A Study on the Prediction of KOSDAQ Index by Comparing Time Series Analysis Models

Abstract

The purpose of this study is to suggest a KOSDAQ index prediction model by estimating and comparing the transfer function model and the multiplicative seasonal ARIMA model, which arethe time-series models. Major findings are summarized as follows.The result of data transformation showed stationarityafter the ADF test.Analysis of sample cross-correlation function (SCCF) in estimating the transfer function model showed that linear dependency relationship exists. As the result of analyzing the goodness of fit of the transfer function model using the impact response weights and the noise time series model, the white noise process was observed in the residual time series, and between the residual time series and the producer price index (PPI).In the multiplicative seasonal ARIMA model estimation, only 5 of the 9 candidate models followed the white noise process. As the result of selecting the model by comparing the values of AIC statistics and SBC statistics among five models and analyzing the goodness of fit, the residual time series follow the white noise process.The comparison of AIC statistics and SBC statistics of fitted models of the two models showed that the goodness of model fit of the multiplicative seasonal ARIMA prediction model was better with AIC=345.5553 and SBC=351.3057.Therefore, the KOSDAQ index prediction value and the predicted interval with 95%` confidence level of the multiplicative seasonal ARIMA prediction modelwere presented by inverting the square root transformed value to the original value, and as the result, the KOSDAQ index was expected to rise sharply in April and May 2020 compared to 2019.