We study whether aggregation residuals in U.S. private investment in information technology (IT) exhibit a predictable pattern that is consistent with Hicks' composite-good theorem and that may be used for forecasting. To determine whether one can extract such a pattern, we apply the general-to-specific strategy developed by Krolzig and Hendry (2001). This strategy combines ordinary least squares with a computer-automated algorithm that selects a specification based on coefficients' statistical significance, residual properties, and parameter constancy. Then, we derive the testable implications from Hicks' theorem and evaluate them with econometric formulations; we find qualified support for these implications. Having obtained these formulations, we evaluate their ex-post predictive accuracy and compare it to that of an autoregressive model. The key finding is that ignoring movement in relative prices results in a loss of information for predicting aggregation residuals.