| Penulis/Author |
Vemmie Nastiti Lestari, S.Si., M.Sc. (1); Prof. Dr. Abdurakhman, S.Si., M.Si. (2); Prof. Dr.rer.nat. Dedi Rosadi, S.Si.,, M.Sc. (3) |
| Abstrak/Abstract |
As investors select stocks more efficiently to form portfolios that meet their objectives, they should be grouped based on their time-varying risk levels. One of the clustering approaches is time series clustering with a model-based approach using hierarchical and K-means clustering algorithms. The distance calculation is based on the estimated parameters of the model used; in this case, the GARCH (1,1) model is used. This paper proposes a modified Piccolo distance that uses the absolute value between two GARCH (1,1) models, which is a development of the Manhattan distance. The modified Piccolo distance improves robustness to outliers and simplifies calculations, resulting in more accurate and efficient time series cluster analysis. Applying hierarchical and K-means clustering with modified Piccolo distance will be compared with other model-based distance modifications for clustering applied to simulated data and case studies using stock data incorporated in the Indonesia Stock Exchange. A measure of cluster validity is calculated using the C index. From the simulated data and case studies, it is found that clustering with Piccolo distance modification and other distance modifications between two GARCH (1,1) models produce clusters with a small C index, both for simulated data and case studies. A small C index value in the clustering results indicates good clustering quality, where the clusters formed have high similarity and are well separated from others. Furthermore, the clusters formed will be considered in making a good portfolio, so it is expected to reduce the risk in the stock portfolio. |
