Experience Weighted Attraction in the First Price Auction and Becker Degroot Marschak

Clocks and Trees: Isomorphic Dutch Auctions and Centipede Games

“…For instance, what if EWA could be augmented with (a) construction of a mapping between institutional features (among them institutional format) and the 5 parameter (governing visualization of payoffs to strategies other than those previously chosen by the agent), and (b) construction of a mapping between institutional features and initial attractions (the initial conditions for the raw material for the probability mass accorded a particular strategy)? We already know, from the simulation results of James and Reagle (2009) for EWA agents in the first-price sealed-bid auction, that S «s 0 and (p ъ 1 yields persistent overbidding in the first-price sealed-bid auction (over the 500-round horizon used), while with 5 = 1 and <p = 1, EWA agents more rapidly converge to expected value maximizing behavior.10 Given that 5 = 0 corresponds to myopia about payoffs to alternative strategies, while 5 = 1 corresponds to being fully informed about them, it is tempting to try to map the tree format to a high value of 5 and map the clock format to a low value of 5; certainly the clear differences in informational content between the two institutional formats suggest that 5 should differ across institutional formats and do so for reasons other than convenience in explaining the particular results in this paper. A similar exercise to seed the initial attractions (to various strategies) might be justified along the same lines and would close the EWA model in this application.…”

Recoverability of parameters from learning models

“…(Camerer and Ho (1999) give an exhaustive description of EWA. James and Reagle (2009) give details on how EWA was applied to the FPSB and BDM 1 . )…”

“…We do this for both the FPSB, and BDM. James and Reagle (2009) was to see whether both violations of the Vickrey (1961) model's predictions for bidding behavior in the FPSB (as observed with human subjects by Cox, Roberson, and Smith (1982), and typical (human) behavior in BDM could be rationalized by a single set of EWA parameters. They found that δ≈0 and φ≈1 (where δ governs apprehension of opportunity cost (of strategies not employed in prior play), and φ governs memory) were able to reproduce the stylized facts on human bidding in these two institutions (overbidding in each).…”

“…They were thus able to use an approach analogous to Gode and Sunder (1993) to explain the cross-institutional patterns of bidding observed in Isaac and James (2000), wherein subjects who behaved as if risk averse in FPSB then behaved as if risk-seeking in BDM. The explanation is that overbidding in each of FPSB and BDM, mapping to risk aversion and riskseeking interpretations respectively, could be an explanation common across the two institutions, and that such overbidding might in turn be due to the common parameterization of EWA uncovered by James and Reagle (2009). 2 We opt for the K-S test instead of the Likelihood Ratio (LR) test due to the additional data available with the Monte Carlo simulations.…”

“…true or population) parameters of the learning model by which those choices were made. This data set -from James and Reagle (2009) -consists of bid data from the first price sealed bid auction (FPSB) and the Becker-DeGroot-Marschak procedure (BDM), where simulated Experience Weighted Attraction (EWA) agents produced the bids.…”