Analysis of the Temporal Structure in Economic Condition Assessments

The Economy Watcher Survey, a market survey published by the Japanese government, features assessments of current and future economic conditions made by individuals from various occupations. While this survey offers crucial insights for economic policy formulation, the specific timing of the events that form the basis for these assessments remains unclear. For instance, one individual might assess future economic conditions based on events that have already occurred, whereas another might base their judgment on potential future events. Understanding the temporal context in which people base their assessments is vital for policymakers. Motivated by this need, we aim to clarify the different points in time that underpin assessments of future economic conditions. In our research, we categorize assessments of future economic conditions into assessments based on current and future events, employing learning from positive and unlabeled data (PU learning). Given that the dataset spans multiple periods, we introduce a novel architecture that enables neural networks to perform PU learning efficiently, drawing on principles of multi-task learning. Our empirical analysis validates that our method effectively classifies these assessments, and we leverage the classification outcomes to derive insights for policy-making.