條件變幅極端值法在期貨保證金估計之應用

Abstract

本文以日內價格變幅資料為基礎，提出兩階段條件變幅極端值法，結合不對稱變幅自我相關（ACARR）模型與極端值理論，藉以刻劃期貨價格變幅的條件異質變異性與極端值行為。本文以臺灣期交所臺指期貨與新加坡期交所摩根臺指期貨作為對比樣本，無論在樣本內回溯測試，或是樣本外預期損失的估計上，本文所提出的條件變幅極端值法，其績效皆優於McNeil and Frey （2000）所提出以報酬率為基礎的條件極端值法。其次，本文亦探討漲跌幅限制對期貨價格行為與模型適用性的影響，評比結果發現7%漲跌幅限制截斷了臺指期貨的極端價格，而且McNeil and Frey（2000）的條件極端值法常會估計出超過漲跌幅限制的價格變化；相對地本文所提出的條件變幅極端值法所估計的價格變化較為合理。這些結果皆顯示本文所提出的條件變幅極端值法較不受漲跌幅限制的影響，支持條件變幅極端值法在刻劃期貨價格過程的優越性。

The article proposes a two-step conditional range-based extreme value approach, which combines the range-based ACARR and extrem value theory. The conditional range-based extreme value approach is applied to examine the hetrosdacasity and the extreme price behavior of futures. Both Taiwan Stock Index Futures and MSCI-Taiwan Stock Index Futures are treated as comparison samples and both the in-sample backtesting and out-of-sample expected loss forecasting indicate that the conditional range-based extreme value approach performs better than McNeil & Frey's (2000) conditional extreme value approach. Furthermore, regarding the impact of price limit mechanism on the price behaviour of futures and the two alternative extreme value approaches, the empirical results show that 7% price limit mechanism trauncates the extreme price of Taiwan Stock Index Futures, and they also support the superiority of the conditional range-based extreme value approach. These findings demonstrate that the conditional range-based extreme value approach is more robust, especially in descrebing the process of the future's price.