Scholarlink

<p>This study advances the methodological integration of the stationary bootstrap into the partial least squares structural equation modeling (PLS-SEM) framework for time series analysis to forecast economic cycles. Traditional PLS-SEM typically relies on the standard sampling bootstrap, which assumes independently and identically distributed (IID) observations—an assumption unsuitable for autocorrelated time series commonly observed in macroeconomic contexts. To address this limitation, this research incorporated the stationary bootstrap, a time series resampling technique that preserves temporal dependence, into the PLS-SEM estimation process. Using quarterly data from 2011Q1 to 2024Q2, the study developed an Early warning model (EWM) for Thailand’s economic cycle. The model specified causal paths among lagged Composite leading indicators (CLIs): Monetary condition (MC), Financial cycle (FC), and Economic sentiment (ES). Comparative results showed that the stationary bootstrap yielded more conservative parameter estimates while enhancing model performance in out-of-sample forecasting using an expanding window evaluation. It outperformed the conventional approach by better preserving the temporal structure of the data. This study demonstrates that integrating the stationary bootstrap into the PLS-SEM framework improves inference validity and predictive robustness. It provides a generalizable framework for structural modeling in dynamic macroeconomic settings, with practical relevance for economic monitoring and policy planning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

PLS-SEM and bootstrap for time series to forecast economic cycle

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Jeerawadee Pumjaroen

摘要

This study advances the methodological integration of the stationary bootstrap into the partial least squares structural equation modeling (PLS-SEM) framework for time series analysis to forecast economic cycles. Traditional PLS-SEM typically relies on the standard sampling bootstrap, which assumes independently and identically distributed (IID) observations—an assumption unsuitable for autocorrelated time series commonly observed in macroeconomic contexts. To address this limitation, this research incorporated the stationary bootstrap, a time series resampling technique that preserves temporal dependence, into the PLS-SEM estimation process. Using quarterly data from 2011Q1 to 2024Q2, the study developed an Early warning model (EWM) for Thailand’s economic cycle. The model specified causal paths among lagged Composite leading indicators (CLIs): Monetary condition (MC), Financial cycle (FC), and Economic sentiment (ES). Comparative results showed that the stationary bootstrap yielded more conservative parameter estimates while enhancing model performance in out-of-sample forecasting using an expanding window evaluation. It outperformed the conventional approach by better preserving the temporal structure of the data. This study demonstrates that integrating the stationary bootstrap into the PLS-SEM framework improves inference validity and predictive robustness. It provides a generalizable framework for structural modeling in dynamic macroeconomic settings, with practical relevance for economic monitoring and policy planning.