92.67%. That is the number a listener got back after running a Monte Carlo simulation on his own portfolio, using AI, across 70,000 scenarios. He is 61, his wife is 58, and he is trying to decide whether he can retire at the end of next year. He read that number as a green light. His wife did not. He wrote in asking what else belongs in the decision, and I have to tell him up front that I am siding with his wife on this one.
I use Monte Carlo simulations regularly. I used to teach quantitative finance and build these models for a living. That is exactly why the two decimal places bother me.
What a Monte Carlo Actually Does
Strip away the name and a Monte Carlo is a what-if analysis. It sounds impressive to say you ran one, but the underlying idea is simple.
Take a portfolio, say $10 million. Instead of assuming a fixed growth rate, you tell the model to pick a random return each year from a distribution you define, usually a normal one with an average and a standard deviation you supply. One year it draws 8%, the next negative 10%, the next 20%. The portfolio rises and falls until it reaches the end of the horizon, and the model records whether you ran out of money. Then it starts over and does it again, thousands of times.
The value of that is real. It lets something genuinely uncertain behave like something genuinely uncertain, instead of pretending markets deliver 7% every year.
Why 70,000 Scenarios Is Not Better Than 2,000
The scenario count is the first place this gets oversold. Somewhere around 1,000 to 2,000 iterations, a Monte Carlo has enough degrees of freedom to give you a stable answer. Running 70,000 does not make the result more accurate. It makes it sound more accurate, which is a different thing, and it is the beginning of the problem with this whole output.
Every Input Carries an Assumption You Can Move
A simple market-return model has one distribution. A retirement model has many. You are letting the portfolio return vary, and spending vary, and inflation vary, and Social Security timing vary, and withdrawal patterns shift not just year to year but decade to decade as goals and liquidity events come and go.
Every one of those inputs needs its own assumed distribution. Once you know which dials exist, you can make almost any plan look strong or fragile. Tighten a distribution, change its shape, understate volatility, and the success rate moves. That is not a criticism of the person running the model; it is how the tool works. Individual stocks are especially difficult here, because we do not know the idiosyncratic risk of any single name well enough to assign it an honest distribution.
There is also the correlation problem. In a standard model, inflation, spending, and market returns all wander independently. Real life is not like that. If the market falls 30%, you are probably not spending the way you planned that year. If inflation runs hot, equities often respond. The model treats as unrelated a set of variables that move together, and that alone should temper how much weight the final percentage carries.
Nobody Actually Runs Out of Money
Here is the assumption buried inside the phrase “probability of success.” A 92.67% success rate means that in about 93 of every 100 simulated paths, your portfolio did not hit zero.
But nobody watches their account drain toward zero and does nothing. You would cut spending. You would postpone a trip. You would change the plan. A Monte Carlo assumes you are frozen at a single point in time, making the same decisions forever regardless of what the market does. The one thing every real retiree has, which is the ability to adapt, is the one thing the model leaves out.
The Risks That Have No Distribution
The model also cannot see the things that most often break a retirement plan, because those things have no probability distribution to draw from. What is the distribution for a change in tax law? For a long-term care event? For a health insurance cost shock, an adult child who needs help, or living to 98?
You cannot put those in a Monte Carlo, so a simulation that appears to cover everything really only covers what happens to be quantifiable. Everything else stays a judgment call, and judgment calls are where the actual planning work lives.
Treat the Output as the First Question, Not the Last
The stress tests matter more than the simulation. Once you have a baseline, the useful work starts: what happens to this plan if the first three years of retirement are bad ones, if a long-term care need arrives at 78, if tax rates rise, if one of you lives much longer than expected? Which single assumption, when you change it, breaks the plan? That is the analysis that tells you whether you can retire at the end of next year.
So run the Monte Carlo. It is a reasonable place to begin. Just do not let a number carried out to two decimal places convince you the question has been answered. Bring the model to us, your advisors, and let us find out where it bends before you make the call.
This post is adapted from a recent episode of the Scholar Wealth Podcast. For more perspective on Monte Carlo simulations and retirement decisions, listen to the full podcast episode here.