Efficiency estimation and duality theory for nonconvex technologies (Q2366717)

The theory of separating hyperplanes, convex sets and dual cones are generalized using the notions of separation by quadrant and \(P\)-convex sets. This is then used in a class of problems in the theory of production where standard quasi-convexity properties of the value function may not obtain but a generalized form of duality results can be derived. This allows extension of non-parametric methods of estimation of the value function to such cases and therefore estimation of degrees of inefficiencies in resource allocation. A brief description follows. Let \(X_ 1,\dots,X_ N\) denote finite-dimensional Euclidean spaces. A set \(F\subset\prod^ n_{i=1} X_ i\) is projectively-convex (\(P\)- convex) if, for each \(x\), \(y\) in \(F\), \(i\in\{1,\dots,N\}\) and \(\lambda\in [0,1]\), there exists \(\lambda_ 1,\lambda_ 2,\dots,\allowbreak\lambda_{i-1}\), \(\lambda_{i+1},\dots,\lambda_ N\in [0,1]\) such that \(((1-\lambda_ 1)x_ 1+\lambda_ 1y_ 1,\dots,(1-\lambda)x_ i+\lambda y_ i,\dots,(1-\lambda_ N)x_ N+\lambda_ N y_ N)\in F\). Let \(f: A\to R\). The upper (lower) level set of \(f\) is a set \(\{a\in A/f(a)\geq (\leq) k\}\) where \(k\) is a constant. A function \(f: R^ n\times R^ m\to R\) is \(P\)-convex (\(P\)-concave) if each lower (upper) level set is \(P\)-convex. The quadrant \(Q_{[\alpha,\beta]}(a,b)=\{(x,y): a\cdot x+\alpha\leq 0\), \(b\cdot y+\beta\leq 0\}\) where \(\alpha\), \(\beta\) are numbers and \(a\in R^ n\), \(b\in R^ m\). Let \(F\) be a nonempty subset of \(R^ n\times R^ m\). If for each point \((x,y)\in F\) there exists a quadrant \(Q_{[\alpha,\beta]}(a,b)\) with \((a,b)\neq 0\) and for which \((x,y)\in Q_{[\alpha,\beta]}\) and \(Q_{[\alpha,\beta]}(a,b)\cap F=\emptyset\) then `\(Q_{[\alpha,\beta]}(a,b)\) separates \(F\) from \((x,y)\)'. A development parallel to that of convex sets, separating hyperplanes and dual cones then follows. Consider the value function \(G(p,b)=\text{Max}\{\phi(x)/p_ i x_ i\leq b_ i;\;i=1,\dots,N\}\) where \(p=(p_ 1,\dots,\allowbreak p_ N)\), \(x=(x_ 1,\dots,x_ N)\) each \(p_ i\), \(x_ i\) are finite-dimensional vectors and \(b_ i\) are scalars. \(\phi(x)\) is the production function. \(x_ i\)'s are the inputs purchased at prices \(p_ i\) by the \(i\)th sub unit and \(b_ 1\) is the budget allocated to the sub unit. It is shown that (1) \(G\) is \(P\)-convex when it is differentiable (Theorem 4) and (2) if \(\phi\) is \(P\)-concave then \(G\) is \(P\)-convex. Dually, if \(G\) is continuous and \(P\)-concave, then \(\phi\) is \(P\)-concave.