In the period 1993 through 2002 examined in this study, quoted and effective spreads declined substantially on Nasdaq and to a lesser degree on the NYSE. At the same time, however, trades outside the quotes increased dramatically on Nasdaq. Because investors would prefer to trade at the quotes rather than outside the quotes, we examine why trades outside the quotes are observed. We focus on how the continuous market mechanism itself influences the outcome of orders and the reporting of trades, and we conclude that slippage exists in the market mechanism. Outside-trades occur on Nasdaq, first, because of delays in reporting trades, second, because the ability of dealers to delay execution of trades creates a look-back option, which when exercised results in outside-trades, and third, because large trades can take place at prices outside the quotes. Outside-trades are rarely observed on the NYSE because the market is more centralized. While the pattern of trading on the NYSE is not inconsistent with the presence of a look-back option, our tests provide no direct evidence that specialists are exercising such an option.

The paper examines the optimal behavior of a single dealer who is faced with a stochastic demand to trade (modeled by a continuous time Poisson jump process) and facing return risk on his stock and on the rest of his portfolio (modeled by diffusion processes). Using stochastic dynamic programming, we derive the optimal bid and ask prices that maximize the dealer's expected utility of terminal wealth as a function of the state in which he finds himself. The relationship of the bid and ask prices to inventory of the dealer, instantaneous variance of return, stochastic arrival of transactions and other variables is examined.

Execution costs, as measured by the quoted spread, the effective spread (which accounts for trades inside the quotes), the realized spread (which measures revenues of suppliers of immediacy), the Roll (1984) implied spread, and the post-trade variability, are twice as large for a sample of NASDAQ stocks as they are for a matched sample of NYSE stocks. The difference is not due to differences in adverse information, in market depth, or in the frequency of even-eighth quotes. Partial explanations are provided by differences in the treatment of limit orders and commissions in the two markets. We conclude that important explanations are the internalization and preferencing of order flow and the presence of alternative interdealer trading systems, factors that limit dealers' incentives to narrow spreads on NASDAQ.

Recent empirical work by Banz (1981) and Reinganum (1981) documents abnormally large risk-adjusted returns for small firms listed on the NYSE and the AMEX. The strength and persistence with which the returns appear lead both authors to conclude the single-period, two-parameter capital asset pricing model is misspecified. This study (1) confirms that total market value of common stock equity varies inversely with risk-adjusted returns, (2) demonstrates that price per share does also, and (3) finds that transaction costs at least partially account for the abnormality.