This paper estimates a nonlinear Threshold-VAR to investigate if a Keyne- sian liquidity trap due to a speculative motive was in place in the U.S. Great Depression and the recent Great Recession. We find clear evidence in favor of a breakdown of the liquidity effect after an unexpected increase in M2 in the 1921- 1940 period. Importantly, this evidence, which is consistent with the Keynesian view on a liquidity trap, is shown to be state contingent. In particular, it emerges only when a speculative regime identified by high realizations of the Dow Jones index is considered. A standard linear framework is shown to be well-suited to test the hypothesis of a Keynesian liquidity trap. An investigation performed with the same data for the period 1991-2010 confirms the presence of a liquidity trap just in the speculative regime. This last result emerges significantly only when we consider the federal funds rate as the policy instrument and we model the Divisia M2 measure of liquidity.