1: \begin{proof}

2: 		Fix $\theta \in \mathcal{P}(E)$.

3: 		Define the sets \[Q_0 := \{\hat{\mu} \in \mathcal{P}(E) : d_W(\hat{\mu},\theta) \leq r\}\]

4: 		and for $i = 1,...,n-1$ and $x_1,...,x_i \in E$

5: 		\[

6: 		Q_i(x_1,...,x_i) := \{\hat{\mu} \in \mathcal{P}(E) : d_W(\hat{\mu},\pi(x_i)) \leq r\}.

7: 		\]

8: 		We note that $M_n(\theta) = Q_0 \otimes Q_1 \otimes ... \otimes Q_{n-1}$, where $Q_0 \otimes Q_1 \otimes ... \otimes Q_{n-1}$ is defined as the set of measures $\mu = K_0 \otimes K_1 \otimes K_2 \otimes ... \otimes K_{n-1} \in \mathcal{P}(E^n)$, where $\mu_0 \in K_0$ and $K_i : E^{i} \rightarrow \mathcal{P}(E)$ are Borel measurable kernels such that $K_i(x_1,...,x_i) \in Q_i(x_1,...,x_i)$ for $\mu$-almost all $x_1,...,x_i$. 

9: 		Since for all $i = 1,...,n$, the set $\{(x_1,...,x_i,\hat{\mu}) \in E^i \times \mathcal{P}(E) : \hat{\mu} \in Q_i(x_1,...,x_i)\}$ is trivially Borel, a measurable selection argument (e.g.~\cite[Prop.~7.50]{bertsekas2004stochastic}) yields for $\nu \in \mathcal{P}(E^n)$

10: 		\begin{align*}

11: 		\inf_{\hat{\mu} \in M_n(\theta)} R(\nu,\hat{\mu}) &= \inf_{K_0\otimes ... \otimes K_{n-1} \in Q_0 \otimes ... \otimes Q_{n-1}} \sum_{i=0}^{n-1} \int_{E^n} R(\nu_{i,i+1}(x_1,...,x_i),K_i(x_1,...,x_i)) \nu(dx_1,...,dx_n) \\

12: 		&\stackrel{(*)}{=} \sum_{i=0}^{n-1} \int_{E^n} \inf_{\hat{\mu} \in Q_i(x_1,...,x_i)} R(\nu_{i,i+1}(x_1,...,x_i),\hat{\mu}) \nu(dx_1,...,dx_n)\\

13: 		&= \beta(\nu_{0,1},\theta) + \sum_{i=1}^{n-1} \int_{E^n} \beta(\nu_{i,i+1}(x_1,...,x_i),\pi(x_i)) \nu(dx_1,...,dx_n)\\

14: 		&= \beta_n^{\theta}(\nu),

15: 		\end{align*}

16: 		where rigorously step $(*)$ works inductively, see the proofs of \cite[Lemma 4.4]{bartl2016exponential} and \cite[Prop.~5.2]{lacker2016non}.

17: 	\end{proof}

18: