Switching years

[cute fuzzy bunny]

In looking through the wreckage of SWR discussions, I found an interesting assertion that might be worthy a separate discussion free of the other flotsam and jetsam. I'm going to say right up front that I do not know if the assertion is true or not and I'm not very motivated to find out. Perhaps someone else has already evaluated this and can affirm or deny. Perhaps this discussion has already taken place in another time zone.

One of the wonder twins noted that if you switched two the returns in any 2 years at random in a 30 year SWR run, you fundamentally changed the SWR for the worse. The derived presumption was that our historic SWR's were therefore based on the luck of things falling the right way and returns coming up the way they did.

So heres the discussion point. Presuming that this case is in fact true (confidence factor: somewhat low) does it say that returns are in fact somewhat random and we've just had good luck, or does it indicate some sort of correlation of returns from one year to the next that creates a positive influence on SWR's?

I have a few opinions, but I'm interested in what others think first.

 

[comixfan]

I would be curious to see what the numbers are after they are crunched. In other words, how much it affects SWR on average, if the 2 are randomly switched in a 30 year period.

It's very interesting that the SWR would be lower. Common sense would say that on average, the SWR should be about the same overall, less when you switch certain years, and more when you switch others which should balance out.

 

[unclemick]

Yikes TH

Shades of Monte Carlo. I ain't going there.

Let the experts come forth.

Heh, heh, heh.

Could be fun.

 

[Hyperborea]

I would be curious to see what the numbers are after they are crunched.  In other words, how much it affects SWR on average, if the 2 are randomly switched in a 30 year period.  

It's very interesting that the SWR would be lower.  Common sense would say that on average, the SWR should be about the same overall, less when you switch certain years, and more when you switch others which should balance out.

Yeah, I agree that for a single 30 year run that the probability of an increase in the withdrawal rate is as likely as a decrease. However, since the SWR is the lowest withdrawal rate that survived any 30 year period then if any switch in year data caused a lowering of the withdrawal rate for the worst such 30 year period it would mean that the SWR (the lowest of the low) would be even lower.

What's at the heart of this matter is reversion to the mean.

 

[comixfan]

Yeah, I agree that for a single 30 year run that the probability of an increase in the withdrawal rate is as likely as a decrease.  However, since the SWR is the lowest withdrawal rate that survived any 30 year period then if any switch in year data caused a lowering of the withdrawal rate for the worst such 30 year period it would mean that the SWR (the lowest of the low) would be even lower.

What's at the heart of this matter is reversion to the mean.

I understand. And I'm also thinking that the longer the time period ie 40, 50 60 years etc, the less the 2 year switch would affect the SWR.

 

[charlie]

We have gone around this horn before, but it
seems to me that if there is no correlation from
year to year in market return, then switching 2
years at random should have no statistical effect.
OTOH, if there is some correlation, then ......

If you give the "Gordon Formula" any credibility,
then you must concede that the looong term
market return is predicted by current dividend
rate plus dividend growth rate. This, to me,
supports the year-to-year correlation theory
in some subtle way that is impossible to predict
on the short term.

The problem is that the shorter the time frame,
the less predictable the outcome.

I hope this does not resurrect the ghost of the
recently departed.

Cheers,

Charlie

 

[comixfan]

We have gone around this horn before, but it
seems to me that if there is no correlation from
year to year in market return, then switching 2
years at random should have no statistical effect.
OTOH, if there is some correlation, then ......

If you give the "Gordon Formula" any credibility,
then you must concede that the looong term
market return is predicted by current dividend
rate plus dividend growth rate.  This, to me,
supports the year-to-year correlation theory
in some subtle way that is impossible to predict
on the short term.

The problem is that the shorter the time frame,
the less predictable the outcome.

I hope this does not resurrect the ghost of the
recently departed.  

Cheers,

Charlie      

What is the worst 30 year period?  Anyone have the returns by year for that period?  I'll crunch them just to take a look.

And I think because there is a correlation from year to year, switching numbers at random doesn't really have the same impact on me that it would've otherwise.

 

[cute fuzzy bunny]

Nah I dont think its going to create any resurrections. NFB is in the process of shutting down due to the unchecked hoco-mania, leaving the wonder twins with no franchise or forum to continue their fun and games.

I dont think there are any correlations directly in very short term periods (couple of years), but I do think market psychology has a short to intermediate term effect...ie...when things are down, people may be likely to buy back in or invest further capital. Markets also move downward on bad economic situations or extended fears and bad news. Since people (in general) will try to resolve poor economic conditions and bad news goes away eventually, things do tend to pick up. The only two times that didnt work out well were the period during and after the '29 crash and that silly 1965-1975ish period when everything stunk. In both cases we went into long term economic funks that took longer to extricate ourselves from.

Longer term, (thanks for taking your cue charlie!) the gordon equation takes hold, or reversion to mean if you want to approach it from that angle.

Switching two years should produce the same results, however if one of the switches results in cratering the portfolio (lets say a 50% decline from sometime in the 70's gets moved into 2003?), and the portfolio is then consumed by the next years withdrawal, you get a failure.

I think the good analogy here is someone repeatedly crossing the street with the walk light on. Theres a chance you get a rogue driver that runs the red light and hits the pedestrian. But the rule of law keeps that likelihood low. If you change that by having the guy occasionally run across the street willy-nilly against the traffic, you've created a higher likelihood of a failure (thump!). But that sort of violates the rule of law in an unnatural manner.

Interesting stuff though...although if its true it sort of says that monte carlo simulations almost automatically create a worse scenario than analysis of data the way it actually happened. There are a couple of hands on the wheel, albeit tenatively...jerking the wheel back and forth artificially creates artificial results not bearing resemblance to the original.

Comixfan, thanks for the offer...knock yourself out if you've got the time, but I'd wait a few hours...my bet is that someone already has beaten this horse to hell and back and they'll be along shortly. BTW, notice your appearance in the past week, why dont you drop by "Hi I am..." and introduce yourself so we know who and what you are?

 

[comixfan]

Ok, I'll post something in the "Hi" section, thanks.

 

[cute fuzzy bunny]

Yep, here we go...
http://raddr-pages.com/Retire Early Web/Sequence.htm

 

[BigMoneyJim]

Some words cut and pasted from a question I had on the same subject earlier:

-----
As far as randomly shuffling returns in the historical record it seems to me self-evident that the averages would be the same yet be more dangerous because historically the big dips were quickly followed by big upswings; shuffle those around and of course ports are gonna fail.

But how do you figure valuation into random returns? If the returns are considered random, then how does valuation matter?
-----

Back to original posting: I haven't looked at raddr's simulations closely yet, and my words above are in response to an assertion--not by raddr as far as I know--that this raddr simulation indicates that past returns were lucky returns. I guess my conclusion for the moment is that valuations and/or correlation matters but we have no reliable way of determining valuations or correlation. However I'm still spinning it around in my head and watching this discussion for more insight.

Please forgive my speaking around a couple of people. At the moment it seems the "100% safe" move.

 

[cute fuzzy bunny]

I think about the only thing I can summarize from this are that there is less randomness in returns than I thought, and that simple monte carlo simulations aint worth doody for stock work.

Surmising that there was some "luck" in the past trends seems a bit of a huge leap to nowhere.

 

[Hyperborea]

I think about the only thing I can summarize from this are that there is less randomness in returns than I thought, and that simple monte carlo simulations aint worth doody for stock work.

There is a small but positive correlation of market returns from one year to the next.  This correlation drops and becomes negative after about 2 to 3 years.  If this wasn't so then rebalancing asset classes in a slice and dice portfolio would provide no benefit.  The historically "best" point to rebalance is about every two years (tested using model portfolios with historical data).  That is just about the time that the correlations start to go negative.

If a Monte Carlo simulation doesn't take this into account then it will produce survival results for withdrawal that are less than historically observed.  If you use Monte Carlo for a portfolio in the accumulation stage the results will also be lower but the difference will be small mostly due to losing the rebalancing bonus.

This is the real beauty of the historical data in that it integrates all the interactions between the asset classes and inflation.  If you try to simulate it then you have to build a very complex model to account for all of this.  If one really wants to try different numbers it would probably be better to use the historical data and slightly perturb it and see what results.  Make the perturbation small and at least somewhat logically consistent or you will push the model too far (the other data will then need to be perturbed some other way that can only be fully determined by using that large complex possibly unknowable model).

 

[cute fuzzy bunny]

Hyper- excellent post...describes why I'm leery of any market simulations or processes that fiddle with the data...

 

[laurence]

Hey, we'll have none of that perterbation on this board! This is a family site!