本文應用時段分割方法(time-splitting method, TS method)，將日內交易資料切割為多個區間再進行估計，在當今交易量巨大和高頻交易盛行的情形下，能夠有效地增加資訊交易機率(PIN)估計的準確度。 本文以2014年TAQ全年資料，抽選的150檔證券作為全市場代表，以驗證時段分割法是否能提高PIN估計的準確度。本文加入傳統的未分割傳統方法(the CL method) 作為對照組，比較將一日分為多個切割區間是否能夠增加PIN的效度，並以均方根誤差(RMSE)衡量不同切割區間下PIN參數和PIN的表現。結果顯示切割時段愈短，PIN的均方根誤差也愈小。同時，使用時段分割法估計PIN參數相較未分割方法能夠有效避免產生靠近上下界的極端值。 另外，本文參考Brennan, Huh和Subrahmanyam (2016)，探討未預期盈餘和PIN的關係。透過檢測以time-splitting method (the TS method)估計發生在宣告日前的PIN，再和未預期盈餘的比較，發現在時段分割法下的結果和參考文獻一致。 ;This thesis applies a time-splitting method (the TS method), which divides intraday transaction data into multiple intervals, to effectively increase the accuracy of estimated probability of informed trading (PIN) under the large trading volume and the prevalence of high-frequency trading nowadays. By selecting 150 securities as a representative of the overall market from the 2014 TAQ data, Iverify whether the TS method improves the accuracy of PIN estimation. In this thesis, I add the traditional undivided-method (the CL method) as a control group to compare the increase the validity of PIN with the time-splitting method (the TS method). I also adopt the root mean square error (RMSE) to measure the performance of PIN parameters and PIN under different intervals. The results show that the shorter the splitting interval, the smaller the RMSE of PIN. Meanwhile, using the TS method for PIN parameters estimation can effectively avoid extreme values close to the upper and lower bounds compared with the undivided-method (the CL method). In addition, this thesis follows Brennan, Huh and Subrahmanyam (2016) to discuss the relationship between earnings surprise and PIN of the previous quarter. By testing the time-splitting method (the TS method) to estimate the PIN that occurs before the earning announcement date and comparing it with earnings surprise, it is found that the results are consistent with the literature.
