Secondary menu

Question Regarding "MC SPENDING BEHAVIOR & PORTFOLIO CHARACTERISTICS" Table

Submitted by jdupton on 5 June 2012 - 10:42am

I have a quick question regarding the table "MONTE CARLO SPENDING BEHAVIOR & PORTFOLIO CHARACTERISTICS" in the PDF Reports. (I am running version 2.20.38)

Does this table reflect the values derived from the Monte Carlo simulation? The table appears in the "Inputs and Assumptions" section of the report so one would be led to believe this is an input to ESPlanner rather than an output of the simulation though I suspect this table is really a calculated output.

I am seeing Mean Real Rates of return that are double the hand calculated mean of my portfolio. I assume this is because of the high variance of a couple of the assets in the portfolio as explained in another thread on the Forum ( http://www.esplanner.com/forum/effect-high-variance-return-calculated-mc ).

The asset income in the Details section of the report show reasonable values given my portfolio. The return rates shown there are about half of the mean hand calculated returns since I am running a Cautious Spending Behavior. The returns in the details are only about one fourth of the value reported in the "MONTE CARLO SPENDING BEHAVIOR & PORTFOLIO CHARACTERISTICS" table in the input section of the report.

Does this make sense? Can I just ignore the high value return rate shown in the "Inputs and Assumptions" section of the report as being too skewed by the statistics of the MC simulation when using high variance assets?

John

‹ Calculation of Total Spending Solving the Inverse Problem ›
Submitted by esplanner on 9 June 2012 - 11:56am.
1

Maybe. This is from an email discussion with the UI engineer on why you would get actual portfolio return numbers that aren't near the weighted mean of your portfolio:

Essentially it means that the weighted mean (if you take the mean rates of returns, hit them by the %age of the asset in the portfolio and sum them) winds up different from the mean that we get when we do all the monte carlo math (choleski) and then take 100k draws and do a mean calculation from those and look at the difference of the two. if it is "large" then this warning is emitted.

Why would this happen?

Lots of reasons, but the three big ones are (1) there isn't enough data in the series to actually get a decent result from the Choleski mathematics, (2) luck of the draw and (3) the Choleski math fails and our work-arounds bugger the results a little (or a lot, depending). The 100K samples SHOULD be enough so that the laws of large numbers apply and you get means that are close to the weighted mean. That doesn't mean this HAS to happen. It's possible, albeit unlikely, that the samples drawn from the distribution just happen to be different from the mean. The test for this is to make another pass and see if the results are, more or less, repeatable. The CE doesn't do this but if you do it a couple of times and see the same warning, then that tends to push things into the (1)/(3) camp.

Of course, there may also be a bug, but we've run some pretty careful tests of the MC and everything seems to work well for the normal cases.

FWIW, none of the math involved in ESPlanner in either MC or regular mode is guaranteed to work. It just happens that things, more or less all the time, converge (with some help from me over that last few years), so the basic module of ESPlanner gets results. It also just happens that we can come up with tweaks for the Choleski mathematics that allow us to generate "results" even if they aren't particularly correct results.

While this isn't necessarily an answer to your question, it's pretty much the best we can do for you. The alternative is to not run MC at all when we have to make adjustments which would be pretty inconvenient.

Best,

Dick Munroe

Share this