Tracing Unemployment Rate of South Africa during the COVID-19 Pandemic Using Twitter Data (Preprint)

BACKGROUND Global economy has been hardly hit by the COVID-19 pandemic. Many countries are experiencing a severe and destructive recession. Unemployment rate is very important to policy makers as it provide a key indicator of overall labour market and wider economic conditions. Despite its relevance, there is usually a delay in the availability of the indicator as it is traditionally based on a survey of households over several months. The speed at which the economy in most countries decline at the onset of COVID-19 highlights the importance of timely information about the labour market during the onset of a recession. In the coming year, there will be uncertainty about the timing and extent of any improvement in labour market outcomes that will also highlight the value of timely information. OBJECTIVE The main goal of this study is to provide policy- and decision-makers with additional and real-time information about the labor market flow during a prolonged pandemic. The first objective of the study is to find the missing unemployment rates in cases where census measurements are incomplete. The second objective is to estimate the unemployment rate in real-time since it usually takes months for formal unemployment data to be published. In this paper, we use social media data, particularly, Twitter to trace and nowcast the unemployment rate of South Africa during the COVID-19 pandemic. METHODS Unemployment rate in South Africa is estimated quarterly. We first used Google mobility index to interpolate it and find the monthly values. Next, we created a dataset of unemployment related tweets in South Africa using certain keywords such as employed, unemployed, and retrench. Principal Component Regression (PCR) was applied to estimate the unemployment rate using the tweets and their sentiment scores. RESULTS Numerical results indicate that the number of tweets is highly correlated with the unemployment rate during and before the COVID-19 pandemic. In addition, the trend of the normalized sum of the sentiment scores of the tweets is negatively correlated with the unemployment rate of South Africa. Moreover, the estimated unemployment rate using PCR is highly correlated with the actual unemployment rate of South Africa and has a low Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). CONCLUSIONS The results of this study show that social media information can be used to reasonably estimate one of the key labor market indicators, especially during disaster events such as a prolonged pandemic. This information can be used to rapidly understand and manage the impacts of the pandemic on the economy and people’s life.