Just curious, if you invest in Total Market or S&P index funds, would you simply ballpark the ROR based on Large, small emerging?
@nc-cpl Basically, yah. Portfolio visualizer says VTSAX is 72% Large, 28% small+mid. So if using the survey figures above I'd ballpark 10y at 4.2%, 20y+ at 4.4%. (In my case, I work in my 30% international allocation, so it's slightly higher.)
But if you look at the span of estimates (either on Greg's or the Horizon survey I cited), you really do need to never forget in the blurry eyes of a spreadsheet result that this is just a guestimate of a median, and at this snapshot moment in time at that, at close to a record valuation high. Figures could change rapidly by next year. But gotta put something in, rather than nothing.
From my comments in other threads, you'll note I don't even think the Pralana monte carlo is conservative enough, since it assumes a normal curve rather than "fat tailed" curve. (That is, it underestimates the chances of black swans.) But I tend to be cautious by nature since I don't have any "safety net" if my plans fails. One reason I prefer an actuarial approach to withdrawal is that it is self correcting as "estimated returns" turn into "real returns". 🙂
The table that I posted was just a quick summary of the reports that are available so there are more asset classes than what I show. I did not bother with the geometric inflation adjustment, I simply subtracted the inflation from the point source ROR. I realize that this is not mathematically correct but given all the uncertainties in these numbers it is close enough. I like to use the individual providers data and read (some or all) their reports so I understand why they think the way they do. The expected ROR modeling methodologies can be very different and I always want to know if there is a bias based on some incentive that pushes their estimates in certain directions. For example, a value oriented firm who sells value funds may have a higher estimate for value vs growth.
My approach is to find the ROR estimates for the asset classes that I own that are as close to the end of the year as possible to correspond to the end of the year portfolio values that I use in Pralana. The Horizon estimates can be way out of date as it reports estimates as far back as October 2023 as seen in the report: "The vast majority of advisors specified effective dates on or around January 1, 2024. However, a few specified effective dates as early as October 1, 2023 and a few specified dates as late as March 31, 2024." This may not seem lie a big deal and if markets and the economy have been very stable it is not a big deal. However, if equity valuations and/or interest rates have significantly changed since they performed their analysis then the estimates may no longer be valid. For example, the YTM for a given duration bond fund is a good proxy for the subsequent ROR for the period of the duration. Also if your portfolio has increased due to the increase in equity valuations, the probability that subsequent 10-year ROR is lower has increased. The opposite is also true - if your portfolio has declined because equity valuations have declined, then the probability that subsequent 10-year ROR being higher has increased. I realize not everyone buys into this concept but Professor Shiller and others have shown that the relationship is pretty good over a 10-year time frame. It is not reliable over short time frames.
Since Pralana is an annual model, most folks use prior EOY portfolio values to start the modeling, therefore, using EOY ROR estimates to keep things consistent makes sense to me, not ROR values that are over a year old especially if your portfolio has increased or decreased significantly. We need to also keep in mind that many providers provide a range of returns around their point estimate. The horizon ranges are not the range of expected returns each provide provides, it is the range around the median provider point estimate so it is likely a narrower range than what the individual provider estimates. So I would be careful not to use the range that Horizon provides as the range of possible outcomes - it may be too narrow.
We have not discussed the Standard Deviations for these estimates, but some providers due provide that data as well. I have found the that the standard deviations do not change much from provider to provider and from year to year. I use the historical standard deviations for each asset class that I own, I do not use the built in function in Pralana. That is another topic by itself.
In summary, I think market conditions at the EOY that match your EOY portfolio values should be taken into consideration. That is one reason I do not rely on the historical analysis. The bond example is one reason and the equity valuation is another. The look back analysis that I have done convinces me that the realized RORs for the first 10-years of your retirement is going to have a huge impact on the success of your plan if you are using your portfolio to fund your retirement. The last 20+ years is not nearly has important. If you have guaranteed income then this is much less important. A boot-strap approach may overcome some of this but that is an area that I know little about.
I look forward to more discussion - I can change my mind if the data supports it!
@golich428 The rub with entering yearly average return estimates is that Pralana's monte carlo generator assumes your entries are geometric. The program derives an assumed arithmetic average from that assumed geometric entry! (The randomizer then generates a random normal return around that derived arithmetic mean and either its own standard deviation or the one your provide. See page 128 in the manual.) I'm not sure in the end it matters all that much, but it's at least not what the programmers intend the number to be.
But you're raising an interesting question, Greg: Do the deterministic results use the number you enter straight up (i.e utilize is as an arithmetic average), or do they "derive" an arithmetic average the way they do in the monte carlo? They can't have it both ways! (I'm going to ask them.)
I hear you about taking valuations into account. If you want to know the truth, I'm nerdy enough that my own private spreadsheet for estimated returns consists of the median of a bunch of different methods. I use (a) Horizon for the "survey" method of the wisdom of the crowd; (b) Research Affiliates' "Yield and Growth" returns estimates, since it's a completely different method of estimation. (They also have valuation estimates of course); (c) regression with the CAPE to find estimated short term and long term returns; and finally (d) the median results of a paper called "The Return on Everything" that looked at stock bond and real estate returns across the whole world since 1900. I include the latter because it's quite pessimistic, and makes the case that the US stock returns 1928 to present are an exception to the rule of world history.
If I have to pick one quick ER, I prefer the CAPE regression cause it's the most "recent" in terms of valuation. (The TPAW site generates CAPE regressions daily.)
And for Bond returns, I simply use the 10 or 20 year TIP rate, since it's more recent than any of the reports, and as far as I know as accurate as any other method.
I utilize a form of the actuarial method for withdrawals, mostly because it adjusts to valuations and is self-correcting. In other words, I think we might agree more than it seems.
FWIW, I'm not personally worried if one or two of Horizon's source data are a year old, since they use such a large sample of 26 sources!
@jkandell I do realize the program is asking for geometric return estimates and the nominal values in the table are geometric, I just did not go to the trouble of calculating the real values with any precision. Do we know what inflation will be with any precision?
As the saying goes, it is better to be roughly right than precisely wrong. I don't think exact precision is really attainable in this situation. I use the results of my model as a reasonable estimate of the future to help me make various decisions but, I try not to consume a lot of energy getting overly focused on minute details given all the uncertainties. However, I will continue to update my assumptions each year and incorporate any new information to make sure my plan is still reasonable.
FYI, I take a safety first approach to retirement and rely on guaranteed income to cover most of my expenses so the ROR expectations are not that important but they still need to be reasonable to help me make reasonable decisions.
I would be interested if you could share how you approach the standard deviations that you use for each asset class in your portfolio.
Thanks for the discussion - I learn a lot from them.
I would be interested if you could share how you approach the standard deviations that you use for each asset class in your portfolio.
@golich428 I do this: I look at the Boglehead asset spreadsheet standard deviations from 1930, 1950, 1970 to present--and use the highest value of those three The standard deviation only matters in Pralana for monte carlo runs. But it's one of the two key variables! A small change in SD can really affect the probabilities, so I use the highest value because I want to capture the "true" extremes. If I were less conservative I'd just average the three. But as you know, for stress testing I favor bootstrapping (which I can't do in pralana) over monte carlo since it doesn't use standard deviation at all and simply generalizes the actual spread of the data as it actually happened.
Thanks for your perspective. I pulled up the Bogleheads spreadsheet but I need to digest it a lot more. One comment I would make about any use of historical data either serial or bootstrap is on the bond portion of a portfolio. It is well established that the YTM on your bond fund is a good proxy for the future ROR for the period that approximates the duration of the bond fund. If we believe that to be true (I do) then using historical data with start dates in the 80's when bond yields were in the 10 to 15% range would not be a very good approximation of the future returns based on the 4.5% yields we have today. How would you propose taking that into consideration if you rely on historical data? It might wash out over 100 years but I don't think it does over a 30 year (typical retirement) period. Again - I enjoy the dialog with someone who thinks about these things.
@golich428 Agree about dangers of relying on historical averages with bonds. I too use SEC rates or YTM rates to predict long term bond returns; most often just use the 10 or 20 year TIP rate since they're the purest. However, history is useful to get a standard deviation to be used with that figure. Having said that, one advantage bootstrapping does have is that it captures bonds and stocks cross-correlation much more accurately than pralana's current system.
I do find historical investigation valuable in other ways though. For instance, after the drop in my "safe" TBM of 25% real rate over two years, which changed my whole investing tactics, I looked back through history to see there was only one other two year stretch since 1871 in which that had happened, around 1918. So it really was a once every 75 year event for what it's worth. The 18.42% real drop in TBM in 2022 was the worst single year Dec/Dec in the whole documented history since 1871. This was way worse than the normal curve probability predicted!
So based on that, I started substituting a bond ladder for bond funds. My current ladder guarantees about 2% real return over 20 years. With zero standard deviation. So that's how I propose to deal with the historical record. 🙂
