In this paper the impact of Hodrick-Prescott (HP) filter on Least Squares Estimation is analyzed. It is shown that HP filter is able to produce spurious autocorrelation of error term and thus change the distribution of t-statistics. The danger consequences may accur not only when a unit root is present in the data but also when the autoregressive root is close to one. In empirical analyses it is thus preferable to use the differencing and methods in conformity with the stochastic properties of data generating process. Moreover, the paper postulates not to use HP filter when it is possible that there exists significant autocorrelation in the series.