Roth Conversion Optimizer
Searches across hundreds of Roth conversion plans to find the bracket and IRMAA settings — independently for each life phase — that produce the lowest present value of lifetime taxes paid. Built around a framework that treats NPV of tax savings as the only defensible primary metric for evaluating a conversion.
How it works
Evaluating a Roth conversion is genuinely difficult. Most intuitive metrics — portfolio value, total taxes paid, even RMD levels — each mislead in common situations. The optimizer is built around a specific reasoning framework: NPV of lifetime tax savings is the primary metric that holds up to scrutiny, with income excluding RMDs as a secondary sanity check and RMDs/IRMAA as informational context. The implementation turns that framework into a discrete search: four bracket ceilings times three IRMAA options, applied independently to each of up to three life phases, for up to 1,728 total combinations. Each combination runs the full projection engine. The combination with the lowest lifetime tax NPV is recommended.
1. Why Evaluating Conversions Is Hard
Before running any numbers, it helps to understand why the obvious metrics fail. Portfolio value misleads because account types are not equivalent — comparing a Traditional IRA and a Roth requires assuming a future tax rate, which picks the winner before any analysis begins. Nominal lifetime taxes mislead because a dollar of tax paid in thirty years is not the same as a dollar paid today — a plan that shifts taxes into the future looks worse on nominal totals even though it is better in economic terms. RMDs are informational: smaller RMDs are often how future taxes fall, but a plan with lower RMDs is not automatically better if it achieved that by converting at an unfavorable rate today. Even lifetime income misleads if RMDs are included, since forced distributions inflate the number without reflecting income the client actually controls.
Example: Two plans both show $500k in lifetime taxes. Plan A paid most taxes early, in a low bracket. Plan B deferred taxes to RMD years at a higher rate. Nominal totals say tie. NPV correctly scores Plan A lower — the early, lower-bracket taxes are worth less in present-value terms.2. Why NPV of Tax Savings Is the Most Defensible Primary Metric
NPV of lifetime taxes paid is the metric that holds up. It captures the time value of money without requiring an assumption about heir tax rates or future account equivalency. It is also the same metric Stratum uses throughout the Strategy Value Breakdown, so the optimizer recommendation reads consistently with the rest of the plan. The secondary metric — NPV of income excluding RMDs — acts as a sanity check: it confirms the conversion is producing more flexible, advisor-controlled income rather than simply eliminating income. RMDs and IRMAA surcharges are shown as context to explain the mechanism behind the recommendation, but neither affects ranking.
Example: A plan that saves $42k in lifetime tax NPV and reduces RMDs by $31k NPV, at the cost of $2.4k in additional IRMAA surcharges, is a net winner. The IRMAA cost is comfortably offset by the tax savings.3. Divide the Plan Into Life Phases
The optimizer splits the projection into up to three phases, because the optimal conversion aggressiveness typically differs across life stages. Pre-retirement: from the current tax year through the year before retirement. Post-retirement: from retirement through the year of first death for couples, or through life expectancy for singles. Survivor: from the year after first death through the surviving spouse life expectancy. Each phase gets its own independent bracket ceiling and IRMAA setting. Phases that do not apply to the scenario are excluded automatically.
Example: Client age 58, retiring at 65, spouse same age, first modeled death at 88, survivor to 90: Pre-retirement 7 years (ages 58-64), Post-retirement joint 24 years (ages 65-88), Survivor 2 years (ages 89-90). Pre-retirement may justify aggressive conversions to 24% before wages end; the survivor phase may justify only filling to 12% given the shorter horizon.4. Build and Search the Grid
For each phase the optimizer tests every combination of bracket ceiling (12%, 22%, 24%, 32%) and IRMAA setting (no constraint, stay below tier 1, stay below tier 2). That is 4 x 3 = 12 combinations per phase. Across three phases the Cartesian product is 1,728 total combinations. Each combination runs the full projection engine. The 10% bracket is excluded because it is too restrictive to produce meaningful conversions for any scenario with a material tax-deferred balance. The 35% and 37% brackets are excluded because conversions at those rates rarely produce a net present-value benefit.
Example: One combination tested: Pre-retirement fill to 24% with no IRMAA constraint, Post-retirement fill to 22% stay below tier 1, Survivor fill to 12% stay below tier 1. The engine runs a complete projection for each combination and records the lifetime tax NPV.5. Score Each Combination
The primary objective is the NPV of total lifetime taxes paid, discounted at the plan assumed investment return rate. The combination with the lowest lifetime tax NPV wins. At an 8% discount rate, $50,000 in taxes twenty years from now has a present value of about $10,700 today.
Score = negative NPV of lifetime taxes paid. NPV = sum over each year Y of [ Tax in Year Y / (1 + Rate)^(Y - Current Year) ]
Example: Plan A: $420k nominal taxes, most in near years (NPV $310k). Plan B: $390k nominal taxes, most in distant years (NPV $305k). Plan B wins on NPV — the time distribution is what matters.6. Apply the Income Tiebreaker
When two combinations produce lifetime tax NPVs within $1,000 of each other, the optimizer promotes the one with the higher NPV of income excluding RMDs. This threshold is intentionally narrow: $1,000 approximates the practical resolution of the search, while being small enough that a meaningful tax-NPV difference is never overridden by the secondary metric. Income excluding RMDs captures whether the client is gaining flexible, controllable income rather than forced distributions.
Example: Two probes both at $305k NPV taxes. Probe A: $180k NPV non-RMD income. Probe B: $195k NPV non-RMD income. Probe B wins the tiebreaker — it produces more income the client actually controls.7. Handle Depleted Plans
For plans where the portfolio runs out before life expectancy, the NPV-of-taxes objective loses meaning — taxes in insolvent years are moot. The optimizer switches to depletion-delay mode: any probe where the portfolio survives beats any probe where it depletes, and among depleting probes, later depletion is better. This is an engineering choice to handle insolvency sensibly; the underlying philosophy does not address this case because the assumption is that a viable plan keeps the portfolio solvent.
Example: Plan A depletes in 2051. Plan B depletes in 2047. Plan A wins in depletion-delay mode. Any plan that avoids depletion entirely beats both, regardless of its lifetime tax NPV.8. Present the Recommendation
The optimizer presents the winning bracket and IRMAA parameters for each phase alongside four NPV metrics: lifetime taxes (the decision metric), income excluding RMDs (the sanity check), RMDs, and IRMAA surcharges (both informational context). All metrics show deltas versus a no-conversions baseline. The advisor can accept, review alternatives, or dismiss.
Example: Recommendation: Pre-retirement fill to 22% no IRMAA constraint, Post-retirement fill to 24% stay below tier 1, Survivor fill to 12% stay below tier 1. NPV tax vs. no conversions: -$42,000. Income (excl. RMDs): +$18,000. RMDs: -$31,000. IRMAA: +$2,400.
Real-world context
The Problem With Using Portfolio Value as the Benchmark
Portfolio value is the most natural thing to look at, and also the most misleading benchmark for Roth optimization. Account types are not equivalent. Comparing a Traditional IRA and a Roth requires assuming a future tax rate, and that assumption determines the winner before any analysis begins. Change the assumed rate by 1% and the winner flips. A Roth strategy funding an income target through tax-free withdrawals may intentionally produce a lower nominal ending balance than the do-nothing case — because drawing from Roth is the point. Portfolio value cannot distinguish this from underperformance. The optimizer computes ending portfolio values and shows them as context on the recommendation card. They do not affect ranking.
What No IRMAA Constraint Actually Means
The purest version of the optimizer philosophy would include a no-surcharge tier — a setting where the conversion stays inside all IRMAA thresholds. The implementation uses no constraint (null) instead, which means the probe is allowed to cross IRMAA tiers if the lifetime tax savings justify it. This is a deliberate engineering choice. A hard constraint that prunes any probe crossing an IRMAA cliff would eliminate plans that are genuinely better on tax NPV — for example, a plan that triggers modest tier-1 IRMAA for two years but avoids $30k of higher-bracket tax in the same period. Advisors who want a definitive IRMAA ceiling should choose stay-below-tier-1 or stay-below-tier-2 instead. The IRMAA metric on the recommendation card shows exactly how much surcharge the recommended plan adds. When no-constraint wins, verify the IRMAA cost is comfortably offset by the tax savings before accepting.
How the Supporting Metrics Explain the Recommendation
Only tax NPV determines the winner. The other three metrics explain the mechanism. Income excluding RMDs rising means more of the client income comes from sources they control — wages, pension, Social Security, scheduled Roth withdrawals — rather than forced distributions. That is generally a better position regardless of taxes. RMDs falling is usually the mechanism by which future taxes fall. A smaller tax-deferred balance at age 73 produces smaller required distributions, which produce lower taxable income, which produces lower tax — particularly for the surviving spouse, who files single at higher effective rates on the same income. IRMAA can move either direction. A modest surcharge increase alongside substantial income tax savings is still a net win. A sound recommendation typically shows: taxes down, non-RMD income stable or rising, RMDs down, IRMAA increase more than offset by the tax savings.
The Optimizer Scores Against No Conversions, Not Current Settings
All deltas compare the recommended plan against doing no Roth conversions at all — Roth strategy disabled, all other strategies active. If the baseline were the current editor configuration, the optimizer would show a zero delta whenever the advisor had not yet configured any conversions. Comparing against no conversions answers the question that actually matters: how much value does a conversion strategy create for this client versus doing nothing? This also matches how the Strategy Value Breakdown attributes value: each strategy contribution is measured against the scenario without it.
Accepting, Reverting, and Manual Adjustments
Clicking Apply writes per-phase bracket and IRMAA settings to the strategy as phase overrides and resets the start and end year to the full lifetime window. The plan recalculates immediately. A Revert link appears after applying, allowing the advisor to return to the exact configuration in effect before the optimizer ran — including any custom start or end year. If the advisor manually changes the top-level bracket or IRMAA dropdowns after applying, the per-phase overrides are cleared and the strategy returns to uniform behavior. A manual edit signals the advisor is taking direct control and does not want the optimizer phase settings active silently.
What drives the result
Writes independent bracket ceiling and IRMAA constraint for each life phase. The engine applies each phase settings only to the years within that phase window. Year-by-year conversion amounts are sized by the auto-mode engine based on actual income and bracket headroom in each year.
Pre-retirement: fill to 24% no IRMAA constraint. Post-retirement: fill to 22% stay below tier 1. Survivor: fill to 12% stay below tier 1.
Sets the NPV discount rate for all optimizer scoring. A higher rate discounts future taxes more heavily, increasing the relative benefit of deferring taxes. A lower rate raises the present value of distant taxes, making it more important to convert now.
At 8% discount rate, $50,000 of taxes in 20 years = $10,727 NPV. At 4%, the same = $22,819 NPV. A lower discount rate makes future taxes look more expensive and can shift the recommendation toward fewer or smaller conversions.
Sets the endpoint of the survivor or post-retirement phase. A longer horizon means more years of RMD exposure — particularly at the survivor filing status, where the same income faces higher marginal rates than under married filing jointly.
Life expectancy 95 vs. 85 adds ten more years of RMD exposure at the survivor single filing status.
Determines future RMD levels. Larger balances create more future tax exposure and make aggressive early conversions more valuable. Smaller balances reduce the potential benefit and may produce a recommendation for conservative or no conversion.
$1.2M tax-deferred produces RMDs at age 75 of approximately $60,000/year. $200k produces approximately $10,000/year. At $60k/year of forced distributions, there is substantial room to reduce future taxable income through early conversion.
Assumptions
- Discount rate for NPV matches the plan assumed investment return rate (default 8%). The same rate used in the Strategy Value Breakdown and Summary Tab NPV columns.
- Search space: bracket ceilings 12%, 22%, 24%, and 32%. The 10% bracket is excluded (too restrictive for scenarios with material tax-deferred balances). The 35% and 37% brackets are excluded (conversions at those rates rarely produce a net present-value benefit).
- IRMAA options per phase: no constraint (null — allows the probe to cross IRMAA tiers if the lifetime tax savings outweigh the surcharge cost), stay below tier 1, stay below tier 2. No constraint is not a no-surcharge guarantee — it means the optimizer is permitted to cross tiers when beneficial.
- Baseline for all deltas: Roth strategy disabled, all other strategies active. Isolates the Roth conversion contribution and avoids a zero delta when the advisor has not yet configured any conversions.
- Phase boundaries use the same retirement year, first death year, and last death year from plan assumptions.
- On accept, the strategy start and end year fields reset to full-lifetime anchors covering the entire projection window.
- Income tiebreaker activates only when two probes are within $1,000 NPV of each other — a pragmatic threshold that resolves genuine ties without letting the secondary metric override meaningful tax differences.
Limitations
- Searches by bracket ceiling and IRMAA tier, not exact dollar amount. A single bracket setting may be too aggressive in some years and too conservative in others when income varies significantly within a phase. Year-by-year sizing is handled by the auto-mode engine within each projection run.
- The no-IRMAA-constraint option does not mean avoid all surcharges — it means the optimizer is permitted to cross IRMAA thresholds when tax savings are sufficient. Advisors who want a hard ceiling should use stay-below-tier-1 or stay-below-tier-2 instead.
- Does not model estate planning objectives. If minimizing heir tax burden or maximizing inheritance is a goal, use the recommendation as a starting point and adjust manually.
- For plans where the portfolio depletes before life expectancy, the optimizer switches to depletion-delay mode (later depletion is better). Lifetime tax NPV is not the ranking criterion in this mode.
- Available on ADVISOR and EDUCATOR plans only.
Watch
Related
This page explains how Stratum models this calculation. It is educational material for financial professionals, not tax or legal advice, and tax law changes. Verify current figures against primary IRS sources before relying on them with a client.