Can there be a market for cheap-talk information?: some experimental evidence