Despite extensive forecasts of the U.S. unemployment rate, state-level forecasts are quite sparse. This study aims to improve the forecasting performance of simpler models at the state-level. I examine the unemployment rate for the four largest U.S. states by population: California, Florida, New York, and Texas. Overall, I compare the performance of two different models: 1) a simple ARIMA(1,1,0) model and 2) a baseline model using an AR(1) term, a lag of the unemployment rate and several exogenous variables or indicators such as initial claims for unemployment insurance (UI), hiring plans, job openings and the U.S. consumer confidence index. I find that initial claims for UI, job openings from the Job Openings Labor Turnover Survey (JOLTS) and new housing building permits are best suited to forecast the unemployment rate, particularly at the state-level. I observe that the baseline monthly model outperforms the ARIMA(1,1,0) model for three out of the four states, California, Florida and Texas. I additionally examine quarterly data and utilize the Survey of Professional Forecasters (SPF) forecast estimates for 1-quarter ahead to improve my baseline model.