ABSTRACT Macroeconomic forecasting is a critical task for policymakers, financial institutions, and businesses, enabling informed decision‐making and strategic planning. Traditional econometric models predominantly rely on official statistical indicators, which are often published with significant lags and limited frequency, creating a fundamental challenge of data scarcity. The recent proliferation of the Internet of Things (IoT) has opened new avenues for economic measurement by generating massive, high‐frequency data that captures real‐time economic activity. Studies have demonstrated the value of using alternative data streams, such as satellite imagery, search engine queries, and maritime traffic, as proxies for traditional macroeconomic indicators. This paper proposes Meta‐IoTNet, a novel framework that combines IoT data with meta‐learning for few‐shot prediction of macroeconomic indicators. Unlike traditional MAML‐based methods and existing IoT‐based forecasting approaches, Meta‐IoTNet incorporates multi‐scale temporal feature extraction to capture both short‐term fluctuations and long‐term trends in IoT data. Additionally, our framework integrates task‐aware context encoding, which adapts the model to new tasks with minimal data, making it particularly effective in volatile economic conditions. Through extensive experiments, Meta‐IoTNet demonstrates significant improvements over existing models, especially in data‐scarce environments.
Sisheng Wan (Thu,) studied this question.