In this paper, we demonstrate that using finite sample correction bootstrapping techniques is advisable in samples that cover less than two complete business cycles, even when high-frequency data seemingly provide a sufficient number of observations to overcome the small sample bias. This is particularly relevant in the current research environment. Because the recent financial crisis is considered as a structural break, research on current problems is often conducted using post-crisis data. That is, the available samples cover only a few years of data, often spanning only one business cycle or even less. We provide ample simulation-based evidence that samples of daily or monthly dynamic data covering periods of this magnitude are prone to a fairly substantial bias. Moreover, we are able to show that standard bootstrap-based bias correction techniques still work in those cases.
El-Shagi, M.: Dealing with Small Sample Bias in Post-Crisis Samples, Economic Modelling, Vol. 65 (September 2017), pp. 1-8