RMDs are now reinvested, and tax brackets now follow inflation

1. What changed

The first change is how the engine treats required minimum distributions that a client doesn't need. Until now, those distributions were withdrawn, taxed correctly, and then dropped out of the projection entirely. From today they are taxed and reinvested, in both the convert and the don't-convert scenario, and they keep compounding for the rest of the projection.

The second is federal tax brackets, which the engine had been holding at their current values for the length of the projection. They now grow with inflation, which is what federal law has done automatically every year since 1985.

Both reduce the reported advantage of converting. We recomputed every case ever run through the system both ways. Of the cases the old model said to convert, about 2% no longer come out ahead. In the other 98% the recommendation is unchanged and only the size of the advantage moved — in the median case, by 57.1%. Bracket indexing reduces it further.

One second effect is worth knowing up front. Reinvested gains occupy tax-bracket space, so the amount the tool recommends converting in a given year can change too, not just the advantage reported for it.

The two changes reach your existing work differently, and section 8 covers that in full. Reinvestment is off for every case that existed before today. Bracket indexing applies everywhere, including cases you have already built.

2. What the model assumed before

Every Roth conversion comparison is two futures placed side by side: the client converts, or the client doesn't. The gap between them is the number the recommendation rests on.

In the future where the client doesn't convert, the IRS eventually requires distributions from the IRA, whether or not the client has any use for the money. The engine took those distributions and taxed them correctly. And then the money left the projection. It was counted once, as income the client received, and never appeared again.

In the future where the client converts, the Roth balance kept compounding, untouched and tax-free, for the rest of the projection.

The comparison was therefore between money that grows for thirty years and money that stops existing the moment it is withdrawn. That gap widens every year, and it widens fastest for the clients advisors most often model conversions for — the ones with no need for the distribution, where nearly every forced dollar was a dollar the projection discarded.

For a 72-year-old with a $1M IRA, no income need, and 6% growth:

The second row is where the flaw is easiest to see. A real client in this position takes the distribution because the law requires it, pays the tax, and puts the remainder back into the market. Thirty years of those distributions are still there at the end. The model was reporting that they weren't.

Several of you have raised a version of this with us over the past year. We've been building it ever since, and this is the result.

3. What it does now

The engine now carries a separate account for reinvested distributions. When a client is required to take money out of an IRA and doesn't need it to live on, the tax is paid and the remainder goes into that account, where it stays invested for the rest of the projection. Money the client actually spends is still spent. Only the unneeded portion is reinvested, and what counts as needed comes from the Net Income field.

It applies to both futures, and that is the part worth reading twice.

A conversion is rarely all-or-nothing. Most recommendations are partial: some of the IRA converts, some of it doesn't, often across a long window. Whatever remains in the IRA keeps producing required distributions on exactly the same schedule under exactly the same rule, and those distributions are reinvested exactly the same way. The convert scenario reinvests too.

Each future compounds at its own assumed rate. Baseline reinvestment grows at the baseline comparison rate, and post-conversion reinvestment at the post-contract rate.

In the example above, the reinvestment account holds $3,034,472 at the end of the projection. That is where the updated legacy figure comes from. It isn't a penalty applied to conversion; it's money the model had been leaving out.

The account doesn't appear on screen; the interface shows the updated numbers, not the mechanism behind them. It's in the Excel export if you want the year-by-year flow. And unlike the IRA, it reaches the heirs with no embedded tax bill attached to it.

4. How the reinvested money is taxed

Reinvested distributions land in a taxable brokerage account, which the tax code treats nothing like an IRA. That left a modeling choice with three possible answers.

Two of them we rejected. Ignoring tax on the growth would have been simple and wrong in a specific direction, overstating what reinvestment recovers. Tracking cost basis lot by lot until death is the realistic model, and it would have turned the death-year math into a number you cannot trace by hand or defend in front of a client.

What the engine does instead is assume the client sells everything and rebuys it every year. Every gain is realized in the year it occurs and taxed as ordinary income at the client's actual marginal rate.

That one assumption is what makes the rest of the math behave. Because the gains land in ordinary income, they carry all the second-order effects that make retirement tax planning difficult in the first place. They push more Social Security into the taxable range. They raise MAGI, which can lift Medicare IRMAA premiums two years later. They erode the senior deduction. And they consume bracket headroom.

That last one matters more than it sounds. Headroom spent on reinvestment gains is headroom no longer available for conversion, so the reinvested money reduces how much can be converted in that year. The correction constrains itself.

It is also deliberately conservative in a specific direction. A client who simply holds the position pays long-term capital gains rates on it, if they realize anything at all, rather than ordinary rates every year. Annual realization is a heavier tax drag than most clients would actually carry, and it falls most heavily on whichever leg has the most IRA left to distribute — usually the one that doesn't convert.

5. Brackets now follow inflation

The second change is smaller to describe and touches more cases.

The engine had been holding federal tax brackets at their current values for the entire projection — the same thresholds in year one and in year thirty. Deductions, IRMAA thresholds and Social Security benefits were already growing with inflation. Brackets were the exception.

That modeled decades of bracket creep the law specifically prevents. Bracket thresholds have been indexed to inflation automatically every year since 1985, under the Economic Recovery Tax Act of 1981. Moving the rates themselves takes an act of Congress, which has happened a handful of times in thirty years. Moving the thresholds happens on its own, every year, without anyone voting on it.

Frozen thresholds made future income look more heavily taxed than the law allows. Future taxable income is concentrated in the scenario where the client doesn't convert and carries a long tail of distributions, so that scenario absorbed nearly all of the distortion.

This one was raised by advisors too.

6. Why the two changes don't cancel out

You would expect a pair of corrections to pull against each other at least sometimes. These don't, and the reason is the same for both.

A conversion pays its tax early, at today's brackets. The scenario where the client doesn't convert carries decades of taxable withdrawals instead. Vanishing distributions and frozen brackets both made that long tail look worse than it is, and neither had much effect on the side that converts.

Two different flaws, both sitting on the same side of the comparison.

7. What this does to the numbers

Before releasing the reinvestment change, we recomputed every case ever run through the system on both the old and the updated engine and compared them.

Of the cases where the old model showed conversion coming out ahead, 95% still show it ahead, most of them by a smaller margin. 3% now land at roughly break-even. 2% now show a net loss.

What moved is the size of the advantage. In the median case the reported advantage fell by 57.1%, and the median dollar decline was $788,796. In roughly four cases out of ten, the overstatement exceeded $1 million.

The 2% were always close calls. Their median reported advantage was $469,774 before the change and their median result after it is a loss of $18,653, a margin thin enough that updating the model was enough to move them to the other side of the line.

Cases with no RMD exposure did not move at all. That is what a targeted correction should look like. It changes what the flaw touched and leaves the rest exactly where they were.

Bracket indexing then reduces what remains. We have not finished measuring it across the full database, so we'll describe the direction rather than put a figure on it: it moves consistently the same way, and by a good deal less than reinvestment did. For the 72-year-old in the example, the reported advantage lands at $799,708 with both changes applied, against $964,321 with reinvestment alone.

8. What happens to cases you've already built

The two changes are treated differently, deliberately.

Reinvestment is switched off for every case that existed before this update. Cases created from now on have it on by default. The control is "Reinvest RMDs," under Advanced, and it works in both directions on any case: turning it off on a new case reproduces the old modeling exactly, and turning it on for an existing case applies the change to that case — useful if a client asks what happens if they reinvest.

Bracket indexing has no toggle and applies to every case, old and new. Whether thresholds move with inflation isn't an assumption about how a client behaves; it's how the tax code works. Freezing it for older cases would mean knowingly showing a projection built on a rule the IRS doesn't follow.

So a case you built last week will read a little differently today, even with reinvestment off. In most cases the difference is small, and it is the bracket change.

We froze reinvestment for existing cases because a number that changes in the middle of a conversation costs more trust than a better number gains. We didn't freeze bracket indexing because the alternative is a number that's wrong about the law.

9. What this doesn't mean

Converting is not suddenly the wrong call. In roughly 98% of cases the recommendation is unchanged; about 2% now point the other way. Scenarios that land at break-even still generally favor converting in our view, because a Roth balance is the position least exposed to a future change in tax law. A modeled advantage near zero is not the only reason to convert.

The old numbers were not fabricated or careless. They were arithmetically correct given an assumption about what a client does with a distribution they didn't need, and the assumption was the part that was wrong. If you presented those numbers, you presented the best output the model could produce at the time, and in all but about 2% of cases the conclusion you drew from them still holds.

And the 57.1% median describes a modeled advantage, not a retirement. No client's outcome changed this week. What changed is the size of the number the software reports when it compares two futures.

10. We're not done

Modeling a retirement means making assumptions, and assumptions are the part of any tool that deserves the most scrutiny. Both of these sat in plain sight for a long time, and it took pointed questions from the people using the software every day to bring them into focus.

We would rather ship a number that holds up under that scrutiny than one that flatters the recommendation, and there is more of this kind of work in progress. When it lands you'll hear about it the same way: what changed, what it does to your numbers, and which of your existing cases it touches.

Keep the observations coming. Both of these started with advisors telling us something didn't look right.