دانلود رایگان مقاله حجم و قدرت آزمون بر اساس مدل و مولفه های دائم موقت

Abstract

The literature has recently proposed a new type of tests for the Efficient Market Hypothesis based on Permanent-Transitory Component Models. We compare the power of these statistics with conventional tests based on linear regressions. Simulation results suggest that the former dominate the latter for a wide range of data generating processes. We propose an application to spot and forward interest rates. Empirical results show that the two types of tests can yield conflicting results which can be explained by the size distortions and reduced power which affect the statistics based on linear regressions.

7. Conclusions

We evaluate the small sample performances of a new type of statistics based on Permanent-Transitory Components Models (PTCMs) used to test for the Efficient Market Hypothesis (EMH) and rational expectations in financial markets. A comparison between these last and conventional tests based on linear regressions is carried out under a wide range of different data-generating processes featuring integrated and near-integrated spot and forward rates, volatility clustering, misspecified term premia, as well as multiple regime shifts. Empirical results suggest that tests based on PTCMs dominate over the full spectrum of data-generating processes considered, as they present either stronger power or better size. We illustrate an application using Eurodollar and Sterling Libor spot and forward interest rates. Empirical results for Sterling Libor rates suggest that both types of tests are concordant in rejecting the null of EMH and rational expectations. However, when applied to Eurodollar rates the two types of tests yield results of more difficult interpretation. On the one hand, tests based on linear regressions soundly reject both the null. On the other hand, when tests based on PTCMs are applied to the same data, the null of EMH is still rejected whereas the null of rational expectations cannot be rejected at the 10% significance level. We resolve this conflicting result by recalling the findings of our simulation exercises which show that, for integrated spot and forward series, conventional tests based on linear regressions are significantly over sized unlike tests based on PTCMs which present approximately correct size and stronger power.