In this article we present four diversified approaches to forecasting main macroeconomic variables without a priori assumptions concerning causality. We include tendency survey data in both the Bayesian averaging of classical estimates (BACE) and the dynamic factor models (DFM) frameworks. With respect to the forecasting models based on BACE we propose two methods of regressors’ selection: frequentist (FMA) and averaging (BMA). Our approaches are a priori atheoretical and we refrain from the theory-based selection of exogenous variables. For comparison between forecasts we apply ARIMA method as well. Our approach is comprehensive with respect to the datasets used. We apply data from the tendency surveys conducted at the Research Institute for Economic Development (RIED) at the Warsaw School of Economics (WSE), Poland, on sentiment in the manufacturing industry, trade and construction as well as among households. We additionally include data from foreign and domestic institutes that construct their own leading indicators. We also use the Purchasing Managers’ Index (PMI) for Polish industry in order to assess the quality of results we check in-sample and out-of-sample performance. The results show that, although the results does not significantly differ, the best results are observed in Bayesian models with frequentist approach.