資源描述：
《!RL for Optimized Trade Execution》由會(huì)員上傳分享，免費(fèi)在線(xiàn)閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、ReinforcementLearningforOptimizedTradeExecutionYuriyNevmyvakayuriy.nevmyvaka@lehman.comLehmanBrothers,745SeventhAv.,NewYork,NY10019,USAYiFengfengyi@cis.upenn.edu1MichaelKearnsmkearns@cis.upenn.eduUniversityofPennsylvania,Philadelphia,PA19104,USAAbstract
2、everystrategy'sprofitabilitywilldependonhowwellitisimplemented.Wepresentthefirstlarge-scaleempiricalapplicationofreinforcementlearningtotheFormostofthehistoryoftheU.S.equitymarkets,theimportantproblemofoptimizedtradeexecutiononlyinformationavailabletoan
3、agentattemptingtoinmodernfinancialmarkets.Ourexperimentsoptimizetradeexecutionwasthesequenceofpricesofarebasedon1.5yearsofmillisecondtime-scalealready-executedtrades,andthecurrentbidandask(thelimitorderdatafromNASDAQ,andbestoutstandingbuyandsellpricesof
4、fered).Butinrecentdemonstratethepromiseofreinforcementyears,electronicmarketssuchasNASDAQhavebegunlearningmethodstomarketmicrostructurereleasing,inrealtime,alloftheoutstandingbuyandsellproblems.Ourlearningalgorithmintroducesandlimitorderpricesandvolumes
5、(oftenreferredtoasorderexploitsanatural"low-impact"factorizationofbookormarketmicrostructuredata).Itisnotdifficulttothestatespace.seethatsuchdatamightbeextremelyvaluablefortheproblemofoptimizedexecution,sincethedistributionofoutstandinglimitordersmayhel
6、ppredictshort-termprice1.Introductiondirection,likelihoodofexecutionatagivenprice,buyorsellsideimbalances,andsoon.Inthedomainofquantitativefinance,therearemanycommonproblemsinwhichtherearepreciselyspecifiedInthispaper,wereportonthefirstextensiveempirica
7、lobjectivesandconstraints.Animportantexampleistheapplicationofreinforcementlearning(RL)totheproblemproblemofoptimizedtradeexecution(whichhasalsoofoptimizedexecutionusinglarge-scaleNASDAQbeenexaminedinthetheoreticalcomputersciencemarketmicrostructuredata
8、sets.Ourbasicexperimentalliteratureastheone-waytradingproblem).Previousworkmethodology(detailedinSection3)consistsoftheonthisproblemincludes(AmgenandChriss,2002),followingsteps:(BertsimasandLo,1998),(Cogginsetal.,2003),(El-1.Thei
