Kitces: Monte Carlo simulations overstate fat tail risk

[walkinwood]

Another interesting post by Kitces:

https://www.kitces.com/blog/monte-c...-withdrawal-rates-rolling-historical-returns/

His assertion is that since Monte carlo simulations treat each period independently, they fail to take into account the historical reversion to mean in market performance. Hence, they tend to overstate the effect of fat tails over long periods (like those used for retirement planning) when compared to historical results. The overstatement is more pronounced when you assume lower than historical market returns. He admits that there are only 114 historical annual (overlapping) periods, but MC does 1000s of simulations.

Of course, it is helpful to keep in mind that there is no "natural" principle that dictates that markets behave in a certain way. After all, there are some markets that dived to zero and no longer exist.

 

[photoguy]

His assertion is that since Monte carlo simulations treat each period independently, they fail to take into account the historical reversion to mean in market performance.

That depends on the particular study and you would need to dig into the implementation details to check. I'm pretty sure there a bunch that don't assume independence.

 

[Blueskies123]

I just finished reading the article. I would love to hear what Nissim Taleb thinks since I am 3/4 of the way through "Fooled by Randomness". In my own opinion, while I understand what Kitces is saying; my thought is that while there are only "114 historical annual (overlapping) periods, but MC does 1000s of simulations" I am thinking there are thousands of Black Swans that have not yet happened. Just because the last 114 historical periods have not produced the worst case MC scenarios I fear that since they can happen in MC simulations they could happen in real life.

 

[ERD50]

... In my own opinion, while I understand what Kitces is saying; my thought is that while there are only "114 historical annual (overlapping) periods, but MC does 1000s of simulations" I am thinking there are thousands of Black Swans that have not yet happened. Just because the last 114 historical periods have not produced the worst case MC scenarios I fear that since they can happen in MC simulations they could happen in real life.

Sure, but what action can we take? Do we decide 3.5% WR is too risky? OK, how about 3.0% then? Nope, better go down to 2, 1, 0, wait - we could have a 30 year period of negative returns in everything, with high inflation, might need to dip in and deplete principal in the first decade! And on and on.

OK, work till you die. There's the answer. (not being sarcastic, just pointing out the real choices)

No guarantees, but I went conservative enough that I should survive the worst of the worst in history. That's about as far as I can take planning, and still enjoy life. I'd feel bad if I could only say I planned for some of the worst, then the worst hits. And if it is worse, I'll probably still be better off than most.

-ERD50

 

[RunningBum]

Sure, but what action can we take? Do we decide 3.5% WR is too risky? OK, how about 3.0% then? Nope, better go down to 2, 1, 0, wait - we could have a 30 year period of negative returns in everything, with high inflation, might need to dip in and deplete principal in the first decade! And on and on.

OK, work till you die. There's the answer. (not being sarcastic, just pointing out the real choices)

Well, one alternative strategy would be to spend at a rate based on your remaining portfolio (like VPW), rather than spend at a certain SWR % + inflation and assume a recovery from a big downturn is coming, when it may not be. You'd theoretically never run out of money this way, at least not until you allowed yourself to take 100% at some presumably unattainable age. It's not immune to failure, since if your portfolio took a beating year after year you probably won't be able to cover your necessary basic expenses, but it would handle a long bad stretch early in your retirement better, assuming you have enough buffer to start cutting expenses early.

 

[haha]

Sure, but what action can we take? Do we decide 3.5% WR is too risky? OK, how about 3.0% then? Nope, better go down to 2, 1, 0, wait - we could have a 30 year period of negative returns in everything, with high inflation, might need to dip in and deplete principal in the first decade! And on and on.

OK, work till you die. There's the answer. (not being sarcastic, just pointing out the real choices)

No guarantees, but I went conservative enough that I should survive the worst of the worst in history. That's about as far as I can take planning, and still enjoy life. I'd feel bad if I could only say I planned for some of the worst, then the worst hits. And if it is worse, I'll probably still be better off than most.

-ERD50

I know you are a careful person, in what you do and what you say. So I will point out that what most of us have is not actually the worst of the worst in history, but the worst of the worst in a fairly limited history of the most successful political system and economy and stock market ever.

And, there may be reason to think that the same conditions that produced this very good run may no longer really hold. I have no better suggestion, just a comment.

Ha

 

[Blueskies123]

Sure, but what action can we take? Do we decide 3.5% WR is too risky? OK, how about 3.0% then? Nope, better go down to 2, 1, 0, wait - we could have a 30 year period of negative returns in everything, with high inflation, might need to dip in and deplete principal in the first decade! And on and on.

OK, work till you die. There's the answer. (not being sarcastic, just pointing out the real choices)

No guarantees, but I went conservative enough that I should survive the worst of the worst in history. That's about as far as I can take planning, and still enjoy life. I'd feel bad if I could only say I planned for some of the worst, then the worst hits. And if it is worse, I'll probably still be better off than most.

-ERD50

I think we have different takes on the same article. You seem to think the article says you need to be more conservative or work until you die. From what I read, Kitces is saying tail risks are over stated by Monte Carlo ergo maybe you should spend more money. My action is to stay pat at 4%. Like other posters have said the next 20 years may not be as nice as the next 20 years. I am not increasing my withdrawal rate.

 

[ggbutcher]

Thing is, the stock market is a rather complex mechanism, and asserting that its performance can be characterized by a mean and standard deviation is just too simplistic. Monte Carlo simulation is not the particular problem, it's the particular statistical model being applied in it.

There's a bit of literature that asserts a Laplace distribution more appropriately represents general stock market behavior than a plain 'ole normal distro, particularly in the extreme outcomes. That might pull the 'fat tails' into something more palatable.

 

[GravitySucks]

Thanks Kitces, you seem like the only guy out there that's optimistic. I need some reassurance sometimes.

 

[ERD50]

I know you are a careful person, in what you do and what you say. So I will point out that what most of us have is not actually the worst of the worst in history, but the worst of the worst in a fairly limited history of the most successful political system and economy and stock market ever.

And, there may be reason to think that the same conditions that produced this very good run may no longer really hold. I have no better suggestion, just a comment.

Ha

Yes, all you say about history and these conditions seems very true. While there was bad stuff in there (Great Depression and 1970's inflation), it was followed by some good runs (post WWII boom, bull run of the 90's) which allowed for recovery. Sure, it might not repeat so cleanly next time.

It's no exact science, nothing deterministic, so we all need to go with our gut, and base it on something, or as I said, work till we die. I think most of us who have gotten to this point have learned to vigilant and flexible, and those traits will serve us if things go very bad.

We'll both make it I think, one way or the other. -ERD50

 

[USGrant1962]

I found his premise that people think MC understates risk to be odd.

I've always seen worse results with MC than say FIRECalc. Most MC models I run show me at 80-95% success rates while historical FIRECalc gives over 100% success.

Have others seen MC results that are more optimistic than FIRECalc?

 

[ERD50]

I found his premise that people think MC understates risk to be odd.

I've always seen worse results with MC than say FIRECalc. Most MC models I run show me at 80-95% success rates while historical FIRECalc gives over 100% success.

Have others seen MC results that are more optimistic than FIRECalc?

The problem I have with MC models, is they can produce just about any result, depending on the implementation the programmer used.

I've heard that some get tweaked until they match historic results - that seems circular and pointless to me. Though as haha pointed out earlier, our history is limited to a fairly small numbers of actual economic cycles, so maybe the odds of seeing something much worse than our recent history worst cases is higher than we might think?

I'll still take comfort in the fact that I'm probably better prepared financially than most, so I'll likely do better than most in any future bad scenario. I don't think we can ever be certain about anything.

-ERD50

 

[Blueskies123]

I found his premise that people think MC understates risk to be odd.

Odd yes; dangerous probably.

 

[Scuba]

IMO Monte Carlo analyses tend to be conservative because they assume one's withdrawal pattern will not change over time. If for example you would reduce spending during a bear market, as many would, MC doesn't take this into account. If you would change your asset allocation (i.e. Use bonds or cash during a bear market rather than withdrawing equities) this is also not taken into account. However MC analysis is useful as a broad indicator of likely success, certainly better than predicting using static future return estimates.

 

[audreyh1]

IMO Monte Carlo analyses tend to be conservative because they assume one's withdrawal pattern will not change over time. If for example you would reduce spending during a bear market, as many would, MC doesn't take this into account. If you would change your asset allocation (i.e. Use bonds or cash during a bear market rather than withdrawing equities) this is also not taken into account. However MC analysis is useful as a broad indicator of likely success, certainly better than predicting using static future return estimates.

Monte Carlo analysis does take into account not drawing from equities during market downturns. Rebalancing takes care of this automatically.

 

[tinlizzy]

I feel like the biggest risk I face is a change in my own behavior in retirement. Will I be taken advantage of in my old age? Will I overreact during a downturn?

 

[audreyh1]

I feel like the biggest risk I face is a change in my own behavior in retirement. Will I be taken advantage of in my old age? Will I overreact during a downturn?

Have you been through a major downturn before? That will give you a clue.

 

[daylatedollarshort]

I know you are a careful person, in what you do and what you say. So I will point out that what most of us have is not actually the worst of the worst in history, but the worst of the worst in a fairly limited history of the most successful political system and economy and stock market ever.

And, there may be reason to think that the same conditions that produced this very good run may no longer really hold. I have no better suggestion, just a comment.

Ha

I think that sentiment is summed up well by the Triumph of the Optimists authors' research. Their point was a country has only one past and many possible futures and historical equity returns from a 17 country analysis don't paint a picture as rosy as just looking at the U.S. stock market returns, though of course it is possible the future could be even better than anything we've seen in the past.

 

[tinlizzy]

Have you been through a major downturn before? That will give you a clue.

I've had nerves of steel in the past but if we're going to speculate about unknown black swan events then I think that I should include my own behavior. Since I have total control over my finances I could be my own worst enemy.

 

[GravitySucks]

I feel like the biggest risk I face is a change in my own behavior in retirement. Will I be taken advantage of in my old age? Will I overreact during a downturn?

Fear and Greed are my main concerns.

 

[audreyh1]

I've had nerves of steel in the past but if we're going to speculate about unknown black swan events then I think that I should include my own behavior. Since I have total control over my finances I could be my own worst enemy.

Nerves of steel in the past indicate good chance of same in the future, IMO.

 

[haha]

I think that sentiment is summed up well by the Triumph of the Optimists authors' research. Their point was a country has only one past and many possible futures and historical equity returns from a 17 country analysis don't paint a picture as rosy as just looking at the U.S. stock market returns, though of course it is possible the future could be even better than anything we've seen in the past.

I read this book also, and it really affected me. Your distillation that their point is that a country has one past, and many possible futures, is a memorable way to express this.

I also agree with sentiments above, one either goes ahead or delays retirement until later is also true. I happen to believe that most of us who quit early are pushed by negative feelings, and that from a purely rational pov we might realize that there is more risk than is commonly stated.

But as has been said, this is an early-retirement board, so I will shut up.

Ha

 

[daylatedollarshort]

Sure, but what action can we take? Do we decide 3.5% WR is too risky? OK, how about 3.0% then? Nope, better go down to 2, 1, 0, wait - we could have a 30 year period of negative returns in everything, with high inflation, might need to dip in and deplete principal in the first decade! And on and on.

OK, work till you die. There's the answer. (not being sarcastic, just pointing out the real choices)

At a zero real return, for 30 years a retiree could safely withdraw 3.33% until portfolio depletion. TIPS currently are at .28% - 1.05% real returns, so depending on the maturities, a TIPS ladder that could possibly increase the safe withdrawal rate to closer to 4%. It may not be the best choice for every retiree, but it is one of the choices out there and provides a 4% safe withdrawal rate with limited risk (U.S. Treasuries).

 

[Scuba]

Monte Carlo analysis does take into account not drawing from equities during market downturns. Rebalancing takes care of this automatically.

Hmmm, didn't think so but not sure. I thought it assumed static prorata withdrawals across asset classes? Also doesn't MC assume a static dollar spend regardless of market performance?

 

[audreyh1]

Hmmm, didn't think so but not sure. I thought it assumed static prorata withdrawals across asset classes? Also doesn't MC assume a static dollar spend regardless of market performance?

However they actually do it, rebalancing occurs annually and even if they take fixed amount from the various asset classes initially in proportion to current ratios, the equities will immediately replaced from fixed income once the rebalancing is done.

The algorithms are probably smarter and do what we do in the real world, take take the withdrawal from each asset class as needed to rebalance the portfolio. If equities are creamed, they won't be touched.

Static (real) spend - that doesn't make a difference. It's the rebalancing that matters.