[Updated] Roth IRA Conversion scenarios - Sensitivity Analysis

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

[Updated] Roth IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Each time after building Roth Conversion Scenarios and inspecting the conversion financial benefits, I always found myself pondering: "What would be the impact of my analysis as my assumptions will likely be wrong?"; hence this sensitivity analysis that I'd like to present to Bogleheads for your thoughts/comment/feedback/advice. Note that this post is primarily about the process of sensitivity analysis, and the results should not be generalized for all or even most cases - run your own numbers, YMMV.

Methodology:
1. Tool used: Retiree Portfolio Model Beta version with Portfolio Asset Allocation feature + custom code for tax-adjusting asset allocation,
2. Key assumptions used as factors for Sensitivity Analysis:
----- Stock average annual returns: 4% low case, 6% base case, 8% high case,
----- Bond average annual returns: 1% low case, 2% base case, 3% high case,
----- COLA/Inflation/Tax bracket adjustment rates (assumed to be all equal): 1%, 2%, 3% for 3 low/base/high cases respectively,
----- Tax changes: (1) no change, (2) changed in 2026 as written, and (3) 2% higher than 2026 as written, for 3 low/base/high cases respectively,
----- Living Expenses: 85%, 100% and 115% of 120K/year for 3 low/base/high cases respectively.
----- Tax-Adjustment discount rates for Taxable and T-IRA accounts: 0%/0% (no tax-adjustment), 15%/15% (base case) and 15%/25% (high adjustment),
3. Build a NO CONVERSION Base Case,
4. Vary one key assumption (in #2 above) at a time from a Low Value to a Base (Expected) Value to a High Value while keeping all else constant,
5. Plot the the resulting Portfolio Ending After-tax Value (PEATV) - after the heirs take their last required distribution and the IRAs are exhausted - used as conversion benefits metrics in a Tornado Diagram,
6. Build a WITH CONVERSION Case and repeat steps 4-5,
7. Compare the WITH and NO CONVERSION cases and perform sensitivity analysis comparison.

Base scenario (fictitious):
MFJ, both 63, model life expectancy 63-92,
SS @ 70 for both = 96K/year
3M portfolio with desired 50% stock / 50% bond portfolio AA,
1M Taxable with desired 100% stock AA, 0.5M ROTH IRA with desired 100% stock AA, 1.5M Trad-IRA with desired 100% bond AA,
Almost financially clone Heirs to inherit T-IRA with 10-year stretch (equivalent to 93-102):
----- retired MFJ with living expenses 25% less than original couple's @ 92,
----- and SS 25% less than original couple's @92,
----- the RMD divisors for the last 10 years of the model (93-102) are set to 10, 9, ... 2, 1 to model that heirs will withdraw 1/10, 1/9, ... 1/2, 1/1 annually from the inherited T-IRA for 10 years under same tax rules.

Results: Financial Benefit of Roth conversion = WITH CONVERSION Portfolio Ending After-Tax Value (PEATV) minus NO CONVERSION PEATV = varies from 5%-12% under these scenarios, so Converting to the top of 22% from 2021-2025 does not seem to be a bad decision - Note that the RPM tool used here does not calculate LTCG/NIIT.

0. Using Base values, converting to the top of 22% bracket from 2021-2025 results in a 504K (6.269Mil - 5.765Mil) or 8.7% gain in PEATV (check out how the AAs change between the WITH CONVERSION and NO CONVERSION cases),
1. Overall higher(lower) stock return result higher(lower) conversion benefit, as stock return in(de)creases, the PEATV of the WITH CONVERSION case in(de)creases faster than the PEATV of the NO CONVERSION case,
2. Overall lower(higher) living expenses result higher(lower) conversion benefit,
3. Higher(lower) bond return result higher(lower) conversion benefit. In this test case bond return variability impact the conversion benefit the most.
4. Overall lower(higher) COLA/Inflation/Tax Bracket Adjustment Rates result higher(lower) conversion benefit in this test case, but these are complex relationships because these rates impact the living expenses and SS and tax brackets in the models so the results are not necessarily indicative of any trend(s),
5. Roth IRA conversion is a tax arbitrage play, so higher tax after 2025 would increase the conversion benefit - no surprise here.
6. Tad-adjusting the Taxable account reduces the conversion benefit but tax-adjusting the T-IRA increases the conversion benefit even more, so the net impact is an increase in conversion benefit when tax-adjusting the portfolios in Roth IRA conversions - more here https://bogleheads.org/forum/viewtopic. ... 4#p6194834

Image

Image

Results: ROTH IRA Conversion up to top of 22% bracket from 2021 to 2025 case:
1. Stock return is the most significant variable that can make the PEATV =3.6M (42% less) or =10.3M (64% more) versus the base case =6.3M,
2. For this portfolio size/AA and the 120K/year budget, the Living Expenses is the 2nd most significant variable that can make the PEATV =3.9M (37% less) or =8.5M (36% more) versus the base case =6.3M,
3. Bond return is the 3rd most significant variable that can make the PEATV =4.9M (23% less) or =8.0M (27% more) versus the base case =6.3M,
4. COLA/Inflation is the 4th most significant variable that can make the PEATV =5.3M (15% less) or =7.0M (11% more) versus the base case =6.3M,
5. Tax changes are the least significant variable that can make the PEATV =6.2M (~0% less) or =6.3M (-0% more) versus the base case =6.3M, a small surprise at least to me,
6. Tax-adjusting portfolio can make the PEATV =6.1M (2% less with 15%/25% discount rates) or =6.5M (3% more) versus the base case =6.3M.

Image

Image

Results: NO ROTH IRA Conversion case:
1. Stock return is the most significant variable that can make the PEATV =3.2M (44% less) or =9.7M (68% more) versus the base case =5.8M,
2. For this portfolio size/AA and the 120K/year budget, the Living Expenses is the 2nd most significant variable that can make the PEATV =3.5M (39% less) or =7.9M (37% more) versus the base case =5.8M,
3. Bond return is the 3rd most significant variable that can make the PEATV =4.6M (21% less) or =7.6M (25% more) versus the base case =5.8M,
4. COLA/Inflation is the 4th most significant variable that can make the PEATV =5.0M (14% less) or =6.4M (11% more) versus the base case =5.8M,
5. Tax changes are the least significant variable that can make the PEATV =5.6M (3% less) or =5.9M (3% more) versus the base case =5.8M,
6. Tax-adjusting portfolio can make the PEATV =5.5M (4% less with 15%/25% discount rates) or =6.1M (5% more) versus the base case =5.8M.

Image

Image

Potential room for improvement:

1. Is there any other potential key input(s) that should be included in the sensitivity analysis?
2. Instead of my own expectations, I would prefer to use results of statistical analysis for the key inputs, such as expected average annual returns for stocks and bonds, and high/low values that would provide say a 95% confidence level. I am however neither an expert in the finance area nor statistics/quantitative analysis area so I have a lot of homework to do, so suggestions for these values from knowledgeable people are very much appreciated.
3. Include LTCG/NIIT calculations in the cases.

Thoughts/comment/feedback/advice on the Sensitivity Analysis presented? Thank you in advance.
Last edited by DSBH on Tue Aug 31, 2021 10:52 am, edited 8 times in total.
John C. Bogle: "Never confuse genius with luck and a bull market".
User avatar
jeffyscott
Posts: 13487
Joined: Tue Feb 27, 2007 8:12 am

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by jeffyscott »

DSBH wrote: Tue Aug 03, 2021 3:35 pm 2. Instead of my own expectations, I would prefer to use results of statistical analysis for the key inputs, such as expected average annual returns for stocks and bonds, and high/low values that would provide say a 95% confidence level. I am however neither an expert in the finance area nor statistics/quantitative analysis area so I have a lot of homework to do, so suggestions for these values from knowledgeable people are very much appreciated.
RA is one source that provides confidence intervals and seems to me to have pretty reasonable figures for expected returns. Plus you get the choice of two different models.

Here's their 10 year expected returns distribution, using the valuation dependent model, for US large cap:
Image

and the same, but using the yield and growth model:
Image

I would think if looking at 10 years or less, either model could be used. But beyond 10 years, only the yield and growth model would be appropriate, since the other one assumes a return to normal valuations in 10 years and so would then expect higher subsequent returns.

They do also have an option for a 50/50 mix of the two models and maybe that would effectively mean a return to normal valuations in 20 years?
https://interactive.researchaffiliates. ... terms=REAL
User avatar
jeffyscott
Posts: 13487
Joined: Tue Feb 27, 2007 8:12 am

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by jeffyscott »

If you have not already reviewed it, the McQuarrie paper that is linked to discussion here, might give you some additional ideas.

I recall that his paper mentioned some other, more complex tax, changes that could potentially occur.
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

jeffyscott wrote: Wed Aug 04, 2021 10:38 am RA is one source that provides confidence intervals and seems to me to have pretty reasonable figures for expected returns. Plus you get the choice of two different models.

I would think if looking at 10 years or less, either model could be used. But beyond 10 years, only the yield and growth model would be appropriate, since the other one assumes a return to normal valuations in 10 years and so would then expect higher subsequent returns.

They do also have an option for a 50/50 mix of the two models and maybe that would effectively mean a return to normal valuations in 20 years?
https://interactive.researchaffiliates. ... terms=REAL
Thanks so much for the reference, I will follow up on the RA models.
jeffyscott wrote: Wed Aug 04, 2021 10:53 am If you have not already reviewed it, the McQuarrie paper that is linked to discussion here, might give you some additional ideas.

I recall that his paper mentioned some other, more complex tax, changes that could potentially occur.
Thank you for the suggestion. I have reviewed the paper, and while I use a different set of assumptions (e.g. paying conversion taxes from Taxable account, tax-efficient investing etc.) I will revisit for potential additional ideas.
John C. Bogle: "Never confuse genius with luck and a bull market".
smitcat
Posts: 13308
Joined: Mon Nov 07, 2016 9:51 am

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by smitcat »

I must be dense and not reviewing this well but do you have a 'base case' that does not include any Roth conversions and what the variables would mean for that?
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

smitcat wrote: Wed Aug 04, 2021 7:28 pm I must be dense and not reviewing this well but do you have a 'base case' that does not include any Roth conversions and what the variables would mean for that?
This is the ‘base case’ with no Roth conversion, as mentioned in Methodology item 1. I’ll try to make it clearer.
John C. Bogle: "Never confuse genius with luck and a bull market".
smitcat
Posts: 13308
Joined: Mon Nov 07, 2016 9:51 am

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by smitcat »

DSBH wrote: Wed Aug 04, 2021 7:42 pm
smitcat wrote: Wed Aug 04, 2021 7:28 pm I must be dense and not reviewing this well but do you have a 'base case' that does not include any Roth conversions and what the variables would mean for that?
This is the ‘base case’ with no Roth conversion, as mentioned in Methodology item 1. I’ll try to make it clearer.
My bad - are you then going to run the same chart(s) with a Roth conversion?
Are you going to then vary the size of the Roth conversions?
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

smitcat wrote: Wed Aug 04, 2021 7:50 pm
DSBH wrote: Wed Aug 04, 2021 7:42 pm
smitcat wrote: Wed Aug 04, 2021 7:28 pm I must be dense and not reviewing this well but do you have a 'base case' that does not include any Roth conversions and what the variables would mean for that?
This is the ‘base case’ with no Roth conversion, as mentioned in Methodology item 1. I’ll try to make it clearer.
My bad - are you then going to run the same chart(s) with a Roth conversion?
Are you going to then vary the size of the Roth conversions?
Excellent questions, I am still struggling with them.

Assuming that there is 1 strategy of Roth conversion. Using the PEATVs as the metrics for comparison, I am thinking about showing the Deltas of the PEATVs between that strategy and the No Conversion base case as each of the key inputs vary.

In other words, you would see the same graph, except the Deltas against the base case of the PEATVs will be shown instead of the absolute PEATVs as the key input vary from Low to Base to High values.

So if there are 4 strategies/sizes for Roth conversions, I am thinking about 4 similar graphs.
John C. Bogle: "Never confuse genius with luck and a bull market".
smitcat
Posts: 13308
Joined: Mon Nov 07, 2016 9:51 am

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by smitcat »

DSBH wrote: Wed Aug 04, 2021 8:06 pm
smitcat wrote: Wed Aug 04, 2021 7:50 pm
DSBH wrote: Wed Aug 04, 2021 7:42 pm
smitcat wrote: Wed Aug 04, 2021 7:28 pm I must be dense and not reviewing this well but do you have a 'base case' that does not include any Roth conversions and what the variables would mean for that?
This is the ‘base case’ with no Roth conversion, as mentioned in Methodology item 1. I’ll try to make it clearer.
My bad - are you then going to run the same chart(s) with a Roth conversion?
Are you going to then vary the size of the Roth conversions?
Excellent questions, I am still struggling with them.

Assuming that there is 1 strategy of Roth conversion. Using the PEATVs as the metrics for comparison, I am thinking about showing the Deltas of the PEATVs between that strategy and the No Conversion base case as each of the key inputs vary.

In other words, you would see the same graph, except the Deltas against the base case of the PEATVs will be shown instead of the absolute PEATVs as the key input vary from Low to Base to High values.

So if there are 4 strategies/sizes for Roth conversions, I am thinking about 4 similar graphs.
This sounds like a much more detailed and valuable data set then what we did with similar future scenarios.
Ours was simpler but we did try to model many potential future variables and their impacts with and without Roth conversions - so we had a handle on both the most likely outcomes aas well as the ousiders.
FWIW - our other variables did include things like: earlier age of demise for one spouse, election age for SS, future SS redux or not, the actual tax situation for heirs (high), etc.
Really looks like your approach is greatly superior in just about every way...
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

OP,

I like your approach and look forward to leveraging the results for my own planning.

You might consider adding “Cumulative Taxes Paid” as a metric for the Tornado Diagram. This could signal a need to shift Conversion Strategies if the Tax Equilibrium has changed due to one of your Input variables.


WoodSpinner
WoodSpinner
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

smitcat wrote: Thu Aug 05, 2021 8:57 am This sounds like a much more detailed and valuable data set then what we did with similar future scenarios.
Ours was simpler but we did try to model many potential future variables and their impacts with and without Roth conversions - so we had a handle on both the most likely outcomes aas well as the ousiders.
FWIW - our other variables did include things like: earlier age of demise for one spouse, election age for SS, future SS redux or not, the actual tax situation for heirs (high), etc.
Really looks like your approach is greatly superior in just about every way...
Nothing superior for sure, it appears that we both try to do similar studies but might present the results differently.

Since I wanted to do Sensitivity Analysis on factors beyond our control, your comment gave me 3 additional factors to look at: earlier age of demise for one spouse, future SS redux or not, and tax situation for heirs. The election age for SS would be part of a strategy under our control, so won't need to be part of my Sensitivity Analysis.
John C. Bogle: "Never confuse genius with luck and a bull market".
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

WoodSpinner wrote: Thu Aug 05, 2021 9:39 am You might consider adding “Cumulative Taxes Paid” as a metric for the Tornado Diagram. This could signal a need to shift Conversion Strategies if the Tax Equilibrium has changed due to one of your Input variables.
Is there a reference you can provide so I can read and learn about this? Thanks.
John C. Bogle: "Never confuse genius with luck and a bull market".
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

DSBH wrote: Thu Aug 05, 2021 9:57 am
WoodSpinner wrote: Thu Aug 05, 2021 9:39 am You might consider adding “Cumulative Taxes Paid” as a metric for the Tornado Diagram. This could signal a need to shift Conversion Strategies if the Tax Equilibrium has changed due to one of your Input variables.
Is there a reference you can provide so I can read and learn about this? Thanks.
Here is a link to Kitces’ blog.

And this is a sample of the Scenario comparison:

Image

Understanding the sensitivities are the various inputs was next on my list — Happy to let you take the lead. PM me if you want to divvy the workload.

WoodSpinner
WoodSpinner
User avatar
Lee_WSP
Posts: 10403
Joined: Fri Apr 19, 2019 5:15 pm
Location: Arizona

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Lee_WSP »

DSBH wrote: Tue Aug 03, 2021 3:35 pm
1. Is there any potential key input(s) that should be included in the sensitivity analysis?
2. Instead of my own expectations, I would prefer to use results of statistical analysis for the key inputs, such as expected average annual returns for stocks and bonds, and high/low values that would provide say a 95% confidence level. I am however neither an expert in the finance area nor statistics/quantitative analysis area so I have a lot of homework to do, so suggestions for these values from knowledgeable people are very much appreciated.

Thoughts/comment/feedback/advice on the Sensitivity Analysis presented? Thank you in advance.
The largest variable that will impact the outcome is the rate of return. It has an extremely outsized impact on how effective conversions are.

Have you looked into Monte Carlo analyses? I'd say you're looking at a potential period spanning 30-40 years from age 60 to 90/100.

It may be easier to do the calculations if you assume a real rate of return and therefore zero inflation.

My own back of the napkin calculator shows your 1.5 million TDA will balloon to 2.5 million if you don't draw it down from age 60 to age 72.5 assuming a real rate of return of only 4%. This will then leave your heirs with about a 1.5 million TDA if the transfer date is age 90, 735k if the age is 100. Divided by 10 that's 150k and 73.5k.
Exchme
Posts: 1335
Joined: Sun Sep 06, 2020 3:00 pm

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Exchme »

When you start to study Roth conversions, one of the things to keep in mind is that RPM, like all programs, has limits. It does not know about your existing LTCGs nor does it understand that when you pull money out of your taxable account to fund Roth conversions, you are likely creating a taxable events, unless you have some losers to sell. Since withdrawals from taxable will stop once RMDs kick in and you get a step-up basis on death, ignoring the taxes on capital gains when selling taxable assets is going to artificially favor doing Roth conversions.

For instance, in my case, most of my taxable assets have 30% existing LTCGs and hopefully that will go up over time, so that is several percentage points of new taxes to fund Roths, that seems too big to ignore in an optimization study.

With RPM, funding conversions from the TDA would give you an easier way to make a fair comparison across cases than trying to figure out the taxes on sales of taxable assets. To fund the conversions from the TDA, I would envision you would have to enter a greater conversion amount and also enter the incremental taxes for that conversion as a Roth withdrawal.

Would like to hear what others do with RPM to make the comparison fair.
User avatar
jeffyscott
Posts: 13487
Joined: Tue Feb 27, 2007 8:12 am

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by jeffyscott »

Lee_WSP wrote: Thu Aug 05, 2021 11:47 am
DSBH wrote: Tue Aug 03, 2021 3:35 pm
1. Is there any potential key input(s) that should be included in the sensitivity analysis?
2. Instead of my own expectations, I would prefer to use results of statistical analysis for the key inputs, such as expected average annual returns for stocks and bonds, and high/low values that would provide say a 95% confidence level. I am however neither an expert in the finance area nor statistics/quantitative analysis area so I have a lot of homework to do, so suggestions for these values from knowledgeable people are very much appreciated.

Thoughts/comment/feedback/advice on the Sensitivity Analysis presented? Thank you in advance.
The largest variable that will impact the outcome is the rate of return. It has an extremely outsized impact on how effective conversions are.

Have you looked into Monte Carlo analyses? I'd say you're looking at a potential period spanning 30-40 years from age 60 to 90/100.

It may be easier to do the calculations if you assume a real rate of return and therefore zero inflation.

My own back of the napkin calculator shows your 1.5 million TDA will balloon to 2.5 million if you don't draw it down from age 60 to age 72.5 assuming a real rate of return of only 4%. This will then leave your heirs with about a 1.5 million TDA if the transfer date is age 90, 735k if the age is 100. Divided by 10 that's 150k and 73.5k.
The scenario has a 50/50 portfolio with the TDA being the preferred location for the bond allocation. So the tax-deferred account has little chance of earning such high returns. Since the TDA is 50% of assets, it will be 100% bonds, even if the allocations are tax-adjusted.

It's not really relevant to this discussion, but for my own much simpler analysis, I assume of low returns. I think it is better to optimize for that scenario. If I get much higher than expected returns, I might end up paying more in taxes than the optimal solution might have, but will have more money than expected, anyway.
User avatar
Lee_WSP
Posts: 10403
Joined: Fri Apr 19, 2019 5:15 pm
Location: Arizona

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Lee_WSP »

I like the simplicity of assuming zero or the current ten year yield for returns. If it goes the other way, I'll be richer and it's not a huge problem unless it comes with stagflation.

The other alternative is not really possible to mitigate against as it'd be pretty devastating to suffer losses for so long.
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Lee_WSP wrote: Thu Aug 05, 2021 11:47 am The largest variable that will impact the outcome is the rate of return. It has an extremely outsized impact on how effective conversions are.
Yes at least in this test case the largest impact on the Portfolio Ending Values for both With and Without conversion cases come from the stock rate of return.

Also in this test case the largest impact on the benefit of Roth conversion (With - Without Conversion portfolio ending values) might come from the bond rate of return (more to follow in a coming update, whether correct or not ...).
Have you looked into Monte Carlo analyses? I'd say you're looking at a potential period spanning 30-40 years from age 60 to 90/100.
I thought about taking advantage of the MC variant of BH Barsoom, but I will need to port over the portfolio AA logic so still in thinking mode.

In my previous life long ago sometimes I used @Risk for work so I might think more about something equivalent in the (perhaps not near) future. If that's doable you will see a probability distribution instead of horizontal bar showing the portfolio values as the variable such as stock return moves from Low to Base to High values as of right now.
It may be easier to do the calculations if you assume a real rate of return and therefore zero inflation.

My own back of the napkin calculator shows your 1.5 million TDA will balloon to 2.5 million if you don't draw it down from age 60 to age 72.5 assuming a real rate of return of only 4%. This will then leave your heirs with about a 1.5 million TDA if the transfer date is age 90, 735k if the age is 100. Divided by 10 that's 150k and 73.5k.
That's a pretty good napkin calculator. Using 2% nominal return for bond the T-IRA in this No Conversion case grew from 1.5 Mil @63 to 1.9 Mil @71, and 762K @92 after the RMDs, so heirs' first year withdrawal is 76K.
John C. Bogle: "Never confuse genius with luck and a bull market".
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Exchme wrote: Thu Aug 05, 2021 1:03 pm When you start to study Roth conversions, one of the things to keep in mind is that RPM, like all programs, has limits. It does not know about your existing LTCGs nor does it understand that when you pull money out of your taxable account to fund Roth conversions, you are likely creating a taxable events, unless you have some losers to sell. Since withdrawals from taxable will stop once RMDs kick in and you get a step-up basis on death, ignoring the taxes on capital gains when selling taxable assets is going to artificially favor doing Roth conversions.

For instance, in my case, most of my taxable assets have 30% existing LTCGs and hopefully that will go up over time, so that is several percentage points of new taxes to fund Roths, that seems too big to ignore in an optimization study.

With RPM, funding conversions from the TDA would give you an easier way to make a fair comparison across cases than trying to figure out the taxes on sales of taxable assets. To fund the conversions from the TDA, I would envision you would have to enter a greater conversion amount and also enter the incremental taxes for that conversion as a Roth withdrawal.

Would like to hear what others do with RPM to make the comparison fair.
I inspected the "Net funds added to or (deducted from) Taxable account" on the Details sheet after my first run, then bump up living expenses by what I think will approximate the tax paid by selling from Taxable. Not sure how close the approximations are, but since I am looking at the deltas between the cases hopefully the approximations are "good enough".
John C. Bogle: "Never confuse genius with luck and a bull market".
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Bogleheads,

I have updated the original post with the:

1. Without Roth IRA Conversion case + Sensitivity Analysis,
2. With Roth IRA Conversion (to top of 22% bracket from 2021-2025) case + Sensitivity Analysis,
3. Comparison (Delta) between the With and Without cases + Sensitivity Analysis (of the Deltas).

Best,
DSBH
John C. Bogle: "Never confuse genius with luck and a bull market".
Chip
Posts: 3994
Joined: Wed Feb 21, 2007 3:57 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Chip »

DSBH wrote: Fri Aug 06, 2021 12:25 pm Bogleheads,

I have updated the original post with the:
I haven't dug deeply into the outputs in your initial post, but am wondering if you are tax-adjusting your portfolio allocations each year based on the amounts that are in the Roth vs. TDA vs. taxable. The 50/50 allocation isn't truly 50/50 if you have 1M in stocks in the Roth/taxable and 1M in bonds in the TDA. That could be a big reason why the equity rate of return has such a large effect.

Apologies if you have noted this in the thread and I've missed it.
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Chip wrote: Fri Aug 06, 2021 1:39 pm
DSBH wrote: Fri Aug 06, 2021 12:25 pm Bogleheads,

I have updated the original post with the:
I haven't dug deeply into the outputs in your initial post, but am wondering if you are tax-adjusting your portfolio allocations each year based on the amounts that are in the Roth vs. TDA vs. taxable. The 50/50 allocation isn't truly 50/50 if you have 1M in stocks in the Roth/taxable and 1M in bonds in the TDA. That could be a big reason why the equity rate of return has such a large effect.

Apologies if you have noted this in the thread and I've missed it.
Yes you are correct, the portfolio allocations are not tax-adjusted, and the impact gets larger as the equity rate of return gets larger.

FWIW, these are the numbers for a 50/40 vs 45/55 cases, both non tax-adjusted:

50/50 AA - PEATV:
Stock return: 4%, 6%, 8%
With conversion: 3.8M, 6.7M, 11.0M
Without conversion: 3.3M, 6.1M, 10.4M
Delta (M): 0.5M, 0.6M, 0.6M
Delta (%): 15%, 10%, 6%

45/55 AA - PEATV:
Stock return: 4%, 6%, 8%
With conversion: 3.5M, 5.9M, 9.3M
Without conversion: 3.1M, 5.5M, 9.0M
Delta (M): 0.4M, 0.4M, 0.3M
Delta (%): 12%, 6%, 3%
John C. Bogle: "Never confuse genius with luck and a bull market".
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

Duplicate
Last edited by WoodSpinner on Fri Aug 06, 2021 2:57 pm, edited 1 time in total.
WoodSpinner
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

DSBH wrote: Fri Aug 06, 2021 12:25 pm Bogleheads,

I have updated the original post with the:

1. Without Roth IRA Conversion case + Sensitivity Analysis,
2. With Roth IRA Conversion (to top of 22% bracket from 2021-2025) case + Sensitivity Analysis,
3. Comparison (Delta) between the With and Without cases + Sensitivity Analysis (of the Deltas).

Best,
DSBH
DSBH,

Interesting work, thanks for your effort!

Questions:
1. How exactly are you calculating PEATV for the Portfolio?
2. How does the sensitivity change if the IRA Equity to Bond allocation changes?
3. How does the sensitivity change if you convert to IRMAA Tier 1 or 2?

WoodSpinner
WoodSpinner
Chip
Posts: 3994
Joined: Wed Feb 21, 2007 3:57 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Chip »

DSBH wrote: Fri Aug 06, 2021 2:38 pm Yes you are correct, the portfolio allocations are not tax-adjusted, and the impact gets larger as the equity rate of return gets larger.
Then wouldn't it be fair to say that some/most of the conversion PEATV gains that you're seeing are the result of an increased allocation to equities?
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

WoodSpinner wrote: Fri Aug 06, 2021 2:57 pm Questions:
1. How exactly are you calculating PEATV for the Portfolio?
2. How does the sensitivity change if the IRA Equity to Bond allocation changes?
3. How does the sensitivity change if you convert to IRMAA Tier 1 or 2?
1. PEATV is the total Portfolio Ending After-Tax Value, calculated at the end of the 10-year stretch after heirs have made the final withdrawal to empty out the inherited T-IRA. In the table showing the year by year itemized cash flows of either the WITH or WITHOUT CONVERSION case, it is the value at the bottom right corner (bottom of grey column titled "Yr-End Total").

2. These cases utilize 50% stock / 50% bond for the portfolio AA, desired 100/0 for ROTH IRA and 100/0 for Taxable and 0/100 for T-IRA. Could you be more specific on your desired portfolio AA and individual account AAs?

3. You're ahead of me. I am thinking about putting sensitivities of multiple comparisons (for instance of these 2 - all against the same No Conversion case) on 1 tornado diagram. Will take me a bit of time.
John C. Bogle: "Never confuse genius with luck and a bull market".
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

DSBH wrote: Fri Aug 06, 2021 4:17 pm
WoodSpinner wrote: Fri Aug 06, 2021 2:57 pm Questions:
1. How exactly are you calculating PEATV for the Portfolio?
2. How does the sensitivity change if the IRA Equity to Bond allocation changes?
3. How does the sensitivity change if you convert to IRMAA Tier 1 or 2?
1. PEATV is the total Portfolio Ending After-Tax Value, calculated at the end of the 10-year stretch after heirs have made the final withdrawal to empty out the inherited T-IRA. In the table showing the year by year itemized cash flows of either the WITH or WITHOUT CONVERSION case, it is the value at the bottom right corner (bottom of grey column titled "Yr-End Total").
What discount rates are you using for the IRA, Roth and Taxable?

2. These cases utilize 50% stock / 50% bond for the portfolio AA, desired 100/0 for ROTH IRA and 100/0 for Taxable and 0/100 for T-IRA. Could you be more specific on your desired portfolio AA and individual account AAs?
Sure, how about an IRA with about a 50/50 Equity to bond ratio. Adjust Taxable and Roth as needed. You are modeling a perfectly positioned portfolio and many don’t have that luxury.
3. You're ahead of me. I am thinking about putting sensitivities of multiple comparisons (for instance of these 2 - all against the same No Conversion case) on 1 tornado diagram. Will take me a bit of time.
You are doing great!

Why do you think Bond returns are having such an influence in your model?

WoodSpinner
WoodSpinner
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Chip wrote: Fri Aug 06, 2021 3:08 pm
DSBH wrote: Fri Aug 06, 2021 2:38 pm Yes you are correct, the portfolio allocations are not tax-adjusted, and the impact gets larger as the equity rate of return gets larger.
Then wouldn't it be fair to say that some/most of the conversion PEATV gains that you're seeing are the result of an increased allocation to equities?
All else being equal, increasing either the equity rate of return or the portfolio equity allocation will likely increase the conversion PEATV gain.

However in this comparison, both the WITH and WITHOUT CONVERSION use the same equity rate of return and same portfolio AA (50/50) and same desired individual account AAs. The table showing how individual account AAs are changing between 2 cases while maintaining the portfolio AA at 50/50 is already in the original post, so I have attached below another table showing the itemized deltas to hopefully show where the 10% conversion gain come from. Among the notables:

1. The WITH CONVERSION case ended with 1,200K less in Taxable, because it had:
----- 803K less in (RMD - Roth withdrawal),
----- even when it has 375K less in taxes paid,
----- and paid 11K more in IRMAA,
----- so it has 760K less in earning,
----- resulting in -803K + 375K -11K - 761K = -1,200K less over the 40-yr study period.
2. But more than made up with 1,800K gain in Roth,
3. So -1,200K + 1,800K = 600K gain, or 600K/6.06Mil ~= 10% gain.

Image
John C. Bogle: "Never confuse genius with luck and a bull market".
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

WoodSpinner wrote: Fri Aug 06, 2021 4:54 pm
DSBH wrote: Fri Aug 06, 2021 4:17 pm
WoodSpinner wrote: Fri Aug 06, 2021 2:57 pm Questions:
1. How exactly are you calculating PEATV for the Portfolio?
1. PEATV is the total Portfolio Ending After-Tax Value, calculated at the end of the 10-year stretch after heirs have made the final withdrawal to empty out the inherited T-IRA. In the table showing the year by year itemized cash flows of either the WITH or WITHOUT CONVERSION case, it is the value at the bottom right corner (bottom of grey column titled "Yr-End Total").
What discount rates are you using for the IRA, Roth and Taxable?
Not sure I understand this question. In the base case I used 6% and 2% for stock and bond average annual returns respectively, and the annual returns of each individual account are the results of the AA allocated to that account by the portfolio allocation logic - see answer below for additional details.
2. These cases utilize 50% stock / 50% bond for the portfolio AA, desired 100/0 for ROTH IRA and 100/0 for Taxable and 0/100 for T-IRA. Could you be more specific on your desired portfolio AA and individual account AAs?
Sure, how about an IRA with about a 50/50 Equity to bond ratio. Adjust Taxable and Roth as needed. You are modeling a perfectly positioned portfolio and many don’t have that luxury.
Not sure I understand the "perfectly positioned portfolio", other than the fact that they are just models. The portfolio allocation logic will maintain a 50/50 portfolio AA while trying to meet the desired individual account AAs - but may not satisfy any of the desired individual account AAs. In addition in these cases per my selection the portfolio allocation logic will try to respect the desired 100/0 AA for the Roth account first, then the desired 100/0 AA for the Taxable account next, and the 0/100 desired AA for the T-IRA account last.

In the original post there is a table showing side by side the individual AAs for both WITH and WITHOUT cases if you'd like to inspect them. In both cases the portfolio AAs were maintained at 50/50 but the individual account AAs do vary. BTW the portfolio allocation logic required that (1) the top priority portfolio AA and (2) the desired individual account AAs be specified, as well as (3) which account has priority to receive stock allocation first - Roth or Taxable.
3. You're ahead of me. I am thinking about putting sensitivities of multiple comparisons (for instance of these 2 - all against the same No Conversion case) on 1 tornado diagram. Will take me a bit of time.
Why do you think Bond returns are having such an influence in your model?
In both the WITH and WITHOUT CONVERSION cases the impact of bond returns are much less than either stock returns or living expenses, so I am not surprised. As far as the bond return impact on the Delta in PEATVs between the 2 cases, I will need to dig deep into the resulting individual AA allocations but I suspect that the 50/50 portfolio AA has a lot to do with it. A 90/10 portfolio AA may show a different sensitivity regarding bond returns.
John C. Bogle: "Never confuse genius with luck and a bull market".
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

DSBH wrote: Fri Aug 06, 2021 6:15 pm
Not sure I understand this question. In the base case I used 6% and 2% for stock and bond average annual returns respectively, and the annual returns of each individual account are the results of the AA allocated to that account by the portfolio allocation logic - see answer below for additional details.
Typically to calculate an After Tax Portfolio Value, you Discount IRA by the Marginal Tax Rate for the Year being Calculated. Tricky part (I haven’t figured out how to model it effectively yet) is to also discount the Taxable Account Gains by the LTCG. This the reasoning behind a dollar in Roth > Taxable > IRA.

Apologies if I am missing the obvious but I don’t see how this is calculated.

Not sure I understand the "perfectly positioned portfolio", other than the fact that they are just models. The portfolio allocation logic will maintain a 50/50 portfolio AA while trying to meet the desired individual account AAs - but may not satisfy any of the desired individual account AAs. In addition in these cases per my selection the portfolio allocation logic will try to respect the desired 100/0 AA for the Roth account first, then the desired 100/0 AA for the Taxable account next, and the 0/100 desired AA for the T-IRA account last.
Your model reflects the perfect asset positioning as described in the Wiki, Tax Efficient Asset Placement. It’s the Asset Location I wish I had. I suspect some of your findings (especially the impact of Bond Rates) are due to having 100% bonds in the IRA. It’s a guess at this point and it would be interesting to see how it plays out.

Many of us have more like 50/50 Equity to Bonds on the IRA. My suggestion was to develop an alternative model with a different equity to bond ratio in the IRA. I would still keep the overall portfolio AA at 50/50 but you will have to adjust the amount of money in Roth and Taxable to make it work.

This may be a future iteration to consider. Just a suggestion.

WoodSpinner
WoodSpinner
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

WoodSpinner wrote: Fri Aug 06, 2021 7:45 pm
DSBH wrote: Fri Aug 06, 2021 6:15 pm
Not sure I understand this question. In the base case I used 6% and 2% for stock and bond average annual returns respectively, and the annual returns of each individual account are the results of the AA allocated to that account by the portfolio allocation logic - see answer below for additional details.
Typically to calculate an After Tax Portfolio Value, you Discount IRA by the Marginal Tax Rate for the Year being Calculated. Tricky part (I haven’t figured out how to model it effectively yet) is to also discount the Taxable Account Gains by the LTCG. This the reasoning behind a dollar in Roth > Taxable > IRA.

Apologies if I am missing the obvious but I don’t see how this is calculated.
This is only ONE final value at the end of the 40th year when the Trad-IRA account balance is zero, so just Taxable and Roth IRA. Perhaps I should have said "Portfolio After-Tax and Tax-Free Ending Value" but no discounting is needed for zero balance T-IRA.

Not sure I understand the "perfectly positioned portfolio", other than the fact that they are just models. The portfolio allocation logic will maintain a 50/50 portfolio AA while trying to meet the desired individual account AAs - but may not satisfy any of the desired individual account AAs. In addition in these cases per my selection the portfolio allocation logic will try to respect the desired 100/0 AA for the Roth account first, then the desired 100/0 AA for the Taxable account next, and the 0/100 desired AA for the T-IRA account last.
Your model reflects the perfect asset positioning as described in the Wiki, Tax Efficient Asset Placement. It’s the Asset Location I wish I had. I suspect some of your findings (especially the impact of Bond Rates) are due to having 100% bonds in the IRA. It’s a guess at this point and it would be interesting to see how it plays out.

Many of us have more like 50/50 Equity to Bonds on the IRA. My suggestion was to develop an alternative model with a different equity to bond ratio in the IRA. I would still keep the overall portfolio AA at 50/50 but you will have to adjust the amount of money in Roth and Taxable to make it work.
Using the WITH CONVERSION case as an example, the "perfect asset positioning as described in the Wiki" only occurs in year 1, then to maintain a 50/50 portfolio AA Taxable became 100% bond in year 5, as the AA comparison table in the original post reflects.

Apologies if I still can't explain myself on my own post, all I can do is to try. Thank you for your feedback.
John C. Bogle: "Never confuse genius with luck and a bull market".
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

DSBH wrote: Fri Aug 06, 2021 8:12 pm
WoodSpinner wrote: Fri Aug 06, 2021 7:45 pm
DSBH wrote: Fri Aug 06, 2021 6:15 pm
Not sure I understand this question. In the base case I used 6% and 2% for stock and bond average annual returns respectively, and the annual returns of each individual account are the results of the AA allocated to that account by the portfolio allocation logic - see answer below for additional details.
Typically to calculate an After Tax Portfolio Value, you Discount IRA by the Marginal Tax Rate for the Year being Calculated. Tricky part (I haven’t figured out how to model it effectively yet) is to also discount the Taxable Account Gains by the LTCG. This the reasoning behind a dollar in Roth > Taxable > IRA.

Apologies if I am missing the obvious but I don’t see how this is calculated.
This is only ONE final value at the end of the 40th year when the Trad-IRA account balance is zero, so just Taxable and Roth IRA. Perhaps I should have said "Portfolio After-Tax and Tax-Free Ending Value" but no discounting is needed for zero balance T-IRA.

Not sure I understand the "perfectly positioned portfolio", other than the fact that they are just models. The portfolio allocation logic will maintain a 50/50 portfolio AA while trying to meet the desired individual account AAs - but may not satisfy any of the desired individual account AAs. In addition in these cases per my selection the portfolio allocation logic will try to respect the desired 100/0 AA for the Roth account first, then the desired 100/0 AA for the Taxable account next, and the 0/100 desired AA for the T-IRA account last.
Your model reflects the perfect asset positioning as described in the Wiki, Tax Efficient Asset Placement. It’s the Asset Location I wish I had. I suspect some of your findings (especially the impact of Bond Rates) are due to having 100% bonds in the IRA. It’s a guess at this point and it would be interesting to see how it plays out.

Many of us have more like 50/50 Equity to Bonds on the IRA. My suggestion was to develop an alternative model with a different equity to bond ratio in the IRA. I would still keep the overall portfolio AA at 50/50 but you will have to adjust the amount of money in Roth and Taxable to make it work.
Using the WITH CONVERSION case as an example, the "perfect asset positioning as described in the Wiki" only occurs in year 1, then to maintain a 50/50 portfolio AA Taxable became 100% bond in year 5, as the AA comparison table in the original post reflects.

Apologies if I still can't explain myself on my own post, all I can do is to try. Thank you for your feedback.
I need to think on the first part …..

As for the second part, I believe (and could be wrong) that having 100% bonds in the TIRA is influencing your findings on the sensitivity of bond rates. This my suggestion of using a 50/50 allocation for the IRA and adjusting the values in Roth/Taxable to maintain an overall 50/50 allocation.

If this isn’t clear, perhaps we should take it private and discuss?

WoodSpinner
WoodSpinner
Chip
Posts: 3994
Joined: Wed Feb 21, 2007 3:57 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Chip »

DSBH wrote: Fri Aug 06, 2021 8:12 pm The table showing how individual account AAs are changing between 2 cases while maintaining the portfolio AA at 50/50 is already in the original post
I'm probably not making myself clear. Let me just ask you this:

Do you believe that a portfolio with 1M in bonds in a tIRA and 1M in stocks in a Roth has the same risk profile as a portfolio with 1M in stocks in a tIRA and 1M in bonds in a Roth?

I believe those risk profiles are quite different, and therefore they are not equivalent 50/50 portfolios. To put a number on it, assuming a 25% tax rate, I would say the first portfolio is 57% stocks [1/(1+1*(1-.25))] while the second is 43% stocks [1*(1-.25)/(1*(1-.25)+1)].
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

WoodSpinner wrote: Fri Aug 06, 2021 11:09 pm ...
Many of us have more like 50/50 Equity to Bonds on the IRA. My suggestion was to develop an alternative model with a different equity to bond ratio in the IRA. I would still keep the overall portfolio AA at 50/50 but you will have to adjust the amount of money in Roth and Taxable to make it work.
...
As for the second part, I believe (and could be wrong) that having 100% bonds in the TIRA is influencing your findings on the sensitivity of bond rates. This my suggestion of using a 50/50 allocation for the IRA and adjusting the values in Roth/Taxable to maintain an overall 50/50 allocation.

If this isn’t clear, perhaps we should take it private and discuss?
Not sure if you're thinking about a scenario where most assets are in tax-deferred. If so let's take a quick numerical example where an MFJ couple has 300K in T-IRA, 0K in both Taxable and Roth. The couple specifies a 50/50 portfolio AA, Wiki-like desired 100/0 for Roth and Taxable, and 0/100 for T-IRA to the Portfolio Allocation Logic.

At the beginning of year 1 the Portfolio Allocation Logic tries to allocate 100% stock to Roth first so zero dollars due to zero balance, then tries to allocate 100% stock to taxable so also zero dollars due to zero balance, then allocates the remainder of the required 50% of portfolio = (50% * 300K) - 0 - 0 = 150K of stock to T-IRA, and the remainder 300K-150K=150K to bond, resulting in a 50/50 AA for the T-IRA account.

During year 1 the couple withdraws 25K from T-IRA, converts to Roth, pays 0 tax and let's assume 0% stock/bond return to make calculations simple. At the end of year 1 the Portfolio Allocation Logic tries to allocate 100% stock to Roth first so 25K in stock and 0 in bond, then tries to allocate 100% stock to taxable so zero dollars due to zero balance, then allocates the remainder of the required 50% of portfolio = (50% * 300K) - 25K - 0 = 125K of stock to T-IRA, and the remainder 300K-25-125K=150K to bond, resulting in a 45/55 AA for the T-IRA account - and so forth.

I can't change the logic to maintain both a specified AA for the T-IRA account and the portfolio AA.

If I misinterpreted your comment again we can take it to PM if you'd like.
Last edited by DSBH on Sat Aug 07, 2021 8:37 am, edited 1 time in total.
John C. Bogle: "Never confuse genius with luck and a bull market".
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Chip wrote: Sat Aug 07, 2021 5:54 am
DSBH wrote: Fri Aug 06, 2021 8:12 pm The table showing how individual account AAs are changing between 2 cases while maintaining the portfolio AA at 50/50 is already in the original post
I'm probably not making myself clear. Let me just ask you this:

Do you believe that a portfolio with 1M in bonds in a tIRA and 1M in stocks in a Roth has the same risk profile as a portfolio with 1M in stocks in a tIRA and 1M in bonds in a Roth?

I believe those risk profiles are quite different, and therefore they are not equivalent 50/50 portfolios. To put a number on it, assuming a 25% tax rate, I would say the first portfolio is 57% stocks [1/(1+1*(1-.25))] while the second is 43% stocks [1*(1-.25)/(1*(1-.25)+1)].
Not sure where this is going - and yes these risk profiles are different. I previously agreed in one of my earlier response that the 50/50 AAs in my examples are not tax-adjusted so the portfolio has more stock than 50% if tax-adjusted. I also agreed in a previous response that the impact on Roth conversion gain gets larger as equity allocation gets larger. In other words, if I rework my examples using a tax-adjusted 50/50 portfolio AA, the impact will be less than presented in the original post. The table just shows that the however non tax-adjusted 50/50 AA is being maintained every year throughout the study.
John C. Bogle: "Never confuse genius with luck and a bull market".
Chip
Posts: 3994
Joined: Wed Feb 21, 2007 3:57 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Chip »

DSBH wrote: Sat Aug 07, 2021 8:33 am Not sure where this is going - and yes these risk profiles are different. I previously agreed in one of my earlier response that the 50/50 AAs in my examples are not tax-adjusted so the portfolio has more stock than 50% if tax-adjusted. I also agreed in a previous response that the impact on Roth conversion gain gets larger as equity allocation gets larger. In other words, if I rework my examples using a tax-adjusted 50/50 portfolio AA, the impact will be less than presented in the original post. The table just shows that the however non tax-adjusted 50/50 AA is being maintained every year throughout the study.
DSBH wrote: Tue Aug 03, 2021 3:35 pm Results: Financial Benefit of Roth conversion = WITH CONVERSION Portfolio Ending After-Tax Value (PEATV) minus NO CONVERSION PEATV = varies from 5%-13% under these scenarios, so Converting to the top of 22% from 2021-2025 is not a bad decision.
Honestly, I'm not trying to be argumentative. All I am saying is that the results showing a large benefit to Roth conversions is primarily because the conversion portfolio is riskier and uses a higher return for risky assets. It's not solely due to the conversions themselves, though they do produce a tax arbitrage benefit.

I think that to isolate the benefit of just the conversions it would be necessary to rebalance the portfolios on a tax-adjusted basis every year.
User avatar
jeffyscott
Posts: 13487
Joined: Tue Feb 27, 2007 8:12 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by jeffyscott »

Chip wrote: Sat Aug 07, 2021 9:04 am
DSBH wrote: Sat Aug 07, 2021 8:33 am Not sure where this is going - and yes these risk profiles are different. I previously agreed in one of my earlier response that the 50/50 AAs in my examples are not tax-adjusted so the portfolio has more stock than 50% if tax-adjusted. I also agreed in a previous response that the impact on Roth conversion gain gets larger as equity allocation gets larger. In other words, if I rework my examples using a tax-adjusted 50/50 portfolio AA, the impact will be less than presented in the original post. The table just shows that the however non tax-adjusted 50/50 AA is being maintained every year throughout the study.
DSBH wrote: Tue Aug 03, 2021 3:35 pm Results: Financial Benefit of Roth conversion = WITH CONVERSION Portfolio Ending After-Tax Value (PEATV) minus NO CONVERSION PEATV = varies from 5%-13% under these scenarios, so Converting to the top of 22% from 2021-2025 is not a bad decision.
Honestly, I'm not trying to be argumentative. All I am saying is that the results showing a large benefit to Roth conversions is primarily because the conversion portfolio is riskier and uses a higher return for risky assets. It's not solely due to the conversions themselves, though they do produce a tax arbitrage benefit.

I think that to isolate the benefit of just the conversions it would be necessary to rebalance the portfolios on a tax-adjusted basis every year.
Perhaps you could get some idea of the effect of tax adjusting by doing a comparison with no taxable account. In that case, I think you can just adjust the TIRA dollars based on an assumed 22% marginal tax rate? The example has the couple pretty solidly in that tax bracket, with $120K per year in spending. So say they have $2 mil TIRA, $1 mil Roth. Then for the tax adjusted case, the effective after tax dollars in TIRA are estimated at $1.56 mil., so they would have about $1.64 mil of the TIRA in bonds as their entire bond allocation, in order to be at 50/50 (1.64 - 22% tax = 1.28 = 50% of tax adjusted dollars). When $10,000 of stock in TIRA is converted, $2200 goes to taxes and $7800 to the Roth and the after tax asset allocation does not change. That could be compared to the same but without tax adjusting, so they would start with $1.5 mil in bonds in TIRA. Maybe also compare those two examples to one with a 50/50 allocation in both the TIRA and Roth. This might give some idea of the sensitivity to both asset location decisions and to tax adjustment decisions.

I think the complexity of modeling conversions with different asset allocations in each account is why most studies just assume the same asset allocation across all accounts. So then the problem with all these studies that do that is they are forcing a sub-optimal asset location choice.
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

Chip wrote: Sat Aug 07, 2021 9:04 am I think that to isolate the benefit of just the conversions it would be necessary to rebalance the portfolios on a tax-adjusted basis every year.
I don't disagree with your above statement. Calculating the "the portfolios on a tax-adjusted basis every year" however is an art (more than science imo) that I am not ready to tackle, so I might approximate a tax-adjusted 50/50 portfolio AA with a non-tax-adjusted 45/50 portfolio AA, which results I have posted in an earlier response.

I will add a qualifier to the statement that you quoted: "Results: Financial Benefit of Roth conversion = WITH CONVERSION Portfolio Ending After-Tax Value (PEATV) minus NO CONVERSION PEATV = varies from 5%-13% under these scenarios, so Converting to the top of 22% from 2021-2025 is not a bad decision - assuming a risk preference equivalent to a non-tax-adjusted 50/50 portfolio AA used in these cases."

Cheers.
John C. Bogle: "Never confuse genius with luck and a bull market".
McQ
Posts: 1425
Joined: Fri Jun 18, 2021 12:21 am
Location: California

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by McQ »

DSBH wrote: Sat Aug 07, 2021 9:49 am
How to set up a Roth conversion sensitivity test

Hello DBSH: in this post I hope to go beyond, “you should do it the way I did it in my paper.” But I admit, this will mostly be a repackaging of what you’ve already heard from me, and from chip, FiveK and others. The goal is something more actionable than anything I’ve written here before --a recipe, a set of instructions

First, treat the conversion as a treatment condition within an experiment. To achieve a good experimental design, the treatment condition must vary from the control in only one key respect, and not in any other element. That is the only way to secure a causal attribution. i.e., that this outcome occurred because of this treatment.
  • That is one reason why chip has hammered the asset allocation question: if you simultaneously convert to Roth and alter the asset allocation, there is no way to attribute the outcome to one or the other, or to apportion it between them.
Second, to start, examine only a TDA and the Roth. Including a taxable account at the outset violates good experimental design—too many moving parts.

3. Set the starting value of the Roth at zero. The conversion creates the Roth. Good experimental design because one less contaminating factor.

4. Examine a one time conversion to start. Good experimental design.

5. Set the conversion amount to a small fraction of the TDA and to an arbitrary number likely to fit into a single tax bracket--up to $100,000 for MFJ in the 24% bracket, $80,000 for the 22% bracket, $50,000 in the 12% bracket.

6. In the spreadsheet, compartmentalize the TDA. The larger portion is shared between no-conversion and conversion conditions. However, in the no-conversion condition, the amount of what would have been converted should be set aside in a separate column and tracked (e.g., $100,000). I call this the counterfactual. It is invested in exactly the same asset as the Roth, but is subject to RMDs same as the main portion. The remainder of the TDA can be invested in any balanced allocation that makes sense to you (30/70, 60/40, won’t matter much, see historical asset returns below).

7. Set the age 72 TDA balance (unconverted) at a level that, with anticipated social security/pension, will put you in the expected tax bracket. For instance, $2 million + age 66 max SS will put you low in the 22% bracket 5-10 years out; age 70 max SS & + $4 million will put you low in the 24% bracket 5-10 years out. Of course, a significant pension (civil servant, military), plus the same SS, will reduce the needed TDA balance from $2M/$4M by quite a bit. Basically, by annual pension dollars X 27.3 for the first RMD.

8. Select one return for stocks and for bonds to start. You can get an introduction to centuries of market history here viewtopic.php?t=353607; or you can cut to the chase per Table 1 in my paper, which distills that history to 5.5% real for stocks, 2% real for bonds; set inflation at 3% (typical post-1926) or 2.5% (the average for the past 30 years). Use nominal returns in the spreadsheet (i.e., stocks nominal = 5.5 + 2.5 = 8%); create a table of future inflation-adjusted tax bracket boundaries for computing tax and brackets. Keep it simple: model a conversion no more than 10 years in advance of RMDs beginning.

9. Keep it simple: Pay the conversion tax out of the converted funds. So for a $100,000 conversion, depending on bracket, the Roth starts with $76,000 or $78,000, 24% or 22%.

10. Everything above should be greeted with fairly wide consent; but the next step is controversial. I haven’t convinced FiveK and others of the recommendation below; you will have to make your own decision.

11. Once RMDs start, what do you do with the portion of the counterfactual RMD that remains after tax? If it is spent, I say you have violated good experimental design: on the no-conversion side of the ledger, you have translated financial wealth into consumption wealth, spending power. But there is no offsetting entry on the with-conversion side of the ledger. Any calculation of the Roth conversion “wealth advantage” is therefore contaminated by neglecting the no-conversion “live well” advantage from spending the corresponding RMD.

So either: reinvest the after tax RMD dollars in the same asset as the Roth, thus opening and introducing a taxable account to the model; or remove from the Roth side of the ledger the after tax amount of the RMD, i.e., spend from the Roth too. Either approach conforms to good experimental design. On balance, if the goal of the Roth is a bequest, re-invest the RMD. If the goal was to have discretionary funds available to spend without tax, deduct from the Roth annually.

Hint: outcomes won’t be the same. Hunch: spending from the Roth will retard/reduce the payoff from the conversion, relative to reinvesting the RMDs. Even though reinvested RMDs add a second source of tax drag. But that’s only a hunch at the moment.

If the RMDs are reinvested, extra tax drag will commence (there is already tax drag from the RMD exposure). Account for it. I assumed dividends taxed at the dividend rate, capital gains accumulating untaxed until the end.

Last, run the spreadsheet out to an age of interest. At that juncture, the remaining TDA is liquidated at the then-prevailing tax rate; the re-invested RMDs, if you went that route, have embedded capital gain, which is taxed as such (if alive, and don’t forget NIIT, or the threshold for the 20% rate) or stepped up (if inherited). In my analysis I mostly ignored the shared portion of the TDA, and focused on liquidating the counterfactual and any reinvested RMDs, relative to what was in the Roth, and assessing the dollar surplus achieved through the conversion. See Woodspinner’s thread for a discussion of metrics parallel to your other thread’s discussion of goals.

Don’t do the ten year period granted heirs on the first pass. Good experimental design.

12. For the initial run, keep the ordinary income tax rate exactly the same at every juncture, conversion and thereafter. That’s good experimental design.

Following this recipe, I found that a Roth conversion always paid off satisfactorily if retiree income was in the 22% or 24% bracket, and if the conversion occurred in same, and if the time frame was long enough (30 years after RMDs begin, +/-), and if the future tax rate was not too many percentage points lower than the conversion rate (four, say), given reinvestment of the RMDs.

YMMV.

Also, the Roth conversion was more lucrative under all the scenarios where conventional wisdom would lead you to expect a better outcome: i.e., lower rate paid at conversion, higher future tax rate assessed on RMDs, larger amounts of conversion (if tax rates favorable), or longer span of time before the analysis concluded.

Back to you. Once you have a robust baseline case set up to your satisfaction, you can start varying elements within the above to commence the sensitivity analysis. Change the tax rates, change the location of assets, change the return on assets, change the number of conversions, change the treatment of RMDs, change where the conversion tax is paid from, add an existing taxable account, introduce an existing Roth balance, yada yada. You can also change income and assets to better correspond to your real-life situation.

Change any and all factors that are of interest, but change them One. At. A. Time.

In summary, before exploring alternatives you have to have a clean experimental design to start; and for it to be a good design, it has to be ruthlessly simplified per the above. The initial goal is to experimentally manipulate just one thing: The. Fact. Of. Conversion. Once you’ve built a working model, you can start altering it; as for instance, adding your actual portfolio value, to see whether those real amounts change anything. And if there is a change, the simplified baseline model may help you to pin down the cause.

Best wishes!
You can take the academic out of the classroom by retirement, but you can't ever take the classroom out of his tone, style, and manner of approach.
Topic Author
DSBH
Posts: 740
Joined: Tue Jul 21, 2020 3:31 pm
Location: Texas

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by DSBH »

McQ wrote: Sun Aug 08, 2021 3:57 pm
DSBH wrote: Sat Aug 07, 2021 9:49 am
How to set up a Roth conversion sensitivity test
Thanks so much for your extensive feedback. I'll be away for a week, will think about your suggestions and reply when I get back.

Cheers.
John C. Bogle: "Never confuse genius with luck and a bull market".
smitcat
Posts: 13308
Joined: Mon Nov 07, 2016 9:51 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by smitcat »

McQ wrote: Sun Aug 08, 2021 3:57 pm
DSBH wrote: Sat Aug 07, 2021 9:49 am
How to set up a Roth conversion sensitivity test

Hello DBSH: in this post I hope to go beyond, “you should do it the way I did it in my paper.” But I admit, this will mostly be a repackaging of what you’ve already heard from me, and from chip, FiveK and others. The goal is something more actionable than anything I’ve written here before --a recipe, a set of instructions

First, treat the conversion as a treatment condition within an experiment. To achieve a good experimental design, the treatment condition must vary from the control in only one key respect, and not in any other element. That is the only way to secure a causal attribution. i.e., that this outcome occurred because of this treatment.
  • That is one reason why chip has hammered the asset allocation question: if you simultaneously convert to Roth and alter the asset allocation, there is no way to attribute the outcome to one or the other, or to apportion it between them.
Second, to start, examine only a TDA and the Roth. Including a taxable account at the outset violates good experimental design—too many moving parts.

3. Set the starting value of the Roth at zero. The conversion creates the Roth. Good experimental design because one less contaminating factor.

4. Examine a one time conversion to start. Good experimental design.

5. Set the conversion amount to a small fraction of the TDA and to an arbitrary number likely to fit into a single tax bracket--up to $100,000 for MFJ in the 24% bracket, $80,000 for the 22% bracket, $50,000 in the 12% bracket.

6. In the spreadsheet, compartmentalize the TDA. The larger portion is shared between no-conversion and conversion conditions. However, in the no-conversion condition, the amount of what would have been converted should be set aside in a separate column and tracked (e.g., $100,000). I call this the counterfactual. It is invested in exactly the same asset as the Roth, but is subject to RMDs same as the main portion. The remainder of the TDA can be invested in any balanced allocation that makes sense to you (30/70, 60/40, won’t matter much, see historical asset returns below).

7. Set the age 72 TDA balance (unconverted) at a level that, with anticipated social security/pension, will put you in the expected tax bracket. For instance, $2 million + age 66 max SS will put you low in the 22% bracket 5-10 years out; age 70 max SS & + $4 million will put you low in the 24% bracket 5-10 years out. Of course, a significant pension (civil servant, military), plus the same SS, will reduce the needed TDA balance from $2M/$4M by quite a bit. Basically, by annual pension dollars X 27.3 for the first RMD.

8. Select one return for stocks and for bonds to start. You can get an introduction to centuries of market history here viewtopic.php?t=353607; or you can cut to the chase per Table 1 in my paper, which distills that history to 5.5% real for stocks, 2% real for bonds; set inflation at 3% (typical post-1926) or 2.5% (the average for the past 30 years). Use nominal returns in the spreadsheet (i.e., stocks nominal = 5.5 + 2.5 = 8%); create a table of future inflation-adjusted tax bracket boundaries for computing tax and brackets. Keep it simple: model a conversion no more than 10 years in advance of RMDs beginning.

9. Keep it simple: Pay the conversion tax out of the converted funds. So for a $100,000 conversion, depending on bracket, the Roth starts with $76,000 or $78,000, 24% or 22%.

10. Everything above should be greeted with fairly wide consent; but the next step is controversial. I haven’t convinced FiveK and others of the recommendation below; you will have to make your own decision.

11. Once RMDs start, what do you do with the portion of the counterfactual RMD that remains after tax? If it is spent, I say you have violated good experimental design: on the no-conversion side of the ledger, you have translated financial wealth into consumption wealth, spending power. But there is no offsetting entry on the with-conversion side of the ledger. Any calculation of the Roth conversion “wealth advantage” is therefore contaminated by neglecting the no-conversion “live well” advantage from spending the corresponding RMD.

So either: reinvest the after tax RMD dollars in the same asset as the Roth, thus opening and introducing a taxable account to the model; or remove from the Roth side of the ledger the after tax amount of the RMD, i.e., spend from the Roth too. Either approach conforms to good experimental design. On balance, if the goal of the Roth is a bequest, re-invest the RMD. If the goal was to have discretionary funds available to spend without tax, deduct from the Roth annually.

Hint: outcomes won’t be the same. Hunch: spending from the Roth will retard/reduce the payoff from the conversion, relative to reinvesting the RMDs. Even though reinvested RMDs add a second source of tax drag. But that’s only a hunch at the moment.

If the RMDs are reinvested, extra tax drag will commence (there is already tax drag from the RMD exposure). Account for it. I assumed dividends taxed at the dividend rate, capital gains accumulating untaxed until the end.

Last, run the spreadsheet out to an age of interest. At that juncture, the remaining TDA is liquidated at the then-prevailing tax rate; the re-invested RMDs, if you went that route, have embedded capital gain, which is taxed as such (if alive, and don’t forget NIIT, or the threshold for the 20% rate) or stepped up (if inherited). In my analysis I mostly ignored the shared portion of the TDA, and focused on liquidating the counterfactual and any reinvested RMDs, relative to what was in the Roth, and assessing the dollar surplus achieved through the conversion. See Woodspinner’s thread for a discussion of metrics parallel to your other thread’s discussion of goals.

Don’t do the ten year period granted heirs on the first pass. Good experimental design.

12. For the initial run, keep the ordinary income tax rate exactly the same at every juncture, conversion and thereafter. That’s good experimental design.

Following this recipe, I found that a Roth conversion always paid off satisfactorily if retiree income was in the 22% or 24% bracket, and if the conversion occurred in same, and if the time frame was long enough (30 years after RMDs begin, +/-), and if the future tax rate was not too many percentage points lower than the conversion rate (four, say), given reinvestment of the RMDs.

YMMV.

Also, the Roth conversion was more lucrative under all the scenarios where conventional wisdom would lead you to expect a better outcome: i.e., lower rate paid at conversion, higher future tax rate assessed on RMDs, larger amounts of conversion (if tax rates favorable), or longer span of time before the analysis concluded.

Back to you. Once you have a robust baseline case set up to your satisfaction, you can start varying elements within the above to commence the sensitivity analysis. Change the tax rates, change the location of assets, change the return on assets, change the number of conversions, change the treatment of RMDs, change where the conversion tax is paid from, add an existing taxable account, introduce an existing Roth balance, yada yada. You can also change income and assets to better correspond to your real-life situation.

Change any and all factors that are of interest, but change them One. At. A. Time.

In summary, before exploring alternatives you have to have a clean experimental design to start; and for it to be a good design, it has to be ruthlessly simplified per the above. The initial goal is to experimentally manipulate just one thing: The. Fact. Of. Conversion. Once you’ve built a working model, you can start altering it; as for instance, adding your actual portfolio value, to see whether those real amounts change anything. And if there is a change, the simplified baseline model may help you to pin down the cause.

Best wishes!
A few thoughts .....

"Second, to start, examine only a TDA and the Roth. Including a taxable account at the outset violates good experimental design—too many moving parts."
I believe you need to model the taxable account as well to reasonably simulate taxes and expenses.

"3.Set the starting value of the Roth at zero. The conversion creates the Roth. Good experimental design because one less contaminating factor."
I believe you need to set the Roth to whatever you have to reasonably simulate future value to cover costs and for value to heirs.

"7.Set the age 72 TDA balance (unconverted) at a level that, with anticipated social security/pension, will put you in the expected tax bracket. For instance, $2 million + age 66 max SS will put you low in the 22% bracket 5-10 years out; age 70 max SS & + $4 million will put you low in the 24% bracket 5-10 years out. ....Of course, a significant pension (civil servant, military), plus the same SS, will reduce the needed TDA balance from $2M/$4M by quite a bit. Basically, by annual pension dollars X 27.3 for the first RMD."
Also any other incomes will affect the needed TDA balance to push tax brackets such as larger after tax accounts, real estate, annuities and any inheritance.

"9. Keep it simple: Pay the conversion tax out of the converted funds. So for a $100,000 conversion, depending on bracket, the Roth starts with $76,000 or $78,000, 24% or 22%."
Pay any conversion taxes with the most efficient funds available in your specific case.

"11.Once RMDs start, what do you do with the portion of the counterfactual RMD that remains after tax? If it is spent, I say you have violated good experimental design: on the no-conversion side of the ledger, you have translated financial wealth into consumption wealth, spending power. But there is no offsetting entry on the with-conversion side of the ledger. Any calculation of the Roth conversion “wealth advantage” is therefore contaminated by neglecting the no-conversion “live well” advantage from spending the corresponding RMD."
Model spending to whatever budgeted expenses fit your future goals - just keep those expenses and goals the same between the conversion and non conversion comparisons.

"Following this recipe, I found that a Roth conversion always paid off satisfactorily if retiree income was in the 22% or 24% bracket, and if the conversion occurred in same, and if the time frame was long enough (30 years after RMDs begin, +/-), and if the future tax rate was not too many percentage points lower than the conversion rate (four, say), given reinvestment of the RMDs."
The word always catches my attention here - this statement would assume a bunch of parameters such as couples ages, specific range for SS, pensions, range of funds in any and all accounts, etc.
Exchme
Posts: 1335
Joined: Sun Sep 06, 2020 3:00 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Exchme »

I ignored all Prof. McQ's advice about simplifying the experimental design and still roughly get answers in line with his paper's conclusions that Roth conversions are valuable but work around the edges and can be overdone. In our case, though we will spend many years in the 25/28% bracket and our heirs will probably average about 22%, I can't find a benefit to convert above the 12%/15% bracket.

I did the asset location optimization similar to DSBH's setup and found that the stock/bond asset allocation skew isn't as much as you might think. The case with the most skew was where lots of conversions happened, filling up the TDA with bonds, so the government then owns bonds and you keep stocks elsewhere. However, since the TDA was being depleted by the conversions and the taxable and Roth were growing rapidly, the overall portfolio skew was not that much. In my case, this location optimization cut the optimum Roth conversion almost in half and the reduction in conversions obviously helps with cash flow.

It is better (at least in my case) to fund the Roth conversion out of taxable, but not as wonderful as it first seems. You have to sell stocks in taxable to pay the conversion taxes and those sales generate capital gains. Those are new lifetime taxes versus using the TDA to pay the conversion taxes, which are just accelerated taxes. The RPM tool is blind to the LTCG taxes, so it has to be corrected for outside and added in as an expense (and then of course there are taxes on these taxes) so if you are at a 15% LTCG bracket, you need to take your estimated LTCGs and multiply by 0.15/(1-0.15) to get the true tax burden (still not quite right as the LTCGs actually affect the various MAGI calculations). Accounting for cash flow cut the optimum conversion some more.

To alleviate the cash flow bottleneck, I found a benefit from accelerating distributions from a stretch inherited IRA to the years prior to RMDs, even though that further reduces the available space for Roth conversions (as does qualifying for ACA subsidies in the early years).

I suspect that the great market returns of the last few years made the folks that did big conversions very happy, so I may be overthinking all this.
yog
Posts: 659
Joined: Wed Jan 15, 2020 11:57 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by yog »

Exchme wrote: Sun Aug 08, 2021 8:33 pm I suspect that the great market returns of the last few years made the folks that did big conversions very happy, so I may be overthinking all this.
The last few years have been a once-in-a-lifetime opportunity for us. We were over-prepared and over-optimized if that's even possible, but it's all played out as I had seen in our stochastic planning. We capitalized fully by using the entire 24% bracket for our conversions, but the magnitude of our results could never be planned.

While not specific to Roth conversions, this is a good Kitces' article on the upside potential from sequence of rewards:
https://www.kitces.com/blog/url-upside- ... al-wealth/
McQ
Posts: 1425
Joined: Fri Jun 18, 2021 12:21 am
Location: California

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by McQ »

smitcat wrote: Sun Aug 08, 2021 7:18 pm
A few thoughts .....

"Second, to start, examine only a TDA and the Roth. Including a taxable account at the outset violates good experimental design—too many moving parts."
I believe you need to model the taxable account as well to reasonably simulate taxes and expenses.

"3.Set the starting value of the Roth at zero. The conversion creates the Roth. Good experimental design because one less contaminating factor."
I believe you need to set the Roth to whatever you have to reasonably simulate future value to cover costs and for value to heirs.

"7.Set the age 72 TDA balance (unconverted) at a level that, with anticipated social security/pension, will put you in the expected tax bracket. For instance, $2 million + age 66 max SS will put you low in the 22% bracket 5-10 years out; age 70 max SS & + $4 million will put you low in the 24% bracket 5-10 years out. ....Of course, a significant pension (civil servant, military), plus the same SS, will reduce the needed TDA balance from $2M/$4M by quite a bit. Basically, by annual pension dollars X 27.3 for the first RMD."
Also any other incomes will affect the needed TDA balance to push tax brackets such as larger after tax accounts, real estate, annuities and any inheritance.

"9. Keep it simple: Pay the conversion tax out of the converted funds. So for a $100,000 conversion, depending on bracket, the Roth starts with $76,000 or $78,000, 24% or 22%."
Pay any conversion taxes with the most efficient funds available in your specific case.

"11.Once RMDs start, what do you do with the portion of the counterfactual RMD that remains after tax? If it is spent, I say you have violated good experimental design: on the no-conversion side of the ledger, you have translated financial wealth into consumption wealth, spending power. But there is no offsetting entry on the with-conversion side of the ledger. Any calculation of the Roth conversion “wealth advantage” is therefore contaminated by neglecting the no-conversion “live well” advantage from spending the corresponding RMD."
Model spending to whatever budgeted expenses fit your future goals - just keep those expenses and goals the same between the conversion and non conversion comparisons.

"Following this recipe, I found that a Roth conversion always paid off satisfactorily if retiree income was in the 22% or 24% bracket, and if the conversion occurred in same, and if the time frame was long enough (30 years after RMDs begin, +/-), and if the future tax rate was not too many percentage points lower than the conversion rate (four, say), given reinvestment of the RMDs."
The word always catches my attention here - this statement would assume a bunch of parameters such as couples ages, specific range for SS, pensions, range of funds in any and all accounts, etc.
We don't disagree about where DBSH or any self-modeler needs to end up; eventually, they have to get to the gnarly specifics of their personal situation. Rather, we disagree about how to start for best results.

In my opinion, if you try to start with the full personal model, and then conduct a sensitivity analysis from that start point, you will wander in a hall of mirrors forever. The number of permutations and combinations (this level of SS, this ratio of TDA to taxable to Roth, yada yada) scales with the factorial. If you start with an instance of the full factorial expansion (i.e., your exact personal situation with its nine or ten interacting choices already made), and then proceed to test, say, making two conversions rather than one, by that gesture you will actually have switched from permutation #362,880 to #362,879, making the outcome uninterpretable in terms of the root cause. Unknowingly, perhaps, you had eliminated a two-factor interaction, and all the n-level interactions that included it. Or whatever: you can't locate the key driver when you alter one element of the complete model.

My recommendation to someone like DBSH is to start simple. Once s/he understands the moving parts, alterations can be made one by one and assessed. It is a matter of following branches in the decision tree until they fall short, and then backtracking to the last node, and then trying another branching from that node. The optimal conversion path for an individual case can only emerge slowly, over much, much scenario planning.

And the baseline case has to be radically simplified, or no subsequent inference can ever be secure.

PS: alternatively, as exchme found, you can trust that Roth conversions generally work out, if made in the middle and lower brackets, and if left undisturbed for decades, and if you don't expect great things.
You can take the academic out of the classroom by retirement, but you can't ever take the classroom out of his tone, style, and manner of approach.
smitcat
Posts: 13308
Joined: Mon Nov 07, 2016 9:51 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by smitcat »

McQ wrote: Sun Aug 15, 2021 11:31 pm
smitcat wrote: Sun Aug 08, 2021 7:18 pm
A few thoughts .....

"Second, to start, examine only a TDA and the Roth. Including a taxable account at the outset violates good experimental design—too many moving parts."
I believe you need to model the taxable account as well to reasonably simulate taxes and expenses.

"3.Set the starting value of the Roth at zero. The conversion creates the Roth. Good experimental design because one less contaminating factor."
I believe you need to set the Roth to whatever you have to reasonably simulate future value to cover costs and for value to heirs.

"7.Set the age 72 TDA balance (unconverted) at a level that, with anticipated social security/pension, will put you in the expected tax bracket. For instance, $2 million + age 66 max SS will put you low in the 22% bracket 5-10 years out; age 70 max SS & + $4 million will put you low in the 24% bracket 5-10 years out. ....Of course, a significant pension (civil servant, military), plus the same SS, will reduce the needed TDA balance from $2M/$4M by quite a bit. Basically, by annual pension dollars X 27.3 for the first RMD."
Also any other incomes will affect the needed TDA balance to push tax brackets such as larger after tax accounts, real estate, annuities and any inheritance.

"9. Keep it simple: Pay the conversion tax out of the converted funds. So for a $100,000 conversion, depending on bracket, the Roth starts with $76,000 or $78,000, 24% or 22%."
Pay any conversion taxes with the most efficient funds available in your specific case.

"11.Once RMDs start, what do you do with the portion of the counterfactual RMD that remains after tax? If it is spent, I say you have violated good experimental design: on the no-conversion side of the ledger, you have translated financial wealth into consumption wealth, spending power. But there is no offsetting entry on the with-conversion side of the ledger. Any calculation of the Roth conversion “wealth advantage” is therefore contaminated by neglecting the no-conversion “live well” advantage from spending the corresponding RMD."
Model spending to whatever budgeted expenses fit your future goals - just keep those expenses and goals the same between the conversion and non conversion comparisons.

"Following this recipe, I found that a Roth conversion always paid off satisfactorily if retiree income was in the 22% or 24% bracket, and if the conversion occurred in same, and if the time frame was long enough (30 years after RMDs begin, +/-), and if the future tax rate was not too many percentage points lower than the conversion rate (four, say), given reinvestment of the RMDs."
The word always catches my attention here - this statement would assume a bunch of parameters such as couples ages, specific range for SS, pensions, range of funds in any and all accounts, etc.
We don't disagree about where DBSH or any self-modeler needs to end up; eventually, they have to get to the gnarly specifics of their personal situation. Rather, we disagree about how to start for best results.

In my opinion, if you try to start with the full personal model, and then conduct a sensitivity analysis from that start point, you will wander in a hall of mirrors forever. The number of permutations and combinations (this level of SS, this ratio of TDA to taxable to Roth, yada yada) scales with the factorial. If you start with an instance of the full factorial expansion (i.e., your exact personal situation with its nine or ten interacting choices already made), and then proceed to test, say, making two conversions rather than one, by that gesture you will actually have switched from permutation #362,880 to #362,879, making the outcome uninterpretable in terms of the root cause. Unknowingly, perhaps, you had eliminated a two-factor interaction, and all the n-level interactions that included it. Or whatever: you can't locate the key driver when you alter one element of the complete model.

My recommendation to someone like DBSH is to start simple. Once s/he understands the moving parts, alterations can be made one by one and assessed. It is a matter of following branches in the decision tree until they fall short, and then backtracking to the last node, and then trying another branching from that node. The optimal conversion path for an individual case can only emerge slowly, over much, much scenario planning.

And the baseline case has to be radically simplified, or no subsequent inference can ever be secure.

PS: alternatively, as exchme found, you can trust that Roth conversions generally work out, if made in the middle and lower brackets, and if left undisturbed for decades, and if you don't expect great things.
"And the baseline case has to be radically simplified, or no subsequent inference can ever be secure."
A baseline case that is radically simplified will not come close to representing a real life situation which holds no interest to us.

"PS: alternatively, as exchme found, you can trust that Roth conversions generally work out, if made in the middle and lower brackets, and if left undisturbed for decades, and if you don't expect great things."
This is related to many variables such as the amount of funds you currently have, their location, your fixed incomes (SS,pensions,etc) , current ages etc. Making blanket statements about middle and lower brackets ,whatever those might be , will only fit a certain group.
smitcat
Posts: 13308
Joined: Mon Nov 07, 2016 9:51 am

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by smitcat »

Exchme wrote: Sun Aug 08, 2021 8:33 pm I ignored all Prof. McQ's advice about simplifying the experimental design and still roughly get answers in line with his paper's conclusions that Roth conversions are valuable but work around the edges and can be overdone. In our case, though we will spend many years in the 25/28% bracket and our heirs will probably average about 22%, I can't find a benefit to convert above the 12%/15% bracket.

I did the asset location optimization similar to DSBH's setup and found that the stock/bond asset allocation skew isn't as much as you might think. The case with the most skew was where lots of conversions happened, filling up the TDA with bonds, so the government then owns bonds and you keep stocks elsewhere. However, since the TDA was being depleted by the conversions and the taxable and Roth were growing rapidly, the overall portfolio skew was not that much. In my case, this location optimization cut the optimum Roth conversion almost in half and the reduction in conversions obviously helps with cash flow.

It is better (at least in my case) to fund the Roth conversion out of taxable, but not as wonderful as it first seems. You have to sell stocks in taxable to pay the conversion taxes and those sales generate capital gains. Those are new lifetime taxes versus using the TDA to pay the conversion taxes, which are just accelerated taxes. The RPM tool is blind to the LTCG taxes, so it has to be corrected for outside and added in as an expense (and then of course there are taxes on these taxes) so if you are at a 15% LTCG bracket, you need to take your estimated LTCGs and multiply by 0.15/(1-0.15) to get the true tax burden (still not quite right as the LTCGs actually affect the various MAGI calculations). Accounting for cash flow cut the optimum conversion some more.

To alleviate the cash flow bottleneck, I found a benefit from accelerating distributions from a stretch inherited IRA to the years prior to RMDs, even though that further reduces the available space for Roth conversions (as does qualifying for ACA subsidies in the early years).

I suspect that the great market returns of the last few years made the folks that did big conversions very happy, so I may be overthinking all this.
"I ignored all Prof. McQ's advice about simplifying the experimental design and still roughly get answers in line with his paper's conclusions that Roth conversions are valuable but work around the edges and can be overdone. In our case, though we will spend many years in the 25/28% bracket and our heirs will probably average about 22%, I can't find a benefit to convert above the 12%/15% bracket."

We had completed a number of similar scenarios a fair time back not related to McQ's paper at all and had differing results with Roth conversions. Our most likley scenarios have larger positive values due to Roth conversions and also converting at higher tax brackets. Anyone who does these scenarios will end up with differing conclusions due to their own specific set of numbers and their specific goals. A few reasons why our results would differ jump out after reading your post:
- our heirs tax brackets are higher
- we anticpate higher taxes in general over time, that is with a conservative portfolio performance
- Roth conversions can be paid for with very little tax consequences out of taxable accounts
- no ACA subsidies or reasons to preserve them
I do agree with you that it's really important to run your own numbers and not rely on rules of thumb or others results.
User avatar
celia
Posts: 16774
Joined: Sun Mar 09, 2008 6:32 am
Location: SoCal

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by celia »

Sorry, but I'm just reading this thread now. I don't think the OP considered the criteria that would have the biggest impact on the final portfolio, which is being widowed, and then filing as Single. In the worst case, that means being widowed (either OP or spouse) this year. I don't think you need to test being widowed in each of the following 40 years, but just run a scenario using Single tax brackets. (Of course, the widowed person could always re-marry. :beer )

The next biggest-impact criteria would probably be in changing the average stock and bond growth rates. The Historical Performance Of Stocks And Bonds shows the average growth rate for stocks being 9.59% for stocks and 5.59% for bonds.


McQ wrote: Sun Aug 15, 2021 11:31 pm In my opinion, if you try to start with the full personal model, and then conduct a sensitivity analysis from that start point, you will wander in a hall of mirrors forever. The number of permutations and combinations (this level of SS, this ratio of TDA to taxable to Roth, yada yada) scales with the factorial....

And the baseline case has to be radically simplified, or no subsequent inference can ever be secure.
I agree with this observation. OP will possibly loose even more confidence in their Roth conversion plan the more that different things are taken into account. He/she needs to acknowledge that "life happens" and unexpected things can change the course at any time. If OP wants to get this decided in their lifetime, sticking with Single/MFJ tax brackets and assuming average stock and bond returns is the way to go. Don't worry about things you can't control.
A dollar in Roth is worth more than a dollar in a taxable account. A dollar in taxable is worth more than a dollar in a tax-deferred account.
sc9182
Posts: 2180
Joined: Wed Aug 17, 2016 7:43 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by sc9182 »

celia wrote: Mon Aug 16, 2021 10:34 am Sorry, but I'm just reading this thread now. I don't think the OP considered the criteria that would have the biggest impact on the final portfolio, which is being widowed, and then filing as Single. In the worst case, that means being widowed (either OP or spouse) this year. I don't think you need to test being widowed in each of the following 40 years, but just run a scenario using Single tax brackets. (Of course, the widowed person could always re-marry. :beer )

The next biggest-impact criteria would probably be in changing the average stock and bond growth rates. The Historical Performance Of Stocks And Bonds shows the average growth rate for stocks being 9.59% for stocks and 5.59% for bonds.
.. If OP wants to get this decided in their lifetime, sticking with Single/MFJ tax brackets and assuming average stock and bond returns is the way to go. Don't worry about things you can't control.
Prof. McQ’s recent thread (and challenge) :-) — still going strong, and no one presented a widely substantive argument against his assertion.
Ref : viewtopic.php?f=2&t=355111

With careful planning (and/or utilizing limited Roth convert windows)., possibly with limited insurance., the much feared widow-tax hit has been largely been deemed - “un-necessary fear” — then again — I haven’t run any/many scenarios around Widow (oft depicted in high tax states) tax-hit situation. I defer it to above Prof. McQ thread — if someone willing to take up that challenge :-)
Exchme
Posts: 1335
Joined: Sun Sep 06, 2020 3:00 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by Exchme »

McQ wrote: Sun Aug 15, 2021 11:31 pm PS: alternatively, as exchme found, you can trust that Roth conversions generally work out, if made in the middle and lower brackets, and if left undisturbed for decades, and if you don't expect great things.
I am certainly wandering in Prof. McQ’s factorial hall of mirrors. I marvel at the complexity and capability of the consumer tools and grind my teeth at the remaining gaps. Not only do they fail to predict future life expectancy, investment returns and tax laws :happy , they all lack key parts of the tax code or cash management needed to make real world comparisons. I mostly spend time spackling over gaps and evaluating whether my spackle job is good enough.

For instance, in my case, the final liquidated values only change by only a small amount whether I pay for the taxes on Roth conversions from tax deferred vs. taxable (staying within the same tax brackets/IRMAA tiers pre-RMD, it actually slightly favored paying from TDA). The usual advice is to pay taxes on conversions from taxable as that “converts” some of taxable to a Roth, so I was surprised by this and being in the hall of mirrors, cause attribution is difficult. The key driver seems to be that paying the conversion taxes from taxable entails selling appreciated assets to get the cash. Once RMDs start, there is plenty of cash and at death capital gains get a step-up basis, meaning those capital gains pre-RMD are new taxes vs. using the TDA where we are just accelerating taxes.

I conclude that paying for conversions from taxable is a true investment and if there are a lot of gains in taxable, the payout may not arrive before you depart.

In the same vein of minimizing lifetime capital gains taxes and to illustrate the complexities, I found I should liquidate my inherited stretch IRA and as much of my HSA as I can prior to RMDs and that a good use of the money is to increase Roth conversions.

Now back to the mirrors.
User avatar
WoodSpinner
Posts: 3505
Joined: Mon Feb 27, 2017 12:15 pm

Re: [Updated] ROTH IRA Conversion scenarios - Sensitivity Analysis

Post by WoodSpinner »

All,

This is an excercise in Sensitivity analysis rather than finding an optimal Roth Conversion strategy. In my mind the goal is to identify the few variables that matter most to your plan so you can more carefully track them. In my case, I am starting with a pretty robust model that matches our circumstances.

At least that is how I plan to use the results.

My thoughts are to start with the best model you have and try and be as accurate as possible (given limitations) and change only 1 variable at a time to understand the implications. Be careful that your model is not misleading you (e.g. not dealing with taxes for selling assets in Taxable to pay for conversions).

I strongly suggest that you start by identifying 1 or 2 variables ahead of time to compare results (e.g. Tax Adjusted Portfolio Value, Cumulative Taxes paid etc.)

Next start identifying the variables that you want to test (e.g. US Stock Returns, International Stock Returns, Marginal Tax Rate of Heirs etc.).

Then build a framework to run the model, adjusting the variables, saving the results and comparing outcomes. This part takes a good bit of work and careful tracking. The approach I used for Conversion scenarios works but it is a bit cumbersome (lots of copy and pasting).

I think McQ comes at this from an academic perspective rather than personal financial planning.

I see no reason that this approach wouldn’t provide useful information for our planning.

Results very well be different for each of us! I think that becomes an interesting data point to explore.

Are the models reasonable accurate?
What circumstances lead to a different result?

Just my $.02

WoodSpinner
WoodSpinner
Post Reply