Interpolation and correlation

Abstract

Historical time series sometimes have missing observations. It is common practice either to ignore these missing values or otherwise to interpolate between the adjacent observations and continue with the interpolated data as true data. This paper shows that interpolation changes the autocorrelation structure of the time series. Ignoring such autocorrelation in subsequent correlation or regression analysis can lead to spurious results. A simple method is presented to prevent spurious results. A detailed illustration highlights the main issues.