This paper examines the role of clustering techniques to assist in the selection of the most indicated method to model survey-based expectations. First, relying on a Self-Organizing Map (SOM) analysis and using the financial crisis of 2008 as a benchmark, we distinguish between countries that show a progressive anticipation of the crisis, and countries where sudden changes in expectations occur. We then generate predictions of survey indicators, which are usually used as explanatory variables in econometric models. We compare the forecasting performance of a multi-layer perceptron (MLP) Artificial Neural Network (ANN) model to that of three different time series models. By combining both types of analysis, we find that ANN models outperform time series models in countries in which the evolution of expectations shows brisk changes before impending shocks. Conversely, in countries where expectations follow a smooth transition towards recession, autoregressive integrated moving-average (ARIMA) models outperform neural networks.