Short-term Gini coefficient estimation using nonlinear autoregressive multilayer perceptron model.

Area of Science:

• Socioeconomics
• Econometrics
• Computational Social Science

Background:

• Poverty is a complex global issue linked to economic, political, and social factors, requiring collective action like the UN's Sustainable Development Goals.
• The Gini coefficient measures income inequality, a critical indicator often correlated with poverty rates. Traditional annual calculations face limitations due to logistical challenges and slow data transformation.

Purpose of the Study:

• To address limitations in traditional Gini coefficient computation by developing a method for short-term forecasting.
• To provide a tool for instantaneous understanding of income inequality shifts, especially during rapid economic transitions like those in the gig economy.

Main Methods:

• Utilized System Identification (SI) principles, specifically the Nonlinear Auto-Regressive (NAR) model, enhanced with Multi-Layer Perceptron (MLP).
• Tested various parameters including output lag space, hidden units, and initial random seeds to optimize the model for Malaysia's Gini coefficient (1987-2015).
• Validated the model using One-Step-Ahead (OSA) prediction, residual correlation analysis, and residual histograms.

Main Results:

• The NAR-MLP model demonstrated high efficacy in estimating Malaysia's Gini coefficient over a 28-year period.
• Achieved a superior model fit with a Mean Squared Error (MSE) of 1.14 × 10^-7.
• Confirmed model validity through uncorrelated residuals, indicating reliability for short-term forecasting.

Conclusions:

• The proposed NAR-MLP approach is a valid and useful tool for predicting short-term variations in the Gini coefficient.
• This method offers a significant improvement over traditional manual approaches by enabling predictions in much smaller time steps.
• The study substantiates the model's capability to capture dynamic changes in income inequality, aiding in timely policy interventions.