Finding a good forecasting model in a data-rich environment is a complex problem which challenges forecasters and statistical methods. In such an environment, automated modelling strategies are necessary for an efficient use of the information in the data. In contrast to frequently applied methods used for large data sets we propose a model selection approach for dynamic single equation regressions that are used to make forecasts. This paper proposes a new approach for quantitative forecasting that is able to deal with both an increasing number of variables that are potentially important for forecasting, as well as an increasing number of observations simultaneously. Another characteristic of the proposed approach is that evaluation of the goodness of forecast models is based on different criteria. As we are interested in finding forecast models with high-quality criteria we define the search for a forecast model as a multi-criteria optimization problem. We define the quality criteria in our goal function by in-sample measures and out-of-sample measures, as well as by a balance between them, and apply a genetic algorithm to solve this complex, global and discrete multi-criteria optimization problem. The efficiency of the approach is illustrated by forecasting German industrial production based on a data set containing key economic indicators and leading indicators. It is shown that, for short forecast horizons, the proposed approach provides forecasts with a high accuracy.