UMIE2003

Because the contents of the experiment were defined more clearly, and agent creation conditions and agents that had participated in UMIE 2002 were analyzed in details, fewer but well-built agents participated in the UMIE 2003. An agent with "on-line leaning ability" focused on the "short-term trend" won first prize, and that reflected the past experiment’s results. At UMIE 2002, scores of agents largely depended on the medium-term trend such as up/down. However, the achievements of the agents participated in UMIE 2003 depended on if they were able to cope with the changes in trend or not.

Date
  June 24, 2003
Place
  Held during NAACSOS 2003 Conference
session as one of related events.
Participants
  7 teams with 18 agents
◆Ratio of engineering vs. economics = 4:7
Agent type
  Machine

【Machine】

Team University Agent name Description
Osaka University of Economics and Law Zaiteku Kenkyukai Osaka University of Economics and Law CrossStrategy NaminoriStrategy DsStrategy ・The following agents participated: an agent that used the rate of deviation from the moving-average of futures price to adjust selling/buying or order quantity, arbitraging-type agent that used spot-futures spread participated, and an agent that ordered when the futures price and spot price crossed.
Deglab Team Tokyo Institute of Technology ClassifierAgent ・Agent that invested by choosing and employing an optimum strategy among four (strategies using arbitrage trade, moving-average, futures trend or spot trend) participated. This agent was created based on the agent using the short-term price changes and with on-line learning ability that had participated and marked high scores in UMIE 2002.
Osaka-city-uni-hk Osaka City University Averaging arbitrage trade
Buy escalation arbitrage trade
Averaging buyescalation
Averaging buy escalation 2
・ Agent that employed the arbitrage trade and dollar cost averaging method
Total five agents with different parameters and initial positions participated
Chuo University Chuo University CK_R10・CK_R20
CK_R30・CK_R40
CK_R50
・ Arbitrage type agent. This agent decided that the index at around the moving-average of spot price, and decided order quantity using spot-futures spread. Five agents with different moving-average periods entered.
OCU_Nakajima Osaka City University PriceMaker
Transaction
・Two agents, one that did speculative dealing, and the other that ordered a bunch of buying/selling at the same time price entered.
TN Kyoto University SimpleProgram ・Arbitrage type agent participated
syn-1only Ritsumeikan University syn ・ Agent that compared spot prices of the last three times with futures prices and ordered when the spread between those were widening

●Comprehensive Pareto-ranking

Rank Agent name MemberID Team University
No. 1 (No. 1 in Ex1, Ex2 and Ex3) ClassifierAgent m4 Degulab Tokyo Institute of Technology
No. 1 (No. 1 in Ex1, Ex2 and Ex3) Strategy4 m7 Osaka-city-uni-hk Osaka City University
No. 3(No. 2 in Ex2 and No. 1 in Ex3, Ex1) Strategy1 m5 Osaka-city-uni-hk Osaka City university
No. 3(No. 2 in Ex2 and No. 1 in Ex3, Ex1) Strategy6 m9 OCU_Nakajima Osaka City University
No. 3(No. 1 inEx1, Ex2and No. 2 in Ex3) Transaction m16 OCU_Nakajima Osaka City university

Correlations in ranking between experiences and time series
(comparison method was the same as we had done with UMIE 2002)

correlate 5% levels of significiance

correlate 1% levels of significiance

Correlation between experiments ((Influence of internal conditions)
How would each agent’s rank alter when its competitor changed?
強いエージェントは、誰と対戦しても強い!

EX2
EX3
EX1
0.31
0.57
EX2
0.33

 

 

Correlation among experiments (Influence of Trends and their changes.)
How would each agent’s rank alter when its competitor changed?

Descent
Oscilation
Reversal
Ascent
-0.80
-0.74
0.24
Descent
0.68
−0.20
Oscilation
−0.32

Differences from UMIE 2002
1)There was no correlation between experiments.
 Results altered largely depending on the competitors.

UMIE2002
EX2
EX3
EX1
0.66
0.69
EX2
0.84
UMIE2003
EX2
EX3
EX1
0.31
0.57
EX2
0.33

 

2)Strong/weak time series per agent changed.
 Last year’s main focus point was down/up, but this year, it was the occurrence of trend changes that influenced an agents’ achievements dramatically.

UMIE2002
Descent
Oscilation
Reversal
Ascent
-0.24
-0.10
0.37
Descent
0.56
0.24
Oscilation
 
0.11
UMIE2003
Descent
Oscilation
Reversal
Ascent
-0.80
-0.74
0.24
Descent
0.68
-0.20
Oscilation
 
-0.32