Forecasting Economic Cycles with Time Series PLS-SEM: Evaluating Reflective vs. Formative Specification of PCA-Derived Indicators

This research examined whether indicators derived from Principal components analysis (PCA) should be modeled as reflective or formative composite constructs in forecasting economic cycles within an Economic early warning model (EWM). The study employed Partial least squares structural equation modeling (PLS-SEM), incorporating Confirmatory tetrad analysis in partial least squares (CTA-PLS), Confirmatory composite analysis (CCA), and was further validated using out-of-sample forecasting performance. Quarterly Thai economic data were collected, in which the in-sample period spanned 2012Q3 to 2022Q4, aligning leading indicators and forecast targets for model development, with out-of-sample evaluation for the final endogenous construct from 2023Q1 to 2024Q3. Results indicate that PCA-derived indicators are better specified as reflective constructs. CTA-PLS did not reject the null hypothesis of reflectiveness, while formative misspecification led to poor performance in the CCA process, resulting in incorrect indicator deletion and reduced forecasting accuracy. These findings underscore the importance of proper model specification for statistical validity and practical forecasting effectiveness, contributing to economic cycle forecasting methodology by integrating composite indicator structure evaluation and bootstrap-enhanced inference within a structural forecasting framework.