A dynamic multi-item two-activity problem

DOI10.1007/BF01720520MaRDI QIDQ684385zbMATH OpenFDO

Publication date 15 September 1993

Published in OR Spektrum (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1007/BF01720520

expected total discounted rewards harvesting sales of assets sequential asset selling

Stopping times; optimal stopping problems; gambling theory (60G40) Stochastic programming (90C15) Dynamic programming (90C39) Markov and semi-Markov decision processes (90C40) Auctions, bargaining, bidding and selling, and other market models (91B26)

Review text

A special class of dynamic programming problems is considered. There are two activities labelled \(\{1,2\}\) exactly one unit of one of these is to be performed in each period. \(i\), \(j\) denotes the total number of times they have to be performed respectively. They yield random rewards \((y_ n,z_ n)\) [possibly negative] which are known before the decision made at time \(n\) and whose joint probability distribution is i.i.d. and have finite means. There may be fixed costs \(c\) (for both activities), \(c_ i\) (for activity \(i\)) per period. The problem involves maximization of the expected total discounted rewards and the paper seeks to provide simple decision rules. Denote by \(g_{ij}\) the maximal value and define \(d_{ij}\equiv\beta(g_{ij-1}-g_{i-1,j})\), where \(\beta\) is the discount factor. \(d_{ij}\) is a measure of preference for activity 1 over 2 when one additional activity is added to the pair \((i-1,j-1)\). Let \(y\), \(z\) stand for the random rewards and \(u\equiv y-z\). An optimal decision rule depends on \(y\), \(z\) only via \(u\). Writing this as \(f(i,j,u)\) it satisfies \(f(\cdot)= 1\) if \(u\geq d_{ij}\) and \(f(\cdot)=2\) if \(u< d_{ij}\) [Lemma 1]. Theorem 1 provides recursive formulae based on parameter values (including the distribution function) for calculating \(g_{ij}\), \(d_{ij}\). Various monotonicity properties are established for the decision rule in special cases. In particular, Theorem 5 shows that \(f(\cdot)\) is decreasing in \(u\), increasing in \(j\) if \(c\leq c_ 1\) and decreasing in \(i\) if \(c\leq c_ 2\). In special cases \(d_{ij}\) have limiting values (with respect to variations, in \(i\), \(j\)) and there may exist a number \(i^*(u)\) [similarly for \(j\)] s.t. for \(i\geq i^*(u)\), \(f(\cdot)=1\) regardless of \(j\) [Theorem 6]. Results regarding bounds for the different limiting values under various scenarios regarding the discount factor is presented. Relation to other papers in the literature and economic examples of sales of assets, harvesting etc. are discussed.

Recommendations

Cites work

This page was built for publication: A dynamic multi-item two-activity problem

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q684385)