To understand decisions by poor households to save in formal banks, and the impacts thereof, we partnered with a formal bank and randomly assigned a variety of treatments to rural households in Mozambique. Formal savings appear to be a normal good: a treatment that persistently raises household income (an agricultural input subsidy) has sustained positive impacts on formal savings. A basic program of financial education also raises formal savings, but a savings match program (generous temporary interest rate subsidies at the partner bank) does not robustly magnify this impact. We also document an information externality: positive treatment spillovers in the form of higher savings at competitor banks. Taken together, these results are consistent with alleviation of information constraints, but not of access constraints on formal savings. All treatments have similar positive impacts on household consumption and total asset accumulation, so the basic financial education program (as the lowest-cost treatment) is the most cost-effective. The information externality provides a rationale for public subsidy or cross-bank collaboration in information provision related to formal savings.