ADAPTIVE MULTIRESOLUTION SEMIPARAMETRIC MODEL INTEGRATING TRUNCATED SPLINE AND WAVELET TO BALINESE VILLAGE CREDIT INSTITUTIONS (LPD)

Abstrak

In this research, an adaptive multiresolution semiparametric regression approach is proposed to examine the financial performance of Lembaga Perkreditan Desa (LPD) in Bali, Indonesia. In the research, the problem that arises from the inadequacy of traditional regression techniques in accounting for the nonlinear pattern in the data and the financial instability is overcome by introducing the truncated spline and wavelet components into the semiparametric regression analysis.This research utilizes a quantitative method based on secondary financial information collected for 50 LPDs between 2015 and 2024, providing around 500 observations. In order to obtain a more consistent dataset, the purposive sampling method will be used. Return on Assets (ROA) is chosen as the dependent variable, and explanatory variables include interest rates, the number of customers, capital adequacy ratio (CAR), total assets, and non performing loans (NPL). The model will be estimated using the penalized least squares with iterative backfitting estimation technique and will be assessed using the RMSE, MAE, and R² criteria based on Kfold cross validation. As it can be seen from the results, the hybrid model significantly improves the predictive power of traditional linear and spline methods, providing smaller error rates and better fit quality. Spline functions help to determine long term trends, whereas wavelets capture short term effects. This shows that multiresolution modeling increases predictability and interpretability. The model has practical utility for financial management and regulation through adaptive risk management and decision-making processes in microfinance firms.

Kata kunci

• Adaptive Multiresolution
• Balinese Village Credit Institutions
• Penalized Estimation
• Truncated Spline
• Wavelet