quarters, our ABM clearly outperforms VAR and ARMA models for almost all forecast horizons and macroeconomic aggregates. Potential applications of this ABM include economic forecasting, stress test exercises and predicting the effects of changes in monetary, fiscal, or other macroeconomic policies. Joint with Michael Miess and Stefan Thurner.