Monte Carlo Simulations Pros and Cons
Monte Carlo Simulations Pros and Cons
Would anyone be able to share on the pros and cons of the "Monte Carlo" simulations with regard to retirement planning? Or share any resources which deal with the pros and cons.
 E
 E
The major cons are that the outcome depends entirely on your inputs, which are predictions of the future and therefore not likely to be accurate (i.e., garbage in, garbage out), and that people will place undue reliance on the results.
The pro is that it can show the relation between return and volatility.
The pro is that it can show the relation between return and volatility.
I use Monte Carlo simulation all the time at work but am not too familiar with the financial planning implementations but I have a pretty good idea of what they are doing.
In a nutshell, what is going on is that they are taking the known inputs (your current financial situation) and applying many many iterations of possible future scenarios according to their model and giving you the statistics based on the results of all the iterations.
The cons are really that this is all model dependent what likelihood did the person who wrote the code give to equities increasing 10% every year forever? What likelihood did they give to inflation being 35%? Of course they probably used past behavior, but at the end of the day we are simply trying to predict the future.
And of course, having a "90% chance" of meeting your goal means 10% of the time the simulation determined you would not. You may attempt to increase your odds, but until you hit 100% (which would mean finding the 100% risk free investment plan having absolutely nothing at all to lose) you cannot guarantee anything. Plus the model could turn out to be just plain wrong increasing your outcome in the model means you are pleasing the model, not reality.
Monte carlo simulation is a good tool, and certainly useful in this case, but I would have to say it is certainly more powerful for physics applications than financial ones where many of the things that affect the global market and its impact on your personal finances may not even be quantifiable or even fully understood/recognized.
I think it's always good for perspective to remember that we humans are the ones who looked at the stars in the sky and determined there were clearly meaningful pictures drawn there that we could interpret.
In a nutshell, what is going on is that they are taking the known inputs (your current financial situation) and applying many many iterations of possible future scenarios according to their model and giving you the statistics based on the results of all the iterations.
The cons are really that this is all model dependent what likelihood did the person who wrote the code give to equities increasing 10% every year forever? What likelihood did they give to inflation being 35%? Of course they probably used past behavior, but at the end of the day we are simply trying to predict the future.
And of course, having a "90% chance" of meeting your goal means 10% of the time the simulation determined you would not. You may attempt to increase your odds, but until you hit 100% (which would mean finding the 100% risk free investment plan having absolutely nothing at all to lose) you cannot guarantee anything. Plus the model could turn out to be just plain wrong increasing your outcome in the model means you are pleasing the model, not reality.
Monte carlo simulation is a good tool, and certainly useful in this case, but I would have to say it is certainly more powerful for physics applications than financial ones where many of the things that affect the global market and its impact on your personal finances may not even be quantifiable or even fully understood/recognized.
I think it's always good for perspective to remember that we humans are the ones who looked at the stars in the sky and determined there were clearly meaningful pictures drawn there that we could interpret.
Re: Monte Carlo Simulations Pros and Cons
elgaeb051 wrote:Would anyone be able to share on the pros and cons of the "Monte Carlo" simulations with regard to retirement planning? Or share any resources which deal with the pros and cons.
 E
I find little use (being generous) in any process that tries to project/predict the future (unknown).
A person that has talked about this is quoted in my signature.
Landy 
Be yourself, everyone else is already taken  Oscar Wilde

 Posts: 657
 Joined: Sun Mar 13, 2011 2:21 pm
I see only one pro or con, and its a con. THis is all complete simulation based on current inputs and using data from the past. I don't see any relation to the future unless you believe the past will be repeated in the future. Greece was once a world power. So was UK. China was, then wasn't, now may be. We were nothing in the US, then a world power for 100 years, and what about the next 50 years? Who knows?
Uninvested wrote:THis is all complete simulation based on current inputs and using data from the past.
There's no reason that one must use data from the past when running simulations. For example, if you believe that expected returns are lower than the historical average looking forward, just plug in your own numbers.
John Norstad
elgaeb051 wrote:Others, I've read said you need 25 times your income? I don't know, how valid is that?
There are a number of studies of past data which concluded that one could safely withdraw 4%, inflation adjusted, of a portfolio for 30 years. 4% means you'd need 25x the needed income.
These studies are based on a about 75 or so years of history. If the future is better than these past years, you could withdraw more, if worse than the past, you wouldn't be able to withdraw as much.
Are you trying to determine if you have enough to retire? How old are you? Do you want to leave much to heirs? If interested in covering spending an annuity can be useful, but not if you want to leave a lot to heirs. A more specific question might prompt a more specific answer.
jln wrote:Uninvested wrote:THis is all complete simulation based on current inputs and using data from the past.
There's no reason that one must use data from the past when running simulations. For example, if you believe that expected returns are lower than the historical average looking forward, just plug in your own numbers.
Just don't have much faith in those numbers and don't believe that the product of a complex calculation is any more accurate than sticking a finger into the wind.
Well, you have to plan, and that requires running some scenarios about what might happen in the future.
Monte Carlo simulation is a way to do this. One can plan by assuming constant returns for each asset class, and running through the calculation for many different combinations of returns. However, this can ignore the risk of running out of money during the period, even if the long term rates of return are high enough. That is where the simulations are helpful.
Since the future is unpredictable, then it is likely that any method of prediction will be incorrect, ex post. MC simulation is no more right or wrong than other approaches. But that is no reason to reject MC simulation.
This would be like saying "the future is unpredictable, so one should not use the mathematical process of multiplication in planning".
If your failure rate is higher than zero, then it does make sense to give some thought to what to do about it. The simplest solution is to revise your assumptions until you get a reassuring result. However, the more conservative approach is to save more or spend less. Increasing your expected returns makes your new plan dependent on things completely out of your control. Revising your asset allocation towards more risk can be appealing, but MC can help you identify the potential consequences. Another approach is to change the inflation rate, and see how you would fare as it goes higher.
One risk of all these approaches is that no one knows the correlation between the returns and inflation, let alone with your earning power. You would like to plan not only for a full career and retirement, but the risk that you will lose your job during a period of prolonged stagflation. I don't know of a calculator that takes that into account.
Monte Carlo simulation is a way to do this. One can plan by assuming constant returns for each asset class, and running through the calculation for many different combinations of returns. However, this can ignore the risk of running out of money during the period, even if the long term rates of return are high enough. That is where the simulations are helpful.
Since the future is unpredictable, then it is likely that any method of prediction will be incorrect, ex post. MC simulation is no more right or wrong than other approaches. But that is no reason to reject MC simulation.
This would be like saying "the future is unpredictable, so one should not use the mathematical process of multiplication in planning".
If your failure rate is higher than zero, then it does make sense to give some thought to what to do about it. The simplest solution is to revise your assumptions until you get a reassuring result. However, the more conservative approach is to save more or spend less. Increasing your expected returns makes your new plan dependent on things completely out of your control. Revising your asset allocation towards more risk can be appealing, but MC can help you identify the potential consequences. Another approach is to change the inflation rate, and see how you would fare as it goes higher.
One risk of all these approaches is that no one knows the correlation between the returns and inflation, let alone with your earning power. You would like to plan not only for a full career and retirement, but the risk that you will lose your job during a period of prolonged stagflation. I don't know of a calculator that takes that into account.
 jasonlitka
 Posts: 270
 Joined: Sat Nov 13, 2010 8:30 am
 Contact:
Re: Monte Carlo Simulations Pros and Cons
YDNAL wrote:elgaeb051 wrote:Would anyone be able to share on the pros and cons of the "Monte Carlo" simulations with regard to retirement planning? Or share any resources which deal with the pros and cons.
 E
I find little use (being generous) in any process that tries to project/predict the future (unknown).
A person that has talked about this is quoted in my signature.
I prefer the Scott Adams approach.
There are, in general, two ways to predict the future. For example, you can use horoscopes, tea leaves, tarot cards, crystal ball, and so forth. Collectively, these are known as the “nutty methods.” Or you can put wellresearched facts into sophisticated computer models, more commonly referred to as “a complete waste of time.” While all these approaches have their advantages, I find it’s a lot easier and economical to simply make stuff up
Jason Litka
There are different types of Monte Carlo simulations. One picks annual returns at random from the past (bootstrapping). This ignores reversion to the mean, and so probably overstates the variation in returns.
An alternate approach is to pick 30 year sequence of returns from the past. How would an investor starting in 1942 have done? The problem with this is that there are few independent 30 year periods. So this probably understates the variation in returns. Something new is probably going to hit the fan.
Another approach is to assume a normal distribution with a certain mean and standard deviation. This ignores fat tails. (see the Black Swan for a discussion)
Overall, I see
pros: avoids recency bias,
lists many of the possible outcomes.
cons: noisy data so hard to draw any useful conclusions,
can lead to a false sense of security (aka the 4% rule).
An alternate approach is to pick 30 year sequence of returns from the past. How would an investor starting in 1942 have done? The problem with this is that there are few independent 30 year periods. So this probably understates the variation in returns. Something new is probably going to hit the fan.
Another approach is to assume a normal distribution with a certain mean and standard deviation. This ignores fat tails. (see the Black Swan for a discussion)
Overall, I see
pros: avoids recency bias,
lists many of the possible outcomes.
cons: noisy data so hard to draw any useful conclusions,
can lead to a false sense of security (aka the 4% rule).
Re: Monte Carlo Simulations Pros and Cons
jln wrote:Uninvested wrote:THis is all complete simulation based on current inputs and using data from the past.
There's no reason that one must use data from the past when running simulations. For example, if you believe that expected returns are lower than the historical average looking forward, just plug in your own numbers.
John Norstad
The future is unknown to me.
 I'm 100% certain.
 An attempt to "simulate" the unknown  with either historical or projectionbased data  is then a futile task.
afan wrote:Well, you have to plan, and that requires running some scenarios about what might happen in the future.
Monte Carlo simulation is a way to do this. Since the future is unpredictable, then it is likely that any method of prediction will be incorrect, ex post. MC simulation is no more right or wrong than other approaches. But that is no reason to reject MC simulation.
There is NO requirement to "run scenarios of what might happen" in the future to plan ahead.
For instance, one plan can be to own Fixed Income to support a 2% withdrawal rate.
Landy 
Be yourself, everyone else is already taken  Oscar Wilde
Re: Monte Carlo Simulations Pros and Cons
elgaeb051 wrote:Would anyone be able to share on the pros and cons of the "Monte Carlo" simulations with regard to retirement planning? Or share any resources which deal with the pros and cons.
Since you asked for references to some written materials discussing pros and cons of Monte Carlo methods, here are some links:
A Critique of Monte Carlo Retirement Calculators, by James Welch (Optimal Retirement Planner): http://www.iorp.com/modeldescription/MonteCarlo.html
Odds On Imperfection: Monte Carlo Simulation, by Eleanor Laise (Wall Street Journal article): http://online.wsj.com/article/SB124121875397178921.html
Monte Carlo Model: Is it Good for Your Client, by Jim Otar: http://www.retirementoptimizer.com/articles/MCArticle.pdf
Before retirement I had once written a Monte Carlo program for predicting electron beam  solid material interactions. Predictions were dead on because the underlying physics was well understood. But for predictions of investment returns there are only very rough estimates of what the future might look like. So the Monte Carlo predictions will only be very rough estimates. If you see someone predicting retirement plan success to the nearest 0.1% (or even to the nearest 1%), then you can be fairly certain they don't appreciate the magnitude of the uncertainties involved.
Investment skill is often just luck in sheep's clothing.
For instance, one plan can be to own Fixed Income to support a 2% withdrawal rate.
Isn't this plan based on the assumption that a 2% withdrawal rate is safe? Isn't that assumption based on an implied real return? What if the real return of these fixed income assets is significantly negative? Ignore that possibility (i.e. explicitly decide not to run those scenarios)?
I don't see how this example says one need not run scenarios in order to plan.
Re: Monte Carlo Simulations Pros and Cons
afan wrote:For instance, one plan can be to own Fixed Income to support a 2% withdrawal rate.
Isn't this plan based on the assumption that a 2% withdrawal rate is safe? Isn't that assumption based on an implied real return? What if the real return of these fixed income assets is significantly negative? Ignore that possibility (i.e. explicitly decide not to run those scenarios)?
I don't see how this example says one need not run scenarios in order to plan.
No.
1. The plan is to save your @$$ off.
2. If Fixed Income saved is >50 times the amount of Income need, the buck stops right there!
3. If Fixed Income is anything less than 100% of assets, there is more than 0% of assets in Equities.
Landy 
Be yourself, everyone else is already taken  Oscar Wilde
I'm sorry, I still don't get it.
The assumption is fixed income= 50 x annual needs, and that is sufficient.
But what is the basis for the assumption that this is sufficient? You spend 1/50th of assets the first year. Now you have high inflation, and you are spending 5% of assets in year 2, 10% of what is left in year 3...
This is still a plan based on assumptions about returns and inflation. You can pretend it is not by refusing to check. But that is like pretending your investments are not volatile because you do not look at share prices. The assumptions are still there, and the plan is based on them.
The assumption is fixed income= 50 x annual needs, and that is sufficient.
But what is the basis for the assumption that this is sufficient? You spend 1/50th of assets the first year. Now you have high inflation, and you are spending 5% of assets in year 2, 10% of what is left in year 3...
This is still a plan based on assumptions about returns and inflation. You can pretend it is not by refusing to check. But that is like pretending your investments are not volatile because you do not look at share prices. The assumptions are still there, and the plan is based on them.
A MC simulation is useful to show that there can be a wide range of outcomes. This should alert the individual to remain flexible and have contingencies for eliminating negative outcomes. I think sometimes people can have a false sense of security when using very simple models. A MC based model can show that life is not as well known as maybe assumed. It is a tool. Like all tools, tools must be used properly. I think that is what the OP is trying to determine. If a person is not comfortable using a particular tool, then use a different one.
Re: Monte Carlo Simulations Pros and Cons
afan wrote:I'm sorry, I still don't get it.
The assumption is fixed income= 50 x annual needs, and that is sufficient.
But what is the basis for the assumption that this is sufficient? You spend 1/50th of assets the first year. Now you have high inflation, and you are spending 5% of assets in year 2, 10% of what is left in year 3...
This is still a plan based on assumptions about returns and inflation. You can pretend it is not by refusing to check. But that is like pretending your investments are not volatile because you do not look at share prices. The assumptions are still there, and the plan is based on them.
Where?
You are assuming to know how income need is calculated. What if income has 5% increases builtin for 30 years for a splurge factor? Flexibility in spending is ALWAYS the best course.
Look, we are headed towards a back/forth about nothing, so I choose to leave it right there!
Back on subject, there's NO requirement to "run some scenarios about what might happen in the future."
afan wrote:Well, you have to plan, and that requires running some scenarios about what might happen in the future.
Landy 
Be yourself, everyone else is already taken  Oscar Wilde
DonSmith wrote:There are different types of Monte Carlo simulations. ... This ignores reversion to the mean, and so probably overstates the variation in returns.
Yes, most (all?) simulators ignore reversion to mean, and this probably overstates the variation in returns.
Most (all?) also ignore fat tails, and this probably understates the variation in returns.
So there you go! It would be cool if they cancelled each other out, wouldn't it? Too much to hope for, I think.
There's so much we don't know. Trying to come up with good estimates for expected returns is the least of it.
John Norstad
What if income has 5% increases builtin for 30 years for a splurge factor? Flexibility in spending is ALWAYS the best course.
Sounds like running scenarios to me.
I suppose no answer to the question of how 2% withdrawal rate avoids scenario planning.
Anyway, since you have to run scenarios to plan, the problem with any method is how exhaustively you investigate the possible outcomes. If you use MC, you still have to decide which relationships to include. As others have noted this may include regression to the mean, the challenge of what distribution to assume, or how many different distributions, with how many parameters, with what range of values for each parameter, to include. Also need to worry about serial autocorrelation, correlations among significant events. If you include foreign stocks, you have to include some estimates of their association with domestic stock returns. These appear to be changing over time.
Simply saying "MC" does not specify how carefully one does this, and it does not tell how reliable the results will be.
That said, MC is a good way to look at the range of outcomes and test the sensitivity of your current plans to changes in these factors. It is a tool, like addition.

 Posts: 657
 Joined: Sun Mar 13, 2011 2:21 pm
jln wrote:Uninvested wrote:THis is all complete simulation based on current inputs and using data from the past.
There's no reason that one must use data from the past when running simulations. For example, if you believe that expected returns are lower than the historical average looking forward, just plug in your own numbers.
John Norstad
I agree, One still has to determine what to put in and since we aren't clairvoyant isn't it an exercise in futility?
No matter what you do it is only an estimate about the future and Monte Carlo is an improvement over what has been used in the past. The worst thing you can do is plug in some historical averages and project the outcome 30 years into the future, and base all your financial decisions on that one outcome. Monte Carlo may enable you to do some sensitivity analysis, and plan for various outcomes, which is what you should be doing. I find Efficient Solutions MCRetire to be useful, but it is only one of many options.
There is no way to predict the future, but any tool that allows to speculate about future outcomes is certainly worth considering.
Sam
There is no way to predict the future, but any tool that allows to speculate about future outcomes is certainly worth considering.
Sam