Abstract A report by the Council of Economic Advisers (1997) is the first of a group of studies, known as caseload studies, analysing the relationship between the US unemployment rate and the welfare participation rate, with special regard to the 1990s. We examine this relationship in a structural VAR model over the period of 1960–2000 and find that the unemployment rate does not help to predict the welfare participation rate while the converse is more likely to hold. These results are robust to state and year heterogeneity over a period of unprecedented positive correlation between unemployment and welfare participation, i.e. 1990–1998. Further, we find that a shock to the welfare participation rate has a contemporaneous impact on the unemployment rate while the converse is less likely to hold. The main conclusion is that several caseload studies may be based on the wrong assumption that the unemployment rate is an exogenous explanatory variable of the welfare participation rate.