Abstract:
The presence of autocorrelated financial returns has major implications for investment decisions. Unsurprisingly, therefore, numerous studies have sought to shed light on whether returns are autocorrelated or not, to what extent, and when. Standard tests for autocorrelation rely on the assumption of strict stationarity of returns, possibly after a suitable transformation. Recent studies, however, reveal that intraday financial returns are often characterised by a subtle form of non-stationarity that cannot be transformed away, namely non-stationary periodicity in the zero-process. Here, we propose tests for autocorrelation that are valid under this (and other forms) of non-stationarity. The tests are simple to implement, and well-sized and powerful as documented in our Monte Carlo simulations. Next, in a study of the intraday returns of stocks and exchange rates, our robust tests document that returns are rarely autocorrelated. This is in sharp contrast to the standard benchmark test, which spuriously detects a substantial number of autocorrelations. Moreover, stability analyses with our robust tests suggest the significance of the autocorrelations is short-lived and very erratic. So it is unclear whether the short-lived autocorrelations can be used to inform decision-making.