Paraphrase of question I was emailed:
How is it possible to model something as unexpected?
Following code shows how to model something unexpected (a zero probability event that then does occur). In this example the unexpected thing is that there is a tax rate tau, and you believe tau will equal zero forever, but then at age 50 (model period 31) you wake up to discover that tau=0.2 (and will stay at this forever). [I use a lightly modified version of Life-Cycle Model 9]
Basic idea is:
first, solve the value and policy function in model where tau is always 0.2.
second, solve for value and policy functions in model where tau is always zero
third, splice together value and policy functions so that it uses those for tau=0 for ages 20-49 (model periods 1-30) and those for tau=0.2 for ages 50+ (model periods 31+)
fourth, now just use spliced policy to create agent distribution
fifth, be careful when computing model statistics (make sure you use right tau values for right things)