Mardika, Zulfa Wahyu (2026) Quantile Regression Neural Network untuk Pemodelan Conditional Value-at-Risk Berbasis Quantile Regression dan Quantile Autoregressive Dengan Stochastic Search Variable Selection pada Emiten di LQ45. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6003241026-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6003241026-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy |
Abstract
Fluktuasi harga saham pada kondisi pasar yang bergejolak dapat meningkatkan risiko kerugian ekstrem dan memicu penularan risiko antar emiten. Pada kelompok saham berlikuiditas tinggi seperti LQ45, pengukuran risiko secara individual menggunakan Value-at-Risk (VaR) belum cukup untuk menjelaskan perubahan risiko suatu emiten ketika emiten lain berada dalam kondisi distress. Oleh karena itu, penelitian ini menggunakan Conditional Value-at-Risk (CoVaR) untuk mengukur risiko ekstrem suatu emiten secara bersyarat terhadap kondisi ekstrem emiten lain. Penelitian ini memodelkan CoVaR pada emiten LQ45 berbasis Quantile Regression (QR) yang direalisasikan melalui Quantile Regression Neural Network (QRNN), dengan komponen VaR berbasis Quantile Autoregressive (QAR), serta menerapkan Stochastic Search Variable Selection (SSVS) untuk memilih variabel input yang paling informatif. Data yang digunakan berupa log return harian 24 emiten LQ45 periode 21 September 2020 hingga 30 Januari 2026, dengan kandidat input berupa VaR-QAR emiten lain, return IHSG, return USD/IDR, return emas, net foreign flow, dan volatilitas historis. VaR diestimasi secara dinamis menggunakan QAR pada kuantil 5% dan 1% sebagai representasi kondisi distress, kemudian CoVaR diestimasi menggunakan QRNN untuk menangkap hubungan kuantil yang berpotensi nonlinear. Hasil penelitian menunjukkan bahwa SSVS mampu menyederhanakan struktur input dari 28 kandidat menjadi rata-rata 2,6 input, dengan IHSG sebagai faktor sistemik paling dominan. Evaluasi menggunakan backtesting dan check loss menunjukkan bahwa model CoVaR berbasis QRNN dengan SSVS menghasilkan model yang lebih ringkas dan efisien dengan kalibrasi pelanggaran yang lebih memadai dan akurasi check yang setara dibandingkan dengan model CoVaR berbasis QRNN tanpa SSVS.
=====================================================================================================================================
Stock price fluctuations under turbulent market conditions can heighten the risk of extreme losses and trigger risk contagion among firms. For highly liquid stock groups such as those in the LQ45 index, measuring risk individually using Value-at-Risk (VaR) is insufficient to explain how a firm’s risk changes when other firms are in distress. Therefore, this study employs Conditional Value-at-Risk (CoVaR) to measure a firm’s extreme risk conditional on the extreme conditions of other firms. This study models the CoVaR of LQ45 firms based on Quantile Regression (QR) realized through a Quantile Regression Neural Network (QRNN), with a VaR component based on Quantile Autoregression (QAR), and applies Stochastic Search Variable Selection (SSVS) to select the most informative input variables. The data consist of daily log returns of 24 LQ45 firms over the period from 21 September 2020 to 30 January 2026, with candidate inputs comprising the QAR-based VaR of other firms; the returns of the IHSG, USD/IDR, and gold; net foreign flow; and historical volatility. VaR is estimated dynamically using QAR at the 5% and 1% quantiles to represent distress conditions, after which CoVaR is estimated using QRNN to capture potentially nonlinear quantile relationships. The results show that SSVS is able to simplify the input structure from 28 candidates to an average of 2.6 inputs, with the IHSG return as the most dominant systemic factor. Evaluation using backtesting and the check loss indicates that the CoVaR model based on QRNN with SSVS yields a more parsimonious and efficient model, with more adequate violation calibration and comparable check accuracy relative to the CoVaR model based on QRNN without SSVS.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Risiko Sistemik, Conditional Value-at-Risk (CoVaR), Quantile Regression Neural Network (QRNN), Quantile Autoregressive (QAR), Stochastic Search Variable Selection (SSVS), Systemic Risk |
| Subjects: | Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression Q Science > QA Mathematics > QA279.5 Bayesian statistical decision theory. |
| Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis |
| Depositing User: | Zulfa Wahyu Mardika |
| Date Deposited: | 04 Aug 2026 09:31 |
| Last Modified: | 04 Aug 2026 09:31 |
| URI: | http://repository.its.ac.id/id/eprint/143473 |
Actions (login required)
 |
View Item |