How to choose an early retirement simulation tool like a boss

By: Feeding Our Fire

Retirement planning is really a risk analysis exercise trying to predict the future. In planning for retirement, we are faced with uncertainty, ambiguity, and variability in what the future will bring, and we can’t accurately predict the future. There are many different online tools available to simulate your retirement (including our own).

With so many options, how do you pick the best tool for your needs? Below we discuss some of the key factors in choosing a good retirement simulation tool. You may find you use multiple tools to help in your planning.

Simulation vs. Calculations

We recommend tools that truly simulated your retirement scenario and don’t just use average market performance to predict the future. Using averages is easy and you could even hand calculate your projected retirement needs. Using averages doesn’t provide a realistic estimate of future success and tend to overestimate future success.

Retirement planning is all about understanding the risk and probabilities for your future. Planning is focused on not running out of money before your expected lifetime. The best tools to do this utilize a Monte Carlo simulation.

Why Monte Carlo Simulation?

A Monte Carlo simulation is a method of testing an outcome over a range of possible variables. It is a “stress test” for your retirement future. Monte Carlo simulations are used in retirement planning to predict the likelihood that you will have your needed retirement income through life expectancy. 
They consider inflation and historic market performance and simulated 1000’s of lifetimes.

The results are all combined to calculate how many life time’s modeled produce income for the whole retirement. Usually represented as a probability of success in percent, the percentage of simulation that provided income for the entire retirement period selected.

Monte Carlo analysis is ideal for modeling situation with uncertainty. In planning for retirement, we are faced with uncertainty, ambiguity, and variability in what the future will bring. We can’t accurately predict the future. Monte Carlo simulations let you see all the possible outcomes of your decisions and assess the impact of risk. Since retirement planning is really a risk analysis exercise, using Monte Carlo modeling really makes sense.

We prefer tools that run a Monte Carlo based simulation and not ones that just rely on average market performance. We also prefer the Monte Carlo option over tools that used backtesting of prior markets. Monte Carlo simulation would produce results beyond what may have been seen in the historical data.

Probabilities of Success for Monte Carlo Simulations

The Monte Carlo simulations are all based on the probability a given scenario will succeed. For most of our posts, we use a 95% chance of success when running a Monte Carlo simulation.

Research shows that a Monte Carlo simulation tends to predict “worse” worse cases and “better” best cases scenarios compared to historic results experienced in the real world. The theory is in the real-world returns tend to revert to their mean values. They don’t trend negative or positive for too long. In a Monte Carlo simulation there is no such bias, so you could see cases with very long sequences of negative or positive returns. 

Derek Tharp wrote in his article “Does Monte Carlo Analysis Actually Overstate Tail Risk In Retirement Projections?”

“… despite the common criticism that Monte Carlo analysis and normal distributions understate “fat tails”, when it comes to long-term retirement projections, typical Monte Carlo assumptions actually overstate extreme outcomes relative to historical returns due to the failure to account for mean reversion – yielding a material number of projections that are worse (or better) than any sequence that has actually occurred in history.” (1)

Derek Tharp

Based on Tharp’s research, a simulated 93.5% success rate may be equivalent to 100% success rate in the real world.

Don’t Over Emphasis Probability!

Too much weight can be placed on trying to meet a 100% success rate. William Bernstein, an American financial theorist who has written many books on investing and modern portfolio theory, commented that “any estimate of long-term financial success greater than about 80% is meaningless.” (2)

His reasoning that these probabilities are merely an imperfect estimate of the investment risk you are taking. Some major event, we can’t plan for, may occur that is not factored in the historic probabilities.

Of course, there are never any guarantees in this process, so pick a value that you feel comfortable with for your own plan. We chose 95% for most of our cases.

Planning for the Worst Case – The “Tails” of Probability

One interesting thing to note in all our retirement planning is we focus on the worst case of possibly running out of money, or what is called the “tail”.

We see in the simulations that achieving a high success rate, the probability your savings outlasting you is significant. In the graph below, we see this scenario has a 4% chance of running out of money but a 95% change the ending balance would be higher than the starting balance at retirement.

Probability Distribution Histogram

This means as your retirement value grows during your retirement phase, you may be able to take our more income than planned.

Historic Market Performance

Simulations really on using the historic market performance to set the probabilities for the future. Using a long market history captures more market variations. We like to see the simulation use at least 50-year data and ideally have the option to use 90-year data. It should also include both the mean and the standard deviation to simulate market returns each year. 

As an example, below is one sample lifetime of market returns for a portfolio using the Monte Carlo method. This uses the historic mean and standard deviation but generates a unique set of returns for each year of this lifetime.

Monte Carlo Simulated Investment Performance

In this case, we see on the year with a market return between -15.4% to -17.9% but another year had a return of 39.6% – 42.1%! So even though the mean is 10.74% the potential return for any given year can vary widely based on the standard deviation, like the real world. Therefore, using just average calculations are dangerous for your retirement planning since the sequence of these returns can produce a wide variety of outcomes.

This will not perfectly align to the historic market it’s based on, it provides one future possibility. In the Monte Carlo simulation, we really on 1000’s of these lifetimes to get a large data set of possible outcomes and predict the success of our plan.

Sequence Risk – Consider Current Market Conditions

Sequence risk is the timing of market returns and having an extended period with low or no growth at the beginning of your retirement. 

Michael Kitces, a financial planner and blogger, has done extensive research about sequence risk and safe withdrawal rates. In his article “Understanding Sequence of Return Risk – Safe Withdrawal Rates, Bear Market Crashes, And Bad Decades” , he writes:

… the real problem is not a bad year or two at the beginning of retirement, but a bad decade to start off retirement. A bad decade outlasts most cash reserve strategies … It’s a slow inexorable grind that whittles down the portfolio to the point there’s just not enough to recover, and there are few places to hide after 10 years of poor returns. 
In fact, as it turns out 10 years really is the “sweet spot” for sequence of returns risk … (3)

Michael Kitces

Sequence Risk: Real World Example

The worst situation is not one or two bad years, the models will typically capture this. It is an extended period of low or no market appreciation. In a recent interview with the Mad Fientist (one of our favorite FI Podcasts), Michael Kitces provides this anecdote:

So, if you imagine being in retirement, and for the first 15 years, the market gives you no capital appreciation whatsoever, you begin to get a sense of what it was like to be a 1966 retiree. Now, on top of that, it gets even worse because inflation went from about two to twelve, which also caused the worst bond bear market of the century at the same time that stocks generated no appreciation for 15 years.

And despite that, or even through all of that, what we find is this 4% initial withdrawal rate adjusting for inflation works. … as bad as it can get when you get these bad sequences, what we still ultimately found is it still doesn’t seem to get any worse than about 4%.  (4)

Michael Kitces

Even during this worst-case 15-year period, the 4% initial Safe Withdrawal Rate worked! You also have flexibility if you really run into a similar scenario. You may decide to Barista FIRE, work part-time or find a side hustle, or cut back on some expenses. 

So, how can we account for this in our planning?

Using Current Market Conditions in Retirement Planning

If your retirement plans are far in the future, using market return mean and standard deviation should provide a good estimation for the performance of your plan. 

Closer to retirement you may want to have the ability to reflect current market conditions in the simulation. Current market valuations can’t determine short term trends, though research suggests they have some correlation to longer-term market performance trends.

More research by Kitces shows there is an inverse correlation between market valuation using the Shiller P/E10 (or CAPE) ratio and future market performance. Kitces takes the inverse of the CAPE (the CAEP) and shows how it is correlated to ten-year future returns.

The graph below shows this correlation. The trend seems to work best for a ten-year outlook. (5)

CAEP vs. 10-Yr Annualized Returns

As of the writing of this article, the Shiller CAPE (6) is sitting at 32.29, very high by historic standards and suggests future returns may be lower. 

We have gone from a CAEP (inverse of the CAPE) of 7.5% in 2008 to 3.1% for August 2018. We had an average return the past ten years (S&P 500) of approximately 12%. The current CAEP is 3.1%. This suggests we will see lower returns in the next ten years, maybe closer to 3-4% or less. 

Don’t Panic! And How to Use this Knowledge

We really don’t know for sure what the future has in store but can use this correlation to help manage our planning and reduce our risks. If you are planning to retire soon with high market valuations, it would be wise to adjust the initial withdrawal rate to be at or below 4%. 

If the simulator has an estimated initial Safe Withdrawal Rate value and it is at or below 4% you are probably in good shape. When the CAPE is very high like it is today (>20), you can set that target a little lower maybe 3.5%. 

You can also reduce the expected returns for the initial ten-year period in your simulation. The simulator tool should allow you to adjust the post-retirement return to account for this. Ideally, it will allow you to adjust the expected returns for the initial ten-year period to account for possible sequence risk of low returns. There is no exact answer to how much to adjust since we are trying to predict the future. Be aware of the risk and lower your initial expectations. 

Considering Social Security

Many simulators ignore other sources of future income such as Social Security. Depending on your retirement window you may leave this out of your planning. Not counting for Social Security does mean you may be taking too conservative of a plan and may postpone your retirement longer than necessary. 

There are many articles on this subject. If you are near a traditional retirement age you should use a tool that allows you to enter your Social Security income. Even if you are in the FIRE mode, you should still consider Social Security in your planning.

Social Security reserve is expected to be depleted within the next 15-20 years, but it will still receive recurring revenue from payroll taxes. Social Security won’t be going away. Unless Congress chooses to raise additional revenue or finds new ways of closing the shortfall, future benefits may need to be cut. Planning for 50% to 70% of the benefits in the future is probably safe.

One note, the estimated benefits assume you are working to your social security retirement age, so you need to adjust that based on your actual years of work. It uses the highest 35-year work history so if you work for less, your benefit will be reduced. The annual estimate also assumes you will continue to earn your last year income till retirement age. The social security administration has some tools to help calculate this.

You may also want a tool that accounts for other income sources such as pensions. If you are taking the pension right away, you can just deduct this from your retirement income needs.

Key Criteria in Selecting a Tool

Based on the above knowledge and some other preferences. Here are some things we like to see in the tools to ensure we understand how they are calculation the results. Our own tool is not yet perfect, and we are working to incorporate most of these features.

Must Haves

1) Uses a Monte Carlo simulation method. Backtesting on historic market performance is acceptable.

2) Easy to Use and clean interface. These tools should be easy to get started and run a basic scenario. You should not need a manual at least for the basic inputs.

3)  Simulate the combined savings phase and retirement phase. Many calculators do one or the other, but you need to look at the holistic period from today through your end of life. You can run these independently but coupling them we believe gives a better result.

4) Allow input for investment performance including the mean and standard deviation. Ideally, support both simple lookup based on a basic portfolio as well as the custom entry of data.

5) Allow selection of savings phase and retirement phase investment performance separately. Once we retire we are no longer planning to accumulate and are more focused on preservation and keeping ahead of inflation. Your investment strategy will switch as you get close to retirement, so the tool should allow changing the investment mix at retirement.

6) Includes inflation and allows adjustment of both the mean and standard deviation.

7) Provides transparency on any assumptions and it is clear how it performs the calculations.

8) Provides output both graphical and tabular data from simulation results. The output should be appealing to the eye and easy to read. It should include both graphical data and have the option to review the tables that make up the graphs.

9) Provide the statistical distribution of savings at the start of retirement and remaining balance at the end of life.

Preferred Features

 1) Allows input of fixed incomes such as Social Security. You can usually adjust for this by adjusting your income needs but if you plan to retire significantly before taking Social Security it is good to have this as an input. This is a feature we plan for our next revision of the tool.

2) Provide the probability the savings at retirement will be at or below a 4% Safe Withdrawal Rate threshold.

3) Allows the ability to save the results and update in the future without re-entering data.

4) Options for more advanced situations. Ability to estimate tax consequences pre and post-retirement is nice to have for more advanced scenarios.

5) Allows for a sequence of returns risk and to be able to adjust expected returns based on current market conditions especially for the initial few retirement years. This can be simply by changing the mean and standard deviation for investment performance but ideally could adjust for the initial retirement period where retirement performance is more susceptible to sequence risk.

6) Provide a savings phase and retirement phase investment “glide slope” that adjusts your investment style each over time. As you approach retirement, the investment mix adjusts to being closer to your retirement mix.

7) Our preference, but not required for everyone, is the ability to modify and adjust the tool for your own needs. This one may not be ideal for everyone, but we like to tweak and test out scenarios and see how things work. Having a tool that is available to tweak is a plus for us.

Before Your Simulate

Before you can start your own simulation, you need to do your own homework. The following are some key pieces of data you need before you can simulate regardless of which tool you use. It’s good to get this together first and may need a little upfront work on your part before you jump into the simulation tool.

A) Your current savings allocated for retirement, both deferred and non-deferred accounts.

B) Your annual savings plans for both deferred and non-deferred accounts.

C) What is your retirement income needed (in today’s dollars)? This one is key and should really be based on a real budget and not some rule of thumb. You hear “estimate 80% of your income” for retirement, but if you have aggressive savings this is not accurate.

D) Have a goal in mind for retirement age but stay flexible!

E) Have an idea of your expected life. This can be based on your own family history or can use an actuary table. The SSA has a tool for this online.

F) If needed, look up your social security estimated income but remember if you retire early it may not be this value. It uses the highest 35-year work history so if you work for less, your benefit will be reduced.

Remember – It’s Just A Model and Stay Flexible

As with all the models the real world may not cooperate, so stay flexible in your plans. We really can’t predict the future, but we can model scenarios and plan accordingly. You should not run an estimate and then stick with those numbers forever. At least annually, you should review your progress and re-estimate your probability for success. 

Retirement planning is a marathon and not a sprint, and over time priorities and personal situations change so stay flexible. Don’t get stuck on a specific number or date. 

One thing the models don’t account for is we can be flexible, so don’t get too caught up in 100% probabilities especially if your retirement date is far in the future. Extending your retirement by one or two years during a down market can make significant improvements in your future success. If you are already retired, maybe you postpone the big vacation till next year and take a few smaller road trips. 

Republished with the permission of

3 replies on “How to choose an early retirement simulation tool like a boss”

The reason why I like to run a retirement calculation based on average market performance is because plugging in an average market performance tends to be quicker and easier to do.

I agree with you that Simulation is the preferred way to go if you have the tool, knowledge and time to run different simulations. Given all the different data points and variables to put in (after all, the model is only as good as the inputs and assumptions), I believe working with a financial planner/adviser is the best approach for running Monte Carlo Simulations than a DIY approach (unless you put in the homework to really understand how the different assumptions/data affect the outcome).

I thought this was a pretty good summary and some good points.

I didn’t agree with point #5 about having a clean line between saving and spendown phases. I think in practice for most people there will not be a bright line between the two. Especially in this increasingly ‘gig’ economy.

Also a nerdy point – a monte carlo simulation can easily (and commonly) have mean reversion built into it. Most of the sophisticated monte carlo models used for investment have some kind of mean reversion built in, since that seems to be the outcome we see in the markets. It’s another argument about whether this is a sensible hypothesis to build into your model.

Thanks for posting.

I think if you want to dig in and understand how to master the Monte Carlo methods here, you’ll want to read “How to Measure Anything” by Douglas Hubbard.

It’s OK to make simple models with shortcomings if you understand them. Often, Monte Carlo is more resilient to mistakes than other modeling approaches because it builds in uncertainty. The biggest mistake to watch out for is using the correct distribution and making sure an assumption of Normal distribution is a reasonable one.

Leave a Reply

Your email address will not be published. Required fields are marked *