We carry out multivariate time series analysis on price indices of individual goods and services collected over the last 35 years in Japan. Adoption of the complex principal component analysis (CPCA) enables us to have a new insight into dynamic correlation structure involved in the price data. The CPCA is based on complexification of real data using the Hilbert transformation; lead-lag relations between individual prices manifest in a form of instantaneous phases of the complex time series. The correlation matrix in the CPCA is purified by adopting the random matrix theory as a null hypothesis for removal of statistical noises. We identify four significant eigenmodes for price movement which are free from seasonal variations. Each of them has different characteristics of dynamical correlations and is shown to be responsive to different economic events. (C) 2015 The Authors. Published by Elsevier B.V.