ͽÄêɽ -¾ÜºÙ¾ðÊó-
·ï̾ | WS Robert Veszteg»á |
³«»ÏÆü»þ | 2009ǯ 10·î 16Æü (¶âÍËÆü) 16»þ30ʬ (GMT+09:00) |
½ªÎ»Æü»þ | 2009ǯ 10·î 16Æü (¶âÍËÆü) 18»þ00ʬ (GMT+09:00) |
¾ì½ê | ¸½ÂåÀ¯¼£·ÐºÑ¸¦µæ½ê |
¾ÜºÙ | Êó¹ð¼Ô¡§Robert Veszteg»á ¥¿¥¤¥È¥ë¡§Learning and playing best response in the hide-and-seek game ¡Ú³µÍ×¡Û Robert Veszteg¹Ö»Õ¤Ï¥Þ¥É¥ê¥Ã¥É¡¢¥«¥ë¥í¥¹Âè»°Âç³Ø¤Î´ü¸ÂÉÕ¤¤ÎÀìǤ¹Ö»Õ¤Ç¤¢¤ê¡¢¸½ºß¡¢ÂçºåÂç³Ø¤Î¼Ò²ñ·ÐºÑ¸¦µæ½ê¤Ë¸¦µæ°÷¤È¤·¤ÆÂںߤ·¤Æ¤¤¤ë¡£¥²¡¼¥àÍýÏÀ¤òÍѤ¤¤¿À©ÅÙ¹½ÃÛ¤ÎÍýÏÀ¤È¼Â¸³¤ÎÀìÌç²È¤Ç¤¢¤ê¡¢GLOPE£²¤ÎËÜ¥×¥í¥¸¥§¥¯¥È¤Î¸¦µæ¤È¤â´ØÏ¢¤¬¿¼¤¤¡£ º£²ó¤ÏƱ¹Ö»Õ¤ÎºÇ¿·¤Î¼Â¸³¸¦µæ¤òÊó¹ð¤·¤Æ¤¤¤¿¤À¤¡¢Æ±»þ¤Ë¡¢ÆüËÜÂÚºßÃæ¤Î¶¦Æ±¸¦µæ¤Î²ÄǽÀ¤òõ¤ë¤â¤Î¤Ç¤¢¤ë¡£ We study short and long-run behavior in hide-and-seek games in which treatments vary with symmetric and asymmetric labeling and with framing. We find that long-run behavior is not affectd by the treatment variables. In particular, we do not find that subjects fall prey to the usual focal point bias. We also study to what extent level-k reasoning, Nash equilibrium, quantal response equilibrium (QRE), fictitious play, reinforcement learning, and experience-weighted attraction (EWA) learning organize the data. All analyzed versions of these models predict the same outcome for this game. For short-run behavior, our data can not be well organized by level-k reasoning. It can be considered as the results of random play. For long-run behavior and behavior over time, the learning models outperform the static equilibrium solutions for hiders and seekers both jointly and separately. Our results suggest that experimental subjects start the game using the available actions randomly and they adjust their strategies to the opponent¡Çs observed behavior. |
¥«¥Æ¥´¥ê¡¼ | ±¡À¸¸þ¤±¹ÖµÁ¡¦Ï¢Â³¹ÖµÁ |