Perbandingan Kinerja Model VARX, QVARX Dan MO-QRNN Berbasis Studi Simulasi Dan Penerapannya Untuk Pemodelan Dan Peramalan Risiko Makroekonomi

Asmaria, Adelia Nur (2026) Perbandingan Kinerja Model VARX, QVARX Dan MO-QRNN Berbasis Studi Simulasi Dan Penerapannya Untuk Pemodelan Dan Peramalan Risiko Makroekonomi. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003241029-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Inflasi bergerak secara fluktuatif sepanjang tahun dengan rangkaian gejala ekonomi, sosial dan politik. Pergerakan yang terlalu tinggi mengindikasikan kemorosotan kekuatan rupiah dan terlalu rendah adalah sinyal melemahnya perekonomian. Inflasi Indonesia berdasarkan 9 kelompok pengeluaran secara spesifik menjelaskan fluktuasi harga di bidang tertentu. Inflasi perumahan dan pendidikan menjadi salah dua kelompok pengeluaran yang menarik untuk diamati mengingat budaya berkeluarga di Indonesia dan bersiap dengan risiko terburuk. Penelitian ini memodelkan dan meramal makroekonomi kondisional bersyarat pada tingkat kuantil untuk memodelkan risikonya. Vector Autoregressive with Exogeneous (VARX) membangun sistem persamaan untuk meramal variabel endogen dengan melibatkan historinya sendiri dan variabel eksogen. Quantile Vector Autoregressive with Exogeneous (QVARX) dikembangkan untuk menangkap kondisi ekstrim melalui kuantil sehingga dapat digunakan untuk menggambarkan risiko sistemik. Kedua model berhubungan secara linear karena didasari struktur model autoregressif, Multi-Output Quantile Regression Neural Network (MO-QRNN) diusulkan untuk dapat menjelaskan hubungan non-linear melalui pendeketan jaringan saraf tiruan. Hasil studi simulasi dengan data bangkitan, baik linear dan non-linear, menunjukkan model MO-QRNN memiliki performa SMAPE yang lebih unggul dibanding VARX dan QVARX di level kuantil 0,5. Adapun nilai QVSS di level kuantil ekstrim atas 0,95 dan bawah 0,05 menunjukkan nilai yang positif bahwa nilai fungsi cost model MO-QRNN lebih kecil dibanding model QVARX. Hasil studi empiris pemodelan dan peramalan risiko makroekonomi secara simultan, yaitu inflasi kategori perumahan dan pendidikan, menunjukkan hasil akhir yang sama dengan hasil studi simulasi. Bahwa performa model MO-QRNN lebih baik dibanding model VARX dan QVARX pada kedua fase prediksi in-sample dan peramalan out-sample. Pemodelan menggunakan VARX menunjukkan nilai SMAPE yang lebih kecil pada inflasi perumahan sebagaimana hasil uji terasvirtanya.
===================================================================================================================================
Inflation fluctuates throughout the year in response to various economic, social, and political events. Excessively high inflation indicates a deterioration in the purchasing power of the Indonesian rupiah, whereas excessively low inflation signals a weakening economy. Inflation measured across Indonesia’s nine expenditure categories provides a more detailed representation of price movements within specific sectors. Among these categories, housing and education inflation are of particular interest, given Indonesia’s strong family-oriented culture and the need for households to prepare for potential economic risks. This study models and forecasts conditional macroeconomic risk at different quantile levels to capture the distribution of inflation risk. The Vector Autoregressive Model with Exogenous Variables (VARX) constructs a system of equations to forecast endogenous variables based on their historical values and exogenous variables. The Quantile Vector Autoregressive Model with Exogenous Variables (QVARX) extends VARX by incorporating quantile regression, enabling the model to capture extreme economic conditions and thereby represent systemic risk. As both VARX and QVARX rely on a linear autoregressive structure, this study proposes a Multi-Output Quantile Regression Neural Network (MO-QRNN) to capture potentially nonlinear relationships using an artificial neural network framework. Simulation studies based on both linear and nonlinear synthetic data demonstrate that the proposed MO-QRNN consistently achieves superior forecasting performance, as measured by the SMAPE, compared with the VARX and QVARX models at the median quantile (τ = 0.50). Furthermore, positive QVSS values at the upper (τ = 0.95) and lower (τ = 0.05) extreme quantiles indicate that the MO-QRNN yields a lower quantile loss function than the QVARX model. The empirical study on the simultaneous modeling and forecasting of macroeconomic risk, represented by housing and education inflation, produces results consistent with those obtained from the simulation study. Specifically, the MO-QRNN outperforms both the VARX and QVARX models in both in-sample prediction and out-of-sample forecasting. However, the VARX model achieves a lower SMAPE for housing inflation, which is consistent with the results of the Teräsvirta nonlinearity test, suggesting that the housing inflation series exhibits predominantly linear characteristics.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Inflasi, FCI, FSI, Multi-Output Quantile Regression Neural Network, QVSS, SMAPE, Inflation, FCI, FSI, Multi-Output Quantile Regression Neural Network, QVSS, SMAPE
Subjects:	H Social Sciences > HA Statistics H Social Sciences > HA Statistics > HA30.3 Time-series analysis
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Adelia Nur Asmaria
Date Deposited:	13 Aug 2026 00:56
Last Modified:	13 Aug 2026 00:56
URI:	http://repository.its.ac.id/id/eprint/144333

Actions (login required)

View Item