Problems Section
Do you have any ideas as to how to explain to clients about the concepts of
probability and randomness?
Here's one idea. You could use dice, and have specific outcomes in mind which have probabilities of occurring corresponding to the key risk numbers. For example, if you roll three ordinary dice the probability of the total score being less than or equal to five is .0463, close enough to .05.
So when appropriate, with clients who need it, you could do the little experiment and observe the outcome. The downside to this or any other such gimmick is, sometimes the unlikely event will occur. Like 5% of the time. And then the client may say, "Aha! I may be that unlucky with your estimate of whether I will run out of money." So you need to have a speech ready for that development. Like, "Well let's repeat it a few times and you'll see it really doesn't happen very often." We will add more material like this from time to time as we think of useful ideas.
( 11/15/02 )
What are some examples of Monte Carlo experiments that researchers have
done that my clients might understand?
Someone published a book quite some time ago on optimal strategies for the board game Monopoly. Here is a perfect example of a pretty simple game which can't be analyzed mathematically, you need to write a computer program which can generate plays of the game, with rules for decision making coded. And run competing strategy ideas 1000s of times and look at the results. The author showed that buying railroads is good, but buying the utilities is bad. Holding out for the most expensive properties and declining to buy the cheaper ones is also bad. If I recall correctly he showed that the green properties were also the best in some sense. So, not a very practical example, but easy to understand and explain.
( 11/15/02 )
Why do you use draws from a lognormal distribution for returns instead of
just using numbers from an historical database?
There are several reasons why structuring the model to accept user inputs on mean and volatility and using an appropriate distribution for the draws is much preferable. First of all a user can simulate the historical data very nicely this way by just putting in the mean and standard deviation of the historical data and out will come numbers consistent with the historical data. Second, if the user wants to do something else, because he doesn't trust the historical data to be representative of what's going to happen in the future, he enters different parameters that he does believe in. The user has a choice.
Finally drawing from a fixed finite set of numbers without replacement can have disastrous results, making the final distribution look concentrated when it should not. As an extreme case, suppose you had 25 returns, and you're simulating a 25 year history, and you wanted to simulate the amount of wealth for a situation with no spending and no deposits. You would get exactly the same number for every replication! Because multiplication is commutative it doesn't matter in what order you draw the 25 returns. But if you take 25 samples from a theoretical distribution with the same mean and standard deviation, and repeat that a few hundred times, you will see the appropriate spread.
( 11/15/02 )
If portfolio returns are generated randomly, why do I get the same answers
whenever I rerun a client case?
The random nature of the Monte Carlo simulations is not completely random, as we use what is known as a pseudo-random number generator. These generate a long repeating sequence of integers which is "seeded" by selecting another integer to determine where in the sequence you start. These integers are then used to determine the pseudo-random returns that are applied to your portfolio.
We use the same seed for all RSP runs, so that you will get identical results each time you run with the same inputs. This is different from what Excel's RAND function would produce, as it seeds itself from the system clock, and produces a new random number each time you recalculate the cell that uses it. We must do it the way we do, so we can reproduce problem cases when you need help.
( 11/15/02 )
Why doesn't RSP model uncertainty in inflation?
It should, but we haven't decided on a model that we are happy with, yet. There are problems: returns will be correlated with inflation, and differently for the various asset classes. There are other modeling problems, such as more difficult conversion of nominal dollar results back to inflation-adjusted, since we'll have to keep track in the program of what inflation was in each year, in each replication. For now we recommend just use a safe highish value for the constant inflation.
( 11/15/02 )
Why doesn't RSP average annual success probability weighted with mortality
data as some other programs do?
We believe using a target age is better for planning purposes. The problem with the actuarial approach is that it counts on the fact that one may die young and thus greater spending is allowed. The actuarial approach is appropriate for an insurance company which counts on people dying according to a certain distribution, they set their prices and their benefits accordingly, and, sure enough, their customers in aggregate die roughly on schedule. An individual is advised to cover as many likely scenarios (like living to age 90) as he can afford.
( 11/15/02 )
There is a lot of discussion in articles recently about using
distributions for returns with fatter tails than the lognormal,
distributions which allow for more extreme outliers. Should
Monte Carlo retirement planners be using these distributions
for returns?
Research done in the '60s by Fama and Mandelbrot indicated market
returns are best fit to a logstable law with the alpha parameter
set at about 1.7. The normal distribution has parameter alpha = 2,
and has finite moments of all orders. The logstable with alpha = 1.7
has fatter tails, so much so that it does not have finite variance.
More recent research
1 indicates that market data are
inconsistent with the infinite variance consequence. That reference
gives evidence for a hybrid model using a truncated stable density,
and point to GARCH models (with time varying volatility) as most
likely the best way to model returns. We do not dispute Fama and
Mandelbrot, we agree that the bulk of the distribution, from the
1st to the 95th percentile fits the stable
law, but the problem is with the size of the tails beyond those
extremes. It's a question of accepting a premise that comes along
with some extreme dubios consequences.
Using tables of the logstable as in reference
2 we see the
following consequences of using that distribution with alpha = 1.7,
and assuming a mean return of 8% per year and a scaling parameter
theta of 12% (the counterpart to standard deviation for stables).
If you did a Monte Carlo projection for 30 years of cash flows
and ran 1000 replications you'd be drawing at least 30,000 random
returns and could expect to get draws as indicated from the 99.9th
percentile, and the 0.1th percentile. The upside and downside
returns that would occur (18 standard deviations from the mean!) are
obviously bizarre and inappropriate for representing typical
diversified investments. (They may be appropriate for a .com
start-up!)
|
Percentile
|
Upside
|
Downside
|
|
95th
|
50%
|
-33%
|
|
99th
|
100%
|
-50%
|
|
99.9th
|
822%
|
-89%
|
Anyone who believes that his portfolio could lose 89% of its
value in a single year should not be focused on retirement
planning methodology. He should be bailing out and
buying T bills.
The corresponding numbers for alpha = 1.9 are a bit more reasonable,
but still not representative for any diversified portfolio:
|
Percentile
|
Upside
|
Downside
|
|
95th
|
45%
|
-31%
|
|
99th
|
68%
|
-41%
|
|
99.9th
|
186%
|
-65%
|
Our conclusion is that the fat-tailed distributions which have
been touted as the preferred way to model returns are inappropriate
to represent investments in anything other than individual stocks
and rather risky ones at that.
When GARCH modeling has progressed, we will revisit the issue.
It may still be the case that lognormal returns are adequate
to model foreseeable risks, but these issues have to be
investigated with numerical experimentation.
[1] An Introduction to Econophysics, Mantegna and Stanley,
Cambridge University Press, 2000.
[2] Stable Non-Gaussian PRandom Processes, Samorodnitsky
and Taqqu, Chapman and Hall, 1994.
( 1/23/03 )
How are the default returns, volatilities and correlations calculated?
The return estimates provided to RSP 3 are intended as starting points for
each advisor's analysis, or to be used as is in the absence of better data.
The returns are based on two methodologies.
The Standard Asset Class returns are based on a model whose inputs include
an estimate for economic growth, change in valuation, asset class growth
rate assumptions relative to economic growth and a starting yield.
The Size-Value Asset Class return estimates are based on the Fama and French
3-factor model risk premiums, an estimate of the market, size and value
coefficients and an estimate of inflation.
Fixed income returns are based on current short term yield to maturity
estimates.
Standard deviation and correlation estimates are based on historic numbers.
( 7/12/02 )
The default portfolio labels A through J are a bit cryptic. Can I change
those labels to be more descriptive?
Sure, just unprotect and type in labels on RebalancingSchedules. You may
then have to widen some columns, depending on how verbose your new names
are, best done using Format -> Column -> Autofit.
When changing a label to a string that looks like a number, please
insert a character like '_' (the underscore) to the beginning of the
label.
( 7/26/02 )
Where does the program assume personal savings contributions comes from?
Here are the rules the program uses for making contributions in any
given year:
Profit Sharing is an employer contribution, so it will be made in full
no matter what your income may be.
Personal Savings, 401-K, IRA, Non-Deductible IRA, and Roth IRA are
individual contributions, so they only are made if the income inputs
for that year will support them.
The income that supports the 401-K, IRA, Non-Deductible IRA, and Roth
IRA contributions is the sum of the following items on the DetailedInputs
sheet: Earnings and Other Taxable Income. If the sum of these
contributions exceeds the sum of these income inputs, then the
contributions will be reduced. These reductions are made in the
following order: Roth IRA, Non-Deductible IRA, IRA, and 401-K.
The net income that supports the Personal Savings contributions is the
sum of Earnings, Other Taxable Income, Social Security, and Tax-Exempt
Income, minus the sum of Annual Expenses, unpaid taxes from the previous
year, management fees, and the tax-deferred contributions made. If the
Personal Savings contribution exceeds this net income, then the
contribution will be reduced. If this net income exceeds the input
Personal Savings contribution, the excess will either get added to the
contribution or spent, depending upon the setting of the "Disposition
of Excess Income" flag.
( 9/24/02 )
I'm struggling somewhat with how to best understand and explain
to clients the Value At Risk concept. Can you help me to better
understand the Value At Risk figure and what it represents?
If you fill in 2.5% as the VaR level, then the One Year Value at Risk
on the Current Portfolio sheet is an amount you might lose from that
account 2.5% of the time. In any given year you might lose that amount
or more 2.5% of the time.
( 11/7/02 )
Are Social Security survivor benefits automatically handled?
RSP does not automatically handle Social Security benefits for survivors. You must manually enter the benefit, in current dollars, on the survivor's detailed worksheet (FirstDetails or SecondDetails). Enter these numbers only in those cells of the Social Security Income column that correspond to the years after the other party is assumed to die.
( 11/15/02 )
My company contributes x% of my income to my 401-K no matter what
I contribute. Explain how to model this.
- Enter the amount of money the employee contributes to his/her 401-K and
set the 401-K Company Match to 0% on the worksheet ClientCashFlows.
- Type in the formula
=CurRA1Savings + 0.06 * CurEarnings
in each element in the Profit Sharing column before age of retirement on the
worksheet FirstDetails. Remember to replace 0.06 with the correct value of x%.
- Make sure to use the same investment accounts for Profit Sharing and for
401-K on the worksheet CurrentPortfolio. Remember to split in half the 401-K
2001 Year-end Balance for Profit Sharing and 401-K 2001 Year-end Balances, too.
The use of Profit Sharing is necessary because the company is not matching
the employee's contribution. The company's contribution is a fixed formula.
( 7/16/02 )
Explain how to make a template case file.
After entering the data, go to the ClientBasics worksheet. Click the Client Cases button and then click the Save As Default button.
To create a new case from this template, go to the ClientBasics worksheet. Click the Client Cases button and then click the Open button. Double on RSP3#Default#Base.xls. You will be prompted to enter the Client Name and the Case.
( 9/24/02 )
For a two person case, how to determine which person should be first and second.
The general rule is that the first person should be the person who will have the
greatest chance of out living the second person.
There is one exception to this rule which only applies when using Accumulation Mode.
Whoever retires last should be the first person.
( 1/15/03 )
Should I include taxes in the Client Expenses?
If Automatic Payment of Taxes is set to Automatic Payment, then do not include taxes in Client Expenses. However
if Automatic Payment of Taxes is set to Ignore Taxes, then you must decide how to include taxes in Client Expenses.
( 11/15/02 )
Where is the information relating to the allocation of the assets?
Unprotect the DetailedTabular sheet (Tools -> Protection -> Unprotect
Sheet...). Next highlight any row from the first column to the "Taxes to
be Paid Next Year" column. Once you have an entire row highlighted, unhide
any columns (Format -> Column -> Unhide).
Columns 19 through 29 and 32 through 42 will contain the average allocations.
( 11/15/02 )
Are management rates progressive or marginal?
Management rates are progressive within RSP.
( 12/02/02 )
Explain how to share assumptions.
- Once you are happy with the values on the AccountPerformance,
RebalancingSchedules, and AccountCorrelation sheets, use the Account
Management button's Save As function to save the Account Performance
data to a new file in the Accounts folder (e.g. AssetClasses.xls).
- Once you are happy with the input values on all of the remaining
sheets, use the Client Cases button's Save As Default to replace
the default client case file in the Cases folder.
- Send the following files to your colleagues:
RSP3\Cases\RSP4#Default#Base.xls
RSP3\Accounts\AssetClasses.xls
- Once your colleagues put these files in their corresponding Cases
and Accounts folders, they will be available for use. Any New Case
will start out like the template.
( 12/02/02 )
I just downloaded the latest Account Performance files.
How do I use them in my existing Client Cases?
Whenever you do a Save Case, the Account Performance data is saved inside
that case. If you want to replace that data in each of your client cases,
you first need to use Open Case to load the case into RSP 3. Next you need
to click Account Management->Open on the AccountPerformance sheet to load
the new performance data (either Standard.xls or SizeValue.xls). Finally,
click Home, Client Cases, then Save Case to save the client case.
To load the new performance data into your default client case, use
New Case to load the case. Click Account Management->Open to load the new
performance data. Click Home, Client Cases, then Save Case As Default
to save the default client case.
( 10/09/03 )
In RSP 3.06, when I set a return on the Account Performance sheet to 0% I get
an error message saying "Management Fee Percentage Must Lie Between 0% and 0%."
How do I get around that?
The code as implemented in RSP 3.06 looks at the Management Fee entered (even
if you say "No" to "Tax Managed Fund?") So you must say "Yes," set the fee
to 0%, then say "No." This flaw is corrected in the current release. Please
e-mail rspsupport@pa.wagner.com
if an upgrade is needed.
( 11/15/02 )
Excel is reporting Run-time error 53 or 76. How is this fixed?
Error 53 is reported when a file that RSP assumes exists does not exist. Two files which RSP assumes exists are Default.xls in the RSP3\Accounts directory, and RSP3#Default#Base.xls in the RSP3\Cases directory. Make sure not to rename or move these files. If they are deleted, check the Recycle Bin. If they are still in the Recycle Bin, please restore them. If you cannot find these files, replacements are available.
Error 76 is reported when a directory that RSP assumes exists does not exist. Two directories which RSP assumes exists are Accounts and Cases. Please do not rename, move or delete them.
( 9/24/02 )
When opening a case or account file, an error message reports that the file cannot be found.
The case and account files must be in the Cases and Accounts directories respectively. For instance if you have a case on floppy, the case must be copied or moved into the RSP3\Cases directory.
( 9/24/02 )
An Excel Run Time Error is reported when using RSP 3.
Please report the error to rspsupport@pa.wagner.com.
Remember to include the case file (located within C:\RSP3\Cases by default), account file (located
within C:\RSP3\Accounts by default), and/or RSP itself as an attachment.
( 9/24/02 )
Copyright 2002 by Daniel H. Wagner Associates, Inc. - All rights reserved.
