This document discusses predictive aspects of partial least squares (PLS) modeling. It makes three key points: 1. PLS is considered a "causal-predictive" approach that allows for testing of causal hypotheses through path coefficients and loadings, but its coefficients may be biased. 2. PLS is primarily an explanatory statistical model focused on theory testing, while predictive models specifically aim to accurately predict new observations. 3. The document questions whether PLS truly has predictive aspects, as predictive power is not evaluated in software or research. It provides an example showing a neural network model outperforms PLS in predicting holdout data.