We analyze the optimal tax choices of a revenue-maximizing government that levies taxes from firms of which the true degree of mobility is ex ante unknown. Differential tax treatment of immobile and mobile firms is ruled out, but the government may learn from the firms’ location responses to past tax rate changes. Firms, however, may anticipate this and adjust their choices accordingly. We derive all symmetric Bayesian equilibria with a focus on the (so far neglected) one where the government sets a tax rate that triggers partial migration but full revelation of the true number of mobile firms. We show that, if tax competition is fierce (i.e., relocation cost and foreign tax rates are low), expected tax rates and expected firm migration are higher if the degree of mobility is unknown. There is a positive value of learning, i.e. commitment on future tax rates cannot increase the government’s expected revenue. However, if the government can commit to a rule-based learning mechanism, i.e. credibly tie its future tax policy to present policy outcomes, it may obtain a Pareto improvement.
