The literature on optimal fiscal policy finds that highly volatile real returns on government debt, for example through surprise inflation, have very low costs. However, policymakers are almost always very apprehensive of this option. The paper discusses evidence concerning features of developing country financial markets that are missing in existing models, and that may suggest why this policy is considered so costly in practice. Most importantly, domestic banks choose to be highly exposed to government debt because the alternative, private lending, is more risky under existing legal and institutional imperfections. This exposure makes banks and their borrowers vulnerable to the government's debt policy.