Data snooping in break tests

(a) Explain why choosing the break date that maximises $F$ and then using the $F(k,n-2k)$ $p$-value over-rejects. Relate this to the multiple-testing ideas of M04. (b) With $n = 200$ and 15% trimming there are 141 candidate dates. Give the Bonferroni-corrected per-date level for a 5% family-wise test and explain why it is conservative. (c) The simulation in 13A.10.2 rejected 37% of the time at a nominal 5%. Design a simulation that would give the correct critical value for your own data, and state what it must hold fixed.