Total portfolio allocation and withdrawal (TPAW)

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
User avatar
Harry Livermore
Posts: 1370
Joined: Thu Apr 04, 2019 5:32 am

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Harry Livermore »

Hi Ben, nice work. Another terrific tool in the toolbox for the DIY crowd.
Cheers
itsmeagain
Posts: 143
Joined: Sat Jan 09, 2016 7:24 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by itsmeagain »

itsmeagain wrote: Sun Dec 18, 2022 3:07 pm I'm confused about one point, and that is the difference between (the asset allocation of) the total portfolio and (the allocation for) the savings portfolio.
I think I understand better after reading more. The total portfolio converts future income streams (social security, annuties, etc) into bonds, discounting for inflation or not depending on whether they are real or nominal streams.

Absent changes in those income streams, the relevant portfolio for managing things over time is what you call the savings portfolio. The savings portfolio concerns the liquid investments (stocks and nominal bonds) that one can change, depending on preferences, needs, etc.

Let me know if I have it hopelessly wrong, slightly wrong, or just about right.

Thanks again!
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

itsmeagain wrote: Sun Dec 18, 2022 3:07 pm This is an amazing tool. Thank you for building and sharing it.
Thanks! Glad to hear you're liking it.
itsmeagain wrote: Sun Dec 18, 2022 3:07 pm I'm confused about one point, and that is the difference between (the asset allocation of) the total portfolio and (the allocation for) the savings portfolio.

When I look at the supplemental screen showing the asset allocation for the savings portfolio for a sample case, I see the median percentage of stocks fall from about 68% to about 60% from age 65 to 100, with a moderate amount of spread. Meanwhile, the allocation for the total portfolio rises from about 40% to about 60% over that time, with much less spread. So is someone supposed to segregate legacy and non-legacy amounts and accounts?
itsmeagain wrote: Sun Dec 18, 2022 5:51 pm I think I understand better after reading more. The total portfolio converts future income streams (social security, annuties, etc) into bonds, discounting for inflation or not depending on whether they are real or nominal streams.

Absent changes in those income streams, the relevant portfolio for managing things over time is what you call the savings portfolio. The savings portfolio concerns the liquid investments (stocks and nominal bonds) that one can change, depending on preferences, needs, etc.

Let me know if I have it hopelessly wrong, slightly wrong, or just about right.
That's right. The savings portfolio is what's normally referred to as the portfolio—the stocks and bonds in the brokerage account, bank accounts etc. The total portfolio adds to this the present value of future savings and retirement income. The asset allocation of the total portfolio is a better measure of the risk to spending, as described in this post and a shorter version in the learn section. Since we care about risk to spending, it makes sense to target the asset allocation of the total portfolio than the savings portfolio per se. The asset allocation of the savings portfolio is recalculated and adjusted as needed to achieve the target asset allocation of the total portfolio.

The asset allocation of the total portfolio will never exceed the target. But it can be below the target when even 100% stock allocation on the savings portfolio is insufficient to achieve the target risk.

The asset allocation of the savings portfolio will be greater than the asset allocation of the total portfolio when there are any future income streams. The asset allocations of both the savings portfolio and the total portfolio are converging to 60% in the example you gave above because towards the end, the portfolio is becoming increasingly dominated by the legacy portfolio whose asset allocation is 60%. In the last year, there is no future savings and retirement income left, so the asset allocation becomes 60%. Here's an example of this happening (open in incognito mode to prevent overwrite of existing plan):

https://tpawplanner.com/plan?params=KMS ... AL3djPcR6e
itsmeagain wrote: Sun Dec 18, 2022 3:07 pm By the way, I'm using a ceiling and floor for spending. So in very good sequences, I'd guess additional income (including unrealized gains) above the amount needed for the ceiling are shifted toward the legacy allocation. Is that what's going in the model, more or less?
Spending above the ceiling is unspent and stays in the portfolio. This will reduce the risk of retirement spending dropping in future years and also increase the distribution of legacy. Usually, most of the impact will be on legacy because retirement spending is already very high and bumping up against the ceiling. So the extra funds are more likely to end up as extra legacy.
Last edited by Ben Mathew on Sun Dec 18, 2022 8:03 pm, edited 1 time in total.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

Harry Livermore wrote: Sun Dec 18, 2022 5:03 pm Hi Ben, nice work. Another terrific tool in the toolbox for the DIY crowd.
Cheers
Thanks!
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

itsmeagain wrote: Sun Dec 18, 2022 3:19 pm Another feature that I and probably many users would find useful, it would be neat if one could press a button at the end to generate a pdf report that shows the various inputs used (basic and advanced, whether preset or modified) and all 5 screens showing the dropdown menu graphs.

Thanks for considering this option!
Thanks for the suggestion. We've thought about doing something like this. Good to know that you'll find it useful. We'll move it up the feature list.
User avatar
Anon9001
Posts: 1779
Joined: Fri Dec 20, 2019 9:28 am
Location: بھارت

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Anon9001 »

Ben Mathew wrote: Mon Nov 30, 2020 9:39 pm
Hello Ben I am wondering why not just use a mixture of 1/CAPE and Arithmetic Mean of Past Annual Returns to get Long Term Expected Return of Asset in the Spreadsheet? (i.e. the Expected Return is calculated from a 50% Weightage of 1/CAPE and 50% Weightage of Arithmetic Mean of Past Annual Returns). I don’t believe for very long time horizons that valuations play a very important role so I believe some weightage of the Expected Return should be given to Past Arithmetic Mean of Asset due to this.

Thanks.
Land/Real Estate:89.4% Equities:7.6% Fixed Income:1.7% Gold:0.8% Cryptocurrency:0.5%
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

Anon9001 wrote: Mon Jan 02, 2023 3:59 am
Ben Mathew wrote: Mon Nov 30, 2020 9:39 pm
Hello Ben I am wondering why not just use a mixture of 1/CAPE and Arithmetic Mean of Past Annual Returns to get Long Term Expected Return of Asset in the Spreadsheet? (i.e. the Expected Return is calculated from a 50% Weightage of 1/CAPE and 50% Weightage of Arithmetic Mean of Past Annual Returns). I don’t believe for very long time horizons that valuations play a very important role so I believe some weightage of the Expected Return should be given to Past Arithmetic Mean of Asset due to this.

Thanks.
The CAPE regressions described above effectively do this. If the slope from the regressions had been zero (i.e. if 1/CAPE had not been correlated to expected return), then the expected return estimate from the regressions would be equal to the historical average return for all values of 1/CAPE. Since the slope is not zero, it deviates from the historical average return on the basis of current 1/CAPE as dictated by the slope.

Regressions are just averages of the dependent variable (historical returns in this case) conditional on the independent variable (1/CAPE).
Last edited by Ben Mathew on Mon Jan 02, 2023 2:53 pm, edited 1 time in total.
User avatar
Anon9001
Posts: 1779
Joined: Fri Dec 20, 2019 9:28 am
Location: بھارت

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Anon9001 »

Ben Mathew wrote: Mon Jan 02, 2023 2:17 pm The CAPE regressions described above effectively do this. If the slope from the regressions had been zero (i.e. if 1/CAPE had not been correlated to expected return), then the expected return estimate from the regressions would be equal to the historical average return for all values of 1/CAPE. Since the slope is not zero, it deviates from the historical average return on the basis of current 1/CAPE as dictated by the slope.

Regressions are just averaging of the dependent variable (historical returns here) conditional on the independent variable (1/CAPE).
Thanks for replying. I do feel though that this Regression suffers from the fact that there are very few Independent Samples of Long Horizon Returns unless you are looking at regressing against 10 Year Returns. The overlapping samples don’t actually provide improvement due to the fact that CAPE or 1/CAPE is highly autocorrelated across months due to the usage of 10 year average for earnings and also due to the fact that price doesn’t change that much from Month to Month. More info here: https://papers.ssrn.com/sol3/papers.cfm ... id=3142575

What I am thinking is a better approach is probably to use a combination of two methods. First method is add the Historical ERP of US Stocks over LT Bonds (AAR of US Stocks-AAR of LT Bonds) (Probably not very far back in time as Steve Reading pointed out in this thread sometime back that LT Bonds had some weird behaviour before 1952) to 30 Year TIPS Yield to get Current Future 30 Year Real Return. This method assumes that ERP is constant. The second method is just 1/CAPE. I believe taking Arithmetic Mean of the Returns produced by both methods is better than just relying on CAPE but that’s my opinion so feel free to disagree.

Thanks.
Land/Real Estate:89.4% Equities:7.6% Fixed Income:1.7% Gold:0.8% Cryptocurrency:0.5%
GAAP
Posts: 1905
Joined: Fri Apr 08, 2016 12:41 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by GAAP »

Anon9001 wrote: Mon Jan 02, 2023 2:53 pm What I am thinking is a better approach is probably to use a combination of two methods. First method is add the Historical ERP of US Stocks over LT Bonds (AAR of US Stocks-AAR of LT Bonds) (Probably not very far back in time as Steve Reading pointed out in this thread sometime back that LT Bonds had some weird behaviour before 1952) to 30 Year TIPS Yield to get Current Future 30 Year Real Return. This method assumes that ERP is constant. The second method is just 1/CAPE. I believe taking Arithmetic Mean of the Returns produced by both methods is better than just relying on CAPE but that’s my opinion so feel free to disagree.

Thanks.
If you're going to use TIPS for this, you're only going to be able to go back to 1997 -- 25 years of data only.
“Adapt what is useful, reject what is useless, and add what is specifically your own.” ― Bruce Lee
User avatar
Anon9001
Posts: 1779
Joined: Fri Dec 20, 2019 9:28 am
Location: بھارت

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Anon9001 »

GAAP wrote: Mon Jan 02, 2023 4:11 pm If you're going to use TIPS for this, you're only going to be able to go back to 1997 -- 25 years of data only.
I am suggesting to calculate ERP using US LT Bonds. The TIPS Yield is added to the Historical ERP to get a Expected 30 Year Real Return. If you want Expected Nominal Return you can add the current Treasury Yield.
Last edited by Anon9001 on Tue Jan 03, 2023 2:23 am, edited 1 time in total.
Land/Real Estate:89.4% Equities:7.6% Fixed Income:1.7% Gold:0.8% Cryptocurrency:0.5%
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

Anon9001 wrote: Mon Jan 02, 2023 2:53 pm
Ben Mathew wrote: Mon Jan 02, 2023 2:17 pm The CAPE regressions described above effectively do this. If the slope from the regressions had been zero (i.e. if 1/CAPE had not been correlated to expected return), then the expected return estimate from the regressions would be equal to the historical average return for all values of 1/CAPE. Since the slope is not zero, it deviates from the historical average return on the basis of current 1/CAPE as dictated by the slope.

Regressions are just averaging of the dependent variable (historical returns here) conditional on the independent variable (1/CAPE).
Thanks for replying. I do feel though that this Regression suffers from the fact that there are very few Independent Samples of Long Horizon Returns unless you are looking at regressing against 10 Year Returns. The overlapping samples don’t actually provide improvement due to the fact that CAPE or 1/CAPE is highly autocorrelated across months due to the usage of 10 year average for earnings and also due to the fact that price doesn’t change that much from Month to Month. More info here: https://papers.ssrn.com/sol3/papers.cfm ... id=3142575
Yes, 30 year periods overlap a lot, and the standard errors of the estimates will be large. In other words, there is a lot of uncertainty about the estimated slope for 30 year regressions because of the paucity of truly independent 30 year periods even in 150 years of data. Still, this data is what we have, and anchoring on simple regression estimates is a reasonably model-agnostic path through the fog of uncertainty. And the story that emerges from 30 year regressions is broadly consistent with that from shorter periods, so that gives some assurance.

Uncertainty about these estimates is a problem for any method, including the one you are proposing. There is no good solution except more data. The amortization based withdrawals used by TPAW are flexible and will self-correct for mistaken assumptions over time. So the harm caused by a mistaken return assumption is lower than in inflexible spending strategies.
Anon9001 wrote: Mon Jan 02, 2023 2:53 pm What I am thinking is a better approach is probably to use a combination of two methods. First method is add the Historical ERP of US Stocks over LT Bonds (AAR of US Stocks-AAR of LT Bonds) (Probably not very far back in time as Steve Reading pointed out in this thread sometime back that LT Bonds had some weird behaviour before 1952) to 30 Year TIPS Yield to get Current Future 30 Year Real Return. This method assumes that ERP is constant. The second method is just 1/CAPE. I believe taking Arithmetic Mean of the Returns produced by both methods is better than just relying on CAPE but that’s my opinion so feel free to disagree.
I don't see this as a better method. One of the estimates I average over to get the default expected stock return for TPAW planner is the raw (i.e. non-regression based) 1/CAPE. So the difference really is in the first method you are proposing. That method requires assuming a constant equity premium. I would be uncomfortable with that. Equity premiums change over time—they shrink during periods of exuberance when no one wants bonds and everyone is piling into stocks. A regression based estimate looks at 1/CAPE and can tell us when that is happening. It tells us when the stock market is overpriced, and favors a countercyclical asset allocation. I think that's important. But that's a judgment call that you will have to make. Good luck.
User avatar
Anon9001
Posts: 1779
Joined: Fri Dec 20, 2019 9:28 am
Location: بھارت

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Anon9001 »

Ben Mathew wrote: Tue Jan 03, 2023 12:39 am So the difference really is in the first method you are proposing. That method requires assuming a constant equity premium. I would be uncomfortable with that. Equity premiums change over time—they shrink during periods of exuberance when no one wants bonds and everyone is piling into stocks. A regression based estimate looks at 1/CAPE and can tell us when that is happening. It tells us when the stock market is overpriced, and favors a countercyclical asset allocation. I think that's important. But that's a judgment call that you will have to make. Good luck.
Yes it is my opinion so maybe your method is better but it also does suffer from the fact that most of the data is in-sample so in future let’s say CAPE of 50 is new normal then the second method would suffer as I presume it predicts Negative Real Return at this high CAPE value.

This issue can be confirmed by doing a Linear Regression on Shiller’s CAPE value from 1871-1988 and seeing what values of CAPE it would suggest give a Negative/0% Real Return. I suspect the CAPE values it suggests would give a Negative Real Return or close to 0% Real Return are lower than the CAPE values it would suggest would give a Negative/0% Real Return if using data from 1871-2022 as US Shiller CAPE has been consistently higher than Median CAPE since 1990s but I am not sure about this.

Another thing I am wondering is why (1/CAPE) used as the Arithmetic Mean of the Stocks Annual Returns here? I believe 1/CAPE is predicting 10-30 Year Real Return and if I am correct in this assumption than it should be used as a Geometric Mean instead of Arithmetic Mean as Arithmetic Mean does not incorporate compounding of returns and hence it is inaccurate to look at Arithmetic Mean of Annual Returns as Long Term Return of Asset.

If you are using 1/CAPE as Arithmetic Mean of Annual Returns then the Geometric Mean (or Long Term Return) you are assuming will be even lower due to Volatility Drag. I want to understand then why the Geometric Mean of Annual Returns is not adjusted (By adjusted I mean reducing each Annual Return by a certain percentage so that the Arithmetic/Geometric Mean of these Annual Returns match 1/CAPE) to match CAPE instead of adjusting the Arithmetic Mean of Annual Returns to match CAPE.

Thanks.
Land/Real Estate:89.4% Equities:7.6% Fixed Income:1.7% Gold:0.8% Cryptocurrency:0.5%
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

NEW RISK INPUTS IN THE ONLINE PLANNER

The risk section in the online planner has been streamlined.

We have changed how risk is entered in TPAW Planner. Previously the inputs were:

- Asset allocation on the total portfolio
- Spending growth (g)

We have now switched to the inputs used in Merton (1969):

- Relative risk aversion (RRA)
- Time preference

There is a one to one correspondence between these inputs, as shown here. So the change is not substantive. The reason for the move is that it helps create better presets. Merton's formulas give us better combinations of asset allocation and spending tilt to try than trying to guess good combinations from scratch. This has allowed us to remove the separation between presets and custom risk entry, which was kind of clunky. The new system is cleaner and easier to use.

Risk tolerance

The risk aversion input is framed as risk tolerance ranging from 0 to 24. Higher values mean more aggressive allocations.

For those interested in the details, the conversion between the risk tolerance input and relative risk aversion (RRA) is as follows:

Risk tolerance of 0 corresponds to infinite relative risk aversion.
Risk tolerances from 1 to 24 correspond to relative risk aversion of

RRA = 1/exp((x-1)/scale+shift)

where x = risk tolerance
scale = 23/(ln(1/.50)-ln(1/16))
shift = ln(1/16)

Here's a table showing the correspondence:

Image

Time preference

The spending tilt input now is simply the time preference parameter. So the meaning of spending tilt has changed. Previously spending tilt had referred to spending growth (g). Now it's time preference. So now even with spending tilt (time preference) set to zero, spending growth (g) will typically be more than zero. (See examples of this in the post linked to earlier).

The default spending tilt (time preference) has been set to zero—meaning you value present and future consumption equally. You need to change spending tilt from the default zero only if you want to increase/decrease spending growth (g) beyond what is given by Merton's formulas for zero time preference. This is the benefit of switching to these inputs. As you change risk tolerance, g is automatically recalculated to account for precautionary savings. This makes it easier to search through the various combinations of asset allocation and g and find the one that best fits your preferences.

Spending tilt is now located in the advanced section of the risk input.

Edit 1/23/2023: Updated table to reflect new migration.
Last edited by Ben Mathew on Mon Jan 23, 2023 10:32 pm, edited 1 time in total.
longratio
Posts: 12
Joined: Wed Aug 11, 2021 1:19 pm
Location: The Netherlands

Re: Total portfolio allocation and withdrawal (TPAW)

Post by longratio »

Hi Ben,

As always its great to see the tool evolve. I do understand the switch from asset allocation and g to RRA and time preference from a modeling perspective. However for me I somehow have a better feel for risk when I think of a 60/40 portfolio than when I think of a risk tolerance of 15. So I tried to reverse engineer the risk tolerance to an asset allocation using your linked post and the linked Google Sheet.

It looks that a RRA of 1.616 maps to a 60/40 stock/bond portfolio using the default settings of the google sheet. That 1.1616 maps to a risk tolerance of something between 11 and 12 based on the table of your latest post. However the TPAW planner transformed my initial 60/40 to a risk tolerance of 15. I guess I am missing something and I will look at the code how this was done later (spending tilt was not used in above exercise).

Is it perhaps possible to have some kind mapping from asset allocation to risk tolerance in the table of your latest post?

Thanks,
Michiel
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

longratio wrote: Fri Jan 06, 2023 5:42 am Hi Ben,

As always its great to see the tool evolve. I do understand the switch from asset allocation and g to RRA and time preference from a modeling perspective. However for me I somehow have a better feel for risk when I think of a 60/40 portfolio than when I think of a risk tolerance of 15. So I tried to reverse engineer the risk tolerance to an asset allocation using your linked post and the linked Google Sheet.

It looks that a RRA of 1.616 maps to a 60/40 stock/bond portfolio using the default settings of the google sheet. That 1.1616 maps to a risk tolerance of something between 11 and 12 based on the table of your latest post. However the TPAW planner transformed my initial 60/40 to a risk tolerance of 15. I guess I am missing something and I will look at the code how this was done later (spending tilt was not used in above exercise).

Is it perhaps possible to have some kind mapping from asset allocation to risk tolerance in the table of your latest post?

Thanks,
Michiel
The relationship between asset allocation and risk tolerance depends on expected returns, so it will change when expected returns change. We can do a mapping between risk tolerance and asset allocation assuming the suggested expected returns just before the most recent update—4.5% stocks and 1.4% bonds. Variance of log stock returns in the data is 0.02969. Entering these inputs into Merton's formula gives us the following mapping between risk tolerance and asset allocation on the total portfolio:

Image

Excel spreadsheet here.

So a 60/40 allocation on the total portfolio translates to a risk tolerance between 15 and 16, in line with the migration of your risk settings to a risk tolerance of 15. In other words, your asset allocation is still in the vicinity of 60/40 like before.

Though you can always back out your AA like this, keep in mind that the risk that ultimately matters is the risk on your retirement spending and legacy. This risk is expressed as the distribution of retirement spending and legacy shown in the results panel. Focusing on that should give you a clearer and more meaningful view of risk than the 60/40 allocation per se. So I would suggest de-emphasizing the 60/40 allocation target and focusing on the risk to spending and legacy displayed there.

Also, since our risk and time preferences are presumably more stable than market parameters, you won't need to change your risk settings in the planner as much. It will automatically update AA and g based on your stable risk aversion and time preference settings. I think it will work well and be easier to use.

Edit 1/23/2023: Updated table and spreadsheet for new migration.
Last edited by Ben Mathew on Mon Jan 23, 2023 10:33 pm, edited 2 times in total.
User avatar
Anon9001
Posts: 1779
Joined: Fri Dec 20, 2019 9:28 am
Location: بھارت

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Anon9001 »

Ben Mathew wrote: Sat Jan 07, 2023 2:16 am
Hello Ben I am still curious on why in your spreadsheet "TPAW for Accumulaters" the Expected Yearly Return i.e. Arithmetic Mean of Annual Simple Returns is equal to 1/CAPE. Considering 1/CAPE is 10 Year Earnings Yield shouldn't the Geometric Mean of Annual Simple Returns be equal to 1/CAPE? If you are setting 1/CAPE as the AAR then the Geometric Mean will be even lower due to "Volatility Drag".

Thanks.
Land/Real Estate:89.4% Equities:7.6% Fixed Income:1.7% Gold:0.8% Cryptocurrency:0.5%
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

Anon9001 wrote: Sat Jan 07, 2023 2:28 am
Ben Mathew wrote: Sat Jan 07, 2023 2:16 am
Hello Ben I am still curious on why in your spreadsheet "TPAW for Accumulaters" the Expected Yearly Return i.e. Arithmetic Mean of Annual Simple Returns is equal to 1/CAPE. Considering 1/CAPE is 10 Year Earnings Yield shouldn't the Geometric Mean of Annual Simple Returns be equal to 1/CAPE? If you are setting 1/CAPE as the AAR then the Geometric Mean will be even lower due to "Volatility Drag".

Thanks.
Can you link to the exact spreadsheet you're referring to?
User avatar
Anon9001
Posts: 1779
Joined: Fri Dec 20, 2019 9:28 am
Location: بھارت

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Anon9001 »

Ben Mathew wrote: Sat Jan 07, 2023 2:44 am Can you link to the exact spreadsheet you're referring to?
This one from the Bogleheads Wiki:https://docs.google.com/spreadsheets/d/ ... 1235921048
Land/Real Estate:89.4% Equities:7.6% Fixed Income:1.7% Gold:0.8% Cryptocurrency:0.5%
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

Anon9001 wrote: Sat Jan 07, 2023 2:46 am
Ben Mathew wrote: Sat Jan 07, 2023 2:44 am Can you link to the exact spreadsheet you're referring to?
This one from the Bogleheads Wiki:https://docs.google.com/spreadsheets/d/ ... 1235921048
I should clarify that in all the ABW and TPAW spreadsheets I've put up, anything in yellow (required inputs) is the responsibility of the user. The numbers I've entered are only meant to be placeholders. I don't update the spreadsheets with new data, so there's no way I can enter sensible values that don't quickly become out-of-date. The goal was simply to create spreadsheets that people can modify with their own assumptions—not to maintain an up-to-date spreadsheets with current market assumptions. For example, I entered current age 25, retirement age 55, etc. in the yellow cells next to it. All yellow cells are meant to be replaced by the users. And I've written "required inputs in yellow" which I hope conveys this intent.

But you may be referring to these instructions next to the expected real return cells:
Consider basing expected stock returns on earnings yield (E/P) measures such as 1/CAPE. The 20 year TIPS yield is a reasonable choice for expected safe bond return, and the same or slightly higher yield can be used for expected risky bond return.
I used the term "base" because the relationship between 1/CAPE and expected returns is open to interpretation. You can use different methods based on 1/CAPE to arrive at expected returns—regressions, theory, etc.. Note that I did not use the term "base" for the 20 year TIPS because I felt more confident that was a reasonable estimate. But again neither for bonds nor for stocks was the numbers I entered in yellow cells meant to be anything other than placeholders for user inputs.

Hope that clarifies what I was trying to do with the spreadsheets.

But I do maintain return assumptions for the online planner. There, the suggested expected stock returns average over the four lowest of nine estimates (eight regression estimates and the raw 1/CAPE estimate). The regression estimates are purely empirical. The regression estimates are done in log returns and volatility drag added to get the expected returns. So subtracting the drag back out when converting expected returns to expected log returns is appropriate. I take the raw 1/CAPE estimate as the arithmetic return average, so volatility drag is coming out of that as well. Here's how I think about that (your model may be different):

CAPE takes a simple average of the earnings of a firm across ten years. That isn't technically a meaningful thing to do because because it's not accounting for retained earnings. Retained earnings grows the capital of the firm over the ten years. So the profits in later years is coming from a larger capital base than the profits in early years. The simplifying assumption under which the CAPE methodology makes sense is that retained earnings is zero. i.e. All the earnings are being paid out as dividends. Then the firm's capital base is not growing and earnings across years are comparable. So a simple average of earnings across several years would give the expected future earnings of the firm. This would then correspond to the expected arithmetic return and would experience volatility drag. Let's look take an example:

Firm A is priced at $100. It makes $10 of profits each year. It pays all of these profits out as dividends, so its capital stock and price does not grow. If you invest in this firm and reinvest the dividends, you will get a return of 10% each year. After 10 years, the median outcome is that $1 would have grown to $2.59.

Firm B is priced at $100. It makes $20 of profits in some years and $0 in other years. It pays all of these profits out as dividends, so its capital stock and price does not grow. If you invest in this firm and reinvest the dividends, you will get a return of 20% half of the years and 0% the other years. After 10 years, the median outcome is that $1 would have grown to $2.49

Both firms have an expected profit of $10 per year. Both have 1/CAPE of 10%. Firm A puts out regular earnings and firm B's earnings are volatile. The median return in firm B is less than the median return in firm A due to volatility drag. So I think of 1/CAPE as an estimate for the mean arithmetic return which is subject to volatility drag.
Last edited by Ben Mathew on Wed Jan 11, 2023 4:41 pm, edited 2 times in total.
longratio
Posts: 12
Joined: Wed Aug 11, 2021 1:19 pm
Location: The Netherlands

Re: Total portfolio allocation and withdrawal (TPAW)

Post by longratio »

Not sure if this is a glitch but after transformation to the new risk parameters my spending tilt is 0.9% on the overview page but -0.9% (minus 0.9%) on the slider on the advanced section of the Risk details.
bikeeagle1
Posts: 153
Joined: Sat Oct 27, 2018 8:55 am

Re: Total portfolio allocation and withdrawal (TPAW)

Post by bikeeagle1 »

Wow. Thanks to the OP! I just have one question, and I apologize in advance if it's already been answered. When I look at the "Graph Card," and I see the various "Spending During Retirement" lines, are they referencing spending from only the stock and bond portfolio, in addition to pension and Social Security, or are they including SS and pension in the spending?

Thanks!
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

longratio wrote: Sun Jan 08, 2023 5:53 am Not sure if this is a glitch but after transformation to the new risk parameters my spending tilt is 0.9% on the overview page but -0.9% (minus 0.9%) on the slider on the advanced section of the Risk details.
That was a typo on the overview page. Fixed.

Thanks for reporting it!
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

bikeeagle1 wrote: Sun Jan 08, 2023 10:07 am Wow. Thanks to the OP! I just have one question, and I apologize in advance if it's already been answered. When I look at the "Graph Card," and I see the various "Spending During Retirement" lines, are they referencing spending from only the stock and bond portfolio, in addition to pension and Social Security, or are they including SS and pension in the spending?

Thanks!
The Spending During Retirement graph shows total spending from all sources—portfolio withdrawals, Social Security, pensions, etc. We will add some language to clarify this.

We want the planner to stand on its own and be easy to use. We don't expect people to read the thread to understand how to use it. So feel free to ask questions about anything that is unclear in the planner itself. That will help us improve it. We have blind spots we don't know about, so this is actually very useful.
itsmeagain
Posts: 143
Joined: Sat Jan 09, 2016 7:24 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by itsmeagain »

@Ben
It looks like the chart showing the asset allocation of the savings portfolio has now been removed. Even though it doesn't reflect the important contributions of SS, pensions, etc. to the overall portfolio, I found it helpful because that's what one can control by rebalancing, shifting asset allocations in accounts, and the like.
itsmeagain
Posts: 143
Joined: Sat Jan 09, 2016 7:24 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by itsmeagain »

@Ben
Another thought. In many cases, the withdrawal rate under worst-case scenarios shoots up to 50% or higher in the last year for the longest living member of a couple. I think that's because the program takes the maximum lifespan as a limit, rather than a guess. I wonder if you could add a couple of optional constraints. One possibility would be to allow the user to specify a maximum withdrawal rate, say, 10% or 20%. Another might be to cap the withdrawal based on the IRS-mandated RMD table for a given age. Or perhaps some combination of the two, say capped at the RMD times C, where the user could set C to vary between, say, 0.5 and 1.5.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

itsmeagain wrote: Sun Jan 08, 2023 1:41 pm @Ben
It looks like the chart showing the asset allocation of the savings portfolio has now been removed. Even though it doesn't reflect the important contributions of SS, pensions, etc. to the overall portfolio, I found it helpful because that's what one can control by rebalancing, shifting asset allocations in accounts, and the like.
It should still be there. The asset allocation graph displayed in the results panel is the asset allocation on the savings portfolio. It should match the asset allocation displayed in the Tasks panel and the risk section. Let me know if you are seeing something different.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

itsmeagain wrote: Sun Jan 08, 2023 2:03 pm @Ben
Another thought. In many cases, the withdrawal rate under worst-case scenarios shoots up to 50% or higher in the last year for the longest living member of a couple. I think that's because the program takes the maximum lifespan as a limit, rather than a guess. I wonder if you could add a couple of optional constraints. One possibility would be to allow the user to specify a maximum withdrawal rate, say, 10% or 20%. Another might be to cap the withdrawal based on the IRS-mandated RMD table for a given age. Or perhaps some combination of the two, say capped at the RMD times C, where the user could set C to vary between, say, 0.5 and 1.5.
Yes, because we are assuming a fixed max age, the withdrawal rate shoots up towards the end, becoming 100% in the last year of life if there is no legacy target. There are two features I plan to add that will take care of this in different ways:

1. Update max age using life expectancy tables

Update expected lifespan each year based on life expectancy tables. So you don't actually ever have 1 year left. As you get older, years remaining shrink but not one-for-one. This would then work the same way as the RMD tables you suggested, except the amortization will be based on the expected return of the portfolio whereas the RMD tables assume an expected real return of zero. We can also allow the user to customize the life expectancy table by specifying a higher or lower life expectancy than the average population. That will let the user control how aggressively to draw down.

2. Annuitization

Have the option to convert bonds to annuities according to some schedule in the latter part of retirement. That will reduce longevity risk. If you specify a legacy goal, then withdrawals from the portfolio will go down to allow for legacy while retirement spending is increasingly financed by the annuity payments.
itsmeagain
Posts: 143
Joined: Sat Jan 09, 2016 7:24 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by itsmeagain »

Ben Mathew wrote: Sun Jan 08, 2023 4:58 pm
itsmeagain wrote: Sun Jan 08, 2023 1:41 pm @Ben
It looks like the chart showing the asset allocation of the savings portfolio has now been removed. Even though it doesn't reflect the important contributions of SS, pensions, etc. to the overall portfolio, I found it helpful because that's what one can control by rebalancing, shifting asset allocations in accounts, and the like.
It should still be there. The asset allocation graph displayed in the results panel is the asset allocation on the savings portfolio. It should match the asset allocation displayed in the Tasks panel and the risk section. Let me know if you are seeing something different.
Ah yes, my mistake. I see you removed the asset aocation for the "total portfolio", which included inferred values for SS, pensions, etc. (I agree with removing that one -- it could cause potential confusion since it's not something the investor directly controls or normally would see.)
itsmeagain
Posts: 143
Joined: Sat Jan 09, 2016 7:24 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by itsmeagain »

Ben Mathew wrote: Sun Jan 08, 2023 5:22 pm
itsmeagain wrote: Sun Jan 08, 2023 2:03 pm @Ben
Another thought. In many cases, the withdrawal rate under worst-case scenarios shoots up to 50% or higher in the last year for the longest living member of a couple. I think that's because the program takes the maximum lifespan as a limit, rather than a guess. I wonder if you could add a couple of optional constraints. One possibility would be to allow the user to specify a maximum withdrawal rate, say, 10% or 20%. Another might be to cap the withdrawal based on the IRS-mandated RMD table for a given age. Or perhaps some combination of the two, say capped at the RMD times C, where the user could set C to vary between, say, 0.5 and 1.5.
Yes, because we are assuming a fixed max age, the withdrawal rate shoots up towards the end, becoming 100% in the last year of life if there is no legacy target. There are two features I plan to add that will take care of this in different ways:

1. Update max age using life expectancy tables

Update expected lifespan each year based on life expectancy tables. So you don't actually ever have 1 year left. As you get older, years remaining shrink but not one-for-one. This would then work the same way as the RMD tables you suggested, except the amortization will be based on the expected return of the portfolio whereas the RMD tables assume an expected real return of zero. We can also allow the user to customize the life expectancy table by specifying a higher or lower life expectancy than the average population. That will let the user control how aggressively to draw down.

2. Annuitization

Have the option to convert bonds to annuities according to some schedule in the latter part of retirement. That will reduce longevity risk. If you specify a legacy goal, then withdrawals from the portfolio will go down to allow for legacy while retirement spending is increasingly financed by the annuity payments.
+2, meaning I like both options!
longratio
Posts: 12
Joined: Wed Aug 11, 2021 1:19 pm
Location: The Netherlands

Re: Total portfolio allocation and withdrawal (TPAW)

Post by longratio »

I think there is a glitch on the my partner age overview settings.
It shows my settings for current age (67) and max age (98) instead of underlying settings section of my partner which are 66 and 97.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

longratio wrote: Thu Jan 19, 2023 3:51 am I think there is a glitch on the my partner age overview settings.
It shows my settings for current age (67) and max age (98) instead of underlying settings section of my partner which are 66 and 97.
Fixed. Thanks for reporting the bug, Michiel.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

I made a mistake in the original migration of the risk inputs. I used expected log returns in the formulas instead of expected returns. Risk inputs have been migrated again to fix this. Like the original migration, there is no substantive change. There is only a change in the mapping from risk aversion and time preference to asset allocation (AA) and spending growth (g). The migration will preserve your underlying AA and g. (But keep in mind that if you have set your expected return assumptions to "suggested", your AA and g will have changed a little due to the automatic expected return updates.)

Tables in the original post and this follow-up post have been updated to show the new mapping.
Circe
Posts: 202
Joined: Thu Apr 21, 2022 7:16 pm

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Circe »

Very nicely done tool. Thanks so much. I had made my own spreadsheets but this tool is quite simple to use.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

Circe wrote: Tue Jan 24, 2023 8:43 pm Very nicely done tool. Thanks so much. I had made my own spreadsheets but this tool is quite simple to use.
That's great to hear. Thanks for the feedback.
longratio
Posts: 12
Joined: Wed Aug 11, 2021 1:19 pm
Location: The Netherlands

Re: Total portfolio allocation and withdrawal (TPAW)

Post by longratio »

Hi Ben,

One thing I really like about the spreadsheet version of TPAW is the overall equation:

Code: Select all

                                             Wealth = spending
Savings portfolio + present value of future savings = present value of essential expenses + risk portfolio

Risk portfolio = savings portfolio + present value of future savings - present value of essential expenses
               = savings portfolio + present value of net future savings
Do you think you can get these numbers somewhere in the web version of the tool?
Especially the NPV of future savings and essential expenses is helpful for me as it gives a good indication how much of our expenses are covered by pension/social securities etc.
User avatar
Topic Author
Ben Mathew
Posts: 2210
Joined: Tue Mar 13, 2018 11:41 am
Location: Seattle

Re: Total portfolio allocation and withdrawal (TPAW)

Post by Ben Mathew »

longratio wrote: Sun Jan 29, 2023 12:48 pm Hi Ben,

One thing I really like about the spreadsheet version of TPAW is the overall equation:

Code: Select all

                                             Wealth = spending
Savings portfolio + present value of future savings = present value of essential expenses + risk portfolio

Risk portfolio = savings portfolio + present value of future savings - present value of essential expenses
               = savings portfolio + present value of net future savings
Do you think you can get these numbers somewhere in the web version of the tool?
Especially the NPV of future savings and essential expenses is helpful for me as it gives a good indication how much of our expenses are covered by pension/social securities etc.
Thanks for the suggestion, Michiel. That will be a neat feature and should be easy enough to do. I will add it to the list.
Post Reply