在過去已經有很多對未拋補利率平價說（uncovered interested parity，UIP）作探討的研究，一般有間接檢定及直接檢定兩種方法。間接檢定主要的分析方法為在假設拋補利率平價說（covered interested parity，CIP）先成立下，若遠期利率是未來即期匯率之不偏估計值假設成立，則表示遠期匯率的變動已經包含所有有關未來即期匯率變動之資訊，遠期匯率可準確預測未來即期匯率，此時外匯市場具有效率性，此效率市場表示CIP與UIP同時成立；而直接檢定是對兩國的資產報酬率直接作其是否相等的檢測。在上述的分析結果，往往無法指出匯率與利差有一對一的變動，於是有其他學者便質疑上述分析之風險中立（即無風險溢酬的存在）假設是否合理，之後在許多UIP的研究中，均不再預先有風險中立的假設，而在模型中加入了風險溢酬的變數。 就針對研究風險溢酬的文章的數量相當多，所提出的模型也是各有優劣，但對於要研究匯率及利率相對關係的研究者，如何在眾多的模型中選擇最適當的模型來分析，便成為一個值得探討的問題。本文採用Durlauf and Hall（1990）所提出之誤差比率（noise ratio），取代傳統評估模型優劣的其他方法，來判定模型與資料的配適程度。 本文主要比較的模型有三種類型，第一種是消費資本資產定價模型（Consumption CAPM model），在其中選取了三個較重要的模型，分別為Hansen and Singleton（1982）所提出之模型（在本文定義為H-S模型）、Brown and Gibbons（1985）之模型（在本文定義為B-G模型）及Epstein and Zin（1991）之模型（在本文定義為E-Z模型）。第二種類型是風險中立模型（risk-neutral model，本文定義為R-N模型），即無風險溢酬的存在。第三種類型是Hansen and Hodrick（1983）所提出之隱藏變數模型（latent variable model，本文定義為H-H模型）。 本篇以美國為主，討探美國與其他工業大國—選取英國、加拿大、法國、德國、義大利以及瑞士之間UIP的關係。最後，由實證結果中可知，在使用不同工具變數下，就三種CCAPM的估計係數及J統計量而言，H-S及E-Z模型的估計係數皆顯著異於0，且無法拒絕變數與殘差成正交的虛無假設，但在B-G模型中，其估計係數不顯著且拒絕變數與殘差成正交的虛無假設；就NR值而言，E-Z模型在三個CCAPM模型中表現較好，平均的NR值皆為最低，在不同工具變數下與R-N模型相較起來約為其8％-15％。當與H-H模型一起比較時，可以發現H-H模型的NR值雖然比B-G及R-N模型要低，但是依然比E-Z模型要來的高，故在本文研究中，E-Z模型相較於其他模型來說其配適資料的程度是最好的，而一般傳統假設風險中立的UIP模型在本文中的表現較差，配適資料的程度較弱。 Substantial empirical literature has rejected the ‘simple efficiency’ hypothesis of the foreign exchange market, i.e. the joint hypothesis that combines a zero risk premium and market efficiency. This rejection does not necessarily imply that the market operates inefficiently. A recognized alternative hypothesis is that a risk premium exists, although inefficiency is certainly an alternative hypothesis. The purpose of this paper is to conduct a race between models in terms of their ability to explain observed violations of uncovered interest parity. The analysis conducted in this paper was motivated by an attempt to identify the importance of risk premiums in explaining violations of uncovered interest parity—a ‘simple efficiency’ hypothesis which assumes risk neutrality and efficiency in the foreign exchange market. This is done by comparing the empirical performance of the ‘simple efficiency’ hypothesis and two types of models which assume the existence of risk premiums—one is implied by the consumption CAPM (of which type this paper suggests three examples), and the other is the HH model, which is implied by the beta CAPM. This paper further uses the hypotheses which assume that people have the same risk-aversion attitude to different countries. The relative empirical performance of each model is assessed using a noise ratio criterion, whose advantage over the J-statistic is that it provides clear insight into which specification is better reconciled with the data among various specifications. No matter people have the same risk-aversion attitude to different countries, the conclusion of this empirical work is that the Epstein-Zin model is more useful than the other consumption CAPMs and the Hansen-Hodrick latent variable model because it best approximates the observed interest rate and foreign exchange rate relationships. The noise ratio of the Epstein-Zin model is about quarter that of the simple efficiency. This implies that the significant components of the repeated rejections of uncovered interest parity could be removed by the risk premiums introduced by the Epstein-Zin model. The other two consumption CAPMs and the HH model are much less successful than the Epstein-Zin model in general.
