Work in Progress! In setting rates, pricing actuaries must forecast future frequency and severity levels. This is often done using simple and often inaccurate forecasting procedures such as the drift method. Can recent advances in time series forecasting using deep learning be leveraged to improve actuarial loss trend selections?

To answer this question, I focus on the frequency of fatal accidents in the US. I derive my dataset from the Federal Highway Administration, NHTSA, BLS, and the Census Bureau.