Strategic information transmission (trojan teaching) in
client-consultant relationships
Alexander Poddiakov1 Stanislav Mos...
Outline
1 Problem statement
2 Experiment
Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 2 / 14
Problem statement
Trojan teaching
Trojan teaching ( c Alexander Poddiakov) stands for the situation when
the informed part...
Problem statement
Examples
Children games with partial communication of information
ForEx trade: teachers benefit not only ...
Problem statement
Literature connections
Crawford and Joel Sobel (1982) study a prototype problem of that
kind, and show t...
Problem statement
Paying for useless advice
Powdthavee and Riyanto, 2012
Experimental test: 5 tosses of coin(s) coming fro...
Experiment
Our experiment
Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 7 / 14
Experiment
Solution
Payoff in the game are alighed in a way that in all sequential
equilibria, sender has to send noisy sig...
Experiment
Experiment
28 paid (16 + 12 subjects, for 15 periods) and 34 unpaid (10 for 15
periods, 12 for 25 and 10 period...
Experiment
Clients’ choices by signals (in rows) and sessions (cols)
0.51
1 2 1 2 1 2 1 2 1 2
1 2 3 4 5
Fraction
decision ...
Experiment
Clients: Results
Cliens made overall 495 choices, of which 1 =A was slightly more
frequent (57% of cases), both...
Experiment
Advisors: signals sent (if any)
.2561
.2035
.214
.3263
.2048
.181
.1667
.4476
0.5
0 1 2 3 0 1 2 3
0 1
Density
f...
Experiment
Advisors: results
Advisors under unpaid conditions send noizy signal more often
Difference among treatments sign...
Experiment
Advisors: signals received and sent
0.2.4.6.80.2.4.6.8
1 2 3 1 2 3
1 2 3
1
2
3
4
5
Fraction
1=send whatever rec...
Experiment
Outcomes, paid (N = 38 × 15 periods)
message signal 1 2 12 21 Total
0 21 17 5 10 53
1 34 0 10 7 51
2 0 41 7 7 5...
Experiment
Results
Prevailing strategy is truth telling (communicate whatever received)
Deliberately trojan teaching (rece...
of 16

PresTrojan0_1212

Published on: Mar 4, 2016
Source: www.slideshare.net


Transcripts - PresTrojan0_1212

  • 1. Strategic information transmission (trojan teaching) in client-consultant relationships Alexander Poddiakov1 Stanislav Moskovtsev2 Alexis Belianin3 1idea 2realization 3this talk December 27, 2012 Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 1 / 14
  • 2. Outline 1 Problem statement 2 Experiment Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 2 / 14
  • 3. Problem statement Trojan teaching Trojan teaching ( c Alexander Poddiakov) stands for the situation when the informed party (sender) communicates to the uninformed party (receiver) less than full information in posession of the sender, to own advantage and at the expense of the receiver. It seems appropriate to speak of trojan teaching (or information transmission) whenever the sender (he) has superior information (finer partition on the set of possible states), knows his information is superior, knows that receiver (she) does not know this (otherwise, cannot act strategically), knows that receiver would have benefited had her information been as complete as that of himself (full transmission) expects to get strategic benefits from incomplete information transmission (has intention and will to do so — ethical issues) Omission of any of these characteristics defines a different (but possibly very interesting!) task from that of trojan teaching in our sense. Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 3 / 14
  • 4. Problem statement Examples Children games with partial communication of information ForEx trade: teachers benefit not only from licensing fees, but also from incremental liquidity brought in by noize traders (Kim e.a., 2012) Consultancy business: consultant allegedly better knows how to solve particular problem of a client (main motivating story) ... Colloqually, this means that trojan teachers communicate truth and only truth, but not all truth (in contrast to blatant lies). Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 4 / 14
  • 5. Problem statement Literature connections Crawford and Joel Sobel (1982) study a prototype problem of that kind, and show that in any Bayesian Nash equilibirium, sender introduces noize in the signal by sending one of possible signal in strategically partitioned information set. Gneezy (2005) experimentally studies deception game, when sender has to communicate to the receiver which of the two actions brings her higher income, when one of the two possible messages is explicitly false. He finds that false information increases when the margin of sender’s gains over receiver is maximal. Rode (2006) extends this setup to 1) three information conditions (high, medium and low uncertainty of the receiver), and 2) cooperation (coordination) vs. competition (matching pennies) games. He reports costly truth-telling of order of 1/5-1/3 of cases (same across treatments), and following the adivce in 2/3-3/4 of cases, with significantly less under high uncertainty and competitive treatment. Recently, Powdthavee and Riyanto (2012) study willingness to pay forAlexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 5 / 14
  • 6. Problem statement Paying for useless advice Powdthavee and Riyanto, 2012 Experimental test: 5 tosses of coin(s) coming from the subjects, with changes of coin and tosses by volunteer participants. Subjects are endowed with 100 tokens, and are to bet consecutively on the outcome of 5 tosses, one after another. All subjects also had 5 envelopes containing randomly generated forecast of the outcome of the next toss, which they can buy for 10 tokens before each round, and check its correctness for free ex post. Of 378 participants from two countries (Thailand and Singapore), 191 received a correct prediction in the first round; 92 received all-correct predictions after the first two rounds; 48 after the first three rounds; and 23 after four rounds. Finding: significant positive, and monotonically increasing effect on probability of purchase of sequences of correct predictions (linear probability coefficients from linear model are 0.0522**, 0.153***, 0.195***, 0.276***). Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 6 / 14
  • 7. Experiment Our experiment Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 7 / 14
  • 8. Experiment Solution Payoff in the game are alighed in a way that in all sequential equilibria, sender has to send noisy signal no matter what he receives himself; hence receiver is indifferent between asking advice and acting on her own. This is the static solution though, although pairs are in partner treatment. Dynamic extension is on the way (?) Experimental data complemented with questionnaire (to be explored) What other payoffs do make sense? Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 8 / 14
  • 9. Experiment Experiment 28 paid (16 + 12 subjects, for 15 periods) and 34 unpaid (10 for 15 periods, 12 for 25 and 10 periods, resp) Moscow students took part in 2012. (time varied to check for time learning effect). Overall, 210 choices under paid conditions, and 285 choices under unpaid conditions. Programmed in z-tree, instructions handled and read aloud Extension to market: planned, but not realized Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 9 / 14
  • 10. Experiment Clients’ choices by signals (in rows) and sessions (cols) 0.51 1 2 1 2 1 2 1 2 1 2 1 2 3 4 5 Fraction decision of client Graphs by group(session) 0.51 1 2 1 2 1 2 1 2 1 2 1 2 3 4 5 Fraction decision of client Graphs by group(session) 0.2.4.6.8 1 2 1 2 1 2 1 2 1 2 1 2 3 4 5 Fraction decision of client Graphs by group(session) Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 10 / 14
  • 11. Experiment Clients: Results Cliens made overall 495 choices, of which 1 =A was slightly more frequent (57% of cases), both overall and when deciding on their own. In all sessions except one, clients who received precise signal strongly follow the advice, whereas clients who received noizy advice favor option 1 =A. Possible framing effect. Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 11 / 14
  • 12. Experiment Advisors: signals sent (if any) .2561 .2035 .214 .3263 .2048 .181 .1667 .4476 0.5 0 1 2 3 0 1 2 3 0 1 Density frequency of advisors’ strategies (0=not asked,1=A, 2=B, 3=AB) Graphs by 0 − unpaid, 1 − paid Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 12 / 14
  • 13. Experiment Advisors: results Advisors under unpaid conditions send noizy signal more often Difference among treatments significant overall (ANOVA F-test= 5.18, p < 0.023) and on advice only (ANOVA F-test= 3.95, p < 0.047). Result robust across time (similar tendency if attention is limited to first 10 periods). Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 13 / 14
  • 14. Experiment Advisors: signals received and sent 0.2.4.6.80.2.4.6.8 1 2 3 1 2 3 1 2 3 1 2 3 4 5 Fraction 1=send whatever received, 2=bet, 3=trojan Graphs by group(session) Paid: Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 14 / 14
  • 15. Experiment Outcomes, paid (N = 38 × 15 periods) message signal 1 2 12 21 Total 0 21 17 5 10 53 1 34 0 10 7 51 2 0 41 7 7 55 12 39 0 36 0 75 21 0 29 0 22 51 Total 94 87 58 46 285 Table: 1 – A, 2 – B, 12 – AB after A, 21 – AB after B Out of 285 cases, 133, or 46% are honest, 53, or 19% do not ask for help, and 68, or 24% are equilibrium Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 15 / 14
  • 16. Experiment Results Prevailing strategy is truth telling (communicate whatever received) Deliberately trojan teaching (received clean, sent noisy signal) is systematic. No significant differences across treatments (ANOVA F-test= 1.83, p < 0.177). Overall frequency of noisy signal is 49.3%, but half of it (24.8%) occurs when signal received was noisy as well. Alexis Belianin et.al. (HSE) Trojan teaching December 27, 2012 16 / 14

Related Documents