資源描述:
《practical dynamic programming》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在工程資料-天天文庫。
1、Chapter4PracticalDynamicProgramming4.1.ThecurseofdimensionalityWeoftenencounterproblemswhereitisimpossibletoattainclosedformsforiteratingontheBellmanequation.Thenwehavetoadoptnumericalapproxi-mations.Thischapterdescribestwopopularmethodsforobtainingnumeri
2、calapproximations.The?rstmethodreplacestheoriginalproblemwithanotherproblemthatforcesthestatevectortoliveona?niteanddiscretegridofpoints,thenappliesdiscrete-statedynamicprogrammingtothisproblem.The“curseofdimensionality”impelsustokeepthenumberofpointsinth
3、ediscretestatespacesmall.Thesecondapproachusespolynomialstoapproximatethevaluefunction.Judd(1998)isacomprehensivereferenceaboutnumericalanalysisofdynamiceconomicmodelsandcontainsmanyinsightsaboutwaystocomputedynamicmodels.4.2.Discrete-statedynamicprogramm
4、ingWeintroducethemethodofdiscretizationofthestatespaceinthecontextofaparticulardiscrete-stateversionofanoptimalsavingsproblem.Anin?nitelylivedhouseholdlikestoconsumeonegoodthatitcanacquirebyspendinglaborincomeoraccumulatedsavings.Thehouseholdhasanendowmen
5、toflaborattimet,st,thatevolvesaccordingtoanm-stateMarkovchainwithtransitionmatrixPandstatespace[ˉs1,sˉ2,...,sˉm].Iftherealizationoftheprocessattisˉsi,thenattimetthehouseholdreceiveslaborincomeofamountwsˉi.Thewagewis?xedovertime.Weshallsometimesassumethatm
6、is2,andthatsttakesonvalue0inanunemployedstateand1inanemployedstate.Inthiscase,whastheinterpretationofbeingthewageofemployedworkers.Thehouseholdcanchoosetoholdasingleassetindiscreteamountsat∈AwhereAisagrid[a17、PracticalDynamicProgrammingendpointsofthegridAisimportant,aswedescribeindetailinchapter18onincompletemarketmodels.Theassetbearsagrossrateofreturnrthatis?xedovertime.Thehousehold’smaximumproblem,forgivenvaluesof(w,r)andgiveninitialvalues(a,s),istochooseapo
8、licyfor{a}∞tomaximize00t+1t=0∞Eβtu(c),(4.2.1)tt=0subjecttoct+at+1=(r+1)at+wstct≥0(4.2.2)at+1∈Awhereβ∈(0,1)isadiscountfactorandris?xedrateofreturnontheassets.Weassumethatβ(1+r)<1.Hereu(c)isastrictlyincreasing,concav