UMIE2004

New faces were appeared from Osaka Prefecture University, Kinki University, Ritumeikan University. In this experiments, 4 agents won the first prize. Two of them were past winners, that is, FuzzyB (The winner of UMIE2002) and ClassifierAgent (the winner of UMIE2003), and two of them were new face, TriDiceP and NN2. TriDiceP uses reinforcement learning, and NN2 uses neural-network. During these 3 years, almost all famous method from artificial intelligence were appeared. Winnes tend to fixed and some kinds of break through should be needed.

･Date
   May 29,2004
･Place
   Session in AESCS04' at Kyoto University
･Participants
   12 teams with 36 agents
･Agent type
  Machine agent

【Machine】

Team University Agent name
OCUNakajima Osaka City University Transaction
Kinki University A Team Kinki University KinkiAsahina
KinkiIkeda
KinkiNakaji
Kinki University B Team Kinki University KinkiMasa01
KinkiMasa02
KinkiMasa03
KinkiMasa04
Kinki University C Team Kinki University KinkiNg001
KinkiNg002
KinkiNg003
KinkiNg004

OsakaCityUnivercityRoom419

 Osaka City University

BreakOut
LastSpreadHunter
MovingAgerageIntersect
M.Kojima Ritsumeikan University TriDice2
TriDiceP
TriDiceR
Zcrossover
TCIT Tokyo Insutitute of Technology RandomLossCutStrategy
MovinAverageStrategy
OPUshu Osaka Prefecture University OPUFuzzyStrategyA
OPUFuzzyStrategyB
OPUPositionControlStrategy
OPUSteadyStrategy
OPUallProbabilityStrategy
negative trader Osaka City University activeRSI
Osaka University of Economics and law Osaka University of Economics and Law KInvestor-20
KInvestor-25
Kinvestor-8
team tar Tokyo Institute of Technology UMIE2003Winner
ClassifireAgent2

●Pareto-ranking

Rank Agent name Member ID Team University
No.1（No. 1 in Ex1, Ex2 and Ex3） TriDiceP m17 Ｍ.kojima Ritsumeikan University
No.1（No. 1 in Ex1, Ex2 and Ex3） NN2 m25 kamlab Ritsumeikan University
No.1（No. 1 in Ex1, Ex2 and Ex3） OPUFuzzyStrategyB m27 OPUshu Osaka Prefecture University
No.1（No. 1 in Ex1, Ex2 and Ex3） KInvestor-25 m33 Osaka University of　Economics and law Osaka University of Economics and Law
No.1（No. 1 in Ex1, Ex2 and Ex3） ClassifireAgent2 m36 team tar Tokyo Institute of Technology
No.6 (No. 1 in Ex1and Ex2,No.2 in Ex3) KinkiNg001 m9 Kinki University C Team Kinki University

Correlations in ranking between experiences and time series
(comparison method was the same as we had done with UMIE 2003)

correlate 5% levels of significiance

 correlate 1% levels of significiance

Correlation among Ex1 ,Ex2 and Ex3(Influence of Set of Agents.）
How would each agent’s rank alter when its competitor changed?
EX2
EX3
EX1
0.77
0.83
EX2
0.92

 

  

Correlation among Time Series.（Influence of Trends and their changes.）
How would each agent’s rank alter when its competitor changed?
Descent
Oscilation
Reversal
Ascent
-0.29
-0.35
0.27
Descent
0.67
0.22
Oscilation
0.35

Differences from UMIE 2003
1)Compare with UMIE2002, UMIE2003, correlation among Exes appeared again.

UMIE2003
EX2
EX3
EX1
0.31
0.57
EX2
0.33
UMIE2004
EX2
EX3
EX1
0.77
0.83
EX2
0.92

2)Almost all correlation is distinct.

UMIE2003
Descent
Oscilation
Reversal
Ascent
-0.81
-0.74
0.24
Descent
0.68
-0.20
Oscilation
-0.32
UMIE2004
Descent
Oscilation
Reversal
Ascent
-0.29
-0.35
0.27
Descent
0.67
0.22
Oscilation
0.35

◆The result of rank correlation is same as UMIE2002.
Rank correlation amoung Ex1, Ex2 and Ex3 all relation is strongly correlated. So we can say, "Strong agent is strong whenever the oposits are". Rank correlation among variation of spot prices is week without the relation between "Discent" and "vibration".