The Retirement Spending Smile Revisited: Cross‐Sectional Patterns Versus Within‐Household Dynamics

The “retirement spending smile” (Blanchett, 2014), a U‐shaped pattern in the rate of retiree spending change, has influenced how advisors project retirement spending. We reconstruct Blanchett's analysis using RAND HRS/CAMS data (2001–2009) and extend it with panel methods through 2021. The smile pattern appears when comparing different households at different ages, but is not statistically detectable within households under Blanchett's specification, which conditions on the starting spending level. In our reconstruction, coefficient signs match Blanchett's specification, but 95% bootstrap confidence intervals for Age 2 and Age include zero, so the curvature is not statistically distinguishable from a simple linear trend. The attenuation also holds under survey weighting. We further document that the ln(Spending) coefficient sign depends on whether spending is measured at interval start or end, an implementation choice unspecified in the original. For individual client projections, a constant real decline (~1%, sensitivity 0%–2% annually) is a more defensible baseline than assuming a smile. More broadly, a pattern in pooled data should not be read as the path an individual household follows. Spending assumptions should be anchored to each client's realized spending and updated over time.