This paper argues that the current framework for predictive ability testing is not suitable for real-time forecast selection. Instead, the authors propose a new framework that evaluates the forecasting method, including the model and estimation procedure, rather than just the model. This approach allows for more relevant conclusions for economic forecasters. The new tests are valid under more general data assumptions and can handle both nested and non-nested models. The authors illustrate the usefulness of the tests by comparing three parameter-reduction methods for macroeconomic forecasting: a sequential model selection approach, the "diffusion indexes" approach, and Bayesian shrinkage estimators. The results show that Bayesian shrinkage performs best for real variables, while the sequential approach performs poorly. The paper also discusses the advantages of the conditional approach, including its ability to handle heterogeneity in economic data and estimation uncertainty. The authors conclude that the conditional framework is more realistic and practical for forecast evaluation.This paper argues that the current framework for predictive ability testing is not suitable for real-time forecast selection. Instead, the authors propose a new framework that evaluates the forecasting method, including the model and estimation procedure, rather than just the model. This approach allows for more relevant conclusions for economic forecasters. The new tests are valid under more general data assumptions and can handle both nested and non-nested models. The authors illustrate the usefulness of the tests by comparing three parameter-reduction methods for macroeconomic forecasting: a sequential model selection approach, the "diffusion indexes" approach, and Bayesian shrinkage estimators. The results show that Bayesian shrinkage performs best for real variables, while the sequential approach performs poorly. The paper also discusses the advantages of the conditional approach, including its ability to handle heterogeneity in economic data and estimation uncertainty. The authors conclude that the conditional framework is more realistic and practical for forecast evaluation.