Curious how Pralana's Monte Carlo algorithm is structured (AAGR, CAGR, something else)? I ask as a competing program very recently announced a significant change to its Monte Carlo calculation theory and it has caused quite the stir within its customer base due to corresponding changes to nearly everyone's "success rates". Here is a portion of the information/explanation-
We Now Use AAGR Instead of CAGR
As part of our latest model updates, we’ve adjusted how we calculate investment returns in Monte Carlo. This change improves how volatility is handled and brings our simulations more in line with real-world behavior.
Why did we switch from using CAGR (Geometric Mean) to AAGR (Arithmetic Mean)?
Using CAGR (which already builds in the long-term effect of volatility) inside a Monte Carlo simulation (which also adds volatility) means volatility gets counted twice. This made projections too conservative, especially over longer time horizons.
Switching to AAGR fixes this and follows the industry best practice:
-
AAGR starts with the average return before accounting for volatility.
-
The Monte Carlo simulation then properly layers in volatility by generating thousands of random paths. This avoids double-counting and makes forecasts more realistic.
Think of it this way:
-
Arithmetic mean (AAGR): starting point, no volatility yet
-
Monte Carlo: adds realistic volatility
-
Geometric mean (CAGR): already includes volatility, so adding more isn’t correct
That’s why we use arithmetic returns in our Monte Carlo simulations—because they help us model the volatility and uncertainty of real-life investing, rather than just showing a simplified average.
Also posted this question in the Pralana Online forum as that is actually what I use. I don't find a way to delete this post or would have done so.
@hh22 Pralana (Gold and Online) has always used geometric mean for deterministic analysis and arithmetic mean for Monte Carlo analysis, and this is explained in the user manuals.
Stuart
I am a new Pralana subscriber, who signed up here after Boldin's algorithm change. My understanding is that Boldin made three changes: they (1) switched to using arithmetic mean for Monte Carlo analysis, (2) changed to assuming correlated results across accounts and (3) adjusted their standard deviation (increasing standard dev. slightly for an intermediate band of returns and reducing it for a higher band of returns). I ended up with a 20% drop in success rate. Entered the same inputs (RoR, inflation, social security, COLA etc.) here (I've tripled checked), and selected the correlated returns setting, and while my Pralana success rate is lower than what Boldin had been giving me pre-Boldin changes, the Pralana success rate is 15% higher than Boldin is now giving me. My chance of success from Vanguard PAS's Monte Carlo sim is closer to Pralana than Boldin's new and supposedly "improved" results. FWIW.
My chance of success from Vanguard PAS's Monte Carlo sim is closer to Pralana than Boldin's new and supposedly "improved" results. FWIW.
I don't think the "correlated ROR" in Pralana's MC is the same as "correlated accounts" in Boldin. As you note, Boldin changed things so that accounts now rise and fall together. They figure out an average return and sd (from a table) for each account behind the scenes. Previously, in their MC one account would go up randomly, the other down randomly, and it would cancel out, not modeling how reality works.
Pralana (and @Smatthews51 can correct me) never fell victim to that error. Pralana uses either a 0 or 100% correlation at the asset level, not the account level, in doing its monte carlos. "0%" box causes a random number drawn for each asset independent of each other. "100% correlated" box causes a single random number to be drawn that affects all assets on each that MC draw.
So e.g. if you have the box checked, then Bonds and Stocks might as a whole over time go up and down together. Since accounts literally consist only of assets, with box checked accounts indeed be 100% correlated--but I think the different methodologies produce different results.
I always recommend most folks with 60/40 type portfolios uncheck the box because the historical correlation between TSM/SP500 and 10yTreasuries/TBM has been about 0.20 and with Treasury Bills about 0.10. Those are much closer to 0 (unchecked) than they are to 1 (checked). (However, if you have very few bonds, e.g. 80%/20%, and mostly differing kinds of stocks assets in your portfolio, this advice would be the opposite, since those as a whole are closer to 1.0 correlation than to 0.)
PS. @Mollerm Where are you able to run Vanguard's MC? They used to have their terrific "retirement nest egg" mc but I can't find it since about a year ago.
Thanks for the info. This is very helpful. And happy to be here--I like Pralana's tool a lot. Re the P.S.: I am a "personal advisor service" subscriber, which gives me the ability to input somewhat more info into Vanguard's Monte Carlo tool. It is nonetheless more of a black box than Pralana's or Boldin's. I understand that it incorporates assumptions of the Vanguard Capital Markets Model about rates of return, volatility, and inflation. FWIW @ Jonathan Kendell, I also incline toward your view (mentioned in another thread) that Boldin's update has a bug.
@jkandell You are correct! Pralana applies random numbers at the asset class level and they can be 0% or 100% correlated. If 0% correlated, we use a unique random number for each asset class; if 100% correlated, we use the same random number for each asset class.
Stuart
this is a test post
@mollerm I am jealous you can run Vanguard's monte carlo. They used to have the tool free. And it was very handy and easy to use for a quick check!
Their proprietary model is much more complex than Pralana: it is a scaled bootstrap, and includes variables like interest rate and inflation projections.
As I've mentioned, using an idealized normal curve like Pralana and Boldin is a very gross approximation of what happens with variance in the real world. It's better than nothing (that is, just using deterministic and historical) though! Sometimes "good enough" is good enough. 🙂
Just take those percentile numbers lightly.
@jkandell Very interesting! Thanks much for the helpful info and advice. Look forward to learning more here.
