On the dual of the solvency cone (Q2345609)

The authors consider a problem of rearranging a portfolio of assets in the presence of transaction costs. They consider \(d\) classes of assets, assuming that a unit of asset \(j\) has price \(\pi_{ij}\) in terms of units of asset \(i\). It is also assumed that some of these assets should be sold and some should be bought. The problem is how to arrange these transactions optimally (i.e., with minimal possible losses), if they should be self-financing. The authors show that all feasible transactions can be represented by a polyhedral convex cone \(K_d\), and that the optimal transactions are on the boundary of the cone \(K_d\). The optimal transactions can be represented by extreme directions of the dual cone \(K_d^+\). In turn, these can be represented by spanning trees of bipartite graphs. The authors consider bipartite graphs \(G(P,N)\), where \(P\) are the indexes of the assets to be sold and \(N\) are the indexes of the assets that should be bought (all edges in the graph are from \(P\) to \(N\)). They prove that the only extreme directions in \(K_d^+\) are vectors \(y\in\mathbb R^d\) that fulfill \(\pi_{ij}y_i\geq y_j\) (for \(i\in P\), \(j\in N\)) with equality if \(ij\) is an edge in a spanning tree of \(G(P,N)\). Such vectors \(y\) are defined as \textit{feasible tree solutions}. In the main result the authors provide the representation in terms of \(P\)-configurations, i.e., the vectors of natural numbers, where subsequent numbers are degrees of the vertices in \(P\). The main theorem states that for every such \(P\)-configuration there exists a feasible tree solution. The proof is by induction and it is constructive. Thus, it provides a recursive algorithm to calculate such a solution. At the end, the authors consider the special case in which relative prices are defined as \(\pi_{ij}=a_j/b_i\), what simplifies the problem.