Asset class correlation & Monte Carlo: Do you use investment asset classes other than Stocks and Bonds?

@thromer I agree with almost everything you say! In particular,

1) I agree as a rule it's best to use broad asset classes (Stocks/Bonds/Cash/RE/Gold) rather than break things down into the minutiae of subcategories within each class (like international, value stocks, etc). It's a rule of thumb in modeling that you want just enough complexity to capture the issue at hand but not more than you need. The chances for error increase exponentially with each new addition/nuance.

I make an exception for a created asset I call "individual treasury bonds" (CDs/TIPs/treasuries) because I feel they act differently than a bond fund, especially with sd. This allows me to model, e.g. that when we get elderly we'll switch exclusively to funds from individual bonds, to adapt to cognitive decline.

Note that just because you lump things assets like "stocks" into a single entry doesn't at all mean you can't reflect the make-up of different sub-types. For instance, my stocks are 38% TSM and 37% International and 15% Small and 15% Emerging markets. So I end up with a weighted average for my estimated geometric return of "stocks" of 5% with sd of 20% (a bit higher on both counts than tsm). Because I use an actuarial method of withdrawal that self adjusts these estimates do not need to be a hundred percent accurate.

2) For asset's standard deviation, I use the standard deviations I get from the Bogleheads "simba" returns database. For each asset I average its sd 1870 on, 1900 on, 1930 on and 1950 on.

3) I agree with you and Kitces that complex correlation matrices lead to fool's confidence. Like you I set Pralana MC checkbox for 0% correlated ("uncorrelated") rather than 100% ("correlated"). The correlation between stocks to treasuries and cash is roughly 10%-20%, which is much closer to 0% than it is to 100%. At the moment because I am 40/60 allocation it is the stock:bond correlation that matters. If my portfolio ever gets to something like 80% fixed income of various types and 20% stocks I may reconsider.

4) I do differ a bit with you about improving Pralana's MC though. While I don't think we need correlation-matrice complexities, I do think Pralana should consider replacing parametric MC with bootstrap MC. A parametric simulation assumes a specific distribution for the data and uses its parameters to generate samples, while bootstrap Monte Carlo resamples the historical data to create new samples. Namely, Pralana assumes a symmetric "normal" curve, which we know does not fit the historical data well. (For instance, it underestimates the chances of "black swans".) I have argued for instead using a scaled bootstrap that would draw block samples (e.g. 5 year) from the historical record, scaled down to match user's expected returns and sd expectations. This preserves the actual "shape" and stistical properties (including autocorrelation, cross correlation, fat tails) of assets like stocks and bonds, while simultaneously reflecting that future returns are likely to be lower than historic. With a bootstrap you don't need to worry about correlation-matrices because the samples themselves, if taken across all assets, generate the correct historical cross-correlation and auto-correlation (return-to-mean). A scaled bootstrap is not more complicated to program than a parametric monte carlo, and mostly changes the results below the 15th percentile and above the 85th percentile bands.

I used to have only stocks and bonds but with the new online options I now have 12 assets. With 12 assets I can play better with Asset Location but cannot use Historical analysis (I do not have historical info of my assets). The problem with Monte-Carlo is the 0% or 100% correlation, both unrealistic. I just run both, assume some average, but mainly work with the Deterministic numbers. Hopefully in the future Pralana will add 50% correlation or even more options?

Thanks for in-depth reply, I very much appreciate your confirmation of my line of thinking from a long-time member with (clearly!) a lot more experience with modeling!

I'm still interested in others' perspectives, don't be shy!

@jkandell your reward is more questions/observations (and my reward is doing a little more due diligence by reading the fine manual...):

• I'm guessing you just have a simple spreadsheet for computing the geometric return and s.d. for "stocks"?
• By actuarial withdrawal do you mean Pralana's actuarial spending method?
• From your comment elsewhere I see you leave TIPS/bonds out of the equation for Monte Carlo entirely by modeling the eventual payout as windfall income. (Aside: from the short description of the "Fixed Income Assets" feature request it sounds like that will continue to be the case if/when the feature is implemented.)
• Thanks for the explanation of bootstrapping! I'm still bootstrapping my own understanding of this particular aspect of MC. Do you use either or both of the tools you mentioned in your wishlist request for bootstrapping to augment Pralana? For others' reference, they were TPAW Planner (free AFAICT), and Portfolio Visualizer ($30/month if you want to save your data).

Ted

@hecht790 What are the advantages of the multiple asset classes in Pralana for you? Do you use the asset location information mainly for tracking, or does Pralana help you notice when you need to rebalance, or something else? I'm new to Pralana so I'm still figuring it all out.

Multiple classes with different taxation help optimizing Asset Location. Why is Asset Location important early in investment life? To reduce the risk of being stuck with large unrealized LTCG in taxable accounts and being unable to change it. (The overall Asset Allocation does not have this risk because we can rebalance it by moving assets within IRA/Roth alone without touching the taxable account.) I am using Pralana to compare various Asset Location options and find the optimum. I will be very happy if Pralana will add an Asset Location optimizer.

Regarding rebalance timing. Pralana rebalances yearly. I personally rebalance when needed (using 5/25 formula of Larry Swedroe). Rebalancing is a complex task for a tool, so I do it manually (with Excel sheet support). I just rebalanced last week, since international equity exceeded the 5% limit.

Posted by: @hecht790

↑

I am using Pralana to compare various Asset Location options and find the optimum. I will be very happy if Pralana will add an Asset Location optimizer.

What is your work-around now to optimize location now? Do you use play around with Account Prioritization, Withdrawal Priority, and Asset allocation with several scenarios?

I admit I just follow simple rules of thumb: taxable is best for stocks, reit and bonds best in tax-deferred. And anything in Roth, but especially things with high expected returns.

Posted by: @thromer

↑

I'm guessing you just have a simple spreadsheet for computing the geometric return and s.d. for "stocks"?

Basically. I track a couple returns estimation sites (Vanguard/Research Affiliates/TPAW), and just do a simple weighted average on paper or spreadsheet every year after eyeballing them. We know from experience that there is a huge difference between predictions and actual returns, so I don't over-do the precision like it might seem; but you have to pick something to do planning so I eyeball the medians.

• By actuarial withdrawal do you mean Pralana's actuarial spending method?

Alas, no; I find the TPAW actuarial too variable, as it does the equations prior to taking out essentials, so discretionary spending changes too much each year. The Pralana actuarial doesn't work with the roth optimizer either!

I personally use the TPAW method (aka "lifecycle model"), where you withdraw taking into account the whole mass of future discounted income (from social security and pension) and the whole mass of discounted future essential expenses. In other words, it's actuarial but diversified across time.

The closest to that kind of actuarial in Pralana is consumption smoothed withdrawal, which is my primary method. For that I set "non-essential fixed amount" at 0 then run the consumption smoothing; I then enter the resulting figure as my "non-essential fixed withdrawal amount" each year, atop changing expenses. I then run the Roth optimizer and Withdrawal Priority optimizer, and re-adjust that amount a little. It's not obvious at first, but consumption smoothing ends up being the same as an actuarial method except by iteration.

• From your comment elsewhere I see you leave TIPS/bonds out of the equation for Monte Carlo entirely by modeling the eventual payout as windfall income. (Aside: from the short description of the "Fixed Income Assets" feature request it sounds like that will continue to be the case if/when the feature is implemented.)

I do exclude my whole liability matching bonds when using TPAW. But in Pralana I include my TIPs/individual bonds as their own asset class (with known return and sd=0) so they can be worked into the optimizers. (I have to be mindful in entering my asset allocation to not assign the "individual bond" asset outside of the span of their ongoing maturity dates!) It's Ibonds I exclude and categorize as windfall income, which as you note, excludes it from monte carlo. (Luckily I only have a few years of those.)

I think in some cases I just included tips as "bonds" but adjusted my overall return and sd in expected returns page for those years in which they dominate.

• Thanks for the explanation of bootstrapping! I'm still bootstrapping my own understanding of this particular aspect of MC. Do you use either or both of the tools you mentioned in your wishlist request for bootstrapping to augment Pralana? For others' reference, they were TPAW Planner (free AFAICT), and Portfolio Visualizer ($30/month if you want to save your data).

I used Portfolio Visualizer for years until they started charging. It's an amazing site. But now I use TPAWs scaled-bootstrapping. (But TPAW's bootstrap charts don't give you returns like Pralana; rather it gives you consumption smoothed income, so is only useful for that model.)

It will be interesting to see the different way people do things; we're all quite different here.

@jkandell

I am not using workaround, just trying various Asset Locations and comparing the result. The tool maintains accurately the overall Allocation for all 12 assets, as requested. However, it has difficulties with Asset Location. I am not sure how the algorithm works, but it seems that the order of the assets’ names in the Asset Allocation/Location table is part of the algorithm.

I am also using some rules of thumb. But it is not always obvious, especially with US vs. Foreign equities.

Posted by: @hecht790

↑

@jkandell

I am not using workaround, just trying various Asset Locations and comparing the result. The tool maintains accurately the overall Allocation for all 12 assets, as requested. However, it has difficulties with Asset Location. I am not sure how the algorithm works, but it seems that the order of the assets’ names in the Asset Allocation/Location table is part of the algorithm. I am also using some rules of thumb. But it is not always obvious, especially with US vs. Foreign equities.

The algorithm for that unscheduled withdrawal priority list is (from the manual):

When a negative cash flow occurs, the tool starts by trying to cover the deficit from the cash account; however, that withdrawal is only allowed to take the cash account balance down to the floor level and if more money is needed, then a withdrawal is made from the first account in the withdrawal priority list. When there is insufficient money in the first account, then the tool goes to the second account, and so on until the spending deficit is fully covered.

But I'm not sure if and how this feature can be part of manually optimizing location manually. It sounds like a daunting task. But I hear you: I feel into the trap you described of having a bunch of stocks in my taxable but now I can't rebalance like I want because it would knock my ACA subsidy out.

@jkandell

I found a workaround for adjusting some Asset Locations. I swapped the location of two assets in the Asset Class Names table (move US REITS from location #6 to location #11 and move EM from location #11 to location #6). I also swapped the data associated with these 2 names. As a result, there are no more REITS in the Taxable account. Overall final saving stayed the same. Taxable went up 1.77%, Roth went down 2.45%.

@hecht790 As a user I'd be surprised if changing the order of items in a list changed the data in the model, maybe there's a bug? Though I'm not quite sure what you meant by "Taxable went up 1.77%, Roth went down 2.45%."

Posted by: @thromer

↑

@hecht790 As a user I'd be surprised if changing the order of items in a list changed the data in the model, maybe there's a bug? Though I'm not quite sure what you meant by "Taxable went up 1.77%, Roth went down 2.45%."

I'm not exactly sure how Gaby did it, but the (unscheduled) Withdrawal Priority list absolutely affects the model. If the cash account doesn't have enough, Pralana draws from the accounts in the order you list. Which creates either taxable vs non-taxable events, different effects on MAGI, and different returns down the line.

@thromer

It is probably not a bug. The user’s Asset Location input may not be possible to achieve mathematically. The algorithm tries placing the assets one by one starting from the top of the list. So, the order of the list will change the results. I suggested in another thread adding an optimizer that will use tax efficiency info to help the locations, so the list order is less important. But it is probably a major task and not beneficial to everyone. Another simpler feature that can help is the ability to drag the names up and down on the list to change their position. (Dragging a name will move also all the associated info related to this name.)

"Taxable went up 1.77%, Roth went down 2.45%." are just to show the effect of swapping 2 assets. Taxable is the final Taxable account balance and Roth is the final Roth account balance.

Hi Ted @thromer Welcome to the forum! Always great to hear another voice 😀. You asked for other views, as @jkandell put it "It will be interesting to see the different way people do things; we're all quite different here." He's not kidding about the vast array of thoughts 🤪. From Pizzaman's bank of 2 cents, I tend to be at the far other end of the spectrum, at least in comparison to the topic of this thread. I do not use any of the techniques being discussed here. (To be crystal clear, there is nothing wrong with them). I am from the keep it simple camp ala JL Collins and his updated book "The Simple Path to wealth" 2025. Well, I don't quite go that simple, but close. I focus on the big things that can really de-rail your retirement planning such as inflation, entering a bear market soon after starting your retirement, and stock market valuations (Shiller CAPE). Add to that the sky high US debt which will eventually lead to higher tax rates, possible reduction in Social Security and Medicare, and the now passed into law reduction in Medicaid. There are plenty of big things to worry about, so I use Pralana Retirement Calculator (PRC) to play around with things such as asset allocation (stocks & bonds) and location (taxes) to help counteract these stressors, which, if you are weird like me, can be kind of fun. Using individual US Treasuries to make a multiyear bond ladder is another way to help counter act all that bad stuff. Plus other things 🤔. Stay turned. (You asked for it 😘)