I will talk about an actuarial model for the wealth of an insurance company with loss-carry-forward taxation. The model consists of a stochastic process whose drift is modified when the process is at its historic maximum. When does this process exist? Is it possible to calculate interesting functionals of the process, such as the expected total tax revenue raised before the company goes bankrupt? Can we determine how to choose the tax rate in order to maximise the total tax revenue? The answers to these questions are 'pretty much always', 'yes' and 'yes'. The results are of interest to actuarial researchers, but I am also fond of tax processes because they offer a way to perturb a stochastic process, in a way which depends on its position and history, while retaining the possibility of finding explicit identities.