A Cost-Based Theory of Profit Split in Transfer Pricing

1. OECD Regulatory Foundation

The Transactional Profit Split Method (TPSM) is codified in Chapter II, Section C of the OECD Transfer Pricing Guidelines (2022), paragraphs 2.114 through 2.177. Three paragraphs supply the specific authority for a cost-based, payroll-driven allocation.

Paragraph 2.166 requires that profits be split on an economically valid basis reflecting the relative contributions of the parties, approximating the division of profits that independent enterprises would have realized at arm’s length. Splitting factors must be based on objective data and be verifiable.

Paragraph 2.171 authorizes splitting factors based on costs, including relative spending and investment in key areas such as research and development, engineering, and marketing, where these factors capture the relative contributions of the parties and can be measured reliably. Employee compensation is explicitly cited as an appropriate factor relating to performing the relevant functions.

Paragraph 2.177 notes that internal data may be used where the profit splitting factor is based on a cost accounting system, including employee costs related to the transactions.

Three cost-based factors satisfy these criteria and are adopted here as allocation keys: payroll W(j), research and development expenditure XRD(j), and advertising and marketing expenditure XAD(j), for controlled party j = 1 to N.

2. Notation

The following symbols are used throughout.

3. Step 1 — The Accounting Identity

For each controlled party j, operating income (OIADP or EBIT) satisfies the accounting identity:

P(j) = R(j) - C(j)

Payroll W(j), R&D expenditure XRD(j), and advertising expenditure XAD(j) are each subsets of total costs C(j). They need not exhaust C(j).

4. Step 2 — Aggregate the Profits

Sum the accounting identity over all N controlled parties to obtain the combined taxable OIADP:

P^{*}
=
\sum_{j=1}^{N} P(j)
=
\sum_{j=1}^{N} R(j)
-
\sum_{j=1}^{N} C(j)

This sum or aggregation is exact. It rests on no behavioral assumption beyond additive accounting. P* may be positive or negative. Define the aggregate allocation bases in parallel:

W^{*}
=
\sum_{j=1}^{N} W(j) 

XRD^{*}
=
\sum_{j=1}^{N} XRD(j)

XAD^{*}
=
\sum_{j=1}^{N} XAD(j)

5. Step 3 — The Split Factor

Let \lambda_W, \lambda_R, \lambda_A be non-negative weights assigned to payroll, R&D, and advertising respectively, subject to the normalization:

\lambda_W + \lambda_R + \lambda_A = 1,
\qquad
\lambda_W \ge 0,
\quad
\lambda_R \ge 0,
\quad
\lambda_A \ge 0

The composite profit split factor for party j is the weighted sum of the three relative cost shares:

\beta(j)
=
\lambda_W \left[\frac{W(j)}{W^{*}}\right]
+
\lambda_R \left[\frac{XRD(j)}{XRD^{*}}\right]
+
\lambda_A \left[\frac{XAD(j)}{XAD^{*}}\right]

Each bracketed ratio is a share: \frac{W(j)}{W^{*}}is party j’s share of aggregate payroll, and analogously for R&D and advertising. The split factors sum to unity by construction:

\begin{aligned}
\sum_{j=1}^{N} \beta(j)
&=
\lambda_W
\sum \frac{W(j)}{W^{*}}
+
\lambda_R
\sum \frac{XRD(j)}{XRD^{*}}
+
\lambda_A
\sum \frac{XAD(j)}{XAD^{*}}
\\
&=
\lambda_W \cdot 1
+
\lambda_R \cdot 1
+
\lambda_A \cdot 1
\\
&= 1
\end{aligned}

The split factors beta(j) thus form a proper probability vector: non-negative, summing to unity, and computable from internal cost accounting records available in the MNE’s Local File and Master File (paragraph 2.173).

Three special cases are noteworthy. Setting \lambda_W = 1 and \lambda_R = \lambda_A = 0 recovers the pure payroll split. Setting \lambda_R = 1 recovers the pure R&D split, appropriate when technology contribution drives value. Setting \lambda_A = 1 recovers the pure advertising split, appropriate in marketing-intensive businesses. Any convex combination of the three is permissible provided it is economically justified by the functional analysis required under paragraph 2.166.

6. Step 4 — Allocate: The Arm’s Length Profit

The reallocated (arm’s length) OIADP for party j is:

\hat{P}(j)
=
\beta(j)\cdot P^{*}

Substituting the definition of \beta(j):

\hat{P}(j)
=
\left[
\lambda_W \frac{W(j)}{W^{*}}
+
\lambda_R \frac{XRD(j)}{XRD^{*}}
+
\lambda_A \frac{XAD(j)}{XAD^{*}}
\right]
P^{*}

This allocation satisfies two conservation properties.

Exhaustion: summing over all parties recovers P^{*} exactly.

\sum_{j=1}^{N} \hat{P}(j)
=
P^{*}\sum_{j=1}^{N}\beta(j)
=
P^{*}\cdot 1
=
P^{*}

Combined profits are fully distributed. No profit is created or destroyed.

Proportionality: for any two parties j and k:

\frac{\hat{P}(j)}{\hat{P}(k)}
=
\frac{\beta(j)}{\beta(k)}

Relative profit shares mirror relative split factors exactly.

7. Step 5 — Implied Profit Intensities

Define three profit intensities — profit per dollar of payroll, per dollar of R&D, and per dollar of advertising — for party j after reallocation:

q_W(j)
=
\frac{\hat{P}(j)}{W(j)}

q_R(j)
=
\frac{\hat{P}(j)}{XRD(j)}

q_A(j)
=
\frac{\hat{P}(j)}{XAD(j)}

In the pure single-key cases (which is unrealistic, except for payroll), the equalization property follows immediately. Under the pure payroll split \lambda_W = 1:

q_W(j)
=
\frac{\beta(j)\cdot P^{*}}{W(j)}
=
\left[\frac{W(j)}{W^{*}}\right]
\frac{P^{*}}{W(j)}
=
\frac{P^{*}}{W^{*}}

The ratio P*/W* is constant across all j. The payroll-based split therefore equalizes profit per dollar of payroll across every controlled party. This is the equalization property: it is not an assumption but a theorem, a logical consequence of the split formula, and it constitutes a testable restriction against external comparable data if they are available.

Under the pure R&D split \left(\lambda_W = 1\right):

q_R(j)
=
\left[\frac{XRD(j)}{XRD^{*}}\right]
\frac{P^{*}}{XRD(j)}
=
\frac{P^{*}}{XRD^{*}}

Profit per dollar of R&D is equalized. Under the pure advertising split \left(\lambda_A = 1\right):

q_A(j)
=
\frac{P^{*}}{XAD^{*}}

In the general composite case with all three keys active, no single profit intensity is equalized across parties. Instead, the allocation reflects a weighted equalization: parties are rewarded in proportion to their combined cost footprint, weighted by \left(\lambda_W, \lambda_R, \lambda_A \right).

8. Step 6 — The Transfer Pricing Adjustment

The arm’s length adjustment for party j, the additional income (or deduction) required to place reported profits on an arm’s length basis, is:

\Delta(j)
=
\hat{P}(j) - P(j)

Substituting P(j) = R(j) - C(j) and P^{*}=\sum R(j)-\sum C(j):

\Delta(j)
=
\beta(j)
\left[
\sum_k R(k)
-
\sum_k C(k)
\right]
-
\left[
R(j)-C(j)
\right]

The bilateral zero-sum property holds by construction:

\sum_{j=1}^{N}\Delta(j)
=
\sum \hat{P}(j)
-
\sum P(j)
=
P^{*}-P^{*}
=
0

Adjustments net to zero across all controlled parties. This is intrinsic to all profit split methods and is required for internal consistency: the combined pre-adjustment profit equals the combined post-adjustment profit.

9. Step 7 — Implied Operating Margins and the Harmonic Mean Bias

Define the implied operating margin for party j after reallocation:

m(j)
=
\frac{\hat{P}(j)}{R(j)}
=
\frac{\beta(j)\cdot P^{*}}{R(j)}

Suppose that under arm’s length conditions the relationship between reallocated profit and revenue is approximately linear (including a nonzero intercept) across the N controlled parties:

\hat{P}(j)
=
a+b\cdot R(j)+u(j)

where b is the market-wide operating margin (slope) and a is a fixed-charge component (intercept). Dividing through by R(j):

m(j)
=
\frac{\hat{P}(j)}{R(j)}
=
\frac{a}{R(j)}+b

This is precisely the structural form that generates the Harmonic Mean Bias Theorem. If the analyst computes a simple arithmetic mean of the operating profit margins m(j) across parties:

\bar{m}
=
\frac{1}{N}
\sum_{j=1}^{N} m(j)
=
b
+
\frac{a}{N}
\sum_{j=1}^{N}
\left[
\frac{1}{R(j)}
\right]

The term \frac{1}{N}\sum\left[\frac{1}{R(j)}\right] is the reciprocal of the harmonic mean H of the revenue values R(j). Substituting:

\bar{m}
=
b+\frac{a}{H}

where H is the harmonic mean of net revenues:

H
=
\frac{N}{
\sum_{j=1}^{N}
\left[
1/R(j)
\right]
}

The arithmetic mean of the implied operating profit margins overstates b by \frac{a}{H}when a > 0, and understates it when a < 0. Since N is small in the profit split context — typically two to six controlled parties — H can diverge substantially from the arithmetic mean of R(j), and the bias a/H is non-negligible. The bias is structural, non-vanishing with sample size, and present in every quartile of the margin distribution, not only in the mean.

The correct estimator of b is the profit-weighted mean of operating margins, which is algebraically identical to the OLS slope with no intercept:

b
=
\frac{
\sum_j \hat{P}(j)
}{
\sum_j R(j)
}

This estimator is unbiased for b regardless of a, because it cancels the intercept contribution through aggregation. It is the metrologically appropriate summary statistic for operating margins under the arm’s length standard, and it corresponds to the harmonic-mean-corrected estimator in the general linear case

10. Summary of Principal Equations

\begin{aligned}
\beta(j)
=
\lambda_W \left[\frac{W(j)}{W^{*}}\right]
+
\lambda_R \left[\frac{XRD(j)}{XRD^{*}}\right]
+
\lambda_A \left[\frac{XAD(j)}{XAD^{*}}\right]
\end{aligned}

\begin{aligned}
\hat{P}(j)
=
\beta(j)\cdot P^{*}
\end{aligned}

\begin{aligned}
\Delta(j)
=
\hat{P}(j)-P(j),
\qquad
\sum \Delta(j)=0
\end{aligned}

\begin{aligned}
m(j)
=
\frac{\hat{P}(j)}{R(j)},
\qquad
\bar{m}
=
b+\frac{a}{H}
\quad
(\text{harmonic mean bias})
\end{aligned}

\begin{aligned}
H
=
\frac{N}{\sum [1/R(j)]},
\qquad
b
=
\frac{\sum \hat{P}(j)}{\sum R(j)}
\quad
(\text{unbiased estimator})
\end{aligned}

11. Critical Observations

On the allocation key. In economic theory (such as unit prices determined by the profit markup based on unit labor cost, ULC), payroll as the primary profit allocation key is unimpeachable.

On the equalization property. In each single-key special case (as stated in economic theory only payroll solo is defensible), the reallocation imposes equal profit intensity per dollar of the chosen cost base across all parties. This is not an assumption embedded in the formula; it is a result or theorem derivable from the formula.

On the harmonic mean bias. Whenever the tax authority or taxpayer computes an unweighted average of operating margins from the reallocated profits across the controlled parties, the bias a/H applies unless the arithmetic mean is replaced by the profit-weighted mean b=\frac{\sum \hat{P}(j)}{\sum R(j)}. This substitution is not a methodological preference; it is required for algebraic consistency with the linear profit model and for compliance with the most reliable measure standard of the applicable regulations.

On the weight vector. The weights \left(\lambda_W, \lambda_R, \lambda_A \right) are parameters to be determined by the functional analysis, not by convenience. In a labor-intensive service business with no proprietary R&D and minimal advertising, \lambda_W near unity is appropriate. In a pharmaceutical MNE where R&D is the dominant value driver, \lambda_R taking a significant weight in addition to payroll is appropriate. In a consumer goods group where advertising producing brand equity drives operating profit margins, \lambda_A taking a significant weight in addition to payroll is appropriate. Mixed cases call for a weighted composite, with the weights themselves documented and defensible under paragraph 2.169. Payroll is the indispensable operating profit allocation key in the cost-based theory of the profit split calculated from verifiable internal data.

References

OECD (2022). Transfer Pricing Guidelines for Multinational Enterprises and Tax Administrations. OECD Publishing, Paris. Paragraphs 2.114-2.177.

OECD (2018). Revised Guidance on the Application of the Transactional Profit Split Method — BEPS Action 10. Incorporated into OECD (2022), Chapter II, Section C.

EU Joint Transfer Pricing Forum (2019). The Application of the Profit Split Method within the EU. European Commission, Taxud/D2, DOC: JTPF/002/2019/EN.