Highly inconsistent results in FireCalc

[dhburton]

Hello everyone. New member here. I have a question about Firecalc that I have searched the forum to see if it has been asked before but haven't found anything so far. I have noticed that when I use the "A portfolio with random performance" selection in the portfolio section, if I change no values whatsoever but click the submit X number of times, each result is very different than the one before. The differences can be vast and are very inconsistent. Is this to be expected and if so why?

 

[Dd852]

Random means random! That just shows how crucial the sequence of returns is to your results. If you hit 10 bad years at the start of your retirement the whole picture will completely different than if you had 10 good years, even if the average over 30 years is the same.

 

[pb4uski]

It does seem to be the case. I ran 20 trials with the default assumptions and the random performance.

Results were 100.0% (10 of 20), 99.5% (2), 99.1%, 98.6%, 97.3%, 97.2%, 94.6%, 95.0%, 90.0%, 89.1%.

 

[Dtail]

This is the Monte Carlo simulation employed by Firecalc. It effectively shows different "random" results each time.
Conceptually, it can show more conservative results, as 10 very bad years from different decades can be randomly spliced together at the beginning of a retirement to create a very conservative result.
Firecalc in general is better off used in the standard historical sequencing concept.

 

[dhburton]

Thanks for the quick replies. I considered that but reasoned it probably would be aggregating random return possibilities using the specified std deviation to apply across the same different historic market periods. Otherwise I can see why the results could swing wildly with each click of the submit button even though you haven't changed any inputs. Given the wildly varying results with this selection, I'm not sure I see much value in using it. Especially so to evaluate changes with my other parameters. The only conclusion I can come to when using that selection is theres not much sense trying to predict any future results at all The lesson it does seem to provide is all we may be able to do is stick to the sound investing strategies, keep planned spending in check, and hope for the best.

 

[REWahoo]

Given the wildly varying results with this selection, I'm not sure I see much value in using it.

^ This.

FIRECalc was designed to look at actual history, allowing you to see if your planned withdrawal would have depleted your nest egg had you retired at some point in the past. A group of posters convinced the guy who created FiRECalc to include a Monte Carlo option, but I agree with you that it isn't all that useful.

 

[USGrant1962]

I believe I read or heard somewhere that FIRECalc's random performance MC only does a limited number of runs, not the thousands that a more robust Monte Carlo sim usually includes. Limited runs would lead to that kind of variance.

 

[dhburton]

Thanks everyone. Really appreciate the replies and thoughts. Really like the tool but intend to use it in conjunction with a few others as a confidence/validation test.

If I may I have one other noob question that I don't recall seeing an answer for. Are the portfolio growth results represented in the report in today's dollars or inflated dollars?

 

[SecondCor521]

If I may I have one other noob question that I don't recall seeing an answer for. Are the portfolio growth results represented in the report in today's dollars or inflated dollars?

See the answers in the thread here:

https://www.early-retirement.org/fo...ut-firecalc-basic-question-really-106009.html

Especially post #10:

https://www.early-retirement.org/fo...basic-question-really-106009.html#post2497951

 

[dhburton]

Awesome! Thought so as otherwise its usefulness would be compromised. Thank you!

 

[Dtail]

I believe I read or heard somewhere that FIRECalc's random performance MC only does a limited number of runs, not the thousands that a more robust Monte Carlo sim usually includes. Limited runs would lead to that kind of variance.

On the flip side, Fidelity uses a Monte Carlo simulation which is fairly robust, but produces the same results with the same inputs.

 

[USGrant1962]

On the flip side, Fidelity uses a Monte Carlo simulation which is fairly robust, but produces the same results with the same inputs.

Yes, it should. Monte Carlo should always be viewed as a range of results - never a single number. It generates a statistical distribution of results and, with enough simulations, should produce the same range each time. Just like rolling dice a thousand times produces the same distribution of results within some very tight error band.

 

[Dtail]

Yes, it should. Monte Carlo should always be viewed as a range of results - never a single number. It generates a statistical distribution of results and, with enough simulations, should produce the same range each time. Just like rolling dice a thousand times produces the same distribution of results within some very tight error band.

Agree, but the number in Fidelity is exactly the same, not just in a tight range.