This paper employs a “transaction” data‐base to study whether observed quote‐revisions are consistent with those predicted by the adverse selection and inventory cost theories of the bid‐ask spread. We find that actual quote‐revisions are consistent with the theoretical prediction in only 25% of the cases. Furthermore, quote‐revision patterns are found to be strongly dependent on the level of the outstanding spread and, to a lesser extent, on the transaction size. These systematic patterns, unrelated to the inventory cost and adverse selection theories, are consistent with the effect on quote‐revisions of the limit order book and the minimum 1/8 price‐change rule.

This paper employs a “transaction” data-base to study whether observed quote-revisions are consistent with those predicted by the adverse selection and inventory cost theories of the bid-ask spread. We find that actual quote-revisions are consistent with the theoretical prediction in only 25% of the cases. Furthermore, quote-revision patterns are found to be strongly dependent on the level of the outstanding spread and, to a lesser extent, on the transaction size. These systematic patterns, unrelated to the inventory cost and adverse selection theories, are consistent with the effect on quote-revisions of the limit order book and the minimum 1/8 price-change rule.

Recent theoretical work on the bid‐ask spread asserts that the dealer should widen the bid‐ask spread when he or she suspects that the information advantage possessed by informed traders has increased. Thus, the dealer's spread can be employed to test for an increase in information asymmetry prior to an anticipated information event. In this paper, the method is applied to earnings and dividend announcements, which have been documented to be information events. The authors study three groups of announcements: (a) joint announcements—i.e., earnings and dividend announcements that are made on the same day, (b) initial (first) announcements—earnings or dividend announcements that were not preceded by another announcement in the prior thirty days, and (c) following (second) announcements—those announcements that follow the first announcement by at least ten days but by no more than thirty days. The authors find a strong increase in information asymmetry only before the second announcements and virtually no increase before the joint and first announcements. This is consistent with the hypothesis that there is, on average, normal information asymmetry before announcements, but that the dealer will suspect a nonroutine announcement (with an attendant increase in information asymmetry) when the second announcement is separated from the first by more than ten days. Other possible explanations for the results are discussed, and suggestions for future research are outlined.

Recent theoretical work on the bid-ask spread asserts that the dealer should widen the bid-ask spread when he or she suspects that the information advantage possessed by informed traders has increased. Thus, the dealer's spread can be employed to test for an increase in information asymmetry prior to an anticipated information event. In this paper, the method is applied to earnings and dividend announcements, which have been documented to be information events. The authors study three groups of announcements: (a) joint announcements—i.e., earnings and dividend announcements that are made on the same day, (b) initial (first) announcements—earnings or dividend announcements that were not preceded by another announcement in the prior thirty days, and (c) following (second) announcements—those announcements that follow the first announcement by at least ten days but by no more than thirty days. The authors find a strong increase in information asymmetry only before the second announcements and virtually no increase before the joint and first announcements. This is consistent with the hypothesis that there is, on average, normal information asymmetry before announcements, but that the dealer will suspect a nonroutine announcement (with an attendant increase in information asymmetry) when the second announcement is separated from the first by more than ten days. Other possible explanations for the results are discussed, and suggestions for future research are outlined.