Applying Time Series Decomposition to Construct Index-Tracking Portfolio

This study proposes a new method for creating an index-tracking portfolio using time series decomposition. First, we construct index-tracking portfolios of stocks chosen because their price movements mimic that of the Dow-Jones Industrial Average. Our method utilizes similarities of constituent stocks to the benchmark that are assessed by distances of time series trends derived from decomposing original series. Although the portfolios chosen by our method reasonably tracked the performance of the benchmark, they did not surpass the clustering approach discussed in earlier studies. Therefore, we examined what causes tracking error and found that two causes for deficiencies in our similarity-based method, which are unintended irregular movements of holding stocks and highly correlated relationships within stocks in the portfolio. To overcome them and to improve tracking performance, we propose a similarity-balanced approach that is another index-tracking method with alternate use of similarity. Doing so improved the tracking performance by avoiding the problem of high correlation among the stocks chosen under the initial method.