Abstract

This dissertation consists of three interrelated essays. The first essay focuses on the adverse selection component of the bid-ask spread. A regime switching model applied to the trading process leads to a parsimonious model of the time-series evolution of the bid-ask spread in which market participants use trade data to answer the following question: Is there currently private information in the market for a given stock? If there is a high probability of private information in the market, this leads contemporaneously to a greater revision in beliefs about the true price. Together with compensation for transactions costs, this leads to a greater revision in transaction price. Using TSE 35 trade and quote data for March and May 1996, the pooled cross-section and time series results support this view. The second essay examines the costs of adverse information and order processing in light of transaction size, type of trader and type of trading method. Specifically, it is found that adverse selection increases with the trade size (consistent with numerous empirical studies relating trade size and the cost components of the bid-ask spread). However, whether the trade was undertaken by the designated market maker, by a principal trader or by an individual belonging to neither of these two categories is also of importance. In addition, we show that trades consummated within the investment dealer's firm have a lower adverse information cost component than trades executed externally. For order processing, it is found that the single most important determinant of cost is whether the transaction is internal or external to the investment dealer firm, with internal trades being more costly. The third essay examines the robustness of the Huang and Stoll (1997) model estimates to the use of different clustering methods and to a minimum quotation increment reduction (MQIR) on the Toronto Stock Exchange. We find that adequate reversal of trade flow is a necessary but not sufficient condition for model performance. We also find that model estimates are quite sensitive to the data clustering method selected. In addition, we show that this model fails to provide adequate cost component estimates of the spread in the post-MQIR period due to a fundamental change in market-maker behavior.