Does an electronic stock exchange need an upstairs market?

We examine the Paris Bourse, whose electronic limit order market closely resembles the downstairs markets envisioned by theorists, to test several theoretical predictions regarding upstairs trading. We present direct evidence in support of the Grossman (J. Business (1992) 509) prediction that upstairs brokers lower execution costs by tapping into unexpressed liquidity, as actual execution costs upstairs are on average only 20% (35%) as large as they would be if block trades were executed against displayed (displayed and hidden) liquidity in the downstairs limit order book. Consistent with prior analyses, the Paris data also support the Seppi (J. Finance (1990) 73) hypothesis that upstairs brokers certify trades as uninformed. We also find that participants in stocks with less restrictive crossing rules agree to outside-the-quote executions for more difficult trades and at times when downstairs liquidity is lacking. These likely represent trades that could not have been otherwise completed, suggesting that market quality can be enhanced by allowing participants more flexibility to execute blocks at prices outside the quotes.