@smatthews51 I Fully appreciate the efforts put into the existing PRC UM but frankly every use seems to result in more frustration than better understanding. Certainly helpful in some cases but overall just something about the style, content seems to miss the mark.
Realize this is not a great response to your request for improvement suggestions. Good documentation is certainly about technical coverage of features but to be effective, "How" the information is presented can also be critical - It can be an art as much as a science.
Just one person's opinion - Might be a worthwhile exercise to start a dedicated thread to ask the Forum for feedback, suggestion...
@bo3b OK, thanks. One of the things I'm doing with the manual for the on-line tool is to organize the information more along the lines of how to use the tool to accomplish a particular thing rather than just a page-by-page explanation. Plus, it'll be integrated into the tool. Also, I think I'm going to develop a companion document with use cases that may replace the examples in the current manual. I like your suggestion of a dedicated thread on the Forum, too, so will get that going soon.
Thanks again,
Stuart
This issue has come back with the 2024 version of the software. I am getting a much better Roth conversion result manually than from the Optimization algorithm. See attached pdf. The top figure is the manual conversion compared to no conversion. The bottom graph show the manual conversion compared to the optimization run for a 20 year duration. I ran the same data with 2023 v4.1.1 and the optimization seems to be working correctly, as it yields essentially the same result as the the manual method. Can email the files if needed.
Thanks
@smatthews This topic is also recently discussed in the thread topic "Asset Allocations in Tabular Projections" by @ricke and @rrkaushik. @ricke notes that performing the optimization in Asset Allocation Mode 1 vs. Mode 2 produces very different results due to the math from asset allocation shifts over time. If true, @ricke further notes that using Mode 1 may lead to "terrible decisions". Therefore, doing a conversion using allocation Mode 2 is suggested.
I do indeed get very different optimization targets in Mode 1 vs Mode 2, with Mode 2 providing similar estimates to a (crude) manual method of estimating Roth conversions. (For example, see: http://www.youtube.com/@NumberCrunchNerds )
Stuart: Any comment on which allocation mode is preferred (more accurate) for Roth Conversions? If there is a preferred Mode, should that be mentioned in the manual?
Doing studies to optimize things like Roth Conversions when using Mode 1 while holding different asset allocations in various types of accounts is the issue. It is not a Pralana problem, it's just a bad science experiment, where you failed to control the relevant variables. You end up measuring the mixed effect of Roth Conversions and an asset allocation shift to more stocks. That will make it appear to the bad scientist that there is an enormous benefit to Roth Conversions and that's where the "terrible" part comes in, if you do a bad comparison, you get a bad result. You are free in reality to hold differing allocations in different accounts, but don't expect to be able to make meaningful comparisons that way.
Using Mode 1 when the asset allocations in all accounts is equal (often called "mirrored" allocations) is the gold standard of knowing your overall asset allocation and therefore your risk. Pralana allows you to reset the allocation 4 times in the model so you can even approximate a glide path.
Mode 2 (many writers call it "tax efficient asset location") allows you to hold different allocations in different accounts and many writers recommend it. This reduces current taxes by putting bonds in your IRA instead of taxable or Roth and increases growth by putting stocks in Roth.
When we do this and use a constant projection of returns, we see that we end with more money using Mode 2 than the same allocation using Mode 1 - Hooray, it's magic! However, if you test a bad historical sequence (Analysis-Run Analysis-then click the button to enable a historic sequence and choose a poor one such as 1966 as the starting year.), then you end up with less money than Mode 1 - Uh-oh, not magic, more risk!
That extra risk is the source of the extra return. When we load our IRAs up with bonds, we are effectively foisting a portion of them on the government (since we owe some fraction of that account in taxes). Meanwhile the stocks we hold in Roth are all ours, so it's effectively an asset allocation shift towards stocks. We could get exactly the same final estate value using Mode 1, equal allocations in all accounts, with some higher allocation to stocks. That equivalence actually holds up pretty well in both bull and bear markets. Pralana is such a powerful tool that you can actually check this yourself.
I happen to currently hold something in between equal allocations in all accounts and the "tax efficient" Mode 2. So I check Roth Conversions with equal allocations in Mode 1 (which will generally favor more Roth Conversions) and with "tax efficient" Mode 2 (which will generally favor smaller conversions) and make the current year's conversion somewhere in between.
@ricke Thanks for your explanation and insight/experience on this important aspect of PRC! Your approach looks very reasonable. I must admit that I would not have appreciated this just based on the manual text. Perhaps a paragraph in the manual on "Roth Conversions and Allocation Mode" would be appropriate to explicitly point out the compromises between the allocation approaches and an example of their application (along the lines of your approach mentioned in the last paragraph above)?
@wallace471 I will take it as an action item to elaborate on this topic in the user manual (and may reach out to @ricke to write it!).
Stuart
@lmolino This case did indeed uncover a failure mode in the new algorithm which has since been resolved in v1.3.
Stuart
@ricke, @smatthews51, Richard, Thanks for pointing that out and the explanation. I have struggled with that the past few years and ended up going to Mode2 for the reasons you point out. We need to keep in mind that we all may have different goals for our Roth accounts. I look at them from a legacy perspective and want to have it 100% invested in equities. Others may view it as a future way to manage tax brackets and may not want all equities. These are such personal topics that it is hard to make blanket statements. Thanks again - I have thought about starting a thread on just modeling ideas, pitfalls and tips. Maybe that is something you could start.
We should be careful here and not make Mode-1 an equivalent option to Mode-2. I believe that mode-1 exist in PRC just because of historical reasons. Initially it was easier to implement Mode-1 than a real Asset Allocation across all accounts (mode-2) which is the common industry understanding for Asset Allocation. Once Mode-1 established in the tool and people used it, it was difficult to remove it when Mode-2 introduced.
I do not see value in Mode-1 in real life. Keeping the same Asset Allocation in all accounts will eliminate Asset Location which is an important tax tool. Not keeping the same Asset Allocation in all accounts in Mode-1 may cause PRC users to make critical financing mistakes, change their investment risk without them knowing and be less tax efficient.
Hello experts - admittedly complete newbie to modeling here so I'm a bit lost with this highly technical discussion. Would appreciate some guidance. Attaching a screenshot of the two different modes.
1. Are we saying that in Mode 1, the allocations for regular, tax-deferred and Roth must be equal for optimization to make mathematical sense? If yes, I'm missing the point. I thought the whole idea was to have *different* allocations per account for tax reasons, resulting in a *weighted* overall allocation to match your risk profile, say 70/30 or 60/40.
2. In Mode 2, if I specify say a 60/40 *overall* allocation but say 100/0 in Roth account and various other allocations per account, how does the tool ensure that the weighted allocation is indeed 60/40?
3. Am I correct in concluding that using Mode 2 is standard best practice unless it's something very specialized?
Sorry if these are questions answered elsewhere - I checked this forum, Bogleheads and the manual but couldn't quite figure it out. Thanks in advance for any explanations.
Hello Ram; I am not an expert, not in the tool and not in investing/planning; but here is my opinion.
- Your observation is correct. The reason to have the allocations equal is to avoid changing of the overall allocation (which is changing the overall risk/return). As you said, by doing so you will lose the tax benefits of different allocations per account types (Asset Location). I do not see a benefit in it just to make the tool work.
- “How”, is a question for the tool developer. The manual explains the difficulties to achieve all conditions set by the user (which are impossible sometimes). The highest priority is the overall allocation and then the priority determined by the order you will place the other accounts’ allocations (put Roth 100/0 in the first column if it is the most important for you). The last accounts’ settings in the priority list may not be fully achievable.
- You are correct in my opinion.
Hello Gaby, I really appreciate your explanations. (and in my eyes, you are certainly an expert power-user!) We'll see if Stuart or others weigh in on this topic and if the new upcoming web version explains this better.
Your point on Mode 1 seems to completely defeat the purpose and skew the likely skew projected cash flows, withdrawals etc. I will switch my approach to Mode 2 and try to adjust the priorities of the different accounts per your suggestion. I can live with the modeling engine trying to achieve the best possible optimization without being mathematically perfect but would be good to get some idea of what it *did* achieve and what it simply *couldn't* given the various constraints. Thanks again!
As Gaby suggested, Mode 1 was much easier to implement than Mode 2 and, therefore, that’s the way PRC was built initially. The realization that many people wanted to model several asset classes made this an even clearer design choice. Mode 2 is a challenging implementation with just two asset classes (stocks and bonds) and doing it for up to 10 asset classes is well beyond the complexity that can be reasonably accommodated in an operational Excel model, particularly with Monte Carlo analysis. Hence, Mode 1 was and still is part of the tool. With that said, you can manually approximate Mode 2 with Mode 1. Here’s how:
First, you need to set up the tabular projections so you can see the overall asset allocation. You do that via the Tabular Allocations > View Mgmt page. On the far right of that very wide page, you’ll find the relevant columns; there’s one column for each of up to 10 asset classes and they’re populated with the overall allocation for each year in the modeling period. For one of your Views (the default wide view already has some these columns checked for the Allocations view page), put a checkmark in each of the columns associated with your asset classes and that will cause those columns to appear in that Tabular Projection > View page. This is now your primary feedback mechanism on your overall asset allocation across all accounts.
Then, go to the Financial Assets > Asset Classes & Allocations page and enter your asset allocations for each account in an attempt to balance your risk and desired tax efficiency. As you do this, you can click the little diamond-shaped icon near the top of the allocation table to see a graph that gives you a quick look at the resulting overall allocation for your asset classes. You can also visit the tabular projections page to see a detailed year-by-year output (as discussed above). Your overall allocation will change over time as the account balances vary. To compensate for this, PRC provides up to five time periods to allow you to vary your account-level allocations over time (again, on the Financial Assets > Asset Classes & Allocations page). So, this necessarily has to be an iterative process to arrive at the desired solution. Sure, this is a pain, but it will allow you to approximate Mode 2 and with more than just the stocks and bonds asset classes.
Mode 2 does this process dynamically (but only for two asset classes) and changes the account-level allocations each year to achieve a desired overall asset allocation, say 60% stocks and 40% bonds. You asked: “In Mode 2, if I specify say a 60/40 *overall* allocation but say 100/0 in Roth account and various other allocations per account, how does the tool ensure that the weighted allocation is indeed 60/40?”
A fundamental requirement of Mode 2 is that you specify the desired overall allocation to stocks and then place your accounts in priority order. Then you specify how you would like to see stocks and bonds distributed across your accounts (i.e., your desired target locations). With those inputs as given, the PRC Mode 2 algorithm does its thing. It starts with the highest priority account and attempts to allocate the desired percentage of stocks to that account. If that account balance x the desired percentage of stocks (let’s call this product “a1”) is less than or equal to your overall portfolio balance x the overall stock allocation (let’s call this product “b1”), then that first account gets the desired target percentage of stocks. Its bonds allocation is simply 100% - the stock allocation. Then, we subtract “a1” from “b1” to get “b2”, and this represents the currently unallocated stock portion of the overall portfolio balance. Then, this process is repeated for the next-highest-priority account. If “a2” is less than “b2”, then the second account gets the desired target amount of stocks and the algorithm moves on to the next account. When this process gets to the point at which the “a” value exceeds the “b” value, then the stock allocation for that account has to be reduced as required to prevent the overall portfolio stock allocation from exceeding the specified percentage. From that point forward, all lower-priority accounts will get only bond allocations. This process has to be repeated for every year in the modeling period to accommodate changing account balances.
The Mode 2 allocation process will usually achieve the desired target locations for the highest-priority accounts, but a user’s targets can be mathematically impossible to achieve even at the outset of the process. Suppose the Roth account is top priority and has a balance of $2,000,000 while the remaining accounts have a total combined balance of only $100,000. Further, suppose the user specifies that the overall portfolio is to have 50% stocks and 50% bonds but he also wants 100% of the Roth account to be stocks. I think you can easily see that the user cannot have it both ways. 50% of the portfolio is only $1,050,000 while 100% of the Roth account is $2,000,000. In this case, the overall allocation of 50% stocks will take precedence and the stock allocation to the Roth account will be cut to 52.5% and the stock allocation to the other accounts will be zero.
This is a top-level explanation of how Mode 2 works, but I’m hoping it’s adequate to give you a fairly good understanding. I realize that the user manual doesn’t go even to this level of detail but I plan to elaborate some in both the PRC2024 and the Pralana Online manual as I get a chance. As part of this I also plan to add a section that explains the relevance of asset allocations and locations to Roth conversions.
Stuart
