This study analyzes whether adding one of three broad Divisia money measures (Divisia M3, M4, or M4-) constructed by the Center for Financial Stability (CFS) and a determinant of their long-run demand—stock mutual fund loads—improves deep learning model forecasts of nominal GDP. We train a long short-term memory (LSTM) neural network on data from 1985q1 to 2013q4, using 8 datasets: a baseline pair of 1,500 FRED variables and a version adding three COVID-19 variables, and three other pairs that add stock mutual fund loads plus one of the three broad Divisia money measures. Adding any one of the Divisia variables plus stock mutual fund loads significantly improved one-quarter- and four-quarter-ahead forecasts of nominal GDP, particularly in capturing long-run trends. Findings imply that broad Divisia money should be among variables used to monitor and forecast nominal GDP, with results slightly favoring adding the broadest measure, Divisia M4, for forecasting nominal GDP four quarters ahead.