A Bayesian estimation approach of random switching exponential smoothing with application to credit forecast
Exponential smoothing techniques are pivotal in forecasting within economic, financial, and operational management domains. The evolution from Holt's initial model to the Single Source of Error (SSOE) and subsequently the Multiple Source of Error (MSOE) frameworks reflects significant advancements in handling the dynamic aspects of time series data. Our research focuses on the MSOE model, specifically its application through Random Coefficient Markov Chain Monte Carlo (RC-MCMC) methods. This method leverages banded precision matrices to enhance the estimation efficiency of model parameters. Our simulations, alongside empirical applications using quarterly credit-to-GDP data from the Bank for International Settlements, demonstrate the RC-MCMC's superior accuracy in parameter estimation compared to the direct RC-SSPACE method. This study underscores the RC-MCMC's practical relevance and robustness in economic time series analysis.
Jun 28, 2024
The Impact of Fintech on the Operational Efficiency of Commercial Banks in China
The Fintech evolution has reshaped commercial banks in China, necessitating a study of its influence on their operational efficiency. Using data from 57 Chinese commercial banks between 2011 and 2020, operational efficiency was gauged via Data Envelopment Analysis (DEA) and further decomposed with the Malmquist index method. A Generalized Moment Estimation model (GMM) assessed Fintech's impact on efficiency. Findings indicated that technological advancements primarily boost bank efficiency. While Fintech aids in enhancing efficiency, its assimilation varies across different bank types.
May 26, 2023
Learning Performance of Nonlinear Classification Models Based on Markov Sampling
Nonlinear classification models excel in handling complex problems and are widely used across various fields. This paper explores the learning performance of nonlinear classification models using Markov sampling, an extension of the traditional i.i.d. sample framework. It introduces a ueMC-NL algorithm specifically designed for these models to facilitate the generation of ueMC samples from a finite dataset. Numerical analyses of the random forest and MLP model show that nonlinear classification models utilizing ueMC samples achieve lower misclassification rates compared to those using i.i.d. samples.
Jan 1, 0001
A Dynamic Programming Algorithm for MSM Based on Markov Loss Function
In this paper, we propose a dynamic programming algorithm based on Markov loss for the state sequence prediction problem of Markov regime switching model. This algorithm aims to improve the accuracy of state sequence prediction by comprehensively considering global and marginal loss. Through numerical simulation experiments and empirical research, the predictive performance of the Viterbi algorithm, marginal loss and Markov loss are evaluated within the MS-AR(1). The derived results exhibit that the algorithm holds considerable superiority in forecasting real state sequences and effectively mitigates state error transition rates.
Jan 1, 0001