We worked with the Massachusetts Department of Economics Research to devise an accurate forecasting model of the Massachusetts unemployment rate one month, two months, and three months in the future. Our input variables included joblessness claims, national macroeconomic sentiment, industry labor flows, and, importantly, Google Trends search indices.

We found that the optimal forecasting algorithm was a Random Forest model that included the Google Trends data, suggesting that although ARIMA remains the gold-standard econometric method for top-line labor prediction, incorporating basic machine learning architectures — trained on non-traditional data, such as Google Trends — can further enhance performance.