Modeling the haemodynamic response in fMRI is still subject of research. Recent developments in accelerated MRI enabled subsecond whole-brain fMRI. In this work we aim to examine the dependency of BOLD signal analysis on the temporal sampling rate and the applied HRF model by numerically undersampling a fast acquisition. Standard GLM fits with different HRF models were performed, followed by group-level statistical analyses and goodness-of-fit computations. The results indicate that variance across HRF models is increased in the undersampled datasets, while the low-pass-filtered undersampling also shows a possibility to enhance the statistical power on the group level.