Imputation Algorithm Based on Copula for Missing Value in Timeseries Data

In this paper, imputation algorithm based on Gaussian copula in time series data is given. The case study is a missing value of 33 years Gross Development Product (GDP) of nine countries (from 1950 to 1983). The missing value was predicted by error model of autoregressive (AR) model assumed following N(mu, sigma(2)) distribution. Since the data is time series and modeled with AR, the recent data is influenced by previous data, added coefficient factor and error. Thus conditional distribution of the measurement at specific time point, which is also conditioned by past measurements, was analyzed. In this research, the conditional distribution, so called joint distribution, was derived by copula. The result shows that the proposed method could predict the missing value with small error.