There is NO history for retirement planning

[smjsl]

FIRECalc, 4% rule, etc. - are they not producing a false sense of security?

I know this was discussed some time ago but I was just analyzing how come I don't feel secure with them when many people seem to rely on these for their planning, at least to some degree. So I think I understand why now. Hope the following line of reasoning is wrong somewhere (although I am sure I will be arguing otherwise).

Say you are planning for 30-year retirement, even though based on the recent thread, most people here try to plan for at least 40 years ahead it seems. Now, FIRECalc, 4% rule, and most if not all of the other research for retirement planning look at the "long" history of the last century. Most start with 1920's. FIRECalc starts with 1879 (i.e. about 130 years of history). This may seem long at first, but you have to keep in mind that compared to the 30 (or more likely 40) years, it's not long at all.

In other words, they look at just 3-4 time spans relative to what you are trying to predict.

How is it any different than trying to guess distribution of market returns over the next year if all you had was historical returns of the past 3-4 years? That's all we seem to be doing with above approaches.

Note how for the 30-year period FIRECalc says that it shows results of 110 runs! Those are not independent results however - they are based on many overlapping time periods. Similarly, if I gave you 3-4 years worth of data, you could produce in the same way hundreds of 1-year runs based on starting your 1-year periods each day in those 3-4 years: Jan 1 of year 1, Jan 2 of year 1, Jan 3 of year 1, etc.

So, do you think there is any point in trying to predict next year returns with any degree of certainty if you have results from just 3-4 prior years? Sure people add some fudge factors and buffers, but they seem pretty small compared to possible variations. Someone who says 'I use 3% instead of 4% in my rule' is (approximately?) saying, if past 4 years averaged 10% return, I will plan for at least 7.5% return during next "retirement" year (i.e. whole 25% lower than past average).

As another example - sometimes you hear : 'we never had 30 years of negative market returns'... well yes, but again, we only had a sample of 4 independent 30-year periods... If during 4 years worth of data you did not have a single year of negative returns (starting at any day), does it mean there is a good chance of not having a negative return over the next year? (I am sure some paper somewhere answers this question )

Some might say, 'Well at least I am looking at 3-4 years worth of data to plan my "retirement" year - that's better than nothing'... Is it? Or is it no more help than looking at your horoscope? (because there is so little data for your 30-40 year horizon that it's meaningless)

 

[Gone4Good]

In other words, they look at just 3-4 time spans relative to what you are trying to predict.

Not exactly. Retiring in 1929 is substantially different than retiring in 1932, even though 93% of the succeeding 40 year periods overlap. I don't see how these two periods could rationally reflect just one data point.

 

[Cb]

A century or so of data includes bubbles/busts, periods of war & peace, several economic cycles, shifts in politics, effects of technological breakthroughs, scandals, etc. I'm inclined to use the longer period rather than just 2006-2009.

Look at it this way..100% TIPS ought to allow a bit over a 3% SWR...market history suggests that a sizeable equity allocation should boost that somewhat.

Cb

 

[chinaco]

There are no guarantees.

Even if the market performs as expected based on the past does not mean that you will (your investments)... you might make detrimental mistakes.

Throw in all of the other possibilities that might cause you problems... now many things could go wrong.

All one can do is make decisions based on information at hand (and hopefully you have good information). The rest is up to you... you can make adjustments along the way!

If you are really concerned... go with a lower % or a TIPs ladder for your joint life expectancy plus 5 years.

 

[smjsl]

Not exactly. Retiring in 1929 is substantially different than retiring in 1932, even though 93% of the succeeding 40 year periods overlap.

Same applies to what I am saying. Since you picked the largest crash, pick a 3-4 years worth of data that includes some crash like the one in 87. You could then say 1-year retirement starting a day or a week before the crash is much different than the one starting a day or a week after the crash.

A century or so of data includes bubbles/busts, periods of war & peace, several economic cycles, shifts in politics, effects of technological breakthroughs, scandals, etc. I'm inclined to use the longer period rather than just 2006-2009.
Cb

When you have 1-year worth of retirement, prior 3-4 years will have analogous "crisis" - at the same smaller scale. Remember that you have to scale down the events 30-40 times as well. E.g. in 2006-2009 - there were a number of "big evens" relative to this scale as well... Remember the 20% "collapse" in emerging markets and the scare of May 2006? Not to mention the whole subprime and other fiasco's in 2008? If you scale these up 30-40 times, they will be more than comparable to various events at the "real-life" scale...

[...] history suggests that a sizeable equity allocation should boost that somewhat.

That's my point. I am not so sure there is enough history to suggest anything, at least not for a 30-40 year span.

 

[smjsl]

All one can do is make decisions based on information at hand (and hopefully you have good information).

Agreed. And my point is that many seem to imply / suggest that last 90 years (130 in case of firecalc) is "good" information. I am saying how can it be good when it only encompasses 3 or 4 independent time periods for your planning... thus, my analogy.

 

[Gone4Good]

You could then say 1-year retirement starting a day or a week before the crash is much different than the one starting a day or a week after the crash.

Yes, because they are.

Crash T-1 50% levered investor owning 100% equity portfolio = failure
Crash T+1 50% levered investor owning 100% equity portfolio = success

How can these not be two different data points if, all else being equal, we get two completely different results?

 

[W2R]

There is no guarantee for anything in life. If you have a job, that job could disappear at any moment (as so many in our country have sadly discovered during the past few years).

When you get a "95% successful" result from FIRECalc, all that means is that given the assumptions of the model (such as historical market behavior continuing), the model has computed a 95% probability of success. The model says nothing about the probability of those assumptions being correct.

Therefore, the best use of models, retirement planning, and so on, are for determining a ballpark level of spending and withdrawal that your nestegg may be able to handle. Flexibility is an important part of retirement planning. Many of us cut back during the stock market crashes of 2008-2009, for example, and a few even went back to work temporarily.

Likewise, just getting a paycheck from a job tells us nothing about the probability of that job continuing.

All promise outruns the performance. We live in a system of approximations. Every end is prospective of some other end, which is also temporary; a round and final success nowhere. - - R.W. Emerson, 1844

 

[chinaco]

Agreed. And my point is that many seem to imply / suggest that last 90 years (130 in case of firecalc) is "good" information. I am saying how can it be good when it only encompasses 3 or 4 independent time periods for your planning... thus, my analogy.

There are other tools and studies.

Using stochastic methods like monte carlo can provide further insight (non determinant)... But the data set is the data set. You might find something that broadens the securities.

It looks like Firecalc has a random option where you specify the return and the variability... to what if.

All bets are off if something fundamentally changes which would make the data set less relevant... or a catastrophic event occurs.

As they say... the market may not perform as it did in the past.

Ultimately you will take some risk retiring! No way around it.

For that matter fundamental changes to the negative would not guarantee you would have a job if you wanted to work.

The only thing you can guarantee is that you will eventually die (and if the govt has any say... pay taxes along the way).

IMO - It is about the degree of confidence!

 

[REWahoo]

Yes, because they are.

Correct.

Smjsl, with one exception, the only thing we can be certain about is that the future is uncertain. With the possible exception of the hammer, no tool is perfect and FIRECalc is nothing more than a tool. But I believe FIRECalc may be the best tool available to give us the opportunity to view the success of a particular withdrawal strategy from a historical perspective.

If you are overly concerned about the results you get from FC, lower your withdrawal rate, delay retirement, work part time, or work until the grim reaper comes calling. His eventual arrival is the one thing we know for certain that the future holds for us.

 

[Rich_by_the_Bay]

There seems to be an assumption here that the usual market history-based 4% SWR is a "set it and forget it" strategy. I don't think it is intended as such. Rather, at least the way I approach it, it is an initial sanity check of what seems to have worked so far so that one's initial plans are not implausible or rash.

But after that, a fair amount of flexibility in implementation is key. I wouldn't treat it as a fixed life plan. Measures such as variable expense control, Clyatt's (and others') percent of total portfolio method, Otar's danger signs with appropriate annuitization of distressed portfolios, modification of 4% withdrawals to avoid reverse DCA during tough years, etc. all contribute to success.

The traditional parameters used for Firecalc type projections are just a starting point, not a guarantee of success. 35 years is too long a period to blindly follow a static plan.

 

[Midpack]

There seems to be an assumption here that the usual market history-based 4% SWR is a "set it and forget it" strategy. I don't think it is intended as such. Rather, at least the way I approach it, it is an initial sanity check of what seems to have worked so far so that one's initial plans are not implausible or rash.

But after that, a fair amount of flexibility in implementation is key. I wouldn't treat it as a fixed life plan. Measures such as variable expense control, Clyatt's (and others') percent of total portfolio method, Otar's danger signs with appropriate annuitization of distressed portfolios, modification of 4% withdrawals to avoid reverse DCA during tough years, etc. all contribute to success.

The traditional parameters used for Firecalc type projections are just a starting point, not a guarantee of success. 35 years is too long a period to blindly follow a static plan.

+1, this would be my answer too.

I don't get the OP's hangup with completely independent periods, overlapping periods are statistically valid.

And critical opinions that offer no potential solutions don't accomplish much IMO. If you have a better idea, I am sure we'd all be interested. We all deal with risks from birth to death, no plan can make risk go away because you choose to retire - no getting around that. You accept the risk that allows you to sleep at night - or not.

You can work until you drop or accept some well thought out risk and be prepared to course correct - a choice each of us will make. Best of luck with your choices...

 

[dex]

Say you are planning for 30-year retirement, even though based on the recent thread, most people here try to plan for at least 40 years ahead it seems. Now, FIRECalc, 4% rule, and most if not all of the other research for retirement planning look at the "long" history of the last century. Most start with 1920's. FIRECalc starts with 1879 (i.e. about 130 years of history). This may seem long at first, but you have to keep in mind that compared to the 30 (or more likely 40) years, it's not long at all.

In other words, they look at just 3-4 time spans relative to what you are trying to predict.

Your issue with Firecal is that it only 3-4 time spans. If you are trying to find a fault with Firecalc it would not be that the number of time spans is small; the issue/question would be does it contain all the outcomes.

Think of the issue as with probability. If you have one dice we know that the probability of one face being on top is 1 out of 6. So you roll the dice 12 times and note that the results are not 1 out of 6. The dice contains the proper number of probabilities you did not roll the dice enough to get to the proper probability.

Do you think that Firecalc does not contain all the market scenarios and are the ones missing significant?

How is it any different than trying to guess distribution of market returns over the next year if all you had was historical returns of the past 3-4 years? That's all we seem to be doing with above approaches.

The answer to this is similar to the person who sees all probabilities/outcomes as 50/50 - either you win or you lose e.g. the odds of winning the Megaball lotto is 50% either you win or you lose - instead of the 1 out of 175 million.

 

[photoguy]

I don't get the OP's hangup with completely independent periods, overlapping periods are statistically valid.

Independent samples are a cornerstone of statistics and almost all tools are based on the assumption that the observations are independently sampled.

Applying the standard statistical methods to data from over-lapping periods (non-independent) may lead to spurious conclusions.

That said, I wouldn't say the historical data is worthless or of no value. Like other posters, i believe needs be considered as a rough guideline, to be factored into plans that change over time in response to personal portfolio conditions.

 

[smjsl]

@Gone4Good: what I am saying is that if you treat year 1 and different data point vs year 2 in the context of 90-130 year history has for 30-40 year retirement plan; then by the same token you should be treating starting each Monday as different data point in the context of 3-4 year history for 1-year retirement plan. I don't mind treating them as different data points! I am saying the two are equivalent. Both cases will have either 3-4 non-overlapping retirement periods, or they will have as many overlapping ones as you wish.

the best use of models, retirement planning, and so on, are for determining a ballpark level of spending and withdrawal that your nestegg may be able to handle

Would you trust the ballpark level you get out of a model based on 3-4 years of data when you plan a 1-year retirement? Would you trust it to the same degree as what FireCalc predicts for your real-life plans?

Regarding staying flexible - yes, I guess that's the only choice then. I agree.

IMO - It is about the degree of confidence!

Yes, and the question is should you have any confidence in the models based on so little data... or if you should - how much of it? It feels like people are much more confident about FireCalc-like outputs than I am, which is why I am questioning this.

But I believe FIRECalc may be the best tool available to give us the opportunity to view the success of a particular withdrawal strategy from a historical perspective.

Don't forget the horoscope - could be more even useful tool... that's the point - I am questioning the usefulness of these tools. Sure you have a lot of complex algorithms (well, not really that complex) running on top of the tid-bit of prior data - but fundamentally, no matter what approach is used, lack of data will result in no useful results...

[...] it is an initial sanity check of what seems to have worked so far so that one's initial plans are not implausible or rash.

I agree with all your points about flexibility. Above statement however again displays some degree of confidence and usefulness of the prior data. I recall reading that some research found out, US stock market prices have most correlation with price of butter in some African country (forget which) - would you use this observation as an initial sanity check for what seems to have worked? So I come back to whether the small set of data we have is so small that it can only mislead as to the original "sanity check"... ?

 

[photoguy]

Do you think that Firecalc does not contain all the market scenarios and are the ones missing significant?

Trivially, we know that Firecalc does not contain all important market scenarios because it misses the one where the market collapsed completely, never to recover. I don't believe we fully understand the markets/economies enough to know what exactly might be missing from Firecalc.

 

[Nords]

FIRECalc, 4% rule, etc. - are they not producing a false sense of security?

My first thought is that if you've ever sent any money to anyone associated with FIRECalc... well golly... they owe you a refund.

This horrible flaw of "past is prologue" seems to be rediscovered on this board about every 18-24 months. I'll ask the questions that pop up every time:
1. Whaddya gonna do about it?
2. Got anything better?

People used to think that a diversified portfolio of dividend payers would be a foolproof retirement, despite needing a bigger portfolio to avoid tapping into principal-- until 2008-2009 saw an unprecedented rash of dividend cuts.

To extend your analogy to its breaking point, all retirement-planning analyses are flawed because anything can happen and we have no data to prove that it won't.

Somehow those unsettling discoveries, once again, are not enough of a concern to make me go find a real job.

Applying the standard statistical methods to data from over-lapping periods (non-independent) may lead to spurious conclusions.

We call that "Monte Carlo". I especially enjoy the debates about somehow "correcting" its random run of data periods to account for the fact that there's persistence to stock market performance that's independent of the calendar's arbitrary cutoff every 31 Dec.

 

[smjsl]

And critical opinions that offer no potential solutions don't accomplish much IMO.

If you were basing your market decisions based on a price of butter in an African country, and I "proved" to you they are worthless, but I have no better idea, would you find this new piece of information useful?

I think it's useful to know when current methods we rely on should not be relied on.

Do you think that Firecalc does not contain all the market scenarios and are the ones missing significant?

Yes, I believe Firecalc (and any other method based on last 90-130 years worth of history) do not necessarily contain all market scenarios for a 40-year plan.

How is it any different than trying to guess distribution of market returns over the next year if all you had was historical returns of the past 3-4 years?

The answer to this is similar to the person who sees all probabilities/outcomes as 50/50 - either you win or you lose e.g. the odds of winning the Megaball lotto is 50% either you win or you lose - instead of the 1 out of 175 million.

I don't follow this at all. I am saying using 100 year history to make predictions about next 30 is the same as using using 10 year history to make predictions about next 3.

 

[REWahoo]

Maybe a Valium would help?

 

[Midpack]

Here's a way to improve on the overlapping data issue. I've contributed to FIRECALC, so I expect these improvements to be incorporated ASAP. Everyone on board?

http://www.qass.org.uk/2009/Vol_3/paper4.pdf

8 Conclusions

We have evaluated different statistical and economic reasons for using overlapping data. These reasons are especially important since they provide the motivation for using overlapping data. With strictly exogenous regressors as well as other standard assumptions, GLS is vastly superior to Newey-West and OLSNO. The Newey-West estimator gave hypothesis tests with incorrect size and low power even with sample sizes as large as 1,000. Unrestricted MLE tends to reject the true null hypotheses more often than it should. However, this problem is reduced or eliminated as larger samples are used, i.e. at least 1000 observations.​

If overlapping data were the only econometric problem, there would appear to be little reason to use overlapping data at all since the disaggregate model could be estimated. The practice of estimating a model with both monthly and annual observations, for example,
would not have any apparent advantage.​

We evaluated several statistical reasons for using overlapping data. If the motivation for using overlapping data is missing observations then GLS is the preferred estimator. Errors in variables with autocorrelated explanatory variables can be a reason to use overlapping data, but even with the extreme case considered, the advantage is small. When overlapping data are used due to nonnormality or errors in variables that are not autocorrelated, then GLS is still preferred compared to Newey-West or OLSNO. However, the GLS estimator provides no improvement compared to the disaggregate model. The GLS estimator would be easier to implement than the Newey-West estimator for varying levels of overlap or imperfect overlap.​

We also evaluated economic reasons for using overlapping data. One such economic reason involves regressions of long-horizon asset returns with overlapping data as in the case of asset returns explained by dividend yields. In this case we proposed a modified rescaling of the errors that produces correct test sizes for different sample sizes and level of aggregation. Another economic reason is the case when lagged dependent variables are used as explanatory variables. In this case the GLS estimator is inconsistent. When aggregate data are used as regressors, consistent parameter estimates can sometimes be obtained with maximum likelihood. In other cases, aggregation makes it impossible to recover the parameters of the disaggregate model.​

It can be reasonable to use overlapping data when the goal is to predict a multi-period change. Results showed no advantage in terms of prediction accuracy from directly predicting the multi-period change rather than using a disaggregate model and a multi-step forecast. But the aggregate model could be preferred if it were more convenient to use.​

Overlapping data are often used in finance and in studies of economic growth. Many of the commonly used estimators are either inefficient or yield biased hypothesis tests. The appropriate estimator to use with overlapping data depends on the situation, but authors could do much better than the methods they presently use.​

 

[smjsl]

1. Whaddya gonna do about it?
2. Got anything better?

See price-of-butter examples above (I hope my lazyness to look up the precise example won't bite me too much ). Anyway, the point is lack of other methods, does not mean the one in use is any good. Recognizing this fact may be useful. Staying "more flexible" is one potential positive outcome for me personally - since the less I trust the other methods, the more flexible I have to be in my mind... (even though by my nature I'd rather have a set-it-and-forget-it kind of plan)

Perhaps periodic reminders of the lack of usefulness of "historical" data is good in that sense?

 

[dex]

Trivially, we know that Firecalc does not contain all important market scenarios because it misses the one where the market collapsed completely, never to recover. I don't believe we fully understand the markets/economies enough to know what exactly might be missing from Firecalc.

Firecalc uses historical USA market information and not things that didn't happen in the past. That would be a completely different model and discussion.

 

[W2R]

Would you trust the ballpark level you get out of a model based on 3-4 years of data when you plan a 1-year retirement? Would you trust it to the same degree as what FireCalc predicts for your real-life plans?

Regarding staying flexible - yes, I guess that's the only choice then. I agree.

Trust? Nothing in life is sure but death and taxes. A model like this is not producing some sort of mathematical/statistical certainty IMO. It just gives you another estimate. The smaller the timeline of historical data available, the less statistically reliable the output over long periods of time, theoretically - - but a longer timeline may not make the model any more reliable due to other problems related to the assumptions of the model. If you trust ANY model, or any financial planner, completely, then you are an even great fool than I am.

Personally I run FIRECalc, look at the results, laugh a little and mutter "yeah, right", and then during this first year of ER I spend about half of what it says I can spend. (If it told me I was spending too much, I would be very alarmed).

As I move into further years of ER, I may spend more depending on what unfolds as I go along. Flexibility will be the key. This is one reason why I choose to make my budget as flexible as I reasonably can during ER.

For those planning to retire, the way to approach these problems (IMO) is to pad your numbers, so that you have a safety margin above and beyond what any models or computations are telling you. Then be prepared to be flexible. But don't take your eye off the goal and remember, life is short.

Maybe a Valium would help?

+1

 

[dex]

Yes, I believe Firecalc (and any other method based on last 90-130 years worth of history) do not necessarily contain all market scenarios for a 40-year plan.

What are the scenarios and how are they significant?

I don't follow this at all. I am saying using 100 year history to make predictions about next 30 is the same as using using 10 year history to make predictions about next 3.

To understand why this is incorrect you need to understand statical sampling and error rate.

Start here
Sampling (statistics) - Wikipedia, the free encyclopedia

 

[Gone4Good]

@Gone4Good: what I am saying is that if you treat year 1 and different data point vs year 2 in the context of 90-130 year history has for 30-40 year retirement plan; then by the same token you should be treating starting each Monday as different data point in the context of 3-4 year history for 1-year retirement plan.

You could do that, and you'd get a somewhat more refined result. It wouldn't yield substantially different conclusions though.

But I don't see how our ability to do the calculation for every day, month, or year supports your assertion that we only have 3 or 4 data points.

More specifically, you've yet to argue persuasively how a retirement pre-crash is the same as a retirement post-crash, which is basically your assertion. Instead of simply claiming that both periods are essentially the same data set because some data overlaps, maybe you could explain how the overlap invalidates what are obviously different results. I think the answer to that question is what Dex has said, that not all possible market scenarios are reflected in the data set. That is quite a different argument then the one you are making, and is one that most here would agree with. We except that a FIRECalc 100% success probability isn't really a 100% probability of success. We understand the model's assumptions and its limitations.