Bill Bengen examines how to orchestrate your clients' retirement withdrawa plans.

Since beginning my research into sustained withdrawals from investment portfolios almost 15 years ago, many papers and articles on this topic have appeared. Virtually all of them, including mine, have treated the client's investment portfolio as if it were a laboratory animal, independent of the rest of the client's retirement resources. Using various analytical techniques, we have poked and prodded the poor critter, testing withdrawal rates under a variety of assumptions about rates of return, inflation, tax rates, time horizon, desired legacy, investor skill and so on. We have emerged with a plethora of numbers-initial withdrawal percentages, such as 4.2%, or 4.7% or 5.8%-which reflect the capacity of the client's portfolio to survive under the stated conditions.
Are we ready to apply these withdrawal percentages directly to a client's retirement portfolio? Can we be confident that these withdrawal rates will work in practice, as well as they have in the laboratory?
My answer to both questions is a resounding "No!" If you feel otherwise, I strongly recommend that you increase the limits on your E&O policy.
The fact is, virtually all prior research has assumed that the client's other cash flows during retirement will vary with inflation. That is probably true for Social Security. But what of fixed pension plans, whose payments do not increase with inflation? Or lump-sum items, such as inheritances or one-time expenses? And how about IRA distributions, whose tax liabilities can accelerate dramatically after age 701/2 because of required minimum distribution (RMD) rules?
The fact is, many clients experience irregular cash flows during retirement. That means the demands for withdrawals from the client's portfolio will not grow smoothly with the inflation rate over time, but will fluctuate. For example, I recently used my "Layer Cake" technology to predict an initial withdrawal rate of 4.3% for a married couple. However, after considering all other aspects of their retirement situation, I adjusted that rate downward to 3.8%.- a very significant difference.
Does this mean we should throw away all our earlier research? Not at all. That research is still essential as the starting point for establishing "raw" withdrawal rates. In this paper, I present a methodology to reconcile our early computations of initial withdrawal rates (IWRs) with the client's total retirement situation so as to develop "adjusted IWRs," which can be used with a reasonable degree of confidence. The techniques will require only the use of financial planning software, which provides a year-by-year forecast of portfolio values throughout the time horizon, as well as a "What-if?" capability. Since most financial advisors are armed with such tools, my approach should be widely applicable.
We shall examine this subject in three parts. First, we will study taxable accounts, then tax-deferred accounts (most notably traditional IRAs). Finally, we will explore the intriguing and little-explored topic of "mixed portfolios": client portfolios with both taxable and tax-deferred accounts.

Taxable Accounts
In my earlier research, taxable accounts presented a particularly challenging problem. Because I assumed that the client's portfolio always paid taxes arising from capital gains and dividends, an estimate had to be made of the magnitude of these quantities. In the end, I used a "modified average tax rate" related to, but not identical to, the client's actual average income tax rate.
However, in this next stage of our research, taxable accounts are far easier to analyze than tax-deferred accounts. That is because we are evaluating primarily what happens outside of the client's investment portfolio, not inside. The financial planning software we use will automatically compute the income taxes arising from capital gains and dividends generated within the investment portfolio. Our primary goal now is to determine whether the withdrawals we postulate (using our IWR) are adequate to accommodate the effects of all the other cash flows anticipated during a client's retirement. If not, we need to make adjustments.
To begin our analysis, let's assume we have the following:
A taxable portfolio with $1 million initial value;
A $600,000 initial cost basis;
Client withdrawals of $40,000 the first year, and increases in dollar withdrawals each year afterward for 30 years;
A constant 3% rate of inflation (applied to tax brackets and dollar withdrawals); and
A constant portfolio rate of return of 8.5%.
Entering the above parameters in a financial planning software program will produce a 30-year forecast similar to Figure 1.
I draw your attention to the portfolio value at the end of the 30th year-approximately $2.67 million. I call this the "finale" value, and it will be crucial to our upcoming analysis. The actual finale value your FP (financial planning) software derives will depend on your typical assumptions about investment rate of return, inflation and the tax computation features of your software. It is not crucial that your finale figure match mine, only that you are consistent in using the same software and same Social Security assumptions in creating client retirement scenarios.
It's not surprising that the portfolio has grown in value from $1 million to $2.67 million after 30 years. After all, during the first year the client withdrew only $40,000, or 4% of the initial value of the portfolio. I call that the "net" or "taxes-excluded" IWR. In addition, approximately 0.9% was withdrawn the first year to pay the taxes on portfolio income (see Figure 1). That's a total withdrawal of 4.9% during year one; I call that the "gross" or "taxes-included" IWR. The gross IWR is less than the 8.5% growth rate of the portfolio, but a 3% inflation rate means that the portfolio only gains about 10% in real value over 30 years.
Now, let's make the following additions to the client's retirement profile:
Social Security receipts of $2,000 monthly, adjusted for constant 3% inflation each year, and
A fixed pension of $2,000 per month (which does not increase with inflation).
Not surprisingly, as both of the above are "additive" to wealth, the new finale value climbs to $6.09 million. Clearly, the client can withdraw more than $40,000 the first year-but how much more?
Some planners might use Monte Carlo modeling to solve this problem. But many planners either do not have the software capability to do so, or prefer not to use such techniques. For those planners (myself included), I offer a quick-and-dirty approach that provides a reasonable degree of accuracy. I recommend that the planner use the "What-if?" capability of his or her financial planning software to modify, via trial and error, the first year's withdrawal so that it returns the "finale" to its original value of $2.67 million.
This approach can be justified if we assume that the client wishes to spend every dollar possible from his pension and Social Security income. Thus, the proceeds of both those income sources will be allocated to either withdrawals or income taxes, with none allocated to capital additions (at least not permanently). Thus, there will be no net growth in portfolio balances over time, compared with the "baseline."
In order to readjust the finale as recommended, the first year's withdrawal must climb from $40,000 to $69,800. Income taxes paid the first year are computed by the software to be $23,300, up from $9,300. Total expenditures are thus $93,100, offset in part by income from Social Security and from the client's pension totaling $48,000. Thus, the first-year dollar withdrawal is $45,100, equivalent to a gross IWR of about 4.5%. This is quite a bit lower than the 4.9% gross IWR we started with. Why?
The answer is that the fixed pension plan declines in real value each year, while we have specified that the client's annual expenses must have a constant real value. As a result, withdrawals must increase faster than inflation each year, beginning with a lower GWR, and ending with a higher GWR, than in the baseline scenario. This is borne out in Figure 2, which charts the growth of GWR over time (the Current GWR for any one year is the total amount withdrawn from the portfolio divided by the portfolio's value at the beginning of that year). By comparison, for our "baseline" scenario, Current GWR actually declined slightly over the 30-year time horizon.
Now let's make one further addition to our client's retirement profile:
A $500,000 inheritance during the 11th year of retirement.
Clearly, in this case, money will be allocated to "capital additions," as well as to expenditures and income taxes. As before, the client desires that his expenditures increase with inflation each year. However, he also specifies that expenditures anticipate, and incorporate from the very first year, the $500,000 inheritance in the 11th year of retirement. The expected consequence of this is that the client's GWR will appear quite high prior to receipt of the inheritance, then return to more normal levels after the inheritance is received.
How do we integrate the $500,000 into our model to achieve the client's goals? There are several possible approaches, none perfect, because of the complexity of the problem and my preference for relative simplicity. One possible approach (and the one I employ in my practice) involves several steps.
First, we observe that the introduction of the $500,000 inheritance has ballooned the "finale" to $4.25 million. By trial and error, I replace the $500,000 inheritance in the 11th year with an inheritance in the first year that produces a virtually identical finale value. In this case, a $281,000 inheritance in the first year of retirement produces the desired finale value. The $281,000 value is, in effect, an adjusted present value of the inheritance reflecting the effects of withdrawals, investment growth, etc.
Second, we note that the $281,000 inheritance has added 28.1% to the original capital of $1 million. I then make the simplifying assumption that, to preserve the same margin of safety as in the original scenario, the finale value also must be decreased from $4.245 million to approximately $3.43 million (which is 28.1% above the original $2.67 million). Using the "What-if?" feature, we determine that the first-year expenditure producing this effect is $76,900-about $7,000 higher than before.
This simplifying assumption is not perfect, as it presumes the scalability of all the components of the retirement model, which is not strictly true. However, after examining a number of
sample cases, I have determined that there are offsets to reduce the error, that the approximation is reasonable, and, in fact, it introduces a desirable element of conservatism into our computations.
The final retirement model for our taxable account is shown in Figure 3. The trend of the annual GWR is plotted in Figure 4. The effect of the inheritance is to "flatten out" the growth of GWR after year ten, as compared with Figure 2.

Tax-Deferred Accounts
Let's now turn our attention to tax-deferred accounts. If you recall, my earlier research specified withdrawals from tax-deferred accounts that grew steadily with inflation. However, as you know, after the client reaches age 70 1/2, traditional IRA accounts (the most popular form of such accounts in client portfolios) are subject to required minimum distributions, which cause accelerating withdrawals (and accelerating tax payments). How do we establish a tax-deferred "baseline" to mimic my earlier research, so we can obtain a suitable "finale"?
My answer is to use a nonqualified tax-deferred variable annuity (NQTDVA) with a zero cost basis. This solves the problem nicely, as there are no RMD requirements, and any desired rate of return may be applied. Let us also assume the following:
That this is a NQTDVA portfolio with a $1 million initial value;
That the client will withdraw $43,000 the first year, and increase dollar withdrawals each year afterward during a 30-year time horizon;
That income taxes also will be withdrawn from NQTDVA;
That there will be a constant 3% rate of inflation (applied to tax brackets and dollar withdrawals).
The "finale" value for this scenario is approximately $3.13 million (remember, your finale value may differ, depending on your assumptions for inflation, ROR and tax computation features). Note that first-year income taxes are computed to be $4,050. Therefore, total withdrawals from the portfolio for the first year are approximately $47,000, yielding a gross IWR of 4.7%. It is apparent now that my original research was based on the gross IWR, as there were no provisions for payment of taxes from anything other than the tax-deferred account being studied. Thus, the IWRs I developed for tax-deferred accounts did not represent fully spendable income for the client. They would have had to be reduced by taxes generated as a result of withdrawals, to generate a "Net IWR."
One has to keep in mind that FP software generally does not require input in terms of IWR percentages; it seeks input in dollars. Furthermore, the dollar amount entered generally corresponds to the net IWR, not the gross IWR, as the software will compute income taxes itself. Since the IWR generated by my research for tax-deferred accounts corresponds to a gross IWR, you may have to arrive at the net IWR desired by the client through trial-and-error entry.
Now we substitute a traditional IRA account, with an initial value of $1 million, for the NQTDVA. This reduces the finale value to $2.68 million because of the accelerating RMD after age 70 1/2. As before, using the "What-if?" capability of our financial planning software, we adjust the discretionary dollar withdrawal until the finale is returned to $3.13 million. This adjustment produces a 30-year model similar to that shown in Figure 5.
This resulting first-year expenditure is approximately $39,900, corresponding to a net IWR of about 4%. The software computes first-year income taxes of $3,300, for a gross withdrawal of $43,200, or a gross IWR of about 4.3%. Obviously, these are both significantly lower than for the NQTDVA. Figure 5 clearly illustrates the rapid annual acceleration of tax payments late in retirement.
As a matter of interest, for this client, with his only source of income a traditional IRA account, the RMD does not force larger-than-expected withdrawals till age 78, the 14th year of retirement (which began at age 65). That is because the RMD rate does not exceed the withdrawal rate until that year.
Figure 6 illustrates how the GWR declines through early retirement, then bottoms out in year 14 and climbs thereafter, reflecting the growing effects of RMD demands.
The IRA baseline model is now ready for the addition of the other elements of the client's retirement profile, using the same approach as we did for taxable accounts.
Remember, great precision should never be expected or promised using this method. In reality, the gross uncertainty in tax rates, inflation and investment returns over 30 years is probably much greater than the errors of approximation encountered in its application. As Warren Buffett has said, it is better to be approximately correct than precisely wrong.

Mixed Accounts
These examples illustrate how I plan for client accounts with single types of investment portfolios, either taxable or tax-deferred. The method is straightforward and relatively simple, and often can be done right in front of the client.
Of course, many clients have portfolios with both taxable and tax-deferred accounts. How does one compute the IWR for such clients? Do we select the rate for one type of account or the other, or do we average the IWRs for the two different types of accounts?
In order to study this question, consider Figure 7.
Figure 7 compares 30-year portfolio growth for two clients. Each initially has a $1 million portfolio value. The portfolio at the top of the chart initially consisted of 20% in a taxable account and 80% in a traditional IRA account. The portfolio at the bottom of the chart has these fractions reversed-80% in a taxable account, 20% in a traditional IRA account. A very low 4.5% ROR is assumed over the entire time horizon (corresponding to a "worst-case" scenario, intentionally causing exhaustion, or near-exhaustion, of both portfolios within the time horizon).
In each case, the client withdrew $40,000 the first year for expenses, plus computed income taxes. The taxable account is always assumed to be the first source of withdrawals; the IRA account is withdrawn from as a last resort or as required by RMD rules.
What is remarkable about these two charts is their extraordinarily similar shapes. That means that the same $40,000 discretionary withdrawal expense-i.e., the same net IWR of 4.0%-resulted in almost the same portfolio longevity for both combinations of taxable and tax-deferred accounts. In fact, a reduction of only 4% in the IWR for the top portfolio would make the two charts virtually indistinguishable.
In conclusion, there really is no problem in choosing the net IWR for portfolios with "mixed" accounts. Given the same degree of risk that the client will outlive the portfolio, the net IWR for taxable and tax-deferred accounts is about the same, all factors considered.