I think the answer is either that I have somehow screwed up the data, though I was very careful, or that there is some sort of bug in ESPlanner, with the former of course far more likely. I can't imagine taxes would lead to such a huge discrepancy. Since I'm almost 75, and my wife is 79, the 20% increase in scenarios 2 and 3 that I get for knocking 5 years off our lives seems about right; it's the first scenario that doesn't make any sense, with our income dropping by dying 5 years earlier. Yet leaving our deaths at 100 produces almost identical results in all three scenarios, which kind of makes sense given the way annuity payouts are determined from mortality tables, and absolutely the only thing I then changed in scenario 1 was the ages. And yes, the suggested discretionary spending is smooth all the way to the end, except a little bit higher in the very last year of my life.

I don't have time right now, but sometime in the near future, probably next weekend, I'll go over everything with a fine tooth comb. If I can't find an error, I'll create a support ticket.

Okay, found the error. The basic profile, from which I copied the other two, changing only the start date of the third annuity from 2018 in the original to 2020 and 2023 in the copies, somehow had its nominal investment return reset to 0%, for all three asset classes (i.e. regular assets, and my and my wife’s retirement assets), from 3% (same as my assumption for inflation), at the same time that I changed the death ages from 100 to 95. I have no idea how this happened, the copies all have the original 3%, and I would never have changed the nominal returns assumptions on the original basic profile to zero deliberately (I’m not that pessimistic, though maybe I should be), and it would be extremely difficult to do accidentally.

Having fixed that error, I then ran the numbers with death ages set to 105. The interesting finding was that the start date of the third annuity (and it is a significant one, would probably account for a bit over 5% of my discretionary spending, hard to come up with an exact percentage given the peculiar tax treatment of immediate annuities) is almost irrelevant. I would have thought that holding off for five years with the consequent 37% increase in annual payout would really pay off if you lived a very long time (e.g. age 105), but for each death age assumption, 95, 100, and 105, the discretionary spending ALWAYS DECLINES with each delay in annuity start date, though the difference between starting in 2018 and starting in 2023 is less than 0.15% in discretionary spending in each case, little more than a rounding error. Kind of startling consistency. Never could have figured this out without ESPlanner.

Okay, found the error. The basic profile, from which I copied the other two, changing only the start date of the third annuity from 2018 in the original to 2020 and 2023 in the copies, somehow had its nominal investment return reset to 0%, for all three asset classes (i.e. regular assets, and my and my wife’s retirement assets), from 3% (same as my assumption for inflation), at the same time that I changed the death ages from 100 to 95. I have no idea how this happened, the copies all have the original 3%, and I would never have changed the nominal returns assumptions on the original basic profile to zero deliberately (I’m not that pessimistic, though maybe I should be), and it would be extremely difficult to do accidentally.

Having fixed that error, I then ran the numbers with death ages set to 105. The interesting finding was that the start date of the third annuity (and it is a significant one, would probably account for a bit over 5% of my discretionary spending, hard to come up with an exact percentage given the peculiar tax treatment of immediate annuities) is almost irrelevant. I would have thought that holding off for five years with the consequent 29% increase in annual payout would really pay off if you lived a very long time (e.g. age 105), but for each death age assumption, 95, 100, and 105, the discretionary spending ALWAYS DECLINES with each delay in annuity start date, though the difference between starting in 2018 and starting in 2023 is less than 0.15% in discretionary spending in each case, little more than a rounding error. Kind of startling consistency. Never could have figured this out without ESPlanner.