08/22/2026
We’d rather be roughly right for twenty years than exactly right for one.
That phrasing isn’t ours. John Maynard Keynes is generally credited with the idea that it’s better to be approximately right than precisely wrong, and Warren Buffett has repeated it for decades. It’s a good sentence. It’s also the least understood sentence in financial planning, because most people hear humility in it and there’s actually arithmetic in it.
Howard Marks has built a career on this distinction. Two of his lines are worth sitting with.
The first: you can’t predict, but you can prepare. Marks borrowed the phrase from an old advertisement and has used it as an organizing principle across four decades of memos.
The second is one he attributes to Elroy Dimson: risk means more things can happen than will happen.
That’s the whole problem with an optimized tax plan stated in nine words.
Here’s the mathematical version.
Say we model your conversion schedule and it produces a projected lifetime tax savings of $340,000. That’s one number. It looks like an answer.
It isn’t an answer. It’s the center of a distribution that nobody printed.
The model assumed a rate environment, an income path, two life expectancies, a set of IRMAA thresholds, and your children’s future brackets. Each is a range, not a point. Multiply five ranges together and the honest output isn’t $340,000, it’s a cloud with $340,000 sitting somewhere in the middle of it.
Marks makes a related observation about forecasting: he’s said the fact that something is going to happen doesn’t mean it’s going to happen soon. Directionally correct and specifically wrong is the most common way to lose money in this business, and it’s exactly what a twenty-year tax projection is exposed to.
Precision in a projection is a claim about the future dressed as a claim about arithmetic.
Now the part that decides the recommendation.
Run the aggressive schedule and the conservative one through a range of outcomes rather than a single path, and something shows up consistently.
The aggressive plan has the higher expected value and the wider spread. In the good scenarios it wins by a lot. In the bad ones it doesn’t just win less, it creates problems. IRMAA surcharges triggered in a year you also realized a gain. Liquidity spent on a tax bill you later needed for care. And no reversal available, since recharacterization of conversions was eliminated by the 2017 tax act. Before then there was a mulligan. There isn’t one now.
The conservative plan gives up meaningful expected value and compresses the range.
So the choice isn’t between a good plan and a worse one. It’s between a higher average and a narrower spread. When the left tail contains structural damage rather than merely a smaller number, the spread matters more than the average.
Marks has a line about this too: he’s noted that his clients don’t need him to deliver more of what everyone else is getting. They need him to be right about the downside.
You cannot spend an expected value. You live in one path.
What “roughly right” actually looks like.
A steadier multi-year schedule instead of a large front-loaded conversion. Bracket ceilings with room left under them. Explicit stops before the IRMAA cliffs. Enough flexibility that a surprise in year four doesn’t force abandoning the plan in year five.
On paper it produces a smaller number. In practice it survives being wrong, which the optimized version does not.
And there’s the behavioral piece. A plan someone abandons in a bad year was never optimal — it only looked optimal in a spreadsheet where the person doesn’t get nervous.
Some history worth knowing.
The 1986 Tax Reform Act took the top marginal rate to 28%. Anyone planning in 1985 on the assumption of high future rates was wrong. Anyone planning in 1988 on the assumption 28% would persist was also wrong; it moved back up within five years.
Rates have moved substantially in 1981, 1986, 1990, 1993, 2001, 2003, 2013, 2017, and again with the 2025 legislation. That’s nine meaningful changes in forty-five years, roughly one every five years.
A twenty-year conversion schedule will encounter approximately four of them.
The plan doesn’t need to guess which four. It needs to still function after they arrive.
All of this is public. Marks’ memos are free at Oaktree’s site, and The Most Important Thing covers the risk framework directly. Search “recharacterization eliminated Tax Cuts and Jobs Act” for the no-undo rule and “IRMAA two-year lookback thresholds” for the Medicare cliff. Tax Policy Center publishes the historical top-rate table free.
No deadline on the reading. There is one on the window; the low-bracket years between retirement and RMDs don’t roll forward, and every year spent deciding closes on its own.
(General education only. Not tax advice, brackets, thresholds, and state rules vary, and the right answer is individual.)
