Historical return data vs monte carlo simulations to estimate probability of success
I have a question about estimates of success in retirement based on historical returns vs monte carlo simulations. We saw a financial planner last year who ran monte carlo simulations to estimate our probably of dying with some money remaining and I was surprised at getting a lower percentage of success than I expected. Looking back now, I can see that they didn't really explain what assumptions went into the simulations and it seems to be overly conservative. Nearly all of the withdrawal percentages are below 3.5% per year (with the vast majority in the 1-3% withdrawal range over 40 years) and they gave us a 78% chance of dying with money. The only thing that they were clear on was that they assumed the last 2-4 years of our life would involve heavier spending - I assume on medical expenses/long term care.
I'm wondering if there are recommendations of using historical returns vs monte carlo simulations to estimate likelihood of success (which will assume each year's returns/inflation are independent when the real world data would indicate that there are correlations between years). I am cautious by nature, but I also don't want to be so cautious that I stay in a job I don't enjoy when we can reasonably live on 3.5% of our investments. Advice/thoughts/suggestions? Thanks!
Join the conversation
Sign up to reply, follow discussions, and connect with the ChooseFI community.
Comments
Monte Carlo simulations will always be more conservative than worst case historic returns. As to how to remain cautious without going overboard (keeping in mind that I am not a financial advisor and this is not financial advice), you may want to consider the EarlyRetirementNow SWR toolkit (referenced in the below links). It uses worst case of backtested historic returns and forward capital market assumptions (based on current CAPE valuations), as well as your own planned cashflows to calculate individualized SWR/SCR (Safe Consumption Rate, which is SWR + cashflows). I would also recommend reading:
and at least one of:
Safe Withdrawal Rates for Early Retirees
or
Safe Withdrawal Rates, Drawdown Strategies, RMDs and 50 Year FI Timelines
which all address realistic SWRs for FIRE timeframes, and will reduce the gap somewhat between backtesting and Monte Carlo analysis.
This week's episode of ChooseFI (ep. 566) has an extensive discussion about the assumptions that go into these sorts of simulations, what the results mean, as well as some discussion of the pros and cons of Monte Carlo versus historical analysis. I highly recommend listening to the episode.
Another way to look at your question is what do historical returns tell us in terms of success odds vs Monte Carlo? Using only backtesting (historical returns) for success odds assumes future sequence of returns risk will be no worse than the past. That isn't a completely solid assumption, though, because "past performance is not indicative of future results" (the US SEC required disclaimer on any forward looking financial statements by regulated entities). So could be overly optimistic.
Consider that investment markets of only a few countries in the world generate high Safe Withdrawal Rates (SWRs) and most are lower: Caution for non-US FIRE adherents: Safe Withdrawal Rates based on simple portfolios vary considerably by country, and are much lower in many countries
On the other hand, Monte Carlo simulations are likely overly pessimistic as they generate completely uncorrelated scenarios that are highly unlikely to occur in reality.
So, likely somewhere in between success rates generated by backtesting and generated by Monte Carlo simulations would match actual likelihood of success.
Is it factoring anything for Social Security?
I like Portfolio Charts for this, though I don't know that they have a way to generate your own portfolio and back test it. It was helpful to see the charts of the historical performance of their sample portfolios, though!