A simple method for variance shift detection at unknown time points

Financial literature considers volatility as a good proxy for the risk level and thus the crucial parameter in many financial techniques and strategies. As such, the aim of this paper is to analyse the evolution of time series volatility and detect significant long-term variance changes. Building up on the variance ratio detection technique introduced by Tsay (1988), our paper extends it in two ways: first, we propose the computation of a moving variance ratio implemented on a selected part of the series, thus reducing the amount of calculus and increasing the reliability and second, as in reality permanent variance changes are almost inexistent, we proceed to an adjustment on a specified part of the series only after the detected variance change. Our moving variance ratio technique proves its efficiency in detecting variance changes and removing them from the series, both on simulated and real financial data. More specifically, two significant variance changes are detected within the series of the Hang Seng daily log-returns between 1994 and 2007: the first one on August 15, 1997 and can be linked to the Asian financial crisis, and the second one on July 27, 2001 corresponding to the beginning of a high volatility regime in emerging markets following the Internet bubble crash along with the first signs of the financial crisis in Argentina.