Michael H. Salama is the Vice President of Tax Administration & Senior Tax Counsel for The Walt Disney Company. A special thank you is due to Karen Mbanefo for her comments and insights regarding particular industry issues and application and Po-Ling Hsiao for his assistance in formatting the content.

I. Introduction

Section 458 allows accrual basis publishers and distributors of magazines, paper backs, records, tapes, and similar specified property to elect to exclude from income qualified sales in one year that are returned in the next year where the returns are made within the merchandise return period. Regs. §1.458-1(b). Qualified sales are those where: (1) the seller has a legal obligation to adjust the price if the merchandise is not sold; and (2) the taxpayer actually makes the adjustment to reflect the failure to sell the covered merchandise. Section 458(b)(5)(A) and (B). The merchandise return period starts after the close of the taxable year. For paper backs and records it is 4 months and 15 days. For magazines it is 2 months and 15 days. Taxpayers may elect a shorter period for tax return purposes, and forgo a portion of the favorable adjustment to income; a change in the period by the taxpayer will be treated as a change in method of accounting. Section 458(b)(7).

The amount of the exclusion from gross income constitutes the lesser of: (1) the amount covered by the legal obligation to adjust the price; or (2) the amount of the adjustment agreed to prior to the end of the merchandise return period. Section 458(b)(6). Regs. §1.458-1(c). Note, correlative adjustments are required where §458 is employed. Specifically, the taxpayer must make correlative adjustments to closing inventory or purchases, as well as to cost of goods sold, for the same year. See Regs. §1.458-1(g)(1).

In a business environment where cash flow is important, which is most environments, it is often necessary to estimate the impact of income exclusion under §458 for returns before the income tax return is filed or even prepared in earnest. There are two fundamental methods to compute expected returns. One method relies upon historical information. The other method examines probabilities of potential outcomes. In either case one is quantifying the weighted average, the statistical nomenclature for expected return.

II. Computing Expected Returns Based Upon the Historic Average

Perhaps the most common method of computing a weighted average to predict future events is to rely upon the average (arithmetic mean) of past events or performance. For example, in the context of magazine returns one might examine the number of returns per year for the publication for the past 5 years (assuming a normal distribution, as discussed later herein, generally, the greater the number of prior periods evaluated the more precision will be found in the estimate). The average is computed as follows:

In other words, the historical average returns of magazine I, read “R-bar sub i” equals one T-th times the sum of the returns for magazine I for the period t. Application is illustrated via the following example.

The average return for the past 5 years was 8.60x. This amount may be used to estimate returns for the current year from a cash flow perspective. This is the most common and perhaps the simplest method to project expected returns based upon historical information.

III. Computing Expected Returns Based Upon Possible Outcomes and Probabilities

The second method of computing the amount of expected returns for §458 concerns the development of alternative scenarios. For example, one may create alternatives for a low, medium, and high level of returns. Each alternative would include an estimate for the level of returns as well as its probability. One benefit of this method is that it employs more than mere historical information. Where historical information is available it may not be perceived as an accurate basis for predicting future performance. This is sometimes the case where there is great variability between the prior population of content and method and means of doing business in prior periods as compared to the current content and method of operation. (See the discussion later herein for more on variability and its import.) It is worth noting that the probabilities assigned may be quite subjective and biased based upon personal perception regarding product performance and/or the ability of the business unit to precisely project and ship/street the amount of product that will be sold in the market place. The negative element of subjectivity is arguably balanced, in part, by the consideration of alternative outcomes and their impact in the model's creation (a positive).

This approach is illustrated as follows. Assume two potential outcomes: (1) low magazine returns; and (2) high magazine returns. Further assume the data has the following characteristics and probabilities.

The expected level of returns is the weighted average, i.e.

In the above example, the weighted average constitutes the sum of each weighted outcome, or 14.4x returns.

Where one seeks to estimate the §458 adjustment at the time of the federal return extension's preparation, assuming an extension for filing is requested, the estimation exercise is a bit different than if the same number is being estimated earlier in the year. In the later case, the preparer already knows all or a portion of what he/she is trying to estimate, i.e. a portion of the actual returns are known (or should be known since where a 4 month 15 day window is used, much of that window has elapsed). There, a lower bound is established for the returns. For example, if at the time of extension there were already 10x returns, the total amount of returns will of course be higher than that. To the extent even greater certainty was desired; one could employ a shorter return window for return and estimate purposes. There, greater certainty is engendered in exchange for a lesser income deferral. That is generally not a desirable exchange from an economic perspective. Whether one employs a historical data or a possible outcome/probability approach, or some combination thereof, is a question of judgment and data quality.

IV. Variance and Standard Deviation

Variance and standard deviation are methods to reflect the range of possible outcomes and the likelihood of outcomes. Calculating these metrics based upon historical performance is sometimes favored since it relies upon factual input rather than unquantified postulation. Variance of a sample of historical returns is computed as:

In the example in Section II herein, the variance is 19.3x.

Where expected returns are used to estimate the §458 returns, variance is computed by: (1) taking the difference between each possible future value and the expected value; (2) squaring each difference; (3) multiplying each squared difference by its probability; and (4) summing the resulting numbers. The formula is written as:

The example from Section III yields the following.

The variance is the sum of the last column, 47.04x.

The variance is the weighted average of the errors squared. Taking the square root of this number converts the data to the same units as the variable. The square root of the variance is the standard deviation. In our historic data example, the standard deviation is 5.31x. In our prognosticated outcome example, the standard deviation is 6.86x.

If the data for the variable is normally distributed, about 68% of the observed values will fall within one standard deviation of the mean, about 95% will fall within two standard deviations of the mean, and about 99.7% will fall within 3 standard deviations of the mean. Usually a 95% degree of confidence is sufficiently high, and it is common for one to forgo stretching the range out to 3 standard deviations in conducting analysis. Where the standard deviation is high with a broad range of data, one may wish to consider stratifying the population of property by genre, e.g. romantic comedy, comedy, non-fiction, etc. to determine whether the wide-range is attributable to returns from one type of property, another subset of the population, or the entire population.

Normal distribution occurs when the data falls into a bell-shaped curve. Where the distribution is normal there exists one mode (i.e. one peak) and the curve is symmetrical, i.e. the median, mean, and mode all have the same value. A line splitting this curve at that point will have two symmetrical halves. Note, normal distributions do not necessarily have the same range or standard deviation. Where the standard deviation is modest the bell-curve will have a more peaked shape with a greater concentration around the expected value. Where the standard deviation is higher the bell-curve will have a flatter curve with less of a concentration around the expected value.

In the §458 estimation context it is typically helpful to evaluate the standard deviation. Where a large standard deviation exists based upon an analysis of purely historical data, one might rightfully conclude that the historical data is not meaningful (or at least will not produce a reliable estimate) and that a scenario/probability modeling approach in connection with some experience regarding returns-to-date or projections is more likely to yield an accurate estimate of returns.

For more information, in the Tax Management Portfolios, see White, 570 T.M., Accounting Methods -- General Principles, and in Tax Practice Series, see ¶3540, Timing of Inclusion.