過去台灣政府公債一向是台灣壽險業者長年期投資資產配置的要角，但近年具有中央公債交易商資格的五大壽險，國泰人壽、新光人壽、南山人壽、台灣人壽與富邦人壽，持續減持公債。根據中央銀行最新統計，五大壽險持債比率在今年3月已跌至9.07% 的歷史新低紀錄。在台債殖利率及波動度持續走低，且政府舉債保守公債供給銳減的現況下，壽險原本買入長期持有的交易策略已無法提供足夠報酬，並帶來長期營運上的風險。本文希望能提升公司原先採用的14天加權移動平均 (weighted moving average，簡稱 WMA) 突破策略交易模型的績效，擬分別運用以「台股變動」、「美國聯邦基金利率」、「美國十年期公債殖利率」及「台灣隔夜拆款利率」等影響台債殖利率的重要因子建立的多元迴歸預測模型及時間序列預測模型，產出台灣十年期公債的預測值，並用其取代市場價格來計算交易訊號，以降低市場雜訊的干擾，達到提升整體報酬的目的。實證結果發現在以2008到2017為樣本的樣本內回測中，以時間序列預測模型的 WMA突破策略績效最優，多元迴歸預測模型的 WMA突破策略績效次之，均優於原先採用的14天 WMA突破策略；而在2018到2019的樣本外回測中亦有相同之結果，顯示在加入量化分析之後，單純的技術指標交易策略在績效上能有顯著的提升。建議後續研究者可搭配更多的技術指標策略及不同的參數設定，搭配不同的量化模型來組成新的交易策略，看是否能達到更好的投資報酬。

Taiwan government bonds have always been the important part of Taiwan’s life insurance industry’s long-term investment assets allocation. But in recent years, the five major life insurances continue to reduce their positions. According to the latest statistics of TCB, the five life insurance holding ratios have fallen to a record low of 9.07% in March. Taiwan government bond’s yield and volatility continue falling, and according to the government's debt-raising conservative, government bond supply has plummeted. The long-term trading strategy has not been able to provide sufficient profit. This paper hopes to improve the performance of the company's original 14-day weighted moving average (WMA) breakthrough strategy trading model. It is intended to use the "Taiwan stock change", "US federal funds rate" and "US ten-year period" respectively. Multivariate regression prediction model and time series prediction model established by important factors affecting Taiwan bond yields, such as the “public debt yield” and “Taiwan overnight interest rate”, which produce the predicted value of Taiwan’s 10-year public debt and use it to replace market prices. Calculate trading signals to reduce the interference of market noise and achieve the goal of improving overall compensation.
This paper hopes to improve the performance of the company's original 14-day weighted moving average (WMA) breakthrough strategy trading model. Trying to use "Taiwan stock change", "US federal funds rate" ,"US ten-year period" ,“public debt yield” and “Taiwan overnight interest rate” establish multiple regression prediction model and time series prediction model. And produce the predicted value of Taiwan’s 10-year public debt and use it to replace market prices. Calculate trading signals to reduce the interference of market noise and improve profitibility. The empirical results show that in the backtesting from 2008 to 2017, the WMA breakthrough strategy performance of the time series prediction model is the best, and the WMA breakthrough strategy performance of the multiple regression prediction model is second, which is better than the original 14 days. The WMA breakthrough strategy; and the same results in the extra-sample backtest from 2018 to 2019, shows that after the addition of quantitative analysis, the simple technical indicator trading strategy can significantly improve the performance.

1. 蔡釗旻(2017)，「臺灣殖利率曲線之建構分析與利率傳遞機制之驗證－兼論臺美利率關聯性」，中央銀行季刊，第39卷第4期，頁47-76。
2. 陳建光、吳紫揚(2009)，「台灣殖利率曲線利差變化與交易策略」，證券公會季刊，頁37-43。
3. 黃富纖(2017)，「殖利率曲線與經濟成長預測－美國與臺灣實證分析」，經濟研究，第17期，頁307-328。
4. 楊淑青(2008)，「影響台灣十年期指標公債殖利率走勢實證研究-迴歸模型運用」，碩士論文，國立成功大學。
5. 陳谷劦(2003)，「公債殖利率與總體變數間關係之探討-台灣的研究實證」，碩士論文，世新大學。
6. 謝介壽(2005)，「台灣公債當沖交易運用技術分析有效性之交易」，碩士論文，國立臺灣科技大學。
7. 袁麗梅(2018)，「應用機器學習預測台灣十年期指標公債殖利率」，碩士論文，國立臺灣科技大學。
8. 江琇貞(2001)，「台灣公債殖利率與台股指數因果關係之實證研究」，碩士論文，國立高雄第一科技大學。
9. 李辛(2009)，「美國次貸風暴前後台債與美債殖利率關聯性之研究」，碩士論文，淡江大學。
10. 許萬宗(2009)，「台灣債券市場與各金融市場之相關性分析」，碩士論文，國立中央大學。