In this paper, we propose a robust evaluation of the information content of microblogging data to forecast useful stock market variables: returns, volatility and trading volume of diverse dataset of indices and portfolios. We analyze a large Twitter dataset, from December 2012 to October 2015, with about 31 million messages related with 3,800 stocks traded in US markets. Also, we apply a sound prediction procedure (e.g., rolling window evaluation, four regression methods) along with a statistical test of predictive accuracy. Furthermore, we explore the diversity of traditional sentiment indicators and assess their complementarity value with microblogging sentiment. A Kalman Filter (KF) procedure is applied to create an unique daily sentiment indicator from a Twitter indicator and four other sentiment indicators (created from surveys). We also predicted two popular survey sentiment indicators using microblogging data. We found that Twitter sentiment and posting volume were particularly important for the forecasting of returns of S&P 500 index, portfolios of lower market capitalization and some industries. Additionally, KF sentiment was informative for the forecasting of returns. Furthermore, Twitter and KF sentiment indicators were useful for the prediction of some AAII and II survey sentiment indicators. These results show that microblogging data are relevant to forecast stock market behavior and can provide a valuable alternative for existing measures (e.g., survey sentiment) with various advantages (e.g., fast and cheap creation, daily frequency).