We carry out a randomized controlled trial to evaluate the effect of three different types of messages sent to taxpayers on their compliance with the rental income tax (direct effect) and the spillovers produced on payments related to the capital gains and the self-employment income taxes. One message highlights detection, another appeals to social norms, and the third type appeals to altruism. This is the first study to evaluate if these messages can produce spillovers across taxes and to perform a long-term follow-up. This is important to determine if the treatment increases tax revenues. We find that the message addressing detection produces a positive and permanent direct effect and a negative but transitory spillover on the other two taxes. Overall, it increases tax revenues by US$3.92 per dollar spent in the long run. The message appealing to social norms has no direct effect but produces a permanent negative spillover on the capital gains tax. Ignoring this spillover would have lead one to conclude that this message is innocuous when in fact produces a loss of US$ 5.20 per dollar spent in the long run. The message appealing to altruism produces a transitory negative effect and no spillovers, and has no effect on tax revenues in the long run.