Here's a statement of the obvious: The opinions expressed here are those of the participants, not those of the Mutual Fund Observer. We cannot vouch for the accuracy or appropriateness of any of it, though we do encourage civility and good humor.
Support MFO
Donate through PayPal
More talk & thoughts on using Monte- Carlo The good & bad
Nice piece. Some things I might state a little differently, but a good presentation.
I would point out that all projections use models. Bengen used a model (past is prologue; assume that the future will repeat one of the past patterns). With some models, it's easy to calculate outcomes directly. For example, with a linear model (x% return per year, $N withdrawn/year), one can calculate exactly how much an investor will have at the end of each year. Other models are harder to "solve", so numeric methods are used to get approximate values.
One numeric method used to solve models is Monte Carlo. It just "runs" the model many times to see what outcomes occur at what frequencies. For example, suppose I model a coin flip as a 50/50 proposition. What is the distribution of outcomes if I flip the coin four times?
I could simply compute the combinations: 4 Heads (1/16) HHHH 3 Heads (4/16) HHHT, HHTH, HTHH, THHH 2 Heads (6/16) HHTT, HTHT, THHT, HTTH, THTH, TTHH 1 Head (4/16) HTTT, THTT, TTHT, TTTH 0 Heads (1/16) TTTT
Or I could run a Monte Carlo simulation, flipping a virtual coin 4 times over and over, and counting the number of times I got 4 heads, 3 heads, etc. Same model, two completely different approaches to "solving" it. Running a Monte Carlo simulation is merely a mechanism to solve the model.
What's key is the model itself. If I'm flipping a biased coin (the stock market wins more than it loses over the long term), then the model above (50/50) is garbage in, garbage out.
The model above was probabilistic, yet I could solve it without using a Monte Carlo simulation. So I try to take care in differentiating the model (hard to get right) from the use of the model to predict outcomes (easy).
This aside, the points made in the article are well stated. It's not just that tweaking parameters of a model can result in major changes in the predicted outcome, but also that the model itself may not be good. When you use a free "Monte Carlo tool", you're getting is a built-in model.
"the client only gets one shot at retirement. ... And if your plan can succeed in the worst-case scenario that should provide some degree of comfort."
This gets us to interpreting the forecasts. Even if they are accurate (i.e. good model, good choice of parameter values), what use does one make of the outputs? On the one hand, a possibility of failure, no matter how slim, can be viewed as unacceptable. OTOH, if one plans for the perfect storm, the once in a century possibility (like, say, a pandemic), then one is almost surely going to underspend/oversave by a lot.
So "all plans, even those with a 95% probability of success, require spending adjustments along the way."
Indeed mutual fund managers almost universally use the Monte Carlo tool. The Monte Carlo method was designed to address uncertain outcomes and nothing is more uncertain than the open investment marketplace.
Here is a Link to an easy to use Monte Carlo tool kit from Vanguard. It is a typical example. I use it. Please try a few what-if cases. I think you will like it.
Monte Carlo is not a perfect tool. It has shortcomings. No tool is perfect. It depends on user inputs and somewhat on his understanding of the limits of such tools. Probabilities are not for everyone. Garbage in, garbage out. That’s the case for all analyses. Good luck!!!
Some things are easy to forecast. I am but the markets are not. Good luck and persistence are essential ingredients. I do persist but my luck is somewhat uncertain.
It's good to hear from you. As you likely know I retired almost 25 years ago. At that time no Monte Carlo-like codes were available to me so I wrote my own simple version of that tool. It served me well but was not all inclusive.
Today's versions of that tool are much more complete and powerful. They will aid potential retirees in making good decisions. I will always take the opportunity to encourage their application for that purpose.
That 's surely a far distant application from their original use which was to help design the atomic bomb. The current applications are a tiny bit more sedate but more widely applied. The investment industry is a most frequent user to help in the retirement decision making process. But you knew that already.
Added comment: If a professional advisor does not make use of a sophisticated Monte Carlo tool, he or she is short changing his client. Monte Carlo is certainly not the total picture but can be a significant input. It is available so it should be exercised.
Comments
I would point out that all projections use models. Bengen used a model (past is prologue; assume that the future will repeat one of the past patterns). With some models, it's easy to calculate outcomes directly. For example, with a linear model (x% return per year, $N withdrawn/year), one can calculate exactly how much an investor will have at the end of each year. Other models are harder to "solve", so numeric methods are used to get approximate values.
One numeric method used to solve models is Monte Carlo. It just "runs" the model many times to see what outcomes occur at what frequencies. For example, suppose I model a coin flip as a 50/50 proposition. What is the distribution of outcomes if I flip the coin four times?
I could simply compute the combinations:
4 Heads (1/16) HHHH
3 Heads (4/16) HHHT, HHTH, HTHH, THHH
2 Heads (6/16) HHTT, HTHT, THHT, HTTH, THTH, TTHH
1 Head (4/16) HTTT, THTT, TTHT, TTTH
0 Heads (1/16) TTTT
Or I could run a Monte Carlo simulation, flipping a virtual coin 4 times over and over, and counting the number of times I got 4 heads, 3 heads, etc. Same model, two completely different approaches to "solving" it. Running a Monte Carlo simulation is merely a mechanism to solve the model.
What's key is the model itself. If I'm flipping a biased coin (the stock market wins more than it loses over the long term), then the model above (50/50) is garbage in, garbage out.
The model above was probabilistic, yet I could solve it without using a Monte Carlo simulation. So I try to take care in differentiating the model (hard to get right) from the use of the model to predict outcomes (easy).
This aside, the points made in the article are well stated. It's not just that tweaking parameters of a model can result in major changes in the predicted outcome, but also that the model itself may not be good. When you use a free "Monte Carlo tool", you're getting is a built-in model.
"the client only gets one shot at retirement. ... And if your plan can succeed in the worst-case scenario that should provide some degree of comfort."
This gets us to interpreting the forecasts. Even if they are accurate (i.e. good model, good choice of parameter values), what use does one make of the outputs? On the one hand, a possibility of failure, no matter how slim, can be viewed as unacceptable. OTOH, if one plans for the perfect storm, the once in a century possibility (like, say, a pandemic), then one is almost surely going to underspend/oversave by a lot.
So "all plans, even those with a 95% probability of success, require spending adjustments along the way."
Indeed mutual fund managers almost universally use the Monte Carlo tool. The Monte Carlo method was designed to address uncertain outcomes and nothing is more uncertain than the open investment marketplace.
Here is a Link to an easy to use Monte Carlo tool kit from Vanguard. It is a typical example. I use it. Please try a few what-if cases. I think you will like it.
https://retirementplans.vanguard.com/VGApp/pe/pubeducation/calculators/RetirementNestEggCalc.jsf
Monte Carlo is not a perfect tool. It has shortcomings. No tool is perfect. It depends on user inputs and somewhat on his understanding of the limits of such tools. Probabilities are not for everyone. Garbage in, garbage out. That’s the case for all analyses. Good luck!!!
Best Wishes
Some things are easy to forecast. I am but the markets are not. Good luck and persistence are essential ingredients. I do persist but my luck is somewhat uncertain.
Here is another Link to yet another Monte Carlo simulator that I find useful and easy to you:
https://www.portfoliovisualizer.com/monte-carlo-simulation#analysisResults
Please give it a sample run. A ton of portfolio candidates can be quickly examined to estimate their robustness under different market returns.
OJ
It's good to hear from you. As you likely know I retired almost 25 years ago. At that time no Monte Carlo-like codes were available to me so I wrote my own simple version of that tool. It served me well but was not all inclusive.
Today's versions of that tool are much more complete and powerful. They will aid potential retirees in making good decisions. I will always take the opportunity to encourage their application for that purpose.
That 's surely a far distant application from their original use which was to help design the atomic bomb. The current applications are a tiny bit more sedate but more widely applied. The investment industry is a most frequent user to help in the retirement decision making process. But you knew that already.
Added comment: If a professional advisor does not make use of a sophisticated Monte Carlo tool, he or she is short changing his client. Monte Carlo is certainly not the total picture but can be a significant input. It is available so it should be exercised.
Best Wishes