2000 May 09
2000May09: Merged the desired output section into the write-up section, concentrating on desired English prose. Also, requested description of what experimental parameters to use, e.g., sending the script you used to run the experiments.
2000May08: Added section with instructions for charts, tables, and text.
2000May04: Added section with suggested timeline and brief description of paper write-ups.
2000May04: Added section listing people and related experiments.
2000May04: Added listings of papers in each experimental section.
Our goal is to write a paper appropriate for a technical journal, not the popular press. An example is the journal version of Cooley, Hubbard, and Walz's work although I think it is a little light on the mathematical description of the calculations.
The paper will be broken into the following pieces:
I plan to use LATEX to typeset the paper and MetaPost to produce the charts so I can easily produce mathematical equations, tables of contents, and HTML, PDF, and PostScript versions. Thus, please submit
Please submit tabular data as tab-delimited files. Each line of the table should correspond to one line of the file. Separate adjacent columns of data using one and exactly one tab character. Hopefully, each line will have the same number of tabs.
If you want to see what your table will look like (before additional formatting applied by Jeffrey), try the following on a UNIX computer:
I plan to use MetaPost to create any necessary charts because the output is extremely precise and clean. See for example, pp. 2-3 of http://www.cs.trinity.edu/~joldham0/1321/hw/hw6/hw6.pdf.
Please submit your chart data as two-column tab-delimited files with blank lines between the data for each line to draw. For example, to draw two lines we could submit
| 1926 tab 8.04 |
| 1927 tab -5.2 |
| 1928 tab 18.0 |
| 1926 tab 18.04 |
| 1927 tab -1.2 |
| 1928 tab 8.0 |
Please add comments regarding the chart's title; annotations of individual lines; the desired type of chart, e.g., bar chart or line chart; axes labels; etc.
Please analyze your results. Among the questions to ask are:
The prose probably should include a description of what experiments were run as briefly as possible but precisely enough so someone else with our code would be able to repeat the experiments. Sending me scripts to run the experiments would also help. Thanks.
As far as I understand, these people are working on the following experiments.
| taxed vs. tax-deferred accumulation | Sara Meyer and Nathan Franklin |
| asset allocation formulae | Collin LeMaistre and John Healy |
| accumulation in an IRA | Danny Johnson and Brent Lew |
| college savings and education IRAs | Jason Meridth and Chris Mitchell |
| asset choice | Jeffrey D. Oldham and Craig Brandenburg |
| short-term saving | ¿Qwinby Swinson?, Scott Schaefer |
Unallocated people include Pete.
We do not a priori know which experiments will yield interesting results. Perhaps only half will be interesting. That's research.
What follows are only suggestions for experiments. Feel free to modify them as appropriate.
Cooley, Hubbard, and Walz avoided the problem of choosing specific values by using percentages. For example, their tables show the probability of success when withdrawing a fixed fraction of one's investment, avoiding the issue of the investment's size and the amount to withdraw.
We might be able to use the same methodology. For example, for withdrawal problems, we can specify an initial investment of, e.g., $1 000 000 and try withdrawing $1000000*3%/12 each month. For accumulation problems, we might specify investment increments as a percentage of the desired target. For example, we could use a desired target of $1 000 000 and specify monthly investment increments of $1000000*2%/12.
Most retirement savings occurs in tax-deferred accounts, e.g., 401(k), 403(b), IRAs, so income taxes and capital gains taxes are not paid while the money is accumulating in exchange for paying a significant penalty for early withdrawal. Young people face many simultaneous financial demands including saving for retirement, purchasing a house, saving for college education, and paying for children, complicating the decision whether money should be saved in a taxable account or a tax-deferred account. In this experiment, we explore the cost of this indecision. How much smaller is one's retirement income when not using a tax-deferred account?
article discussing portfolio turnover rate and its tax implications. apparently one of few articles noting the effect of taxes on portfolios [JA93].
discusses effects of choosing tax-deferred vs. taxed [GR96].
effect of taxes and inflation [SM95].
Bergen has written several papers advocating asset allocation formulae that vary with a person's age. Conceptually, as retired people age, their investments shrink and also their ability to earn additional income through work decreases. Thus, asset allocation formulae move from more aggressive, e.g., stocks, to less aggressive, e.g., bonds.
In this experiment, we compare the results of using a fixed asset allocation with using three different age-varying formulae for a person approaching retirement. To perform a fair comparison between the fixed asset allocation(s) and the age-varying allocations, we should probably choose fixed asset allocations using values that are close to the average of the age-varying allocation throughout the timeframe. For example, if an age-varying formula moves an allocation from 80% to 50% through the timeframe, it might be a good idea to compare against a fixed allocation of 65%. I do not feel strongly about this; if you have better ideas, use them.
Perold and Sharpe discuss different asset allocation formulae including constant proportion [PS95].
The annual IRA limit of $2000 has remained unchanged since 1981. In this experiment, we address the issue of how much retirement money one can accumulate using only IRAs during one's working life.
1999 January, Representative McCollum introduced House Resolution 188 of the 106th U.S. Congress to index the annual IRA contribution for inflation. (See Section 2 of the bill in Thomas.) It would be interesting to show the change in accumulation when the annual contribution is indexed for inflation.
Many parents face the task for saving for a child's college education while simultaneously saving for retirement, purchasing housing, and raising the child. Exacerbating the task is the fact that this savings is taxed as it accumulates. (The newly-enacted Education IRAs permit savings only $500 per child per year, an amount significantly smaller than college expenses.) In this experiment, we address the amount of money to save to pay for college expenses.
According to the College Board, the average four-year public and private budgets are $10,909 and $23,651, respectively (source: Trends in College Pricing, Table 3, College Board). In the experiment, we simulate the necessary monthly savings to accumulate four times these annual amounts.
Asset allocation: Most financial writers recommend starting with an aggressive asset allocation and modifying it as the child nears entering college. See, e.g., Figure 7 of Vanguard's ``Plain Talk'' booklet on financing college. Modify the assetAllocation formulaes to reflect these recommendations. Perhaps the two assets to consider would be large-company stocks and U.S. Treasury bills since very short duration assets are recommended while the child is in high school. Contrast these age-varying formulae with one or two fixed asset allocations. Most financial writers would say that one's college savings should be moved to cash during the last two years of high school and during college to prevent the risk of losing all the savings. However, maintaining a fixed asset allocation and assuming the risk of a sharp decrease in its value may be better than the decrease in the investment's value by gradually moving to less risky assets through time.
Criticisms: College costs seem to increase faster than inflation. If you are willing to do some programming, modify the program to run experiments where college expense inflation increases faster than the consumer-price inflation index by fixed amounts. See Table 5 and Figure 4 of the College Board's Trends in College Pricing to estimate how much higher. It might be useful to modify only the financial target, not the monthly investment increment, which presumably increases at a rate similar to one's salary increase. Thus, modify InflateTarget in the investment class.
When simulating with taxes, the asset allocation is basically ignored during the asset-holding period. Using too long an asset-holding period (compared with the timeframe) will eliminate the effect of the age-varying formula. I do not have recommendations what values to use.
Maria Crawford Scott writes about gradually changing the asset allocation as the child ages but does not show many numerate examples [Sco96b].
Education IRAs permit saving $500 per year in a tax-free account. Simple math shows that even saving for eighteen years will not yield any significant fraction of college expenses. The real advantage of the Education IRAs that the saved money is never taxed, not when contributed and not when withdrawn. Assuming the maximum contribution limit of $500 does not change (a safe assumption since the normal IRA contribution limit has not changed since 1981), what percentage of a student's four-year college expenses could be paid using an Education IRA?
Performing this experiment may require modifying the statistics object to record the average and standard deviation of four-year college expenses and the average and standard deviation of the fraction of the expenses that the Education IRA would pay.
More more information on Education IRAs, see Section 3 of IRS Publication 590, Individual Retirement Arrangements (IRAs) (Including Roth IRAs and Education IRAs).
Cooley, Hubbard, and Walz considered only large-company stocks and long-term corporate bonds (with one additional table that incorporate U.S. Treasuries), ignoring the other asset classes. One question to explore is whether using other asset classes affects the results. There are too many different combinations of assets to try them all. Instead, we might try substituting one asset for one of Cooley, Hubbard, and Walz's assets to look at the effects. One possibility is replicating the experiments of Cooley et al., substituting
different assets' influence on mutual fund returns [IK00].
Financial writers usually recommend investing in stocks and long-term bonds only if the money is needed at least five years in the future. Thus, they are recommending trading-off the possibility of higher returns for decreasing the risk of not meeting one's financial target. Instead of making this blanket statement, we seek to simulate the return-risk trade-off.
We consider saving for various short-term financial goals using different assets. For example, if one saves 8% of one's financial target each month for twelve months using large-company stocks, what is the probability of reaching the target?
Criticism: As written in the experiment on college expenses, the asset allocation is basically ignored during the asset-holding period. Since we are using such a short timeframe, perhaps we should just ignore taxes? What do you think?
[Stocks are] extremely risky in the short term. Investing in the stock market for the short term is a fool's errand. People should plan to stay with a stock investment a rock-bottom minimum of five years. But risk means nothing except in a comparative sense. And stocks and bonds are, at the very least, equally risky over the long term.James K. Glassman, author of the book Dow 36,000: The New Strategy for Profiting from the Coming Rise in the Stock Market, as quoted in [gla00, p. 3].
[Risk is defined] as the variation in returns over time. Standard deviation is a common way to measure volatility. Stocks have a standard deviation of roughly 20 percentage points, which means that two-thirds of the time, stock returns will vary by 20 points from their long-term average of about 11% a year--or in a range from -9% to 31%. That's very scary in the short term. But if you hold them for 15 years, the standard deviation drops down to about 2%, which means that two-thirds of the time the average annual return for 15-year periods varies between 9% and 13%.James K. Glassman, author of the book Dow 36,000: The New Strategy for Profiting from the Coming Rise in the Stock Market, as quoted in [gla00, p. 3].
Historical return information before 1926 available in [CA39].
Capital gains tax rates through the years and how to compute them [Lin87].
Additional work on probability of success in retirement: