An oddball idea?

[sgeeeee]

. . .

[REWahoo]

Quote:

Originally Posted by sgeeeee That's one of my 15 citrus trees behind me. It looks pretty healthy, doesn't it? Perhaps I'll have my next photo taken with my head hidden by the branches. That's so much more classy than a bag.

Well, you've tried the floppy hat, long hair, beard, moustache and sunglasses 8)...yeah, I guess hiding in the citrus branches is the logical next step.

Post that photo and we'll poll the forum as to whether or not you need to don the bag.

[cute fuzzy bunny]

Quote:

Originally Posted by REWahoo! True. Short of a bag over your head, there isn't much more you can do.

Had i known he was a hobbit, i'd have gone easier on him.

Not much of a search really. He linked to his website, his name was at the top, its not a common name, typed it into google image search and there it was. AIYEEEE!

[modlair]

Dory, another idea. Maybe this does not belong in this thread. Dunno.

A gentleman just posted to the main board about firecalc observations and a point he made is somewhat relevant to the discussion just completed here, re: what makes market-like data.

We have the 130 data points that are market like data (because they are the market and thus must be market-like), but the gentleman points out that there is an absence of independence in the 90 or so samples of 30 yrs segments because when you move 1 yr, 29 of the 30 samples are a repeat of the previous examination.

At some point, maybe in the thread, someone said something about shuffling sequences and maybe what I'm about to say is what was meant.

What makes the historical record market-like? Is it the discrete numbers of annual market performance? Or is it the order of their occurence?

If you were to use a random number generator to . . . not create numbers, but to select a year's performance from your array of 130, and thereby invent sequences of 130 out of the historical data record, can we then achieve a goal of increased sample size of market-like data?

The fact that the historical record seems to have some human behavior identified that prevents SWRs significantly > 4% is supported by your previous pointing out that success %'s collapse when SWRs very little greater than 4% are tried -- but will that still be true for other data that does not perhaps share the absence of independence of 29 points being repeated in the measurement?

My wording is not clear on this. And neither is my concept, I confess. Just thought it might be worth tossing out.

[cute fuzzy bunny]

The simple problem with that approach (which firecalc doesnt invoke) is that you break all of the year to year correlations. What you're proposing is basically a monte carlo approach only with the 130 year data set as the range of possible 'plug in' return sets.

[modlair]

Quote:

The simple problem with that approach (which firecalc doesnt invoke) is that you break all of the year to year correlations.

Yes, I know. The point being that what makes those numbers "market-like" is that human behavior created them. Was the human behavior that created them more defined by day to day stimuli to their decisions or year to year?

I don't know. I would somehow suspect that day to day responses are more overt than . . . not year to year, but surely every other year.

[cute fuzzy bunny]

Its easy. When things go bad, people are market averse for a time. Then they're not. When things are good, people are market friendly. Then they're not. There is no mathematically reproduceable mechanism to create these psychological streams, although there are plenty of fair examples of them in the historic data. Short good ones. Long good ones. Short bad ones. Long bad ones.

While its not a bad idea, I'm not sure it changes the results much, or gives you a significant difference from a run of the mill monte carlo situation.

[modlair]

I don't know the ramification either. I am just sensitive to the notion pointed out that 29 of the 30 samples in successive calculations are not independent. Hmmm, the calculation is not statistical because it is a single calculation of the performance of human behavior over 130 yrs. Therefore the 29 of 30 samples not being independent . . . perhaps is less offensive.

Meaning that because it is not a statistical calculation, the lack of independence may mean less.

There is a desire/goal to have more samples. But the single sample present is not a sample, it is the universe.

Have to think on this more.

[wabmester]

Just to show how fruitless I think this is, let me throw something random out there.

Somebody just published another book about the Great Depression. One of this guy's arguments is that the depression lasted as long as it did because people had lost faith in the government. People didn't know what the hell was going to happen. Things were so confused that people found it plausible that property rights might disappear and that the government might start seizing property and nationalizing businesses, so nobody was willing to invest.

How do you model *that* based on just looking at annual return data? All of the inputs that move markets are hidden variables.

[dory36]

Well, 2007 is not going to be independent of 2006, and July is not going to be independent of June. If we tried to model what was going to happen in 2007 or in July, and we abandoned the conditions leading up to those times, we might be able to claim "independence" but we won't learn more -- we'll have shuffled the data and pretended to get more data points by using them in a different order, but by doing so, we will have given up what made the data meaningful in the first place.

[Nords]

Quote:

Originally Posted by dory36 we'll have shuffled the data* and pretended to get more data points

Hey hey hey, watch your vocabulary on this family board.

We call that "applying an inverse fast Fourier transform to selected eigenvalues of the applicable differential equations, thereby revealing fundamental harmonic distribution correlations".

Now that I've explained that question it should be pretty easy to turn into PHP code over a couple of workdays. Can I take a look at the first run next Tuesday?

[REWahoo]

Quote:

Originally Posted by Nords We call that "applying an inverse fast Fourier transform to selected eigenvalues of the applicable differential equations, thereby revealing fundamental harmonic distribution correlations".

Dory!!!! Make him stop!!! :P :P :P

[unclemick]

And the Norwegian widow translation is: ?

Hmmmm - perhaps there isn't one.

Oh well - what the heck - drum roll please:

PSST WELLESLEY!

heh heh heh heh - 4.53% current yield. And Wellington is 3.25% for the er ah younger crowd.

[dory36]

Quote:

Originally Posted by unclemick2 And the Norwegian widow translation is: ?

4%.

[unclemick]

Yeah - you rite!

heh heh

[cute fuzzy bunny]

Give or take. Unless a lot of stuffs different. In which case nevermind.

[dory36]

The problem that causes almost all failures is a prolonged (3-5 year) down market shortly after retirement. *Eating into the principal so much during that prolonged period just doesn't leave enough to recover.

Randomizing the selection of years would make meaningless the one primary cause of failure -- sequential years with down markets, occuring shortly after retirement.

[modlair]

Quote:

Well, 2007 is not going to be independent of 2006, and July is not going to be independent of June. If we tried to model what was going to happen in 2007 or in July, and we abandoned the conditions leading up to those times, we might be able to claim "independence" but we won't learn more -- we'll have shuffled the data and pretended to get more data points by using them in a different order, but by doing so, we will have given up what made the data meaningful in the first place.

To some extent, my point above is musing. I don't know what this technique expands the quantity of available market-like data. Let's get that preface out first.

But your last phrase, Dory, is suspect. We do not know what makes that data meaningful (or even if it is meaningful, I suppose, but we presume that it is for all the discussion). I alluded to this above. The definition of "market-like" is not determined.

In fact, there would seem to be a tendency to look at the numbers as entities in and of themselves, when they are not. They are a measure of human behavior. They are the correlation coefficients of weather in that Wisconsin county that just happens to sync up to the moons of Jupiter -- i.e., a raw measurement absent an explaining algorithm.

So, not to be argumentative, but there is no algorithmic reason to, say, not prefer a definition of "market-like" to be how often over 130 yrs there were gains of 0-5%, 5-10%, 10-15%, etc. histographically vs in what order the numbers occur. Maybe humans behave a certain way more often than other ways and they do so more reliably than they behave in certain ways sequentially.

I don't claim that the histogram is the definition of market-like. I don't know, any more than I knew above ahead of time if a particular spectrograph of the waverform might describe market-like. I'm just pointing out that we don't know what makes that waveform market-like, and it could be something other than the order of the numbers.

Quote:

The problem that causes almost all failures is a prolonged (3-5 year) down market shortly after retirement. Eating into the principal so much during that prolonged period just doesn't leave enough to recover. Randomizing the selection of years would make meaningless the one primary cause of failure -- sequential years with down markets, occuring shortly after retirement.

. . . and that is a significant truth in the discussion.

Like I said, I don't know exactly what this means or how to expand it. I'm puzzled by this item, Dory, in that MC runs yield lower SWRs, yet you're saying randomizing the data will result in more frequent success. Hmmm. I know both are true, but what does this mean algorithmically for the mechanism that causes all this to happen?

[dory36]

I didn't say that randomizing the data would increase the rate of success -- I was saying it would remove the one thing that we do know about the data, and that is that it really happened.

[dory36]

For those concerned about the independence issue, consider this.

A 30 year retiree in 1973, another in 1974, and a third in 1975 experienced exactly the same conditions for 28 years of their retirement -- the years 1975 - 2003 were shared years.

It's tempting to say that this overlap makes any results too similar to be meaningful as separate instances. But that simply isn't the case, since it is, as I have said, the initial few years that have such big impact. And every single starting year is a separate and distinct situation, as the graph below shows.

This graph is the year-to-year portfolio balances after taking $35,000 (and adjusting for inflation) from a $750,000 portfolio every year for 30 years, starting in 1973 (red line), 1974 (blue line), or 1975 (green line).



This makes it pretty clear that the first year or two overwhelmingly sets the stage for the future, far more than the 28 years that all three retirements had in common.