Unemployment Rate Estimation with Administrative Big Data: Case Study in a City of 4-Million Population
Abstract
Among one of the major indicators of the economy, the unemployment rate figures yet confront lots of skepticism in China. The register-based urban unemployment rate has been stable at around 4% with very low fluctuations over time, while the survey-based unemployment rate is calculated from a relatively small sample which is difficult to deliver accurate provincial or prefecture-level estimates. In this paper, we address these challenges by combining machine learning algorithms with traditional national account systems to estimate the unemployment rate. Essentially, we train a machine learning model to predict an individual’s monthly employment status based on administrative big data in a city with a four million population and estimate the unemployment rate of the city with a national account method. At the individual level, our model achieves an accuracy of 96.7% on the out-of-sample test set. Summing up in the national account system, the estimated local unemployment rate fluctuates in a reasonable range and exhibits periodic characteristics. Our estimates provide a better economic indicator than the register-based unemployment rate released once every year. We also study the gender, educational level, and the pattern of reemployment of the local unemployed population with individual-level data. Our paper proposes a new approach of using administrative big data to understand the economic conditions and to facilitate policy-making in the age of big data.
