Do decision-makers suffer from the sunk cost fallacy in high-stakes situations?  We develop a behavioral model of usage of a durable good with mental accounting for sunk costs.  It predicts that the usage increases with the sunk cost, and attenuates with time at a rate that increases with the sunk cost.  The model nests conventionally rational behavior as a special case. 

We take the model to a panel of 6,474 cars between 2001-2011 in Singapore.  During that period, the sunk cost involved in a new car purchase varied substantially with the continuing government policy.  We found robust evidence of a sunk cost fallacy.  The elasticity of usage with respect to the sunk cost was 0.563(±0.072). An increase in the sunk cost by $4,500 (the outcome of government policy between 2009 and 2010) would have been associated with an increase in monthly usage by 147 kilometers or 8.8%.  Our results were robust to various checks including alternative controls for selection, differences in specification, and allowing for heterogeneity in engine size and target cumulative usage.