Optimization challenge idea

Just curious if others would be interested to see how/how much various profiles could be optimized. Sort of a challenge or game where we could learn from each other on how to get more out of ESPlanner and potentially to apply these insights for our own profiles. If this worked out, a new challenge could be posted once every couple of months or some other frequency.

For example, profile "X" would be known to us all in sufficient detail, and certain variables are fixed or limited to certain ranges (e.g. inflation, rate of return, demographics, years working, wages, only sell/downsize home within certain ranges/times, maximum indebtedness, etc.). How much could profile "X" consumption be increased? To be consistent, the latest version should be used to cover taxes and various benefits and have apples to apples comparisons. Perhaps the challenge profile(s) could be downloaded so users wouldn't have to program in as many details to get started.

Users could post on how big an increase could be achieved along with details on how to optimize for that specific profile. Or perhaps the optimized profiles could be uploaded for others to view.

Anyway...this isn't thought out in detail, but could be interesting and perhaps help users get more out of the software.

Best,
Brian

Comments

dan royer's picture

That would be fun. . . .

That would be an interesting source of new strategies to consider.

Hopefully others will be interested and join in as they become aware of this. I'd be happy to set up the first couple of these and can work with Dan on any logistics beyond posting.

Part of the untapped power of ESPlanner is in the experience of users across a wide range of scenarios. The more people participate and at a deeper level, we all gain in knowledge and insights.

Best,
Brian

dan royer's picture

Yes, I suppose the way to proceed would be to just create a blank database with the basic profile that we all download--then create a "due date," (say three or four weeks or whatever), and then back here to discuss.

I'm guessing some of the "rules" of the game might get complicated?

That makes sense Dan. Pardon in advance for the lengthy reply below.

The key, at least for me, is to attempt to increase consumption without taking on significant "new" risk beyond that in the default profile. For example, anyone could change the rate of return assumption to 100% or some other unrealistic number, but that makes the results meaningless for serious users. This is the "spirit" of the challenge.

I'll draft a profile and "challenge" and share with others who express interest here for feedback. We could leave the specific rules/guidelines for each challenge to the person who drafts it and will probably have to adjust after the first couple based on our experience to date.

Here are some example variables that I think are reasonable to leave either fixed or within specific constraints:
- Inflation, rate of return, years working, wages/earnings, if/when to sell/downsize home (primary, vacation, or other real estate) and state to which you'll move, maximum indebtedness, current assets, age of last contributions, contribution amounts (or target amounts at retirement), Medicare start age, estate plans/life insurance, property taxes/insurance (may decrease if downsizing), house appreciation rate, 529 plans (or target amounts at year "x" for college), standard of living, cost of children, taxes, demographics, Social Security/Medicare benefits (if/when they will be changed and how much), two can live as cheaply as "x", special expenditures/receipts, etc.

Perhaps that's too strict to start, but the goal is to have close to apples to apples comparisons. Even with all the variables listed above, there are a ton of things that can be adjusted to increase consumption in various combinations.

I'm really hoping to learn and generate some new ideas that can applied in the future. The more attractive this is to users and the more thoughtful participation, the better it is for everyone. If there's enough interest down the road, we could have some sort of actual prizes to build up excitement and participation, but that's probably not an issue yet.

One question (of many) for Dan, could we be allowed to post the profiles here? Right now the only allowed file types on the forum are for images.

Best,
Brian

dan royer's picture

I was wondering if we could just all start with the same database (upload it here) and we each download it. That way we're all working with the same plan to begin with.

I wonder too if there are different kinds of contests here . . . not just highest consumption, but perhaps also best solution to a financial problem such as wife wants to retire as soon as possible or something like that. In the end, it is possible that our various values we express in the case are as interesting to others as who has the highest consumption. I don't if that makes sense.

Totally agree on the database. Makes it much easier. That's what I was trying to get at on uploading the profiles. We need a common database for everyone to start with. Then, we could have people post either their results (and how they did it) or their optimized database. Perhaps it would be best to post results while the contests are in progress and then the final, optimized profile / database, for the "winner" so other users can dig into it if they want to afterwards.

I also agree on the other types of contests. There is a wealth of insights across the ESPlanner community that, IMO, is largely untapped outside of forum posts. I don't think we know what potential contest areas are of most interest to the community, but can reach out to others and experiment over time to see. Areas that are compelling and of sufficient interest should present themselves.

Perhaps it will make sense to set up a new contest-centric web page (as part of the main esplanner.com site) that logged in users can see, post, download / upload, etc. so it doesn't get in the way of the forum?

Best,
Brian

I would be in favor of limiting downside potential but not in favor of stipulating a certain rate of return. Part of the choices you make in retirement is how to invest your assets. Anybody could put 100% in risky assets and, by luck, come out okay. But doing that while also limiting your downside risk is not the same thing.

I suggest a shared database to start with, having a few different families at different income levels.

Hi Chris,

Perhaps we are talking about different things. My 100% comment above (Oct. 15th) was related to rate of return (RoR) and not % of risky assets. How can we compare optimization if one person uses a 100% RoR and another person uses 6%? Of course, the 100% profile will have a much higher consumption, although unrealistic and unachievable in real life.

As you suggest, there are multiple aspects to optimization. Increasing consumption, limiting downside risk along with other non-systemic risk, increasing flexibility, maintaining a sufficient fund of liquid assets, planning for certain life events such as LTC, healthcare, college, etc. all matter. From a practical standpoint, each of us will make different tradeoffs.

My suggestion, to try and have rough apples to apples comparisons, is to have a number of variables either fixed or within certain constraints set by the person proposing each challenge. We'll learn soon enough what's reasonable / practical.

I'm close to having a draft ready and can share with Dan and yourself for feedback. It's a new database/profile and I can either email it to you or post here, if Dan is okay with that.

Best,
Brian

dan royer's picture

I think just post the DB here. I'll make sure that upload is permitted. I think we'll just see how it goes here the first time and then modify our rules if we like next time. I'll add permission to upload .mdb file. You'll need to include a fake earnings history in order to get some SS benefits. You could complicate things a bit by using different ages for H and W. And you could give one a define pension maybe or something like that. (perhaps it's too much to show options for defined pension at different ages). Even if we end up with strategies that are "unfair" to the contest, (e.g., somebody slips in a refinance at some lower rate or some such thing) it will still be illustrative and interesting. I think to keep the easy things across the board, we could stick to 6% on retirement assets and 2% on regular assets?

Posting the database here makes it much easier Dan. Thanks.

My basic plan, for the first draft, is to do much as you suggest. I'll document in more details ASAP and post. This has a married couple with kids. Different ages, different wages and retirement savings for each. A mortgage. College special expenditures. I plan to put in a pension option. As far as the RoR for regular vs. retirement assets, for this profile, I'm assuming they have some regular assets in mutual funds/stocks so am planning higher than 2% for regular assets. Anyway...I'll try to get this up by EOD Sunday.

Best,
Brian

Hi Dan and Chris,

Attached is my draft database and profile for the first optimization challenge. Here's the summary:

Married (John and Sue), live in CA with 25 years left on a mortgage, 2 kids who will go to college (with special expenditures and a 529 plan), different ages, earnings, and retirement contributions for each, $150K in regular assets, $175K in retirement (tax-sheltered) accounts, both retire at age 65, currently planning to take Social Security at full retirement age (67), only 1 change of home allowed (at age 95 for youngest spouse) to free up home equity for end of life expenses (this is fixed), "assumptions" tab has all default values except for rate of return on regular assets (now at 4.5%) and these are fixed with one exception for maximum indebtedness which can range between $0 and $30,000.

There is a small pension ($10K annually) for Sue that is not covered by Social Security and no inflation indexing or survivor benefit. The reason I'm calling this out is that it increases by 5% each year. So users can choose $10K in 2038, $10.5K in 2039 or $11,025 in 2040. The user can pick any option, but must match the format in the database.

Fixed variables include:

- Everything in the "Assumptions" tab with the exception of "maximum indebtedness" which can range between $0 and $30,000
- Earnings, retirement dates, employer retirement contributions (set to $1,000 each per year), current assets, Medicare start age
- Housing (all variables), estate plans, ESPlanner recommendations for life insurance (e.g. use whatever the program recommends)
- Special expenditures for college for both kids ($20K per year * 4 years each)
- Contingent planning is disabled (although the fields have been populated)

Constraints include:
- "Maximum indebtedness" which can range between $0 and $30,000
- Pension option described above

Use release 2.28.0 (most current) and the challenge ends on date XX (I'm thinking 3 or 4 weeks).

Let me know what you think. Even though there are a number of fixed / constrained variables, there are plenty of ways to optimize available. I tried to make it fairly realistic. Hope it isn't too complicated.

Edit - I've updated the database and profile. Please use the version lower in this thread. I'm deleting this old version.

Edit - There is an updated challenge below. Please use that version.

Best,
Brian

dan royer's picture

I'll give it a shot!

Great! Before you get too far into it, let me know if you think I should tweak the profile.

Otherwise, I'll give it a shot too.

I'd say allow 2 changes of home, one within a few years of younger child graduating college. Downsizing when the kids move out is realistic and common. The second change of home would be at an old age, as you describe.

We'll need to specify an agreed interest rate for future mortgages, if borrowing to pay the primary home is useful or necessary. That would be a guess, but we should agree on the guess.

Personally, I'm downsizing now, as my second child recently left for his freshman year. We'll still have designated bedrooms for both until they're really gone, but we need and want less space. We plan on staying in that house (3/2, single family) for about 10 years, after which we anticipate moving to a nice 2/2 condo.

If I could record a third housing change late in life, I would. I'll have to wait until we complete the transaction on the current house before I can update ESP to reflect that. I won't bother with special expenses and income until then. It's not worth the hassle.

Thanks for the feedback Chris.

Why don't we include a 2nd change of home in the next challenge? I considered it here, but was worried that it was too complicated for the first one. I think a lot of people are considering the path you describe above so it would be useful to include.

Best,
Brian

Dan and Chris,

I made a mistake in the profile regarding future wages for both spouses. John's retirement age is set to 12-31-2035 (meaning he worked in 2035), but no wages are shown in the profile for 2035 which are supposed to grow by 1% annually in real terms. So...John should earn $98,591 in 2035 (in today's dollars).

Sue's retirement age is 6-1-2038 (meaning she works the first 5 months of 2038). The old version incorrectly showed she only working the first 5 months of 2037. With a 1% real growth rate, her wages should be $54,909 in 2037 and $23,107 in 2038. I've also updated the future earnings for both in the Social Security tab.

This seemed big enough to update the profile which is attached below. Please use this version and delete the old one. Sorry about that.

Dan - the "grow" button on the "earnings" tab only goes up to the year prior to their retirement dates. I mistakenly assumed it would go through 1 more year which led to the error.

Edit - there is an updated challenge below. Please use that version.

Best,
Brian

dan royer's picture

OK. thanks. So we begin with some liquidity constraint: $31,862 in the early years and then $48,779 from the year 2028 onward. If we are comparing our results for highest and smoothest, do we then need to assume that we are trying to smooth per-adult living standard from 2014 on?

I'm glad you brought that up. There seems to be 3 straight forward ways to measure for this profile:

1. Consumption (combination of lifetime and fairly smooth annual levels or "highest and smoothest")
2. Per adult living standard (highest and smoothest)
3. Lifetime maximum consumption

#2 is probably the fairest option as you suggest.

Since this is our first challenge, if you happen to optimize on a different option, perhaps in a different profile, it would be interesting to capture. I started experimenting with #1 and #3 so will have to adjust for the challenge. I'm assuming we will eventually post the optimized databases here. I'd like to see how you, Chris, and hopefully others use different strategies that I haven't considered.

Best,
Brian

dan royer's picture

Boy, I'm having trouble getting smooth per adult living standard even with some rather radical moves. It may not be possible. If this couple wants smooth living standard, I believe they'd have to severely downsize the home immediately. Perhaps there's another way I'm not imagining?

So OK, I moved them to a whole different side of town. :) I also changed their retirement profile quite a bit, but I now at least have smooth per adult living standard: $44,404. I think I'm cheating! :)

I'm in pretty much the same boat. I was able to smooth consumption to a decent extent, but am running into the same problems that you are on living standard per adult (LSPA) although I'm still experimenting.

Why don't we park this one and I'll come up with option #2? I can have this ready by EOD today.

Hi Dan and Chris,

Sorry about the hassle with the first database. I've edited this somewhat and it is possible to smooth living standard per adult (LSPA) or very close to this although I haven't tried to optimize it. Here's the new summary:

Married (John and Sue), live in CA with 25 years left on a mortgage, 2 kids who will go to college (with special expenditures and a 529 plan), different ages, earnings, and retirement contributions for each, $250K in regular assets, $185K in retirement (tax-sheltered) accounts, both retire at age 65, currently planning to take Social Security at full retirement age (67), only 1 change of home allowed (at age 93 for youngest spouse) to free up home equity for end of life expenses (this is fixed), "assumptions" tab has all default values except for rate of return on regular assets (now at 4.5%) and these are fixed with one exception for maximum indebtedness which can range between $0 and $30,000.

There is a small pension ($10K annually) for Sue that is not covered by Social Security and no inflation indexing or survivor benefit. The reason I'm calling this out is that it increases by 5% each year. So users can choose $10K in 2039, $10.5K in 2040 or $11,025 in 2041. The user can pick any option, but must match the format in the database.

Fixed variables include:
- Everything in the "Assumptions" tab with the exception of "maximum indebtedness" which can range between $0 and $30,000
- Earnings, retirement dates, employer retirement contributions (set to $1,000 each per year) and assumed to go along with $2,000 in individual contributions, current assets, Medicare start age
- Housing (all variables), estate plans, ESPlanner recommendations for life insurance (e.g. use whatever the program recommends)
- Special expenditures for college for both kids ($20K per year * 4 years each)
- Contingent planning is disabled (although the fields have been populated)

Constraints include:
- "Maximum indebtedness" which can range between $0 and $30,000
- Pension option described above
- All retirement contribution amounts can be up to current legal limits. E.g. Roth contributions can be up to $5,500 / year for ages 49 or less and $6,500 / year for ages 50+.

Optimization based on:
Living standard per adult (highest and smoothest). The default profile has a low LSPA of $41,969 for several years and a lifetime LSPA of $3,003,690.

Edit - Use release 2.28.0 (most current) and the challenge ends on Nov. 30th.

This is open to all ESPlanner users so please feel free to join in. The database is attached below.

Best,
Brian

Sorry, gents, but I've had to focus my attention elsewhere. I do hope to participate (and benefit from others ideas) but can't, immediately. I doubt I'll be able to do so before 11/7.

Chris

No worries Chris. Sorry you can't make this one, but we should have more to come. If you get time, please join in.

Best,
Brian

Figured I'd share a progress update. This hasn't been the easiest to optimize. Anyway...here's where I am right now. The figures below are all for Living Standard Per Adult (LSPA).

2014-2022 47,249
2023 47,568
2024 48,185
2025-EOL 51,760

Lifetime LSPA of $3,108,994. It's not too hard to raise the lifetime total by a decent amount, but it skews the early years so this is my smoothest version so far.

If others want to participate (besides Dan or Chris) and can't make the 11/7 date, we can extend it.

Dan - What would be really valuable is an optimization engine with dynamic programming to try 1,000s of options. Perhaps with certain user controlled constraints along with legal limits, etc. Although this was not practical in the past, today it could run in the cloud (for practical reasons). It would be great to be able to have this in either a private ESPlanner-controlled cloud or extremely high confidence in security/ability to permanently delete everything after completion. My assumption is that a decent number of people would pay at least $1,000 to gain the benefits of real optimization, but that's just a guess. Regarding my comments about a download version on the other thread, I still stand by them. I'm just assuming it would be nearly impossible to achieve the level of performance needed to do the optimization described here for users at home.

Best,
Brian

I seem to be tapped out on this one. Here's my best Living Standard Per Adult (LSPA) based on the "highest and smoothest" criteria:

2014-2022 $47,248 (this can be a bit higher, but it lowers the lifetime total)
2023 $47,566
2024 $48,184
2025-EOL $51,778

Lifetime LSPA of $3,109,882

The lifetime total can easily be over $100K higher, but this caused ~20 years of much lower LSPA while younger so didn't seem to be worth the tradeoff.

If anyone needs more time, we can extend the deadline. I'm hoping someone comes up with a way to boost this up a lot more so we can compare approaches!

Best,
Brian

dan royer's picture

I've just been too busy to try this, but I will next week.

Let's push the end date to the end of the month. There's no rush and hopefully Chris and others can join us.

If anyone is interested in participating, please join in so we can all benefit.

This round of the contest ends on November 30th.

Just bumping this to see if anyone is interested in joining this round. If the timing doesn't work out this time, we can try again at some point.

dan royer's picture

Ugh. I just haven't had the free time in the last few weeks. But I'm still interested. Had several piles of papers to grade. . . .

dan royer's picture

OK. this was complicated. I don't like using the max indebtedness feature. :)

My first efforts ended this way:

2014-2024: $41,863 (2K and 3K more than that in the last two years in that series)

2025-2043: $51,316

2043-EOL: $56,247

That column total is 3,188,986 (though I don't think that's the right way to measure the total value of that life thread since there may be some weighting needed to account for future value of the early dollars--but I'm not sure about that).

I guess in the end I came up with this:

2014-22: 46,055
2023: 46,834
2024: 47,434
2025-2044: 50,633
2044-EOL: 51,665

I think you beat me. :(

There's a lot of variables there. I didn't experiment with the pension like perhaps I should have. I could have gotten smoother, but it comes at the expense of higher standard later. I was able to get $47,619 through 2022 but then it levels off at 49,016 through EOL.

Thanks for doing this Dan. I threw the kitchen sink at this as it was more complex than I intended. I'm not a fan of using max indebtedness either, but did use it. I started exactly the same with your first edits (to the exact $). :-)

Here are some of the changes I used:

- Age 70 SS for both, file and suspend for John
- No future funding for 529 (beyond current amount) and expense to optimize withdrawal and taxes
- $30K max indebtedness
- Adjust key ages for retirement withdrawals along with contributions (multiple accounts), special withdrawals and withdrawal order
- Pension start date delayed

There were a lot of variables and some constraints which maybe should have been lessened (e.g. downsizing, regular assets early in life due to college funding, etc.).

If you used some other options, I'm curious to learn about them.

Best,
Brian

I guess we'll close this one out. Thanks for joining Dan! I learned some new things so it was definitely worth it for me.

For what it's worth, this profile (which was entirely made up, but meant to be somewhat realistic) was improved by >12.5% for the first 9 years and >3.5% lifetime. It was also much smoother with <10% gap from lowest to highest living standard per adult compared to >20% for the original profile.

If there is interest in optimizing another profile, I'd be willing to set it up or participate.

Best,
Brian

dan royer's picture

Yes, I learned some things there too. I had not experimented much with indebtedness and I am surprised how often a solution of sorts involves contributing less to retirement--especially when funding the delay of SS.

I've been thinking of a related type challenge--but instead of optimizing so much, try to compare values on a path for a young person--say 30 year old married couple. Let me think about this a bit. But the challenge might be something more like solving a particular problem--how to fund education. It might be easier to solve a specific problem any way we like but require a short paragraph explaining the rationale. It would reveal our biases or values, but that would be interesting too. That way there's no "wrong answer" so much as there is different values to reveal.

Sounds interesting. I'm up for it.

My question for the ESPlanner community is how much is optimization worth to you?

For instance, I probably spent ~5 hours experimenting on the challenge and found >$100K in lifetime benefits along with smoothing the SOL.

That's ~$20,000 benefit per hour of effort :-) Sounds pretty good to me!

With more users participating and sharing insights we all win and can apply these to our own profiles.

Best,
Brian