Axiomatizations of the proportional Shapley value

Manfred Besner

doi:10.1007/s11238-019-09687-7

Theory and Decision

pp 1–23 | Cite as

Axiomatizations of the proportional Shapley value

Authors
Authors and affiliations

Manfred Besner

Article

First Online: 31 January 2019

20 Downloads

Abstract

We present new axiomatic characterizations of the proportional Shapley value, a weighted TU-value with the worths of the singletons as weights. The presented characterizations are proportional counterparts to the famous characterizations of the Shapley value by Shapley (Contributions to the theory of games, vol. 2. Princeton University Press, Princeton, pp 307–317, 1953b) and Young (Cost allocation: methods, principles, applications. North Holland Publishing Co, 1985a). We introduce two new axioms, called proportionality and player splitting, respectively. Each of them makes a main difference between the proportional Shapley value and the Shapley value. If the stand-alone worths are plausible weights, the proportional Shapley value is a convincing alternative to the Shapley value, for example in cost allocation. Especially, the player splitting property, which states that players’ payoffs do not change if another player splits into two new players who have the same impact to the game as the original player, justifies the use of the proportional Shapley value in many economic situations.

Keywords

Cost allocation Dividends Proportional Shapley value (Weighted) Shapley value Proportionality Player splitting

This is a preview of subscription content, to check access.

Notes

Acknowledgements

We would like to thank André Casajus, Winfried Hochstättler, Jörg Homberger, Frank Huettner, Hans Peters, and especially an anonymous referee for their helpful comments and suggestions.

Appendix

Additional lemmas and a remark, used in the proofs

Remark 4

We can consider the collection of all TU-games $v \in \mathcal {G}^N, N\in \mathcal {N},$ as a vector space $\mathbb {R}^{2^N-1}$. Each game v is represented by a vector $\overrightarrow{v} \in \mathbb {R}^{2^N-1}$, where the entries in the $2^{|N|-1}$ coordinates of the $2^{|N|-1}$ coalitions $ S\subseteq N,\, S\ne \emptyset ,$ get the worth v(S) of the respective coalition S. Hence, there exists for every game v a vector $\overrightarrow{\varDelta _v} \in \mathbb {R}^{2^{N}-1}$, which corresponds to the vector $\overrightarrow{v}$, where the entries of the coordinates get the dividends of the respective coalitions. By Eq. (1), we obtain with $v, v_1, v_2 \in \mathcal {G}^N,$ and for all $S \subseteq N$:

$$\begin{aligned}&\overrightarrow{\varDelta _v}&=\overrightarrow{\varDelta _{v_1}}+\overrightarrow{\varDelta _{v_2}}\\ \Leftrightarrow&\varDelta _v(S)&= \varDelta _{v_1}(S) +\varDelta _{v_2}(S)\\ \Leftrightarrow&v(S)-\sum _{R\subsetneq S}\varDelta _v(R)&={v_1}(S)-\sum _{R\subsetneq S}\varDelta _{v_1}(R)+{v_2}(S)-\sum _{R\subsetneq S}\varDelta _{v_2}(R)\\ \Leftrightarrow&v&= v_1 +v_2. \end{aligned}$$

Lemma 4

Equation (4) in WCSE can be replaced equivalently by the following:

$$\begin{aligned} \varDelta _{v}(S)={\left\{ \begin{array}{ll}\varDelta _w(R)+c,\text { if } S=R, \\ \varDelta _w(S), \text { otherwise. }\end{array}\right. } \end{aligned}$$

Proof

Let the notation and the preconditions as in WCSE. By (2), if $v(S)=w(S)$ for all $S\nsupseteq R$, we have $\varDelta _{v}(S)=\varDelta _w(S)$ for all such S and vice versa. Hence, by (1), $v(R)=w(R)+c$ is equivalent to $\varDelta _{v}(R)=\varDelta _w(R)+c$. By induction on the size $s:=|S|$, we now show that $v(S)= w(S)+c\;\Leftrightarrow \;\varDelta _{v}(S)=\varDelta _w(S)$ for all proper supersets $S \supsetneq R$.

Initialization: Let $S\supsetneq R$ and $s=|R|+1$. R is the only proper subset of S where there is a difference of the related dividends in both coalition functions and we obtain the following:

$$\begin{aligned}\begin{array}{rcrcl} v(S)=w(S)+c&{}\underset{(1)}{\Leftrightarrow }&{}v(S)-\displaystyle {\sum _{T\subsetneq S}\varDelta _v(T)}&{}=&{}w(S)+c-\displaystyle {\sum _{\begin{array}{c} T\subsetneq S,\\ T \ne R \end{array}}\varDelta _w(T)}-(\varDelta _w(R)+c)\\ &{}\underset{(1)}{\Leftrightarrow }&{}\varDelta _{v}(S)&{} =&{}\varDelta _w(S). \end{array} \end{aligned}$$

Induction step: Assume that equivalence holds for $s'=s-1$, $|R|+1\le s'\le n-1$ (IH). Then, by (IH), R is again the only proper subset of S with not equal related dividends in v and w. By (1), we get $v(S)= w(S)+c\;\Leftrightarrow \;\varDelta _{v}(S)=\varDelta _w(S)$ as before and Lemma 4 is shown. $\square $

Lemma 5

(Casajus and Huettner 2008). If $i \in N$ and $v,w \in \mathcal {G}^N$, then ${\text {MC}}_i^v(S)={\text {MC}}_i^w(S)$ for all $S \subseteq N\backslash \{i\}$ iff $\varDelta _v(S \cup \{i\}) = \varDelta _w(S \cup \{i\})$ for all $S \subseteq N\backslash \{i\}$.

Proofs

Proof of Lemma 1

Let $i,j \in N$ and $v \in \mathcal {G}^N$. If $S= \emptyset $, we have $v(S\cup \{k\})=v(S)+v(\{k\})$. By induction on the size $s:=|S|$ of all coalitions $S \subseteq N\backslash \{i,j\},\,S \ne \emptyset ,$ we show that

$$\begin{aligned} v(S\cup \{k\})=v(S)+v(\{k\})\quad \Leftrightarrow \quad \varDelta _v(S \cup \{k\})=0. \end{aligned}$$

Initialization: Let $s=1$. For $k \in \{i,j\}$ we have the following:

$$\begin{aligned}{}\begin{array}[b]{crcl} &{}v(S\cup \{k\}) &{}=&{}v(S)+v(\{k\})\\ \underset{(1)}{\Leftrightarrow }&{} \varDelta _v(S\cup \{k\})+\varDelta _v(S)+\varDelta _v(\{k\})&{}=&{}\varDelta _v(S)+\varDelta _v(\{k\})\\ \Leftrightarrow &{}\varDelta _v(S\cup \{k\})&{}=&{}0. \end{array} \end{aligned}$$

Induction step: Assume that equivalence and equality in the first and last line of the system above hold for all coalitions $S'$ with $s'\ge 1$ (IH), and let $s=s'+1$ and $k \in \{i,j\}$. We get the following:

$$\begin{aligned}\begin{array}{crcl} &{}v(S\cup \{k\}) &{}=&{}v(S)+v(\{k\})\\ \underset{(1)}{\Leftrightarrow }&{} \varDelta _v(S\cup \{k\})+\displaystyle {\sum _{R\subsetneq (S\cup \{k\})}\varDelta _v(R)}&{}=&{}\displaystyle {\sum _{R\subseteq S}\varDelta _v(R)}+\varDelta _v(\{k\})\\ \underset{(IH)}{\Leftrightarrow }&{}\varDelta _v(S\cup \{k\})+\varDelta _v(\{k\})+\displaystyle {\sum _{R\subseteq S}\varDelta _v(R)}&{}=&{}\displaystyle {\sum _{R\subseteq S}\varDelta _v(R)}+\varDelta _v(\{k\})\\ \Leftrightarrow &{}\varDelta _v(S\cup \{k\})&{}=&{}0. \end{array} \end{aligned}$$

Thus, the equivalence is shown. $\square $

Proof of Theorem 1

I. Existence: By Béal et al. (2018), ${\text {Sh}}^p$ satisfies E and D.

$\bullet $P: Let $v \in \mathcal {G}_{0}^N$ and $i,j \in N$, such that i and j are weakly dependent in v. We have the following:

$$\begin{aligned}{}\begin{array}[b]{rcl} {\text {Sh}}^{p}_i(v)&{} = &{}\displaystyle {\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{k \in S}v(\{k\})}\varDelta _v(S)}\underset{Lem. 1}{=}v(\{i\})+ \sum _{\begin{array}{c} S\subseteq N,\\ {\{i,j\} \subseteq S} \end{array}} \frac{ v(\{i\})}{\sum _{k \in S}v(\{k\})}\varDelta _v(S)\\ &{}=&{}\displaystyle {\frac{v(\{i\})}{v(\{j\})}v(\{j\})+\frac{v(\{i\})}{v(\{j\})} \sum _{\begin{array}{c} S\subseteq N,\\ {\{i,j\} \subseteq S} \end{array}} \frac{ v(\{j\})}{\sum _{k \in S}v(\{k\})}\varDelta _v(S)}\\ &{}\underset{Lem.1}{=}&{}\displaystyle {\frac{v(\{i\})}{v(\{j\})} \sum _{\begin{array}{c} S\subseteq N,\\ {S\ni j} \end{array}} \frac{ v(\{j\})}{\sum _{k \in S}v(\{k\})}\varDelta _v(S)}=\frac{v(\{i\})}{v(\{j\})}{\text {Sh}}^{p}_j(v). \end{array} \end{aligned}$$

$\bullet $WA: Let $v,w \in \mathcal {G}_{0}^N$ with $w(\{i\})=c\cdot v(\{i\})$ for all $i \in N,\, c>0$. We have the following:

$$\begin{aligned}{}\begin{array}[b]{rcl} {\text {Sh}}_i^p(v)+ {\text {Sh}}_i^p(w)&{} \underset{(3)}{=}&{} \displaystyle {\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{j \in S}v(\{j\})}\varDelta _v(S)+\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ w(\{i\})}{\sum _{j \in S}w(\{j\})}\varDelta _w(S)} \\ &{}=&{}\displaystyle {\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{j \in S}v(\{j\})}\varDelta _v(S)+\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ c\cdot v(\{i\})}{\sum _{j \in S}c\cdot v(\{j\})}\varDelta _w(S)} \\ &{}=&{}\displaystyle {\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{j \in S}v(\{j\})}\big [\varDelta _v(S)+\varDelta _w(S)\big ]} \\ &{}=&{} \displaystyle {\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ (1+c)v(\{i\})}{\sum _{j \in S}(1+c)v(\{j\})}\big [\varDelta _v(S)+\varDelta _w(S)\big ]} \\ &{}=&{}\displaystyle {\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ v(\{i\})+w(\{i\})}{\sum _{j \in S}\big [v(\{j\})+w(\{j\})\big ]}\varDelta _{v+w}(S)}\\[1ex] &{}=&{} {\text {Sh}}_i^p(v+w). \end{array} \end{aligned}$$

II. Uniqueness: Let $N\in \mathcal {N}, \,n:=|N|$, $v \in \mathcal {G}^{N}_{0},$ and $\varphi $ a TU-value which satisfies all axioms of Theorem 1. To prove uniqueness, we will show that $\varphi $ equals ${\text {Sh}}^p$.

For $n=1$, $\varphi $ equals ${\text {Sh}}^p$ by E.

Let now $n\ge 2$. For each coalition $S \subseteq N,\,S\ne \emptyset ,$ we define, corresponding to Remark 4, a TU-game $v_S \in \mathcal {G}^{N}_{0}$ through a vector $\overrightarrow{v_S}\in \mathbb {R}^{2^N-1}$ by assigning the coordinates of the related vector $\overrightarrow{\varDelta _{v_S}}\in \mathbb {R}^{2^N-1}$ in the entry of a coalition $R \subseteq N,\, R \ne \emptyset ,$ the dividend:

$$\begin{aligned} \varDelta _{v_S}(R):= {\left\{ \begin{array}{ll}\dfrac{v(\{j\})}{2^n-1}, \text { if } R=\{j\} \text { for all } j\in N, \\ \varDelta _{v}(S), \text { if } R=S, \,|S| \ge 2,\\ 0, \,\text {otherwise.} \end{array}\right. } \end{aligned}$$

Thus, each vector $\overrightarrow{v_S}\in \mathbb {R}^{2^N-1}$ gets in the coordinates of coalitions $R\subseteq N,\,R \ne \emptyset ,$ the entry:

$$\begin{aligned} v_S(R)= {\left\{ \begin{array}{ll} \varDelta _v(S)+\sum \nolimits _{j\in R}\dfrac{v(\{j\})}{2^n-1}, \text { if } R\supseteq S,\,|S| \ge 2,\\ \sum \nolimits _{j \in R}\dfrac{v(\{j\})}{2^n-1},\,\text {otherwise.} \end{array}\right. } \end{aligned}$$

(5)

We have $\displaystyle {\overrightarrow{\varDelta _{v}}=\sum _{\begin{array}{c} S \subseteq N, \\ S\ne \emptyset \; \end{array}}\overrightarrow{\varDelta _{v_S}}}$, and so, by Remark 4, $\displaystyle {v=\sum _{\begin{array}{c} S \subseteq N, \\ S\ne \emptyset \; \end{array}}v_S.}$

By D, we obtain the following:

$$\begin{aligned} \varphi _i(v_S)={\left\{ \begin{array}{ll}v_S(\{i\})=\dfrac{v(\{i\})}{2^n-1} \text { for all } i \in N \text { and }|S|= 1, \text { and} \\ v_S(\{i\})=\dfrac{v(\{i\})}{2^n-1} \text { for all } i \in N,\, i \notin S,\,|S|\ge 2. \end{array}\right. } \end{aligned}$$

(6)

By Lemma 1, all players $i \in S,\,|S|\ge 2,$ are pairwise weakly dependent in $v_S$. We get for an arbitrary $i \in S,\,|S|\ge 2,$ and by $v_S(N)\underset{\begin{array}{c} (5) \end{array}}{=}\varDelta _v(S)+\sum _{j\in N}\dfrac{v(\{j\})}{2^n-1}$:

$$\begin{aligned}{}\begin{array}[b]{rcl} \sum \nolimits _{j \in S}\varphi _j(v_S)&{}\underset{(\mathbf P )}{=}&{} \displaystyle {\sum _{j \in S}\frac{v_S(\{j\})}{v_S(\{i\})}\varphi _i(v_S)}= \sum _{j \in S}\frac{v(\{j\})}{v(\{i\})}\varphi _i(v_S)\\ &{}\underset{\begin{array}{c} (\mathbf E ) \end{array}}{=}&v_S(N)- \displaystyle {\sum _{j \in N\backslash S}\varphi _j(v_S)} \underset{\begin{array}{c} (6) \end{array}}{=}\varDelta _v(S)+\sum _{j\in S}\dfrac{v(\{j\})}{2^n-1} \\ \Leftrightarrow \;\;\varphi _i(v_S)&{}=&{} \displaystyle {\frac{v(\{i\})}{\sum _{j \in S}v(\{j\})}\varDelta _v(S)}+ \frac{v(\{i\})}{2^n-1}. \end{array} \end{aligned}$$

(7)

Therefore, we have by (3), (6), and (7) for all $S \subseteq N,\,S\ne \emptyset :$

$$\begin{aligned} \varphi _i(v_S)={\text {Sh}}^p_i(v_S) \text { for all } i\in N. \end{aligned}$$

${\text {Sh}}^p$ and $\varphi $ satisfy WA. It follows:

$$\begin{aligned} \varphi _i(v)={\text {Sh}}_i^p(v) \text { for all }i \in N, \end{aligned}$$

and uniqueness is shown. $\square $

Proof of Proposition 1

$\Rightarrow $: We show that WM implies WCSE: Let v and w two TU-games satisfying the hypotheses of WCSE, i.e. for a coalition $R \subseteq N, \,|R|\ge 2,$$c \in \mathbb {R},$ we have the following:

$$\begin{aligned} v(S)={\left\{ \begin{array}{ll}w(S)+c, \text { if }S\supseteq R,\\ w(S),\text { if }S\nsupseteq R.\end{array}\right. } \end{aligned}$$

Let $\varphi $ be a TU-value which obeys WM. By Lemma 4, we have

$$\begin{aligned} \varDelta _{v}(S)={\left\{ \begin{array}{ll}\varDelta _w(R)+c,\text { if } S=R, \\ \varDelta _w(S), \text { otherwise.}\end{array}\right. } \end{aligned}$$

Thus, we have $\varDelta _v(S \cup \{i\}) = \varDelta _w(S \cup \{i\})$ for all $i \in N\backslash R$ and $S \subseteq N\backslash \{i\}$. It follows from Lemma 5 that ${\text {MC}}_i^v(S)={\text {MC}}_i^w(S)$ for all $S \subseteq N\backslash \{i\}$. Therefore, we can use WM and get $\varphi _i(v)= \varphi _i(w)\, \text { for all } i \in N\backslash R$ and WCSE is satisfied.

$\Leftarrow $: we show that WCSE implies WM: let $N \in \mathcal {N},\,i \in N,\,v,w \in \mathcal {G}^N$ two coalition functions satisfying the hypothesis of WM, i.e., ${\text {MC}}_i^v(S)={\text {MC}}_i^w(S)$ for all $S \subseteq N\backslash \{i\}$ and $w(\{k\})= v(\{k\})$ for all $k \in N$ and $\varphi $ a value satisfying WCSE. Then, by Lemma 5, we have $\varDelta _v(T) = \varDelta _w(T)$ for all $T \subseteq N,\,T\ni i$. Let $\mathcal {R}=\{R_j\subseteq N:\varDelta _v(R_j)\ne \varDelta _w(R_j) \}$ an indexed set of all subsets of N with different dividends in v and w, $1 \le j \le |\mathcal {R}|$. We inductively define a sequence of coalition functions $w_j,\,0\le j \le |\mathcal {R}|,$ by $w_j:=w$ if $j=0$, and, if $1\le j \le |\mathcal {R}|:$

$$\begin{aligned} \varDelta _{w_{j}}(S):={\left\{ \begin{array}{ll}\varDelta _{w_{j-1}}(R_j)+\big [\varDelta _v(R_j)-\varDelta _{w_{j-1}}(R_j)\big ],\text { if } S=R_j, \\ \varDelta _{w_{j-1}}(S), \text { if }S\subseteq N, \,S\ne R_j.\end{array}\right. } \end{aligned}$$

Then, we have $w_{|\mathcal {R}|}=v$ and, by Lemma 4 and WCSE, we get $ \varphi _i(w_{j})=\varphi _i(w_{j-1})$ for all $j,\, 1 \le j \le |\mathcal {R}|,$ and therefore, $\varphi _i(v)=\varphi _i(w)$ and WM is satisfied. $\square $

Proof of Theorem 2

I. Existence: by Theorem 1, ${\text {Sh}}^p$ satisfies E and P.

$\bullet $WCSE: By Lemma 4, we have for v and a coalition R from WCSE:

$$\begin{aligned} \varDelta _{v}(S)={\left\{ \begin{array}{ll}\varDelta _w(R)+c,\text { if } S=R, \\ \varDelta _w(S), \text { otherwise}.\end{array}\right. } \end{aligned}$$

Thus, we obtain for all $i\in N\backslash R$ by the following:

$$\begin{aligned} {\text {Sh}}^{p}_i(v) = \sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{j \in S}v(\{j\})}\varDelta _v(S)=\sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i} \end{array}} \frac{ w(\{i\})}{\sum _{j \in S}w(\{j\})}\varDelta _w(S)={\text {Sh}}^{p}_i(w). \end{aligned}$$

II. Uniqueness: Let $N\in \mathcal {N}, \,n:=|N|$, $v \in \mathcal {G}^{N}_{0}$, and $\varphi $ a TU-value which satisfies all axioms of Theorem 2. We will show that $\varphi $ satisfies Eq. (3).

For $n=1$, Eq. (3) is satisfied by E.

Let $n\ge 2$. We use an induction on the size $r:=|\{R \subseteq N: R \text { is active in } v$ and $|R|\ge 2\}|$.

Initialization: Let $r=0$. By Lemma 1, all players $i,j \in N$ are pairwise weakly dependent in v. We get for an arbitrary $i \in N$:

$$\begin{aligned} \sum _{j \in N}\varphi _j(v)\underset{(\mathbf P )}{=} \sum _{j \in N}\frac{v(\{j\})}{v(\{i\})}\varphi _i(v)\underset{(\mathbf E )}{=}v(N). \end{aligned}$$

With $v(N)= \sum _{j \in N}v(\{j\})$ it follows that $\varphi _i(v) =v(\{i\})$ and Eq. (3) holds to $\varphi $ if r = 0.

Induction step: Assume that Eq. (3) holds to $\varphi $ if $r\ge 0,\,r$ arbitrary (IH), and let exactly $r+1$ coalitions $Q_k \subseteq N, \,|Q_k|\ge 2,\, 1\le k\le r+1,$ active in v. Let Q be the intersection of all such coalitions $Q_k$:

$$\begin{aligned} Q=\bigcap _{1\le k\le r+1}Q_k. \end{aligned}$$

We distinguish two cases: (a) $i \in N\backslash Q$ and (b) $i \in Q$.

(a) Each player $i \in N\backslash Q$ is a member of at most r active coalitions $Q_k,\, |Q_k|\ge 2,$ and v gets at least one active coalition $R_i,\,|R_i|\ge 2$, $i \notin R_i$. Hence, there exists a coalition function $w_i\in \mathcal {G}^{N}_{0}$, where all coalitions get the same dividend in $w_i$ as in v, except coalition $R_i$ which gets the dividend $\varDelta _{w_i}(R_i)=0,$ and there is a scalar $c \in \mathbb {R},\,c\ne 0, $ with the following:

$$\begin{aligned} \varDelta _{v}(S)={\left\{ \begin{array}{ll}\varDelta _{w_i}(R_i)+c,\text { if } S=R_i, \\ \varDelta _{w_i}(S), \text { otherwise. }\end{array}\right. } \end{aligned}$$

By Lemma 4 and WCSE, we get $\varphi _i(v)=\varphi _i(w_i) \text { with } i \in N\backslash R_i$ and because there exists for all $i \in N\backslash Q$ a such $R_i$, we get $\varphi _i(v)=\varphi _i(w_i) \text { for all } i \in N\backslash Q$. All coalition functions $w_i$ get at most r active coalitions with at least two players and Eq. (3) follows by (IH). Thus, we have the following:

$$\begin{aligned} \varphi _i(v)={\text {Sh}}_i^p(v) \text { for all }i \in N\backslash Q. \end{aligned}$$

(8)

(b) If $Q=\{i\}$, we get, by E of $\varphi $ and ${\text {Sh}}^p$ and case (a), $ \varphi _i(v)= {\text {Sh}}^p_i(v)$. If $|Q|\ge 2$, each player $j \in Q$ is a member of all $r+1$ active coalitions $Q_k \subseteq N, \,|Q_k|\ge 2,\, 1\le k\le r+1,$ and therefore, by Lemma 1, all players $j \in Q$ are weakly dependent. By P and E of $\varphi $ and ${\text {Sh}}^p$, we get for an arbitrary $i \in Q$ the following:

$$\begin{aligned} \sum _{j \in Q}\varphi _j(v)&\underset{(\mathbf P )}{=} \sum _{j \in Q}\frac{v(\{j\})}{v(\{i\})}\varphi _i(v) \underset{\begin{array}{c} (\mathbf E )\\ (8) \end{array}}{=}v(N)- \sum _{j \in N\backslash Q}{\text {Sh}}^p_j(v)\underset{\begin{array}{c} (\mathbf E ) \end{array}}{=}\sum _{j \in Q}{\text {Sh}}^p_j(v)\\&\underset{(\mathbf P )}{=} \sum _{j \in Q}\frac{v(\{j\})}{v(\{i\})}{\text {Sh}}^p_i(v) \;\;\Leftrightarrow \;\;\varphi _i(v)={\text {Sh}}_i^p(v) \end{aligned}$$

and together with I. the proof is complete. $\square $

Proof of Proposition 2

Let $(N,v) \in \mathcal {G}_{0}^N,\,j \in N,$ and $(N^j,v^j) \in \mathcal {G}_0^{N^j}$ a corresponding split player game to (N, v). We point out that we have for all $S \subseteq N\backslash \{j\},\, S \ne \emptyset ,$$\varDelta _{v^j}(S)=\varDelta _{v}(S),$$\varDelta _{v^j}(S\cup \{k,l\})=\varDelta _v(S\cup \{j\})$, and $\varDelta _{v^j}(S\cup \{k\})=\varDelta _{v^j}(S\cup \{\ell \})=0$. Then, we get for all $i \in N\backslash \{j\}$ the following:

$$\begin{aligned} {\text {Sh}}^{p}_i(N,v) =&\sum _{\begin{array}{c} R\subseteq N,\\ {R\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{m \in R}v(\{m\})}\varDelta _v(R)\\ =&\sum _{\begin{array}{c} S\subseteq N\backslash \{j\},\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{m \in S}v(\{m\})}\varDelta _v(S)\\&+\sum _{\begin{array}{c} S\subseteq N\backslash \{j\},\\ {S\ni i} \end{array}} \frac{ v(\{i\})}{\sum _{m \in S\cup \{j\}}v(\{m\})}\varDelta _v(S\cup \{j\}) \\ =&\sum _{\begin{array}{c} S\subseteq N^j\backslash \{k,\ell \},\\ {S \ni i} \end{array}} \frac{ v^j(\{i\})}{\sum _{m \in S}v^j(\{m\})}\varDelta _{v^j}(S)\\&+\sum _{\begin{array}{c} \begin{array}{c} S\subseteq N^j\backslash \{k,\ell \},\\ {S\ni i} \end{array} \end{array}} \frac{ v^j(\{i\})}{\sum _{m \in S\cup \{k,\ell \}}v^j(\{m\})}\varDelta _{v^j}(S\cup \{k,\ell \})\\ =&\;\sum _{\begin{array}{c} R\subseteq N^j,\\ {R\ni i} \end{array}} \frac{ v^j(\{i\})}{\sum _{m \in R}v^j(\{m\})}\varDelta _{v^j}(R)\\ =&\;{\text {Sh}}^{p}_i(N^j,v^j). \end{aligned}$$

$\square $

Proof of Lemma 2

Let $N=\{1,2,..., n\},\, |N|\ge 2$, $v \in \mathcal {G}_{0}^N$, $\varphi $ a TU-value which satisfies E and PS for all $v \in \mathcal {G}_{0}^N$, and w.l.o.g., player 1 and player 2 be symmetric in v. If we split player 1 according to PS into two new players, player $n+1$ and player $n+2$, $N^1= \{2,3,...,n, n+1, n+2\}$, we have the following:

$$\begin{aligned} \varphi _{2}(N^1,v^1)= \varphi _{2}(N,v), \end{aligned}$$

(9)

and, if we split player 2 according to PS into the same players as before, player $n+1$ and player $n+2$, instead, $N^2= \{1,3, 4,...,n, n+1, n+2\}$, we have the following:

$$\begin{aligned} \varphi _{1}(N^2,v^2)= \varphi _{1}(N,v), \end{aligned}$$

(10)

where we choose $v^2(\{n+1\}):=v^1(\{n+1\})$ and $v^2(\{n+2\}):=v^1(\{n+2\})$.

In the same manner, we split now in the game $(N^1, v^1)$ player 2 into two new players, player $n+3$ and player $n+4$, and analogous in the game $(N^2, v^2)$ player 1 into the same players as before, player $n+3$ and player $n+4$, and choose ${v^{2}}^1(\{n+3\}):={v^{1}}^2(\{n+3\})$ and ${v^{2}}^1(\{n+4\}):={v^{1}}^2(\{n+4\})$. We have ${N^{1}}^2={N^{2}}^1=\{3,4,..., n,n+1,n+2,n+3,n+4\}$ and ${v^{1}}^2={v^{2}}^1$ and get by E, according to Remark 1:

$$\begin{aligned} \varphi _{n+3}\Big ({{N^1}^2,{v^1}^2}\Big )+\varphi _{n+4}\Big ({{N^1}^2,{v^1}^2}\Big )&= \varphi _{2}(N^1,v^1)\underset{(9)}{=}\varphi _{2}(N,v),\\ \varphi _{n+3}\Big ({{N^{2}}^1,{v^2}^1}\Big )+\varphi _{n+4}\Big ({{N^{2}}^1,{v^2}^1}\Big )&= \varphi _{1}(N^2,v^2)\underset{(10)}{=}\varphi _{1}(N,v). \end{aligned}$$

Hence, we have $\varphi _{1}(N,v)= \varphi _{2}(N,v)$ and S is shown. $\square $

Proof of Lemma 3

Let $N \in \mathcal {N},\, |N|\ge 2$, $v \in \mathcal {G}_{0_{\mathbb {Q}}}^N$ a TU-game, and, w.l.o.g., players $i,j \in N$, such that i and j are weakly dependent in v. Furthermore, let $\varphi $ a TU-value which satisfies E and PS for all $v \in \mathcal {G}_{0_{\mathbb {Q}}}^N$, and therefore, by Lemma 2, also S. Due to $v(\{i\}),v(\{j\}) \in \mathbb {Q}\backslash \{0\}$, the worths of the singletons $v(\{k\}),\,k \in \{i,j\}$, can be written as a fraction. We distinguish two cases: (a) $v(\{k\})>0$ and (b) $v(\{k\})<0$.

(a) We have the following:

$$\begin{aligned} v(\{k\})=\;\dfrac{p_k}{q_k} \;\text { with }\; p_k,q_k \in \mathbb {N}. \end{aligned}$$

We choose a main denominator q of these two fractions by $q:=q_iq_j$. With $z_i:= p_iq_j$ and $z_j:= p_jq_i$, we get the following:

$$\begin{aligned} v(\{i\})= \;\frac{z_i}{q}\;\text { and }\;v(\{j\})= \;\frac{z_j}{q}. \end{aligned}$$

(11)

Now, we define a player set $N'$ and a coalition function $v'$ by “splitting” each player $ k \in \{i,j\}$ into $z_k$ players $k_{1}$ to $k_{z_k}$, such that we have $N'=(N\backslash \{i,j\})\cup \{i_m: 1\le m \le z_i\}\cup \{j_m: 1\le m \le z_j\}$. Each player $k_{m}\in N'\backslash (N\backslash \{i,j\}),\,\, 1\le m \le z_k,$ gets a singleton worth $v'(\{k_{m}\})=\frac{1}{q} \text { for } k \in \{i,j\}$, synonymous with

$$\begin{aligned} v'(\{\ell \})= \frac{1}{q} \text { for all } \ell \in N'\backslash (N\backslash \{i,j\}), \end{aligned}$$

where $|N'\backslash (N\backslash \{i,j\})|=z_i +z_j$ and $v(\{k\})= \sum _{1\le m \le z_k}v'(\{k_{m}\}),\,k \in \{i,j\}$. We define $v'(R'):=v(R)$ for all $R'= R\backslash \{i,j\} \cup N'\backslash (N\backslash \{i,j\}),\, R \subseteq N, \{i,j\}\subseteq R,$ and $v'(S):= v(S)$ for all $S\subseteq N'$ with $S\subseteq N$. All other coalitions $T\subseteq N'$ are defined as not active in $v'$.

Applying PS (repeatedly) to $v,\; \varphi $ and the two players $i,j \in N$, we can get the coalition function $v'$ defined just before and, by Remark 1, we have the following:

$$\begin{aligned} \varphi _k(N,v) = \sum _{1\le m \le z_k}\varphi _{k_{m}}(N',v') \text { for } k \in \{i,j\}. \end{aligned}$$

(12)

All players $\ell \in N'\backslash (N\backslash \{i,j\})$ are symmetric in $v'$. Hence, it follows from S that

$$\begin{aligned} \varphi _{\ell }(N',v')=\,\frac{\varphi _i(N,v)+\varphi _j(N,v)}{z_i+z_j}\text { for all } \ell \in N'\backslash (N\backslash \{i,j\}). \end{aligned}$$

We get

$$\begin{aligned} \varphi _k(N,v)\underset{(12)}{=}\,\sum _{1\le m \le z_k}\varphi _{k_{m}}(N',v')=\frac{z_k}{z_i+z_j}\big [\varphi _i(N,v)+\varphi _j(N,v)\big ]\text { for } k \in \{i,j\}. \end{aligned}$$

It follows:

$$\begin{aligned} \varphi _i(N,v) = \frac{z_i}{z_j}\varphi _j(N,v)\underset{(11)}{=} \frac{v(\{i\})}{v(\{j\})}\varphi _j(N,v) \end{aligned}$$

and P is shown.

(b) We have the following:

$$\begin{aligned} v(\{k\})=\;\dfrac{p_k}{q_k} \;\text { with }\; (-p_k),q_k \in \mathbb {N}. \end{aligned}$$

We choose a main denominator q of these two fractions by $q:=-q_iq_j$. With $z_i:= -p_iq_j$ and $z_j:= -p_jq_i$, we get

$$\begin{aligned} v(\{i\})= \;\frac{z_i}{q}\;\text { and }\;v(\{j\})= \;\frac{z_j}{q}. \end{aligned}$$

The remaining part of the proof equals the related part in case (a). $\square $

Logical independence

Finally, we want to show the independence of the axioms used in the characterizations.

Remark 5

Let $v \in \mathcal {G}^N_0,\; N\in \mathcal {N}.$ The axioms in Theorem 1/Corollary 2 are logically independent:

E: The TU-value $\varphi $ defined by
$$\begin{aligned} \varphi _i(v)= v(\{i\})+2\cdot \sum _{\begin{array}{c} S\subseteq N,\\ {S\ni i,\, S\ne \{i\}} \end{array}} \frac{ v(\{i\})}{\sum _{j \in S}v(\{j\})}\varDelta _v(S)\;\text { for all } \,i \in N \end{aligned}$$
satisfies D, P/PS, and WA but not E.
D: The proportional rule$\pi $ (Moriarity 1975), given by
$$\begin{aligned} \pi _i(v) = \frac{ v(\{i\})}{\sum _{j \in N}v(\{j\})}v(N)\;\text { for all } \,i \in N, \end{aligned}$$

(13)
satisfies E, P/PS, and WA but not D.
P/PS: ${\text {Sh}}$ satisfies E, D, and WA but not P/PS.
WA: The TU-value $\varphi $ defined for all $i \in N$ by
$$\begin{aligned} \varphi _i(v)= {\left\{ \begin{array}{ll} v(\{i\}) , \text { if}~ i~ \text {is a dummy player,}\\ \dfrac{v(\{i\})}{\sum _{\begin{array}{c} j \in N,\\ j {\textit{ is no dummy}} \end{array}}v(\{j\})}\Big [v(N)-\sum \nolimits _{\begin{array}{c} j \in N,\\ j {\textit{ is a dummy}} \end{array}}v(\{j\})\Big ], \text { otherwise},\end{array}\right. } \end{aligned}$$
satisfies E, D, and P/PS but not WA.

Remark 6

Let $v \in \mathcal {G}^N_0,\, N\in \mathcal {N}.$ The axioms in Theorem 2/Corollary 1 are logically independent:

E: The TU-value $\varphi $ defined for all $i \in N$ by
$$\begin{aligned} \varphi _i(v)= {\left\{ \begin{array}{ll}0, \text { if } |N|=1, \\ {\text {Sh}}^p_i(v), \text { otherwise,}\end{array}\right. } \end{aligned}$$
satisfies P/PS and WCSE but not E.
P/PS: ${\text {Sh}}$ satisfies E and WCSE but not P/PS.
WCSE: The proportional rule $\pi $ [Eq. (13)] satisfies E and P/PS but not WCSE.

References

Amer, R., Carreras, F., & Magaña, A. (2007). Two main methods for utility sharing in joint business: A discussion. Journal of Mathematics and Statistics, 3(1), 28–35.Google Scholar
Banker, R. D. (1981). Equity considerations in traditional full cost allocation practices: An axiomatic perspective. Pittsburgh: Carnegie-Mellon University.Google Scholar
Barton, T. L. (1988). Intuitive choice of cooperative sharing mechanisms for joint cost savings: Some empirical results. Abacus, 24(2), 162–169.Google Scholar
Béal, S., Ferrières, S., Rémila, E., & Solal, P. (2018). The proportional Shapley value and applications. Games and Economic Behavior, 108, 93–112.Google Scholar
Besner, M. (2016). Lösungskonzepte kooperativer Spiele mit Koalitionsstrukturen, Master’s thesis at the Chair of Discrete Mathematics, FernUniversität in Hagen (in German).Google Scholar
van den Brink, R., Levínský, R., & Zelený, M. (2015). On proper Shapley values for monotone TU-games. International Journal of Game Theory, 44(2), 449–471.Google Scholar
Casajus, A., & Huettner, F. (2008). Marginality is equivalent to coalitional strategic equivalence. Working paper.Google Scholar
Chun, Y. (1989). A new axiomatization of the Shapley value. Games and Economic Behavior, 1(2), 119–130.Google Scholar
Chun, Y. (1991). On symmetric and weighted Shapley values. International Journal of Game Theory, 20(2), 183–190.Google Scholar
Dehez, P., & Tellone, D. (2013). Data games: Sharing public goods with exclusion. Journal of Public Economic Theory, 15(4), 654–673.Google Scholar
Feldman, B. (1999). The proportional value of a cooperative game. Manuscript. Chicago: Scudder Kemper Investments.Google Scholar
Gangolly, J. S. (1981). On joint cost allocation: Independent cost proportional scheme (ICPS) and its properties. Journal of Accounting Research, 299–312.Google Scholar
Harsanyi, J. C. (1959). A bargaining model for cooperative n-person games. In A. W. Tucker & R. D. Luce (Eds.), Contributions to the theory of games IV (pp. 325–355). Princeton, NJ: Princeton University Press.Google Scholar
Hart, S., & Mas-Colell, A. (1989). Potential, value, and consistency. Econometrica, 57(3), 589–614.Google Scholar
Huettner, F. (2015). A proportional value for cooperative games with a coalition structure. Theory and Decision, 78(2), 273–287.Google Scholar
Kalai, E., & Samet, D. (1987). On weighted Shapley values. International Journal of Game Theory, 16(3), 205–222.Google Scholar
Leng, M., & Parlar, M. (2009). Allocation of cost savings in a three-level supply chain with demand information sharing: A cooperative-game approach. Operations Research, 57(1), 200–213.Google Scholar
Moriarity, S. (1975). Another approach to allocating joint costs. The Accounting Review, 50(4), 791–795.Google Scholar
Myerson, R. B. (1980). Conference structures and fair allocation rules. International Journal of Game Theory, 9(3), 169–182.Google Scholar
von Neumann, J., & Morgenstern, O. (1944). Theory of games and economic behavior. Princeton, NJ: Princeton University Press.Google Scholar
Nowak, A. S., & Radzik, T. (1995). On axiomatizations of the weighted Shapley values. Games and Economic Behavior, 8(2), 389–405.Google Scholar
Ortmann, K. M. (2000). The proportional value for positive cooperative games. Mathematical Methods of Operations Research, 51(2), 235–248.Google Scholar
Radzik, T. (2012). A new look at the role of players’ weights in the weighted Shapley value. European Journal of Operational Research, 223(2), 407–416.Google Scholar
Roth, A. E., & Verrecchia, R. E. (1979). The Shapley value as applied to cost allocation: A reinterpretation. Journal of Accounting Research, 295–303.Google Scholar
Shapley, L. S. (1953). Additive and non-additive set functions. Princeton: Princeton University.Google Scholar
Shapley, L. S. (1953). A value for n-person games. In H. W. Kuhn & A. W. Tucker (Eds.), Contributions to the theory of games (Vol. 2, pp. 307–317). Princeton: Princeton University Press.Google Scholar
Shubik, M. (1962). Incentives, decentralized control, the assignment of joint costs and internal pricing. Management Science, 8(3), 325–343.Google Scholar
Spinetto, R. D. (1975). Fairness in cost allocations and cooperative games. Decision Sciences, 6(3), 482–491.Google Scholar
Thomas, A. L. (1969). The allocation problem in financial accounting theory-studies in accounting research no 3. Evanston, IL: American Accounting Association.Google Scholar
Thomas, A. L. (1974). The allocation problem in financial accounting theory (No. 9). Evanston, IL: American Accounting Association.Google Scholar
Tijs, S. H., & Driessen, T. S. (1986). Extensions of solution concepts by means of multiplicative $\varepsilon $-tax games. Mathematical Social Sciences, 12(1), 9–20.Google Scholar
Young, H. P. (1985). Monotonic solutions of Cooperative Games. International Journal of Game Theory, 14(2), 65–72.Google Scholar
Young, H. P. (1985). Cost Allocation: Methods, Principles, Applications. : North Holland Publishing Co.Google Scholar

Copyright information

Authors and Affiliations

Manfred Besner
- 1
Email authorView author's OrcID profile

1.Department of Geomatics, Computer Science and MathematicsHFT Stuttgart, University of Applied SciencesStuttgartGermany

Axiomatizations of the proportional Shapley value

Abstract

Keywords

Notes

Acknowledgements

References

Copyright information

Authors and Affiliations

Personalised recommendations

Cite article