I have read that Shapley values satisfy the balanced contribution axiom (e.g., p. 6).
However, I have not been able to show that this is indeed the case. I tried to prove it using the definition of the Shapley value: $$ \phi_{i}(v)-\phi_{i}(v \backslash j)=\phi_{j}(v)-\phi_{j}(v \backslash i) \\ \sum_{\mathcal{C} \subseteq N \backslash\{i\}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !}(v(\mathcal{C} \cup\{i\})-v(\mathcal{C})) - \sum_{\mathcal{C} \subseteq N \backslash\{i, j\}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{(n -1)!}(v(\mathcal{C} \cup\{i\})-v(\mathcal{C})) = \sum_{\mathcal{C} \subseteq N \backslash\{j\}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !}(v(\mathcal{C} \cup\{j\})-v(\mathcal{C})) - \sum_{\mathcal{C} \subseteq N \backslash\{i, j\}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{(n-1) !}(v(\mathcal{C} \cup\{j\})-v(\mathcal{C})) $$
But the only thing that can be reduced seems to be the $v(C)$ in the sum over $N \setminus \{i, j\}$. Is this the wrong approach?
LHS expands to
$$\sum_{\mathcal{C} \subseteq N \backslash\{i\}\ \color{blue}{ \wedge\ j \in \mathcal{C}}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !}v(\mathcal{C} \cup\{i\}) - \sum_{\mathcal{C} \subseteq N \backslash\{i\}\ \color{blue}{ \wedge\ j \in \mathcal{C}}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !}v(\mathcal{C}) - \sum_{\mathcal{C} \subseteq N \backslash\{i, j\}} \frac{|\mathcal{C}+1| !(n-(|\mathcal{C}|+1)-1) !}{n !} v(\mathcal{C} \cup\{i\}) + \sum_{\mathcal{C} \subseteq N \backslash\{i, j\}} \frac{|\mathcal{C}+1| !(n-(|\mathcal{C}|+1)-1) !}{n !} v(\mathcal{C})$$
The idea is to divide subsets of $N \setminus \{i\}$ into two: subsets without $i$ but must have $j$ (the colored text) and subsets without both $i$ and $j$. The first subsets are unaffected by $\phi_i(v \setminus j)$, hence the first and second summations are expansions of $\phi_i(v)$ by simply separating $v(\mathcal{C} \cup \{i\})$ and $v(\mathcal{C})$. Now for the second subsets, perform
$$\frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !} - \frac{|\mathcal{C}| !(n-|\mathcal{C}|-2)!}{(n-1) !}$$
The third and fourth summations are also from separating $v(\mathcal{C} \cup \{i\})$ and $v(\mathcal{C})$.
Similarly, RHS expands to
$$\sum_{\mathcal{C} \subseteq N \backslash\{j\}\ \color{blue}{ \wedge\ i \in \mathcal{C}}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !}v(\mathcal{C} \cup\{j\}) - \sum_{\mathcal{C} \subseteq N \backslash\{j\}\ \color{blue}{ \wedge\ i \in \mathcal{C}}} \frac{|\mathcal{C}| !(n-|\mathcal{C}|-1) !}{n !}v(\mathcal{C}) - \sum_{\mathcal{C} \subseteq N \backslash\{i, j\}} \frac{|\mathcal{C}+1| !(n-(|\mathcal{C}|+1)-1) !}{n !} v(\mathcal{C} \cup\{j\}) + \sum_{\mathcal{C} \subseteq N \backslash\{i, j\}} \frac{|\mathcal{C}+1| !(n-(|\mathcal{C}|+1)-1) !}{n !} v(\mathcal{C})$$
For the first summation of LHS and RHS: Now let $\mathcal{C}_1$ satisfy the condition of $\mathcal{C}$ in LHS. Let $\mathcal{C}_2 = \mathcal{C}_1 \cup \{i\} \setminus \{j\}$. Observe that $\mathcal{C}_2$ satisfies the condition of $\mathcal{C}$ in RHS, $|\mathcal{C}_1| = |\mathcal{C}_2|$, and $\mathcal{C}_1 \cup \{i\} = \mathcal{C}_2 \cup \{j\}$. This implies a one-to-one correspondence between the $\mathcal{C}$'s in the first summation of LHS and in RHS, and the addends of the correspondence are equal.
For the second and third summation of LHS and RHS: observe that the second summation of LHS is identical to the third summation of RHS, and vice versa!
For the fourth summation of LHS and RHS: identical.
Note: in the summation over $N \setminus \{i,j\}$, $(n-|\mathcal{C}|-1)!$ should be $((n-1)-|\mathcal{C}|-1)! = (n-|\mathcal{C}|-2)!$.