Small sample properties of alternative tests for martingale difference hypothesis

Abstract

A Monte Carlo experiment is conducted to compare power properties of alternative tests for the martingale difference hypothesis. Overall, we find that the wild bootstrap automatic variance ratio test shows the highest power against linear dependence; while the generalized spectral test performs most desirably under nonlinear dependence.

Will Precious Metals Shine? A Market Efficiency Perspective

“…These tests perform highly desirably in small samples as a means of testing for the martingale difference property (no return predictability) of asset returns (see Charles et al, 2011). In particular, these tests are robust to non-normality and (conditional) heteroscedasticity that are stylized features of precious metals returns (see, e.g., Hammoudeh et al, 2011;Cochran et al, 2012).…”

Eficiência adaptativa nos mercados futuros agropecuários brasileiros

“…A escolha do janelamento deu-se com base nas simulações de Monte Carlo de Charles et al (2011), as quais não encontraram distorções amostrais para os testes acima de 100 observações. Ao serem implementadas as rolagens em subamostras, essa abordagem permite inferências robustas contra possíveis mudanças estruturais, além de viabilizarem a comparação sobre a eficiência relativa dos mercados.…”

Time-Varying Return Predictability in the Chinese Stock Market

“…Charles et al compared di®erent MDH testing methods through conducting Monte Carlo experiments. 10 They concluded that the wild bootstrap automatic variance ratio test (hereafter, AVR test) 33 and the generalized spectral test (hereafter, GS test) 17 are more favorable to test the linear and nonlinear dependencies in the return time series. Hence, we use both the AVR test and the GS test to investigate the return predictability in the Chinese stock market.…”

“…These results are not sensitive to the two exchanges and data sample frequency. Since the Monte Carlo shows that AVR tests are not always performed better than other test approaches when the data generating process is AR(1) with GARCH errors and the power of AVR might be smaller than the AQ test, 10 we also perform the AQ test. In our case, similar¯ndings are obtained.…”