Pengembangan Financial Risk Meter untuk Perbankan di Indonesia dengan Model LASSO Quantile Regression (LASSO-QR) dan LASSO Quantile Regression Neural Network (LASSO-QR-NN)

Khoirunissa, Husna Afanyn (2023) Pengembangan Financial Risk Meter untuk Perbankan di Indonesia dengan Model LASSO Quantile Regression (LASSO-QR) dan LASSO Quantile Regression Neural Network (LASSO-QR-NN). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003211007-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (3MB) | Request a copy

Abstract

Risiko sistemik merupakan faktor yang sangat menentukan pembangunan stabilitas sistem keuangan di suatu negara. Kegagalan satu perusahaan dapat berdampak pada perusahaan lain dengan efek domino yang menimbulkan risiko sistemik terhadap sistem pasar keuangan. Dalam melakukan pengukuran risiko sistemik, penelitian ini menggunakan metode quantile regression (QR) dimana log return satu perusahaan keuangan menjadi variabel respon dan log return perusahaan keuangan lainnya menjadi variabel prediktor pada nilai kuantil tertentu sehingga terdapat model QR sebanyak perusahaan keuangan. Nilai log return dari perusahaan-perusahaan yang digunakan masing-masing memiliki banyak prediktor sehingga setiap model memiliki dimensi yang tinggi sehingga diperlukan seleksi variabel. Metode seleksi variabel yang digunakan adalah metode LASSO. Rangkaian deret rata-rata parameter penalti LASSO dari setiap perusahaan yang diamati memiliki bentuk dan volatilitas sesuai dengan volatilitas pasar dan peristiwa keuangan dengan dampak besar pada risiko sistemik. Rangkaian rata-rata parameter penalti tersebut digunakan sebagai ukuran risiko sistemik yang disebut dengan Financial Risk Meter (FRM). Untuk mengakomodasi adanya hubungan non-linier antar risiko perusahaan, dalam penelitian ini, dikembangkan model quantile regression neural network (QR-NN) dengan menambahkan parameter penalti pada fungsi tujuan sehingga didapatkan solusi untuk model LASSO-QR-NN. Aplikasi model LASSO-QR dan LASSO-QR-NN pada data nilai log return perusahaan-perusahaan sektor perbankan di Indonesia beserta variabel makroekonomi yang turut menjadi variabel prediktor. Pemodelan dengan pendekatan moving window pada model LASSO-QR menghasilkan FRM tinggi pada pertengahan 2020 dan kuartal pertama 2021 sedangkan pada model LASSO-QR-NN menghasilkan FRM tinggi pada pertengahan 2020, kuartal pertama 2021, dan kuartal ketiga 2021. Klasifikasi tingkat risiko pada hari terakhir dataset, yaitu 30 Desember 2022, disimpulkan tingkat risiko rendah krisis dengan model LASSO-QR dan tingkat risiko krisis agak lebih tinggi dari biasanya dengan model LASSO-QR-NN. Model LASSO-QR-NN lebih banyak dalam menangkap krisis meskipun memiliki nilai jangkauan yang lebih kecil daripada model LASSO-QR.
======================================================================================================================================
Systemic risk is a factor that highly determines the development of financial system stability in a country. The failure of one company can impact other companies with a domino effect that causes the systemic risk to the financial market system. In measuring systemic risk, this study uses the quantile regression (QR) method, where the log return of one financial company becomes a response variable, and the log return of other financial companies becomes predictor variables at a specific quantile value. Hence, there are as many QR models as financial companies. The log return values of the companies used each have many predictors, so each model has a high dimension, then the variable selection is required. The variable selection method used is the LASSO method. The series of the average LASSO penalty parameters of each observed company have the shape and volatility according to market volatility and financial events with impact on systemic risk. The series of the average penalty parameters is used as a measure of systemic risk called the Financial Risk Meter (FRM). To accommodate the non-linear relationship between firm risks, in this study, a quantile regression neural network (QR-NN) model was developed by adding a penalty parameter to the objective function to obtain a solution for the LASSO-QR-NN model. Application of the LASSO-QR and LASSO-QR-NN models to data on log return values of banking sector companies in Indonesia along with macroeconomic variables that also become predictor variables. Modeling with the moving window approach produces a high FRM LASSO-QR model in mid-2020 and the first quarter of 2021, while the LASSO-QR-NN model is in mid-2020, first quarter 2021, and third quarter 2021. Classification of risk level on the last day of the dataset, 30 December 2022, it is concluded that the level of crisis risk is low with the LASSO-QR model and the level of crisis risk is somewhat higher than usual with the LASSO-QR-NN model. The LASSO-QR-NN model is more capable of capturing crises even though it has a smaller range value than the LASSO-QR model.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Financial Risk Meter, LASSO, quantile regression, neural network, Financial Risk Meter, LASSO, quantile regression, neural network
Subjects:	H Social Sciences > HA Statistics > HA31.3 Regression. Correlation H Social Sciences > HA Statistics > HA31.7 Estimation H Social Sciences > HG Finance Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Husna Afanyn Khoirunissa
Date Deposited:	06 Aug 2023 07:04
Last Modified:	06 Aug 2023 07:04
URI:	http://repository.its.ac.id/id/eprint/103957

Actions (login required)

View Item