Wildly different simulation results for FI date

[SimpleCycle]

I am trying to project an FI date, and I get wildly different results using the same parameters when I use different sources.  I'm using our actuals for spending and savings and some projected numbers for college spending.

Personal Capital runs monte carlo with my current AA and its historical returns and SD.  This simulation says we're 15 years out from a 95% success rate.

CFireSim simulates all the different time periods of retirement using historical data.   It says we are 8 years out from a 95% success rate.

Networthify doesn't simulate at all, just does a straight calculation with savings, spending, and rate of return.  It says we are 12 years away from FI (using a 6% ROR and 4% withdrawal rate), although it adds an additional three years for reasons I'm not clear on, since per the table we hit our target net worth in 9 years.

Do you have a preferred source for doing projections?  8 to 15 years is a pretty big range, and I'm not sure how to weigh the pros and cons of different methods.

[bacchi]

Markets just don't work like (truly random) monte carlo. I'm not fond of them.

[secondcor521]

I like historical calculators best personally, but I think it's a personal preference.

Monte Carlo simulations I find to be too conservative, and I consider myself a conservative person.

Static straight line calculations usually use too conservative estimates (IMHO) for rates of return and inflation.

If you're that far out (and a ~decade is far out in my opinion), then there will be quite a bit of variance in the projections anyway.  But I'd guess you'd be able to retire in about 8-9 years just based on how you wrote about it, because that's where your mental target seems to be, and those have a way of becoming reality if you're focused on a goal.

[SimpleCycle]

Oh, well I messed up Networthify and forgot to put in a starting portfolio value.  It now says 7.6 years.

@secondcor521, I do think you're right that where we set our sights matters quite a bit in terms of motivation toward a goal.  I didn't model any income increases adding to our savings rate, and I made some conservative spending assumptions, so there are a few levels of "conservative" built into the assumptions.  There's also no way to reasonably predict the next 8-10 years of market returns.

[Monkey Uncle]

I don't care for Monte Carlo simulations for reasons outlined in the early part of this thread: https://forum.mrmoneymustache.com/investor-alley/trinity-study-in-another-dimension/
Basically, most financial planning uses of the tool violate the basic assumption that performance in one year is independent of performance in adjacent years.

Straight-line return simulations are too simplistic; you can't ignore the wild variability in financial markets and expect to come up with an accurate assessment.

That leaves us with historical simulators, like cFiresim and Firecalc.  But you can't just look at the mean output, you have to look at the full range of outputs.  Because reality might lie anywhere within that range.  Assuming, of course, that the future is no worse than the worst times of the past, and no better than the best times of the past.  Which basically means that there is no accurate way of predicting how many years you have until you reach your number.

Another thing to consider is how your number might change throughout the market cycle.  When the market is hitting new highs, you are wise to calculate your number using a high historical success percentage.  But in the depths of a bear market, you might be willing to relax those standards a bit.

[maizefolk]

Just joining the chorus pointing out that monte carlo is not a good fit for simulating financial returns because market returns in a given year are not independent of those in the year before and the year after. Monte carlo simulations get used in a lot of tools slapped together by people who want to have that functionality, because:

1) They're easy to implement and give clean and detailed probability distributions (not necessarily correct, but definitely detailed) because it is possible to run as many monte carlo iterations as needed while there are a limited number of real 30 year historical intervals to sample in the time period for which we have data.

2) The terminology has a good reputation because in other use cases (for example predicting elections) it works better than other models.

[ender]

Honestly, I've just accepted that any ER I do will be "4% and adjust."

In nearly no scenario ever will someone retire and do the withdrawal strategy of blindly withdrawing the same thing. I can barely imagine my and the global financial picture next year, let alone in 20 or 50 years.

The reality is the further out your projections go, the less likely the simulation results are likely to happen.