Abstract:
|
Financial literature has documented that long-term asset return exhibits predictability if regressing aggregated-forward returns on aggregated backward volatility. This return predictability increases with the time horizon. We investigate this phenomenon by proposing an asset price model based on a pure-jump point process for tick-level prices. The model generates calendar-time return and volatility directly from the tick-level data and is able to capture stylized facts such as long memory in volatility of returns. Under the framework of the proposed model, we show that if the return process has long memory, there is return predictability over the long-term horizon; while if it is a short memory process, then there is no return predictability. In addition, we propose a new hypothesis test to detect the long-term return predictability in empirical data. We show that the proposed test has a higher power than the tests proposed in other literature.
|