<beliavsky[at]aol.com> wrote
Elle wrote
- quote -

snip
- quote -

Mathematical algorithms can be helpful, but they can be
basic ones that are merely adjusted over time and one's
needs. That's all I think this consumption-smoothing
software is.

Re a Monte Carlo approach vs. a more basic allocation
strategy, based on an average of some number of years of
asset performance, the question to me is: Is the output
clearly superior? First, I don't think anyone can prove it
is. Second, the assumptions of the complex models and the
simpler ones are usually similar. Third, no one will rattle
off for me the uncertainty, say, on any of these more
"sophisticated" (sic) models' asset allocation percentages,
probably because the assumptions are either (a) so broad
that it's not meaningful to do so; or (b) the uncertainty is
so large that many basic models already suffice, providing
similar guidelines.

Monte Carlo models are wonderful when they're used with
truly random processes such as radioactive decay. The fact
that you're applying Monte Carlo model to a questionably
random process perhaps defines the divide between us.

If asset returns were truly predictable (or just conformed
to a random process), the math would truly be applied and
useful. But asset returns are not truly predictable. The
financial future is arguably nothing more than a collective
hunch about ever unpredictable human behavior.

Elizabeth Richardson wrote:
- quote -

I don't really consider this an issue for modeling purposes. If I could
get the model to behave nicely, then I would need to take this into
consideration. For now, I just assume that I could convert some
non-liquid asset (like my house) into capital to live on (the models I'm
playing with only include liquid assets as retirement resources, not
total net worth). I also assume that my wife and I will live to the
same age - if we're getting too close to the end and we're still
healthy, I'll just bump her off to reduce my expenses. Or I could just
set the end age to 130 - might be a little easier on the wife...

-Will

<beliavsky[at]aol.com> wrote
- quote -

You do realize that "confidence level" in statistics has a
meaning different than that you imply above, don't you?

Statistical models come up so often here that I think it's a
very important point. "Confidence level" is not so easily
explained.

- quote -

In theory, every additional year (or of course even moment)
that one lives pushes one's terminal age up. Which in theory
means one's drawdown rate should be steadily adjusted
downward. (Practically speaking, most might probably wave
their hands at much of this and just make the adjustment
yearly.) But then some reality should inject itself. For
example, health care costs for the elderly may increase far
out of proportion to inflation. Or one year a person may
live cheaply by him/herself at age 85, then have a stroke,
and his/her yearly costs skyrocket. Meanwhile s/he's
squandered (so to speak) money from age 65-85, drawing down
and spending it at a fairly high rate.

I think the answer to Elizabeth's query (which I actually
think was politely rhetorical; she doesn't strike me as
having been born yesterday) is more like either

(a) do not draw down from principal; instead, live from the
income from the principal for as long as possible; ensure
the principal is "large" so Medicaid can be put off as long
as possible.
(b) plan to become comfortable with the worst case: drawing
down and running out of money, then going on Medicaid.

Not that the worst case will happen. It's just that the
inexact nature of financial planning can't do much better.
Despite all efforts to plan to the contrary, one may either
die with a nice inheritance for heirs, or one may die
penniless.

Will Trice wrote:
- quote -

To answer Elizabeth Richardson's question about the risk of running out
of money,
you could, in the simulation, recalculate every year the age to which
the money is supposed to last.

Suppose one adopts a confidence level of 10%, and at age 65 one thinks
that there is only 10% chance of living to age 90. (I am just making up
this number -- one should consult a life expectancy table and consider
individual factors). If, 10 years later, one is still in good health,
the 10% age will have increased beyond 90, and the withdrawal rate can
be changed accordingly.

"Will Trice" <wwtrice[at]paragondynamics.com> wrote in message
news:445D35E0.3010506[at]paragondynamics.com...

- quote -

What happens if you live to the "some age in the future" plus one year?

Elizabeth Richardson

Dave Dodson wrote:

- quote -

I am surprised to find that this is a "simple" optimization engine. It
does not consider the volatility of returns (or anything else for that
matter) at all. Still, it does have some cool features.

-Will

beliavsky[at]aol.com wrote:

- quote -

I'm using both of these now, with inflation also treated as a random
variable. The amount of money at the beginning of retirement is such
that I will have $0 at some age in the future assuming constant average
returns and constant average inflation, with drwadowns increasing with
inflation. I start with a desired income with a full year's drawdown
taken at the beginnning of the year. Given average or better returns, I
increase this drawdown by the inflation amount, but given less than
average returns I reduce the drawdown amount to the point where I again
would have $0 at the same age in the future given average returns/inflation.

The idea here is that I have some ideal income I want in retirement, but
that number has plenty of pad so if it needs to be reduced because of
inadequate returns, I just take one less trip to the south of France.
At the same time I want to retire as soon as possible, so I want to find
the minimum amount likely to meet my goals.

- quote -

I've just started playing with the idea of including these in the
models. I hesitate because annuities lock-up so much money and the
returns on inflation-indexed bonds are relatively low meaning I need
more money and must retire later. Maybe I should take two less trips to
the south of France...

- quote -

Thanks for these, I see Evensky's name a lot so I should probably check
that book out.

You know the funny thing is that I started playing with all this about a
month ago because I remembered a post you made a long while back asking
for a closed-form solution to the retirement drawdown problem and I
needed an excuse to brush up on my stats. So I've been playing with the
equations and have learned some things along the way (always hate it
when that happens).

-Will

Will Trice wrote:
- quote -

I don't know of an easy answer to that question -- one could try to use
conservative estimates of future returns.

What spending rule are you using in the simulations? It is common to
assume that the one increases spending in line with inflation, but it
would make sense to scale back spending somewhat if returns are lower
than expected.

One could try to devise a financial plan so that minimum needs are met
with inflation-indexed bonds, and later in life, with immediate
annuities.

There is a paper and an associated spreadsheet by Mosh Milevsky on
"sustainable withdrawal rates" at http://www.ifid.ca/research.htm . He
has recently published a book "The Calculus of Retirement Income:
Financial Models for Pension Annuities and Life Insurance" --
information is at http://www.ifid.ca/cri.htm .

Another recent book is
Retirement Income Redesigned: Master Plans for Distribution
Harold Evensky and Deena B. Katz
Bloomberg Press (2006)

beliavsky[at]aol.com wrote:

- quote -

The estimation problem is one I've been struggling with lately. One
output of a Monte Carlo simulation could be the probability of success
of a retirement drawdown plan, but in playing with these simulations I
find that the answers are highly sensitive to the distributions used for
future returns. Since these are not known, how good is the answer?
Just wondering what your thoughts are on this.

-Will

Elle wrote:
- quote -

Academics have known for some time that return distributions are not
strictly normal.
The distribution of stock returns does get closer to normal as the
sampling frequency decreases, so that monthly returns are less
non-normal than daily returns. All models simplify reality, but some
are still useful. One can simulate from distributions with a bigger
left tail (more downside risk) than the normal.

- quote -

If one has a financial plan (savings rate and asset allocation), how
would you estimate the likelihood of success of that plan? Would you
simulate from historical returns instead of from a parametric
distribution? It is worth considering the results of historical
simulation as a complement to a parametric simulation, but
(1) there may be insufficient historical data for some assets (for
example, U.S. inflation-linked bonds have not been around for too long)
(2) one may wish to consider scenarios that have not happened before
(3) one may think the distribution of returns has changed. Most
academics who have studied the equity risk premium think that it will
be lower going forward than it has been historically.

- quote -

I don't agree that all one needs is "common sense". A U.S. investor can
allocate assets among

(1) stocks (large or small cap, value or growth, domestic or foreign)
(2) bonds (government, mortgage, muni, or corporate, short/medium/long
duration, domestic or foreign, nominal or inflation-indexed)
(3) cash
(4) REITs
(5) many other asset classes

It is "common sense" that I should not invest all money in only one of
these asset classes, but in determining a good allocation, I think a
mathematical algorithm is helpful.
I now work on such problems for a living as an analyst at a global
macro hedge fund.

- quote -

savings decision. Even if future investment returns were known (suppose
the only investable asset was an inflation-indexed government bond),
this is a nontrivial problem for which a computer model can be helpful.
Some retirement calculators on the web ask one to work backwards from a
desired level of income in retirement. Higher targeted income in
retirement means less consumption now, and a mathematical model can
help one decide on the optimal trade-off, even if the results are
estimated with some error.

<beliavsky[at]aol.com> wrote
- quote -

"The Problems with Monte Carlo Simulation," by David
Nawrocki, Ph.D. (Professor of Finance, Villanova), 2001,
Journal of Financial Planning
http://www.fpanet.org/journal/articl...1101-art12.cfm

The following excerpts, among others, echo some of my
thoughts:
---
Evensky [2001] gets to the heart of the matter-that is, risk
versus uncertainty: "The problem is the confusion of risk
with uncertainty. Risk assumes knowledge of the distribution
of future outcomes (i.e., the input to the Monte Carlo
simulation). Uncertainty or ambiguity describes a world (our
world) in which the shape and location of the distribution
is open to question. Contrary to academic orthodoxy, the
distribution of U.S. stock market returns is far from
normal."
...
Evensky notes that Monte Carlo simulation is an effective
way of educating people regarding the uncertainty of risks,
but rather than reducing uncertainty, it increases the
guesswork manyfold because of its assumption set.
...
The benefit that Monte Carlo simulation promises to provide
might be better achieved by using common sense in the
financial planning process.
---

Thanks for the reference B, you're always good for those.
I haven't seen you post in a while, good to have you back.

-Will

beliavsky[at]aol.com wrote:
- quote -

Four comments:
-- I don't reject this approach. In fact, I have advocated, for one,
Monte Carlo methods (using historical data) applied to financial
planning several times here. I expect I will have occasion to do so in
the future as well.
-- I do not consider this approach particularly sophisticated. I write
this as someone who has in fact published engineering research where
the featured tool in the research was Monte Carlo simulation. Tack on
some understanding of statistical correlation and distributions, and we
have concepts that are certainly within the grasp of someone competent
in high school math. (Though I don't expect such folks to be able to
teach the concepts; only grasp them.)
--Where I take particular issue is that the assumptions used in such
Monte Carlo modeling are so different as to make a useful distinction
between what it advocates and what, say, basing one's allocation on,
say, a simple average of historical returns (for different asset
categories) advocates.
--Then to me there is the clincher: Can you prove (rigidly) that your
approach will yield superior performance compared to, say, an
allocation based on a simple average of historical returns? We could
certainly back test both approaches. We could also then comment on
things like the (numerical) volatility of each approach compared to
some benchmark and suggest one approach was less riskier. But //prove//
it's less riskier in the future? That's the linchpin.

- quote -

What approach do you think I advocate? Because I wouldn't call my
approach anymore heuristic than your approach. I think "heuristic" is a
loaded descriptor here, not loaning much meaning at all to the
discussion.

What is the margin of error on the allocations your approach
recommends, anyway? (Obviously presenting an precise number will
require listing the assumptions used to obtain that number. First and
foremost, what time period for historical data are you using? Why that
set and not another? Why do you assume a random distribution of this
data? Is that assumption a good one? And so forth.)

Will Trice wrote:
- quote -

I read the section of the ESPlanner documentation describing the
methodology and think it is better than simpler, more commonly used
methods. There ARE many "moving parts" in the program, such as future
income, consumption, tax rates, Social Security benefits, etc., but the
financial situation of a middle-class American IS complicated. It is
better to make reasonable assumptions and test the sensitivity of the
results to those assumptions than to use an oversimplified model that
ignores relevant effects.

Some of the mathematical buzzwords would be Monte Carlo, dynamic
programming, utility maximization, portfolio optimization, and Bellman
Equation. One book describing the mathematics involved is

"Introduction to the Economics and Mathematics of Financial Markets"
by Jaksa Cvitanic, Fernando Zapatero MIT Press (2004)

in particular Chapter 4, "Optimal Conumption/Porfolio Strategies".

The implementation of these algorithms is complicated, but the basic
principle of "consumption smoothing" accords with common sense. One
saves money during one's working years to the extent that the
resulting capital in retirement provides more "utility" than would be
provided by consuming now. Put more simply, don't buy luxuries now at
the cost of not affording necessities later. The definitions of
"luxuries" and "necessities" partly depend on one's wealth and
expected income.

Elle wrote:

<snip
- quote -

I think Monte Carlo simulations of multivariate return distributions of
assets, together with expected income and consumption, are a good tool
for financial planning when used properly. You ooze contempt for
practioners of approaches different from your heuristic one.

Will Trice writes:
- quote -

At least for a start, check out
www.esplanner.com/Download/QVESPFinalDraft.PDF

Dave

Dave Dodson wrote:

- quote -

You may have a point, but I think the comparison is accurate. In fact,
I think the problems are quite similar given that the hard part of the
computation in both cases is determining future rates of return, the
volatility of same, and the probabalistic effects they have on the final
answer. The rest is simple math and a massive amount of assumption.
I'm not sure what complex dynamic programming you may be referring to.
I haven't found a good source describing the details behind these models
yet, so perhaps there is more than meets the eye.

-Will

Will Trice writes:
- quote -

Perhaps, but because of the complexity of the dynamic programming
problem involved, there is only one consumption smoothing package right
now. The comparison with asset allocation software is moot, anyway,
because they are quite different problems. There is no ideal asset
allocation, just a bunch of pretty good, acceptable, or mediocre ones.

Dave

Dave Dodson wrote:
- quote -

I would guess that if I ask these questions of 10 different consumption
smoothing software packages, I would get 10 widely varying answers. I
have observed this with asset allocation tools, some of which are hardly
simplistic.

-Will

BRH wrote:
- quote -

It seems to me that the goal is to start early and be on a path toward
the right numbers. At 20, projecting out 40 years is certainly a bit of
a crap shoot. But as a starting point, I advised a 20 year old that if
he saved 10%/yr and his employer matched it with 5%, and then assumed 3%
annual raises, and 8% rate of return, that by 61, he'd have nearly 20
times his salary, and with a conservative withdrawal rate of 4%, he'd
draw 80% of his final year's salary as a retirement income, not
including Social Security.
But here's my point, change the saving rate to a total of 20%, and the
20yr income in bank (20YIB) drops to age 56, 5 years younger.
Assume the rate of return is 10% (saving rate back to 15%) and the 20YIB
age is 54. Also, people in higher income brackets who live under their
means can likely retire on 60% or less of their historical income.

I understand the point of the ESP is to maximize one's standard of
living throughout their lives, but I'd still like to see it in action.
The example I read through suggested a couple cut their spending from
$65K to $50K, over a 20% cut. How realistic is that?

I'd think the plans produced may be a good start, but anything so cookie
cutter makes me wonder what the sucess rate is.

JOE