Applications (see Crawford+ 2010 review) • Matrix steps – Hedden, Zhang Cog 02, TICS 03 • 2D matrix beauty contest – Chen, Huang, Wang (NatlTaiwanU) 09 • Hide-and-seek – Crawford, Iriberri AER 06; Alec Smith, Forsell+ in prep • Coordination games & cheaptalk (theory) – Ellingsen, Ostling AER 12? • Auctions – Crawford, Iriberri Ecma 07; Gneezy MS 05; Ivanov Ecma 09; Nunnari+ in prep • Private-information betting games – Brocas+ REStud in press • Global games (theory) – Kneeland (UBritishColumbia) 09? • Heterogeneous CH/QRE splice – Rogers, Palfrey, Camerer JEcTheory 09 RSF 4.July.2014 9 RSF 4.July.2014 10 2. Lowest unique positive integer game (LUPI) • • • • • Swedish lottery n=53,000 players Choose k from 1 to 99,999 Lowest unique number wins 10,000€ If n is Poisson distributed… – mixed equilibrium solves enp(k+1) = enp(k) – np(k) Östling, Wang, Chou, Camerer AEJ: Micro 12 RSF 4.July.2014 11 Poisson equilibrium is a surprisingly good approximation (week 1)… RSF 4.July.2014 12 …but Q-cognitive hierarchy fits deviations CH τ=1.80 RSF 4.July.2014 13 Lab replicates direction of field deviations RSF 4.July.2014 14 RSF 4.July.2014 15 RSF 4.July.2014 16 Imitation learning produces convergence over time (week 7) 60 40 20 0 2000 4000 6000 Number guessed RSF 4.July.2014 8000 0 10000 17 18 4. Field application: Cold opened (unreviewed) movies (Brown, Camerer, Lovallo AEJ 12, Mgt Sci 13) • Film distributors choose: – Show movie to critics before opening – Withhold (7% 2000-06; 25% 20072009) RSF 4.July.2014 35 Logical “unravelling argument” predicts no cold openings (Milgrom Bell J 81, Grossman JLE 81 ) • Suppose quality is U ~ [0,100] Movies below q* are opened cold --> E(q|cold)=q*/2 Movies with q [q*/2,q*] are misjudged judged as bad, they are “not so bad” …. open all movies except very worst • CH: Naïve moviegoers overestimate q, box office gross is higher than predicted RSF 4.July.2014 36 RSF 4.July.2014 37 Low quality movies are cold opened 400 350 300 Cold Openings 250 200 150 100 50 0 0-20 20-40 40-60 Cold Openings 60-80 80-100 Regular Openings Average Rating of 30 critics RSF 4.July.2014 38 Low quality movies are cold opened 400 Number of movies 350 Sophisticated moviegoers know 250 cold openings have this quality 200 300 150 100 50 0 0-20 20-40 40-60 Cold Openings 60-80 80-100 Regular Openings Average Rating of 30 critics RSF 4.July.2014 39 Low quality movies are cold opened Naive moviegoers think cold openings have this quality 400 Number of movies 350 Sophisticated moviegoers know 250 cold openings have this quality 200 300 150 100 50 0 0-20 20-40 40-60 Cold Openings 60-80 80-100 Regular Openings Average Rating of 30 critics RSF 4.July.2014 40 Why open cold? • “If you screen [a bad movie] for critics all they can do is say something which may prevent someone from going to the movie.” • Greg Basser, CEO Village Roadshow Entertainment Group • “…if negative reviews are expected, the studio may decide not to screen a picture hoping to delay bad news.” – Mark Litwak, Reel Power RSF 4.July.2014 41 OLS estimation strategy • Bm = αE(qm) + ΣkβkXmk + εm Box office = f(expected quality,other controls) • What is E(qm)? –Reviewed movies: E(qm)=qm –Cold opened movies: E(qm)> qm (from CH) •Bm = αCCOLD + αRqm+ ΣkβkXmk + εm •Coefficient αC > 0 indicates CH naivete •Coefficient αC = 0 indicates sophistication RSF 4.July.2014 42 Residuals: Cold vs. Reviewed Movies (based on log of 1st wkd BO) 2 1.5 1 Residual 0.5 0 -0.5 -1 -1.5 -2 0 10 20 30 40 50 60 70 80 90 100 Metacritic Rating Cold Opened Screened for Critics Linear (Cold Opened) RSF 4.July.2014 Linear (Screened for Critics) 43 Cold opening variable is significantly positive in US 15% increase in revenue insignificant in UK, Mexico, US rental (DVD) markets Can be fit with CH model with 1.63 (Brown+ Mgt Sci in press) RSF 4.July.2014 44 No effects in UK, Mexico, US rentals (word leaks out) RSF 4.July.2014 45 Propensity score matching RSF 4.July.2014 46 PSM similar to OLS regression RSF 4.July.2014 47 Studios learning that cold opening pays? RSF 4.July.2014 48 frontier 2: What is level 0? • My current view – Level 0 is fast, salient – Fundamentally an empirical question: – Cf. Schelling: • “one cannot, without empirical evidence, deduce whatever understandings can be perceived in a non-zero sum game of maneuver any more than one can prove, by formal deduction, that a particular joke is bound to be funny.” RSF 4.July.2014 51 But what’s salient? • • • • “personal” numbers ends + center of a number line visual: Itti-Koch “low level” algorithm Private info: Strategy = known state – e.g. bid your value in an auction – e.g. report state honestly in sender-receiver RSF 4.July.2014 52 Neurally Based Models of Visual Salience (Itti Koch Nature 05) RSF 4.July.2014 53 RSF 4.July.2014 54 Schelling (1960) map RSF 4.July.2014 55 RSF 4.July.2014 56 RSF 4.July.2014 57 Fails on categorical distinctiveness RSF 4.July.2014 58 Meta-model approach: Level 0’s focus on strategy features (Wright, Leyton-Brown subm 14) RSF 4.July.2014 59 Georganas+ 14 (UG1) Nash (30%) Level 1 (33%) RSF 4.July.2014 60 RSF 4.July.2014 61 Conclusions • Cognitive hierarchy approach – Lab, field, eyetracking, fMRI • Many open questions – Are there distinct types? – Closer link to ToM regions • Beliefs, intentions, attributions – Disorders of strategic thinking • Paranoia, gullibility, autism(s) – Experience and expertise – Endogenized steps RSF 4.July.2014 62