There is NO history for retirement planning

[smjsl]

What are the scenarios and how are they significant?

I'd rather not speculate about it in this thread - otherwise it'll go off on a tangent. Overall you can imagine any sequence of returns not represented by Firecalc and I am sure some events may or may not lead to them. Which 30-40-year outcomes are more likely and which ones are less likely is NOT represented by firecalc because it does not have anywhere enough data to work with. That's my only concern here.

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

There is the same number of samples and same error rate in both. In 100/30-year case and in 10/3-year case, the only difference is scale (i.e. 10 years vs 1 year). The rest is the same...

 

[smjsl]

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

I agree it would be nice to fix this but the issue of overlapping data is not central at all to this thread. If you like overlapping data (as Gone4Good indicated), I don't mind you using it in 3-4 year example as well. The point was that predicting 30-40 year retirement based on currently used and available history is the same as predicting 1-year retirement based on 3-4 year history. This applies whether or not you will use overlapping data (in both cases).

 

[Rich_by_the_Bay]

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"... ?

Do a Montecarlo if you want random, unrelated year simulations - there are programs out there that do so. Personally, I don't think sequential years of historic market are independent of one another -- the gains of 2009 were not unrelated to the losses of 2008, for example, IMHO.

You are letting perfect be the enemy of good. Taken to the extreme, it's like saying that just because the sun has risen every morning for millions of years, that doesn't mean it will happen tomorrow. True, but I'll take the odds.

 

[smjsl]

@Gone4Good: you are concentrating on the overlapping periods being significant to my argument. They are not. As I mentioned, you can overlap all you like in the 3-4 year data as well...

 

[Gone4Good]

@Gone4Good: you are concentrating on the overlapping periods being significant to my argument. They are not. As I mentioned, you can overlap all you like in the 3-4 year data as well...

Um, yeah, but it's not the same at all. 3-4 year periods don't even include a full political cycle, let alone a business cycle. Can you name any known and relevant 30-40 year cycles that make this a valid comparison?

 

[smjsl]

Taken to the extreme, it's like saying that just because the sun has risen every morning for millions of years, that doesn't mean it will happen tomorrow. True, but I'll take the odds.

To the contrary, if I had a large enough sample pool I would also take the odds. My point is that because we don't have a large enough history, there is not much value I can assign to the outcome based on such history.

Regarding your other point that consecutive years are not independent - it's actually probably true but the only reason we can conclude it is true is BECAUSE we have a large number of consecutive years from which we can now make that conclusion... If you lived in a world where you had only 3-4 consecutive years of returns, you would not know whether this statement is true or not. (and to gone4good & midpack: ... even if you used overlapping periods in those 3-4 years ;-) )

 

[ronin]

I’ve always relied on this:

Ask the Magic 8 Ball

until I stumbled upon this:

Crystal Ball | Oracle

 

[smjsl]

Um, yeah, but it's not the same at all. 3-4 year periods don't even include a full political cycle, let alone a business cycle. Can you name any known and relevant 30-40 year cycles that make this a valid comparison?

Of course at 30-40 years we have different scale of events as well. As I mentioned in an earlier post, you have to scale down your events as well in 3-4 year example. See my reply to CB in the 5th post on this thread (hope the link works)

Now, you could say that at 30-40 year scale, the events are much more predicable and will have a tendency to repeat themselves, thus making prior data more reliable; but I would guess it's the opposite actually... countries and economies collapse and get reborn as you start approaching such large scales.

 

[MasterBlaster]

2B or not 2B - That is the Question

The OP is exactly correct. There are not enough statistically independent data cycles to accurately predict the likelihood of sucessful retirements going foreward. Whatever estimates Monte Carlo calclators come up with are highly varaiable.

Instead we hope that (relatively recent) past is prologue and base our predictions on that. Note that the past century or two includes massive growth for the US economy. One could pontificate whether or not that is likely going foreward.

It still is a giant crapshoot. Your little retieement timeframe may or may not perform as Firecalc and other calculators suggest.

 

[Gone4Good]

Of course at 30-40 years we have different scale of events as well. As I mentioned in an earlier post, you have to scale down your events as well in 3-4 year example. See my reply to CB in the 5th post on this thread (hope the link works)

Now, you could say that at 30-40 year scale, the events are much more predicable and will have a tendency to repeat themselves, thus making prior data more reliable; but I would guess it's the opposite actually... countries and economies collapse and get reborn as you start approaching such large scales.

Your answer is a cop-out. Your assertion that using 3-4 years of market data to project a 1 yr retirement is the same as using 130 yrs of data to project a 30-40yr retirement assumes that the cycles that are apparent over short periods, are equally apparent over long periods. If that is true, you should be able to list them. If you can't list them, then the comparison is invalid.

Said another way, I know precisely why I wouldn't use 3-4 years of financial market data to make projections. The same is not true for longer periods.

 

[photoguy]

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.

You asked about what scenarios firecalc was missing. Events that did not occur in USA market but have occured elsewhere is directly relevant to the discussion because it speaks to the completeness/assumptions on which firecalc is based and hence how much confidence we can put in the results.

As you state, firecalc is based on US historical data (about 100 years). If it is missing important events that have occured in the past (in other markets), then we need to adjust our confidence in Firecalc results appropriately.

 

[dex]

You asked about what scenarios firecalc was missing. Events that did not occur in USA market but have occured elsewhere is directly relevant to the discussion because it speaks to the completeness/assumptions on which firecalc is based and hence how much confidence we can put in the results.

No post should have to put in the caveats you require. And the 'elsewhere' is in the USA information to the degree it affected the USA data.

As you state, firecalc is based on US historical data (about 100 years). If it is missing important events that have occured in the past (in other markets), then we need to adjust our confidence in Firecalc results appropriately.

The confidence is represented by the success rate result.

 

[dex]

I'd rather not speculate about it in this thread - otherwise it'll go off on a tangent. Overall you can imagine any sequence of returns not represented by Firecalc and I am sure some events may or may not lead to them. Which 30-40-year outcomes are more likely and which ones are less likely is NOT represented by firecalc because it does not have anywhere enough data to work with. That's my only concern here.

Your first post introduced the speculation.

So the question below remains. You should be able to identify the scenarios, approximately how often they would occur over the years covered by Firecalc and their significance to the model.

Originally Posted by dex
What are the scenarios and how are they significant?

There is the same number of samples and same error rate in both. In 100/30-year case and in 10/3-year case, the only difference is scale (i.e. 10 years vs 1 year). The rest is the same...

Research statistics, error rates and related topics and you will see the error in the above.

 

[photoguy]

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.

When I go to Firecalc, I notice it doesn't actually produce any statistical estimations. For example, it states that "FIRECalc found that 6 cycles failed, for a success rate of 94.5%." This is not a statistical estimate but purely a statement of fact.

Where overlapping periods comes into play, statistically, is if you tried to put confidence bounds on future success rates based on the historical data.

Firecalc does not do this, but internally we all look at the number and try to figure out how reliable it is. Since firecalc is based on overlapping data I might personally put wider bounds, say +/- 10% on the estimate whereas if Firecalc had more non-overlapping data I might use +/- 5%.

 

[TooFrugal]

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.

I tend to agree. I personally don't use Firecalc. I wrote my own retirement program where I can enter in varying inflation rates, varying TIPS returns, varying social security income, etc.

 

[REWahoo]

I tend to agree. I personally don't use Firecalc. I wrote my own retirement program where I can enter in varying inflation rates, varying TIPS returns, varying social security income, etc.

And if I recall correctly, you don't invest in equities either, right?

 

[smjsl]

Your assertion that using 3-4 years of market data to project a 1 yr retirement is the same as using 130 yrs of data to project a 30-40yr retirement assumes that the cycles that are apparent over short periods, are equally apparent over long periods.

No. For 3-4 year cycles, I am saying there is not enough data to derive the next year (although, I think some research does suggest that actually some derivations could be done to a limited degree - and this is only BECAUSE we had a lot of samples of this length).

For longer periods, I am saying there is NOT ENOUGH data to decide whether 130 yrs of data tells us something about the performance of the next 35-year period.

If that is true, you should be able to list them. If you can't list them, then the comparison is invalid.

Are you aware of any larger-scale cycles that 4-year presidential one, or unknown-number-of-years business one? Who knows what other cyclical or acyclical evens may or may not affect market performance over the next 30-40 years? Are you sure there are no prolonged bear markets that can last for 100 years? Do you have any data to back that up?

Said another way, I know precisely why I wouldn't use 3-4 years of financial market data to make projections.

I assume it's because you've see enough of them to know they are not very predictable.

The same is not true for longer periods.

I don't think we've seen enough of them to make this conclusion.

What are the scenarios and how are they significant?

I still don't get your question. How about a 50-year bear market with deflation along the way as one of countless examples? How about (inflation - 1)% average market returns for the next 50 years? It's significant because that would ruin a portfolio or a plan derived on the premise of the past 130 years. How likely is it? I don't know - I don't enough data to tell you. Is it less than 50% likely? I don't know - I don't enough data to tell you.

There is the same number of samples and same error rate in both. In 100/30-year case and in 10/3-year case, the only difference is scale (i.e. 10 years vs 1 year). The rest is the same...

Research statistics, error rates and related topics and you will see the error in the above.

dex, I will disregard your arrogant comment and try to explain my position again in the hopes of getting a better reply with an actual explanation of what you are thinking. Since you did not say, I suspect you are thinking in the first case you have 10 times as many samples as in the second. But you don't, because your samples all have to be intervals of time 10 times larger as well.

It's analogous to the following two problems having the same number of samples and other statistical information:
(1) You have a dice with numbers 1,2,3,4,5,6. You throw it 10 times and have to predict the expected value of the next 3 throws.
(2) You have a dice with numbers 10,20,30,40,50,60. You throw it 10 times and have to predict the expected value of the next 3 throws.
Units are different but the rest is the same. Same thing with retirement problem - in one case your "dice" outcomes are measured with 1 year interval based on outcomes of 3-4 years, in the other it's measured in 35 years based on outcomes of 90-130 years.

 

[dex]

No. For 3-4 year cycles, I am saying there is not enough data to derive the next year (although, I think some research does suggest that actually some derivations could be done to a limited degree - and this is only BECAUSE we had a lot of samples of this length).

For longer periods, I am saying there is NOT ENOUGH data to decide whether 130 yrs of data tells us something about the performance of the next 35-year period.

Are you aware of any larger-scale cycles that 4-year presidential one, or unknown-number-of-years business one? Who knows what other cyclical or acyclical evens may or may not affect market performance over the next 30-40 years? Are you sure there are no prolonged bear markets that can last for 100 years? Do you have any data to back that up?

I assume it's because you've see enough of them to know they are not very predictable.

I don't think we've seen enough of them to make this conclusion.




I still don't get your question. How about a 50-year bear market with deflation along the way as one of countless examples? How about (inflation - 1)% average market returns for the next 50 years? It's significant because that would ruin a portfolio or a plan derived on the premise of the past 130 years. How likely is it? I don't know - I don't enough data to tell you. Is it less than 50% likely? I don't know - I don't enough data to tell you.

We are talking about numbers and outcomes. "significant" means how would those events affect Firecals' survivability results.

dex, I will disregard your arrogant comment and try to explain my position again in the hopes of getting a better reply with an actual explanation of what you are thinking. Since you did not say, I suspect you are thinking in the first case you have 10 times as many samples as in the second. But you don't, because your samples all have to be intervals of time 10 times larger as well.

It's analogous to the following two problems having the same number of samples and other statistical information:
(1) You have a dice with numbers 1,2,3,4,5,6. You throw it 10 times and have to predict the expected value of the next 3 throws.
(2) You have a dice with numbers 10,20,30,40,50,60. You throw it 10 times and have to predict the expected value of the next 3 throws.
Units are different but the rest is the same. Same thing with retirement problem - in one case your "dice" outcomes are measured with 1 year interval based on outcomes of 3-4 years, in the other it's measured in 35 years based on outcomes of 90-130 years.

A forum such as this is not the place to teach or learn statistics and the related issues.

I'll step out of your discussion now.

 

[smjsl]

We are talking about numbers and outcomes. "significant" means how would those events affect Firecals' survivability results.

Clearly, Firecalc results would be very much affected in the situations I described.

 

[Gone4Good]

Are you aware of any larger-scale cycles that 4-year presidential one, or unknown-number-of-years business one?

The question was whether you can identify any such long-term cycles. The fact that you can't, means you're comparing apples and oranges, regardless of how much you wish otherwise.

Who knows what other cyclical or acyclical evens may or may not affect market performance over the next 30-40 years? Are you sure there are no prolonged bear markets that can last for 100 years? Do you have any data to back that up?

That is your argument? That there could be 30-40 year cycles, even though we can't identify any or even think of a good reason why they should exist? That's pretty weak.

Meanwhile most of your argument still rests on the notion that every period with overlapping data should be viewed as a single observation. It' is an assumption that runs through this entire thread and is one that you've failed to defend. When asked to defend it, you've dismissed it as irrelevant, but yet it keeps popping up as a central assumption in your point of view.

I'm guessing we've already passed the point of useful discussion, so I'll follow Dex's lead and bid this thread a farewell.

 

[Gumby]

I'm guessing we've already passed the point of useful discussion, so I'll follow Dex's lead and bid this thread a farewell.

smjsl has convinced me that the future is quite hopeless and that I should just slit my wrists now, so at least that was useful.

 

[BigNick]

Either the OP is trolling, or s/he needs to quit worrying and start enjoying things, for a number of reasons:

1. Nobody (I hope) is using FIREcalc results to go out and write a month-by-month spending plan for the next 480 months, with no possibility of changing, no thought of saying "yes, I can see that market conditions are uniquely bad, but I'm going to spend that $72K this year because darn it, I decided 22 years ago that that would be what I spend this year".

2. FIREcalc "only" covers 120 years or so. What would you like it to cover? How much further back in legal and economic history can it go before the results become totally meaningless in the 21st Century? Heck, I'm pretty skeptical about the results before the ending of the gold standard, because the world is so different today.

3. Suppose the economy does totally collapse in a way which has not been seen in the past 120 years. How can any model predict that? If you sincerely believe that that's likely, well, invest in islands, concrete, ammunition, and canned food, because any scenario worse than anything we've had in that time would probably require quite a bit of all of those. In fact, in such a scenario, I suspect that it wouldn't matter how much money you have in the bank. In a state of general economic meltdown, why do you imagine that anyone will be standing around waiting to sell your pizza or fix your elective medical problems, when they themselves will presumably also be stockpiling tuna and buckshot?

 

[smjsl]

The question was whether you can identify any such long-term cycles. The fact that you can't, means you're comparing apples and oranges, regardless of how much you wish otherwise.

If you only had 3-4 years worth of data, you could potentially miss on bull/bear market cycles because you may not have noticed them to begin with. It does not mean it's not there. It just means you did not have enough data to recognize these. My argument is that looking at 30-40 year returns requires way more than 90-130 years worth of data because just as well you don't know the larger cycles that may affect this...

You know some idea about how business cycles affect the markets BECAUSE you have more than enough experience / data with it. It's hard to say what's out there for the markets when you really don't have the data for it on the larger scale. (As an example you could look at other nation's markets, although people question whether those apply to US.)

That is your argument? That there could be 30-40 year cycles, even though we can't identify any or even think of a good reason why they should exist? That's pretty weak.

You could certainly think of them and try to speculate what they are. You can imagine markets like Japanese over last 20 years times 2. You could imagine hyper inflation scenarios. You could imagine slow growth below an average inflation level. I don't want to get into reasons why some of these are valid or not valid. Point is we don't have enough data to figure out true probabilities of these events in the next 30-40 years based on available history.

Meanwhile most of your argument still rests on the notion that every period with overlapping data should be viewed as a single observation. It' is an assumption that runs through this entire thread and is one that you've failed to defend. When asked to defend it, you've dismissed it as irrelevant, but yet it keeps popping up as a central assumption in your point of view.

No, my argument does not depend on this. I told you you could use a 1000 observations in the 3-4 year period just like you could use a 1000 observations in 100 year period. Scale is different. Number of observations is not. How is it central to my point of view?

 

[smjsl]

smjsl has convinced me that the future is quite hopeless

unknown != hopeless

1. Nobody (I hope) is using FIREcalc results to go out and write a month-by-month spending plan for the next 480 months, with no possibility of changing, no thought of saying "yes, I can see that market conditions are uniquely bad, but I'm going to spend that $72K this year because darn it, I decided 22 years ago that that would be what I spend this year".

Question is not whether you should use FIRECalc blindly. Question is how much better FIRECalc compared to your horoscope.

FIREcalc "only" covers 120 years or so. What would you like it to cover? How much further back in legal and economic history can it go before the results become totally meaningless in the 21st Century? Heck, I'm pretty skeptical about the results before the ending of the gold standard, because the world is so different today.

All good and valid points. It's not a problem with Firecalc, it's the problem with no data to base it on.

Suppose the economy does totally collapse in a way which has not been seen in the past 120 years. How can any model predict that?

I don't know, some models try I am sure. When you plan for future though, you have to decide whether your source of data is any good for such planning.

 

[Bikerdude]

FIREcalc "only" covers 120 years or so. What would you like it to cover? How much further back in legal and economic history can it go before the results become totally meaningless in the 21st Century? Heck, I'm pretty skeptical about the results before the ending of the gold standard, because the world is so different today.

Exactly. Should we include the asteroid that destroyed the dinosaurs? Or maybe the impact on the markets of another ice age. Brrrr...is it getting cold in here or is it just me?