This paper develops a new approach to isolate and quantify the extent to which deposit withdrawals are due to liquidity, exposure to policy risk, or expectations about how other depositors will behave. We use high frequency micro-data on insured time-deposits from a large Greek bank over a long time period that spans quiet periods as well as events with large policy uncertainty. We use variation induced by maturity expiration of time deposits around the large policy uncertainty events to filter deposit withdrawals due to direct exposure to policy risk from those due to expectations about behavior of other depositors. In response to a policy uncertainty shock that doubled the short-run CDS price of Greek sovereign bonds, the early deposit withdrawal probability quadrupled. About two-thirds of this increase is driven by direct exposure to policy risk with the remainder due to changes in expectations of behavior of other depositors. We quantify these effects in terms of forgone interest rates and changes in short-run CDS prices. Our estimates imply effects that compare well with anecdotes from other recent prominent episodes of depositor withdrawals.