Uncontrolled, and Comparable to What?

1. The IRS set carries everything

The Commissioner’s expert tested six concentrate manufacturers, the Supply Points (tested parties), in Brazil, Chile, Costa Rica, Ireland, Mexico and Swaziland. A seventh, in Egypt, was adjusted in the taxpayer’s favor and drops out of what follows. The audit report of July 2015 selected eighteen uncontrolled Coca-Cola bottlers as comparables. The litigation report of June 2017 added six more, five of them private companies, for a set of twenty-four.

The selected profit indicator was the return on operating assets (ROA): operating income, augmented by imputed interest on non-interest-bearing liabilities, over average operating assets, pooled across 2007 to 2009 by summing numerator and denominator so that each entity contributes one weighted-average ratio.

The interquartile range (IQR) of that profit ratio decided whether a tested party is inside the arm’s length range. The median decided the size of the adjustment when it is not. Section 1.482-1(e)(3) is explicit: where the interquartile range is used, an adjustment “will ordinarily be to the median of all the results.” On that arithmetic, the notice of deficiency reallocated $9.494 billion; the six Supply Points at issue account for $9.588 billions of it, the difference being Egypt.

Everything downstream rests on the “comparable” set. A profit level indicator is a ratio computed on comparables. If the comparables are neither uncontrolled nor comparable, the arithmetic that follows is about the wrong population.

2. What “uncontrolled” was taken to mean

Two searches, two standards of comparability.

The audit report assembled a universe of 508 companies from two industrial classifications, SIC 2086 and 2087, supplemented by bottler agreements and bottling-association rosters, and screened the survivors on their trade descriptions. Selection by industry code (the comparability of the selected companies assumed rather than demonstrated) is impeachable, on the authority of Nissho Iwai and of Westreco.

The eighteen that emerged are described in one clause: “The 18 companies in the comparison group are Coca-Cola bottlers in which TCCC did not own a controlling interest, and for which sufficient financial data was available from public sources.”

The six added for litigation passed three numerical screens, disclosed in Appendix D. A candidate was excluded if it was more than 50 percent owned by The Coca-Cola Company (TCCC); or more than 75 percent owned by a bottler already in the set; or more than 50 percent owned by a non-holding company and private throughout 2007 to 2009.

Take the first screen. A company half owned by the taxpayer passes. That is a test for non-consolidation. Section 1.482-1(i)(4) prescribes a different test. Control “includes any kind of control, direct or indirect, whether legally enforceable or not, and however exercisable or exercised,” and “it is the reality of the control that is decisive, not its form or the mode of its exercise.”

The second screen is the stranger of the three. It admits a parent and its 75-percent subsidiary to the same comparable set, so that intercompany balances between two comparables would themselves be transfer prices. Whether that occurred here the exhibit does not disclose. That the rule permits it is defect enough.

The third is a data-quality screen in an ownership screen’s clothing, and the label matters, because the reader is told that ownership was tested when what was tested was disclosure.

Now the figures. Table 1 of the audit report discloses The Coca-Cola Company’s direct or indirect equity share in each of the eighteen, under a heading that reads “Uncontrolled Coca-Cola Bottlers” above a column headed “TCCC Ownership.”

Bottler	Country	TCCC %
Coca-Cola Embonor S.A.	Chile	45.5
Coca-Cola Enterprises, Inc.	United States	35.4
Coca-Cola Amatil Limited	Australia	32.2
Coca-Cola FEMSA, S.A.B. de C.V.	Mexico	31.6
Embotelladoras Coca-Cola Polar S.A.	Chile	29.4
Coca-Cola Bottling Co. Consolidated	United States	27.3
Coca-Cola Hellenic Bottling Company S.A.	Greece	23.7
Grupo Continental, S.A.B.	Mexico	20.8
Coca-Cola Icecek A.S.	Turkey	20.1
Nigerian Bottling Co. PLC	Nigeria	15.7
Embotelladora Andina S.A.	Chile	11.0
Coca-Cola West Holdings Company, Ltd.	Japan	3.7
Compania Nortena de Bebidas Gaseosas, S.A.	Spain	0.0
Embotelladoras Arca S.A.B. de C.V.	Mexico	0.0
Haad Thip Public Company Limited	Thailand	0.0
Hokkaido Coca Cola Bottling Co., Ltd.	Japan	0.0
Mikuni Coca-Cola Bottling Co., Ltd.	Japan	0.0
Shikoku Coca-Cola Bottling Co., Ltd.	Japan	0.0

Twelve of the eighteen carry a stake. Eleven exceed ten percent. Six exceed twenty-five. The largest is 45.5 percent, four and a half points below the threshold that would have excluded it. The six Latin American bottlers that furnish the benchmark for four of the six tested parties run from zero to 45.5 percent and average about twenty-three.

Fourteen of the twenty-four carry the Coca-Cola name in their corporate title. It is the kind of fact that shifts the burden of demonstration onto the party asserting independence, and no demonstration is offered.

The Court disposed of the question in a footnote:

“TCCC held minority equity interests in Coca-Cola FEMSA and certain other bottlers. In no case did these stock holdings permit petitioner to control those bottlers’ activities or dictate their decisions.” (Slip op. at 16 n.7.)

Two observations. The footnote answers a question about the reality of control by measuring the form of it, which is the criterion the regulation declares not decisive. And it addresses equity alone: not the board representation, not the franchise agreement terminable by the franchisor, and not the fact that the franchisor sets the price of the supposed comparable’s principal input.

That last is the circularity from which none of this escapes. Every comparable buys concentrate from the party whose income is being tested. Its operating profit is the residual after a price the tested party’s parent administers. The expert’s own footnote records that the parent sought to raise concentrate prices for certain Latin American bottlers because it judged their profitability higher than it needed to be, and that the bottlers resisted by threatening to reduce their marketing spending. Returns that are administered cannot discipline the administrator. Returns that are negotiated measure the bargaining.

3. Geographic comparability, disclaimed and then decisive

The report is candid on the point:

“Based on the information available to me, I cannot determine whether the reasons for geographic variation in bottler profitability would necessarily lead to similar geographic variation in Supply Point profitability, if the Supply Points were uncontrolled.” (Exh. 8294-R at 49.)

The geographic split is then offered as a sensitivity test. It is not a sensitivity test. It is the operative benchmark. The IRS adjusted the four Latin American Supply Points to the median of the Latin American bottlers, and the Irish and Swazi Supply Points to the median of the thirteen non-East-Asian bottlers of the audit report. The table showing that these medians reproduce the notice to within five percent is the geographic table. A condition the expert declined to verify carries the load.

Where, then, do the comparables sit?

Jurisdiction	Comparable bottlers	Supply Point tested
Spain (all private)	5	no
Japan	5	no
Chile	3	yes
Mexico	3	yes
United States	2	no
Australia	1	no
China (private)	1	no
Greece	1	no
Nigeria	1	no
Thailand	1	no
Turkey	1	no
Brazil	0	yes
Costa Rica	0	yes
Ireland	0	yes
Swaziland	0	yes
Total	24	6 tested parties

Two of the six tested jurisdictions contribute comparables. Four contribute none. The Irish and Swazi Supply Points are benchmarked against a group of eleven that is, on the expert’s own description from the stand, “the greatest number European bottlers,” with one United States bottler, Coca-Cola Enterprises, a Nigerian bottler, Coca-Cola Icecek, and Coca-Cola Amatil.

The report anticipates the objection to Table 2, and the objection is sound: the country shown is “the country of incorporation or headquarters,” and “a number of these bottlers were large multinational enterprises with operations in many countries.” Coca-Cola Hellenic served twenty-eight national markets across Western and Central Europe, the Balkans, Russia and Ukraine. Coca-Cola Enterprises operated in Belgium, France, Great Britain, Luxembourg, Monaco, the Netherlands, Norway, Sweden and North America. Icecek operated in Turkey, Pakistan, Central Asia and the Middle East. Amatil in Australia, New Zealand and Indonesia.

Whether any comparable served Brazil, Costa Rica, Ireland or Swaziland as a market of consequence the exhibits do not say, because the financial data are consolidated and no segment disclosure accompanies them. A bottler whose returns consolidate twenty-eight markets is a weighted average of twenty-eight economic environments, and the weights are nowhere given. Multi-nationality does not repair a geographic mismatch.

Section 1.482-1(d)(3)(iv) makes economic conditions a comparability factor and enumerates eight: the similarity of geographic markets; the relative size of each market and the extent of overall economic development in each; the level of the market; relevant market shares; the location-specific costs of the factors of production and distribution; the extent of competition in each market; the economic condition of the particular industry; and the alternatives realistically available to buyer and seller.

Three of the eight were reached. The Court rested its economic-conditions holding on the extent of competition, subdivision (F), and on the alternatives realistically available, subdivision (H): the Supply Points were replaceable at low cost and held no bargaining position, while the bottlers enjoyed de facto exclusivity and could secure as much as 55 percent of system profit. The level of the market was raised on cross-examination and rejected.

Two were not reached by anyone. The relative size of each market and the extent of overall economic development in each: Swaziland against Spain, Costa Rica against Japan. And the location-specific costs of the factors of production and distribution. The local sovereign interest rates differed a lot.

On the second of those the report supplies its own evidence and then declines to read it. To impute a return on non-interest-bearing liabilities the study applies a local cost of funds to each company: 1.0 percent for the three-month yen rate in 2007, and 17.4 percent for the Turkish overnight rate plus spread in that same year, 17.6 percent in the next. Those are the expert’s measurements of how far apart these financial markets stood in the years at issue. They were material enough to move the numerator of every ratio in the study. They were not treated as evidence about the comparability of the markets that produced the ratios.

4. A partition chosen by inspecting the returns

No test for a country or a region effect appears in the report. The words regression, standard error and confidence interval do not occur in it. Economics is absent. The grouping rests on inspection, and the expert said so on cross-examination:

“Because I could see the geographic variation – it was apparent in the observed returns – I thought it was a conservative approach to apply the high-profit set of Latin American bottlers as a benchmark against the Latin American supply points.” (Tr. 8355.)

Twenty-four bottlers become three groups: six Latin American, seven Asian discarded, eleven remaining. The interquartile spread falls from 16.4 points across the twenty-four to 8.8 points within the six and 6.5 points within the eleven.

What the narrowing does is not uniform, and the expert’s own tables record it.

Supply Point	Adjusted ROA %	Against all 24 (9.3 – 25.7)	Against its own group
Ireland	25.7	at the upper quartile	above (14.4 – 20.9)
Swaziland	24.8	inside	above (14.4 – 20.9)
Brazil	33.5	above	inside (31.8 – 40.6)
Mexico	35.1	above	inside (31.8 – 40.6)
Chile	35.2	above	inside (31.8 – 40.6)
Costa Rica	34.6	above	inside (31.8 – 40.6)

The partition moved the four Latin American Supply Points from above the range into it, close to the median. For Ireland and Swaziland, it moved them out. Adjusting every Supply Point to its segment median would have raised the aggregate by $296 million, three percent above the notice.

Ireland’s adjusted return on operating assets is 25.7 percent. The upper quartile of the twenty-four is 25.7 percent. The upper quartile of the eleven is 20.9 percent. Nothing about the Irish Supply Point differs between those two lines. Only the reference set differs. A benchmark whose verdict on a taxpayer turns on an unmodelled choice of comparison group is not a benchmark. It is the choice, restated as a number.

Nor does the report claim otherwise, and its silence is the whole of the objection. The interquartile range is presented as the arm’s length range of section 1.482-1(e)(2)(iii)(B), a regulatory construct that makes no assertion about sampling. Fair enough. But a range whose width decides a ten-figure question is then reported without a standard error, without an interval, and without any test of the partition that generates it. No goodness of fit between the components of ROA is present; again, economics or defensible data analysis is not present. The regulation supplies the rote recipe. It does not supply the inference, and the study does not supply it either. Almost USD 10 billion adjustment based on Pygmalion comparables, no demonstrable appeal to economics, and the most rudimentary statistical summaries based on quartiles.

5. What was contested, and what was not

The record does not show a taxpayer that failed to fight. It shows a taxpayer that fought everywhere but one place, and the place was chosen.

Counsel took the expert through section 1.482-1(d)(3) factor by factor over three days and closed with several concessions: no adjustment for differences in functions beyond the liabilities imputation; none for location; none for risks; none for asset composition or profiles; and, for economic conditions, only “the geographics groupings, which may reflect that to some extent, but other than that, no.” Counsel put the 1994 preamble’s “method of last resort” language to him. The taxpayer’s experts attacked the return on operating assets as a profit level indicator, one of them decomposing it into margin and turnover on a scatter the expert answered from the stand. The expert called his own instrument “crude” and defended this application as “on the side of relatively more refined.”

What went uncontested occupies one sentence of the opinion, and that sentence is the reason this series exists:

“Petitioner does not contend that Dr. Newlon erred in the search process he employed. Nor does petitioner challenge the representativeness of Dr. Newlon’s 24-bottler sample or the reliability of the data he extracted from bottler financial statements.” (Slip op. at 136.)

The sentence covers three things: the search process, the representativeness of the sample, and the reliability of the data extracted from bottler financial statements. It does not say that comparability went unchallenged. Part IV.B of the opinion, pages 120 to 133, takes up the taxpayer’s comparability arguments and rejects them one by one. The distinction is worth marking, because the silences in sections 2, 3 and 4 above are of two kinds. The search and the sample are conceded in terms. The ownership screens of Appendix D and the instability of the partition are absent from the opinion altogether, which is a silence and not a concession, and rests on the record rather than on this sentence.

Two of those silences merit an explanation that is not negligence, and an opponent will offer it before I do. The taxpayer’s case required the bottlers to be independent. Its architecture — that the bottlers created and owned marketing intangibles, that they bargained the parent to an equitable division of system profit and once to 55 percent in their favor, that they could not be replaced –collapses the moment the bottlers are recharacterized as controlled. Counsel could not argue that the comparables were creatures of the taxpayer and in the same breath that they were formidable arm’s length adversaries. The concession on independence was the price of the dubious theory, and it was paid with open eyes.

What the theory cost is measurable, and the cost is not distributed as one would expect. For five of the six tested parties a challenge to the set was unwinnable on its face. The reported returns on operating assets ran from 94.8 percent for Mexico to 214.7 percent for Ireland; the highest of the twenty-four comparables was 43.6 percent. No re-screening carries a tested party from 214.7 into an interquartile range built out of that population, and under a standard of abuse of discretion the attempt is worse than futile. But one needs to examine the effect of the dissimilarity of asset turnovers.

For the sixth it was expensive. Ireland’s adjusted return is 25.7 percent and the upper quartile of the twenty-four is 25.7 percent. At that margin the composition of the set is outcome-determinative: one comparable added or struck moves the quartile and moves Ireland across it. Ireland carries $6.19 billion of the $9.49 billion in the notice. Two-thirds of the money turned on a boundary that no one put to the witness, in a case where every other issue was pressed to exhaustion.

Geography and the partition had no such excuse. Neither contradicted anything in the taxpayer’s theory; attacking either cost that theory nothing. Both were left alone, and the Court filled the silence itself, attributing East Asian returns to “local economic conditions or financial reporting rules” and Latin American returns “in part to local accounting rules and (apparently) to having a more atomized retail customer base.” No party had tested either proposition. The parenthesis is the Court’s own, among many candid sentences in the opinion.

Section 1.482-1(c)(2)(ii) makes the completeness and accuracy of the data and the reliability of the assumptions a factor in identifying the best method. Comparability under section 1.482-1(d) is a condition on the method, not an ornament attached to it. A method applied to a set whose composition nobody examined has not been shown to be the best method. It has been shown to be the only one attempted.

A case argued as a dispute over who owns a brand treats the comparable set as a technical annex to somebody else’s report. The taxpayer scotched the snake and not killed it: the analysis was wounded at every joint above the base and left standing on the base. Elsewhere, I stated that Coca-Cola failed to rebut, and my view has not changed after another reading of the available record. Call it legal zeal and inattention to facts. The legal misconception (bait and switch) and neglect of facts show up again in the protest. Daniel Kornstein, Shakespeare’s Legal Appeal (“Kill all the lawyers”), Princeton University Press, 1994 can be consulted for literary precedents.

6. What follows

Comparability is the first of several separable failures and the one requiring the least machinery to see. The instalments to come take up the others: the source regimes and accounting frameworks mixed into a single denominator; the tested party’s own denominator, and what “operating assets” was taken to mean on each side of the ratio; the algebra of a ratio whose expectation is not the parameter anyone wants, and the sampling uncertainty of quartiles nobody computed; and the frequency-distribution display, which locates a concentrate manufacturer inside a population of bottlers by a difference in asset turnovers rather than in operating profit margins. We’ve seen this error before in DuPont.

They are separate failures rather than several versions of one failure, and the first is sufficient on its own. If the comparables are not uncontrolled and the geographic condition is untested, the quartiles that follow describe that set and not much else.

Principiis obsta; sero medicina paratur. Resist the beginnings; the medicine is compounded too late, once the ills have gathered strength through long delay. Ovid was writing about love. He might have been writing about the comparable set, which is the beginning challenge of every profit level indicator ever computed.