Abstract Based on the sequential auction model of Kyle (1985) and embedded the formulation of positive feedback traders in De Long et al. (1990), our model formulates a recursive market game of insiders, noise traders, and market makers. In particular, the submitted demands of positive feedback inertial traders are influenced by previous own trading quantities. I prove the existence and uniqueness of a recursive linear equilibrium with positive feedback inertial trades. Further, the equilibrium calibrates that the strategies of insider and market makers are also influenced by positive feedback trades. Finally, we conduct a simulation analysis to get a price-volume pattern with some empirical interesting implications. Finally, this thesis takes trading strategies to trade the individual stock in TSEC. Although the market mechanism of TSEC has no market makers, it is still expected that these trading strategies are useful for traders which implies the information is filtrated by these trading strategies.

本文以三篇論文討論三個與財務報表向下重編盈餘公司有關的議題：（1）宣告財務報表重編後的長期股票績效；（2）重編公司內部人交易行為；（3）重編公司如何操弄盈餘以及操弄動機為何？ 本文樣本是從1984年1月至2000年12月，557家因為財務報表違法、舞弊或錯誤而宣告需要重編以前財務報表的公司。 第一篇論文討論公司宣告重編後三年長期股票績效。在不同配對組合下，實證結果顯示，三年長期持有異常報酬（buy-and-hold abnormal return）達-34％。文中也討論公司宣告重編時以及後續分析師預測行為。結果發現，分析師對於重編資訊有反應不足的現象，而且三年的盈餘預測修正與長期異常報酬有顯著正相關。 第二篇論文探討盈餘品質與內部人交易行為。本文假設內部人擁有重編公司盈餘品質不良的私有資訊，內部人會利用此私有資訊從事異常交易。實證結果顯示，內部人早在重編前兩年就已經開始異常出售持股，但是為了避免被發現，愈接近重編期間則交易量愈少，而且內部人異常交易與重編金額成正相關。 第三篇論文採取應計項目分項的方法（disaggregate approach）探討重編項目與特定應計項目的關連性以及重編公司違反一般公認會計原則的動機。實證結果顯示，特定重編項目公司的總應計項目比特定應計項目更顯著。很可能是管理當局為了避免操弄特定項目以致於被發現，所以操弄各種應計項目以達到總金額的目標。尤其在不同盈餘管理動機的成本效益的考量下，如果操弄主要應計項目的效益大於成本，管理當局還是會操弄特定應計項目以達到操弄的目的。因此，應計項目分項的方法可以作為進一步探討盈餘管理的方法。 / This dissertation examines three different aspects of downward earnings restatements in three essays: (1) the long-run stock performance of restatement firms following the announcements of restatements; (2) insider trading activities of earnings restatement firms; and (3) how earnings manipulations of restatement firms are effected and what are the incentives for earnings manipulations? Using extensive keyword (i.e., “restatements,” “restate,” “restated,” “restates,” and “restating”) searches over the period from January 1, 1984 through December 31, 2000, 557 firms are identified as having restated their previously published or filed financial statements due to accounting irregularities, fraud, or errors. The first essay examines the post earnings restatement announcement of long-term stock performance. Using various benchmark portfolio formulation strategies, I document an average buy-and-hold abnormal return of -34% over the 36-month horizon. I then investigate analysts’ forecast behavior around and after the restatement announcements. I find that market underreactions are associated with a sluggish forecast revisions by financial analysts. This study sheds light on how restatement information is transmitted to the capital markets and provides evidence that the market under reacts to externally initiated corporate events. The second essay examines the relationship between earnings quality and insider trading. Using downward earnings restatement firms to identify low-quality earnings, I find that insiders outsell non-earnings restatement firms of their holdings over the period from two years before to one year before the beginning of the restatement period. In addition, the amounts of restatement are positively associated with the excess insider selling. I also provide evidence that excess insider selling predicts excessive earnings manipulations that eventually lead to GAAP violation. In the third essay, I take advantage of the disclosed manipulation of items and approach the earnings manipulation issue by a disaggregate approach. Given that management considers cost/benefits of specific accruals to be manipulated, I examine whether management chooses different items to manipulate under different goals. Overall, the empirical results support the equity offering hypothesis and weakly support the meeting earnings threshold hypothesis. However, the results fail to support the avoidance of debt covenant violation hypothesis, indicating that manipulation under certain monitoring conditions can be conducted in a very subtle manner.