EcoSta 2019: Start Registration
View Submission - EcoSta2019
Title: Refined RBFN test for martingale difference hypothesis: Application to predictability of exchange rate returns Authors:  Jinu Lee - King's College London (United Kingdom) [presenting]
Abstract: A regression-typed test for a martingale difference hypothesis (MDH) based on a radial basis function network (RBFN) is revisited. New testing procedures are discussed to construct more rigorous RBFN specifications. A Monte Carlo experiment is conducted to show that the new test has improved finite sample properties in terms of size and power. Further, the proposed method is applied to examine time-varying predictability of major foreign exchange rates from January 1999 to April 2015 with a moving time window of nearly two years at a daily frequency. The empirical results confirm that the exchange rate returns are not consistently unpredictable or weak-form efficient with continued fluctuations. There are statistically significant evidences to more often deviate from the martingale behaviour since the recent global financial crisis. The findings may be in line with an implication of the adaptive market hypothesis due to market condition changes.