Each year millions of low-to-moderate income (LMI) households receive tax refunds due to an array of refundable tax credits and over withholding. Using data from a sample of two large free public volunteer income tax assistance (VITA) sites, this paper explores the expectations those LMI taxpayers have about their refunds amounts as they enter the income tax filing process, and compare that to actual their refunds calculated shortly later. On average people slightly underestimated the size of their refunds. This suggests many receive a surplus over their initial expectations. This surplus is not mentally mapped or designated for consumption or bill payments and presents an opportunity to encourage savings.

We explore how refund underestimates are related to the taxpayer's willingness to commit to one of the savings products (US Savings Bond, CD, Savings Account) offered at the tax site after the actual refund is known. Higher rates of savings products is consistent with the notion that the surplus over expectations is treated differently and that people are more willing to save the more their refund exceeds expectations.

Clients show other aspects of mental accounting. Clients were asked at the onset of the tax preparation process to identify their plans for the expected refund in one (or more) of three categories: to pay off bills, for meeting current expenses, and for savings for the future. In addition, filers were likewise asked to respond "are you willing to save some part of your refund this year?''. Those items had a significant effect on later savings choices. As such, the mental accounting process of labeling the tax refund for savings purposes becomes the main driver on their propensity to save (spend) in case of a windfall (shortage). These results were robust to site and day fixed effects as well as controls for marketing of savings products at the VITA sites.

The policy implications of this work are important for policymakers and program managers to understand. Lump sum payments generated by the tax code or other programs serve to form expectations among people. These expectations can generate incentives to spend or save. The potential of ''found money'' to be saved can be an important strategy for asset building programs. More research is needed to understand how expectations are formed, and how systematic differences in expectations can be used to enhance savings - or debt repayment - by LMI households. While underestimation may offer opportunities to save, as expectations are formed over time we may discover more overestimation, which could be problematic. Overestimation might result in people borrowing in advance of the expected payment only to receive less than expected and an unfunded liability.  Overestimation might signal a positive judgment bias towards other uncertain future incomes as well.

We conclude with a discussion of how over and under estimation may influence behavior, and how VITA and asset building programs can better use expectations to promote positive financial behaviors.