Can there be a market for cheap-talk information? An experimental investigation