We investigate why only some banks use regulatory arbitrage. We predict that banks wanting to be riskier than allowed by capital regulations (constrained banks) use regulatory arbitrage, while others do not. We find support for this hypothesis using trust-preferred securities issuance, a form of regulatory arbitrage available to almost all U.S. banks from 1996 to Dodd-Frank. We also find support for predictions that constrained banks are riskier, perform worse during the crisis, and use multiple forms of regulatory arbitrage. We show that neither too-big-to-fail incentives nor misaligned managerial incentives are first-order determinants of this type of regulatory arbitrage. (JEL G01, G21, G28)