Assessing the impact of financial asset structure on household consumption decisions in the digital age using neural networks

Abstract

The rapid digitization of finance has transformed household asset structures, introducing new financial instruments such as cryptocurrencies and digital investment platforms. However, the implications of this evolving asset composition on household consumption behavior remain underexplored. Traditional models often fall short in capturing the temporal and nonlinear dynamics of such relationships. Research aims to assess the impact of financial asset structures, including both traditional and digital assets, on household consumption decisions by leveraging advanced neural network architectures. The dataset comprises household-level financial and consumption data drawn from finance surveys, enriched with detailed metrics on digital asset ownership and usage. Key variables include household income, traditional and digital asset composition, demographic attributes, and monthly expenditure patterns. Data pre-processing involved handling missing values and outliers, followed by Z-score standardization to ensure normalization. A stacked Long Short-Term Memory (SLSTM) network was employed to model time-series consumption trends and capture underlying temporal dependencies. The Enriched Satin Bowerbird (ESB) algorithm was used to enhance training efficiency and convergence. The Enriched Satin Bowerbird mutated Stacked LSTM (ESB-SLSTM) model effectively identified complex nonlinear interactions and temporal patterns between asset structures and consumption behavior. Implemented using Python, the proposed ESB-SLTSM model accurately analyzes household consumption based on financial asset structures. It achieves over 93% in F1-score, accuracy, recall, and precision, ensuring robust and insightful behavioral predictions. Results indicate that households with significant digital asset holdings display more dynamic and sensitive consumption responses, particularly in response to market fluctuations. The model achieved superior predictive performance compared to baseline econometric models. These findings offer valuable insights for policymakers and financial planners aiming to promote stable consumption and responsible digital asset integration.