Abstract
This paper models success probability in imperfectly discriminating contests involving multiple players and multiple prizes. This, in fact, turns out to be a generalization of Tullock’s contest success function to a multi-player, multiple prizes. The model can be used to analyze efforts exerted by individuals in various real-life situations, like obtaining seats in congested public transportation vehicles or obtaining admission into elite educational institutes. We propose a “holistic” probability model, derive the equilibrium efforts exerted, and analyze those efforts, the associated total costs and total dissipation, and explore pricing and number of ‘seats’. The derivation provides a new rational for the multinomial Logit Model. It also derives formula for rent dissipation. We also discuss two extensions of the model.
Similar content being viewed by others
References
Anderson SP, de Palma A, Thisse J-F (1992) Discrete choice theory of product differentiation. The MIT Press, Cambridge
Arnott R, de Palma A, Lindsey R (1993) A structural model of peak period congestion: a traffic bottleneck with elastic demand. Amer Econ Rev 83:161–179
Blavatskyy PR (2010) Contest success function with the possibility of a draw: Axiomatization. J Math Econ 46:267–276
Chiappori PA, McCann R, Nesheim L (2010) Hedonic price equilibria, stable matching, and optimal transport: equivalence, topology, and uniqueness. Econ Theory 42:317–354
Clark DJ, Riis C (1998a) Competition over more than one prize. Amer Econ Rev 88:276–289
Clark DJ, Riis C (1998b) Contest success functions: An extension. Econ Theory 11:201–204
Chowdhury SM, Sheremeta R M (2011) A generalized Tullock contest. Public Choice 147:413–420
de Palma A, Picard N, Waddell P (2007) Discrete choice models with capacity constraints: An empirical analysis of the housing market of the greater Paris region. J Urban Econ 62:204–230
de Palma A, Munshi S (2013) A generalization of Berry’s probability function. Theor Econ Lett 3:12–16
de Palma A, Lefèvre C (1981) Simplification procedures for a probabilistic choice model. J Math Sociol 6:43–60
de Palma A, Lefèvre C (1983) Individual decision-making in dynamic collective systems. J Math Sociol 9:103–124
de Palma A, Lefèvre C (1988) Population systems with (non-) extensive interaction rates. Math Model 10:359–365
de Palma A, Monchambert G, Lindsey R (2017) The economics of crowding in rail transit journal of urban economics. J Urban Econ 101:106–122
Dixit A (1987) Strategic behavior in contests. Amer Econ Rev 77:891–898
Gaudry MJI, Dagenais M G (1979) The dogit model. Transp Res 13:105–112
Gradstein M, Nitzan S (1989) Advantageous multiple rent seeking. Math Modell 12:511518
Hillman AL, Riley JG (1989) Politically contestable rents and transfers. Econ Polit 1:1739
Hirshleifer J (1989) Conflict and rent-seeking success functions: ratio vs. difference models of relative success. Publ Choice 63:101–112
Hwang SH (2009) Contest success functions: theory and evidence. Economics Department Working Paper Series, pp 11
Laffont JJ, Martimort D (2005) The theory of incentives: The principal-agent model. Princeton University Press, Princeton
Mohring H (1972) Optimization and scale economies in urban bus transportation. Amer Econ Rev 62:591–604
Moldovanu B, Sela A (2001) Optimal allocation of prizes in contests. Amer Econ Rev 91:542–558
Münster J (2009) Group contest success functions. Econ Theory 41:345–357
Nitzan S (1994) Modeling rent-seeking contests. Eur J Polit Econ 10:41–60
Nti KO (1997) Comparative statics of contests and rent-seeking games. Int Econ Rev 38:43–59
Pucher J, Korattyswaroopam N, Ittyerah N (2004) The crisis of public transport in India: Overwhelming needs but limited resources. J Publ Transp 7:1–20
Rosen S (1986) Prizes and incentives in elimination tournaments. Amer Econ Rev 76:701–709
Skaperdas S (1996) Contest success functions. Econ Theory 7:283290
Spence M (1973) Job market signaling. Q J Econ 87:355–374
Szymanski S (2003) The economic design of sporting contests. J Econ Lit 41:1137–1187
Tullock G (1980) Efficient rent seeking. In: Buchanan J M, Tollison R D, Tullock G (eds) Toward a theory of the rent-seeking society. College Station. Texas A&M University Press, TX, pp 97–112
Vickrey W (1969) Congestion theory and transport investment. Amer Econ Rev Papers Proc 59:251–61
Acknowledgments
The first author would like to thank the seminar participants at ETHZ (Civil Engineering), Cambridge University, Catholic University of Leuven (Economics), University of Copenhagen (Economics), University of Louvain-La-Neuve, as well as Simon Anderson (University of Virginia), Mogens Fosgerau (Technical university of Copenhagen), Robin Lindsey (University of British Colombia), Nathalie Picard (University of Cergy-Pontoise) and Jean-Luc Prigent (University of Cergy-Pontoise). He would also like to thank the following organizations: Tarification des transports individuels et collectifs à Paris. Dynamique de l’acceptabilité: Predit and Ademe. Surpriceproject, Scheduling, trip timing and scheduling preferences, Predit. He would also like to thank his former student Sue Wang (MIT) and Charles Maurin (Columbia) and Lucas Javaudin, Florence Helft (ENS-Paris Scaly) who helped us to edit the present paper. The first author would like to thank French ANR (Elitisme) for financial support. The second author would like to acknowledge the hospitality of the faculty and students of École Normale Supérieure de Cachan - this project was initiated during her visit there. She would also like to thank Prof. Barry Sopher of Rutgers, the state University of New Jersey, for his academic as well as non-academic help, and Professors Kalyan Chatterjee of Pennsylvania State University, Joan Walker of University of California Berkeley, and Krishnendu Ghosh Dastidar of Jawaharlal Nehru University, for fruitful discussions about the paper. Last, but not the least, she would like to thank Prof. Tomas Sjöström of Rutgers University for his versatile and helpful advice.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Derivation of the Probability Model with an Example
Consider the basic probability model with \(\lambda = 0\). Let \(n \quad =\) 3, \(\bar {n} \quad =\) 2. Then the possible outcomes and associated probabilities are as follows:
Hence summing over all outcomes we get,
Now
Hence substituting we get
And \(2=\left ({\begin {array}{l} 3 \\ 2 \end {array}} \right )\), for this example.
Now consider an example with \(n = 3,\,\bar {{n}}= 2,\,e_{1} = 8,\,e_{2} =e_{3} = 1\). Then we get the following:
Again the sum equals \(\bar {{n}}= 2\). This means (roughly) that person 1 exerting high effort gets to sit with a very high probability in one of the seats, while the other two people exerting the same low probability gets to sit with almost equal probability in the remaining 1 seat.
Appendix B: Proof of Proposition 2
Proof
We can write the objective function as:
Now\(P_{i}=\frac {\bar {{n}}-1}{n-1}\mathrm {+} \left [ {\frac {n-\bar {{n}}}{n-1}} \right ]\,\,\left [{\frac {e_{i}} {\sum \limits _j {e_{j}} } } \right ].\)
Therefore we can rewrite the objective function as
Consider a change of variable \(e_{i} =\exp \,(E_{i} )\),so that
Alsolet
Hence the objective function can be written
Therefore, the F.O.C.s are
Thus Eq. 23 defines the best reply of agent i, with respect to the strategy of the other agents,\(E_{i}^{br} \).
Now assuming \(\chi _{i} =\chi \forall _{i} \), we get the symmetric Nash equilibrium as follows (refer to appendix C for the asymmetricsolution):
or
The S.O.C.s are (which are true even for the asymmetriccase):
Substituting the F.O.C. in the S.O.C., we get,
For convexity of the objective function, we need the above expression to be positive. Therefore, a sufficient condition wouldbe \([2P_{i} +\alpha ]>0\), for any\(P_{i} \). For this,\(\alpha >0\) is a sufficient condition.
In the symmetric case, the condition becomes,\(\left [ {\frac {2}{n}+\alpha } \right ]>0\) which reducesto \(\alpha >-\frac {2}{n}.\) Forvery large n, this implies again that the cost function is convex.
Uniqueness. The condition for uniqueness is
Let
Thus, by the implicit function theorem, weget,
Hence
Hence summing over all \(j\ne i\), weget
which is a decreasing function of \(\alpha \). Hence, for\(\alpha =-1,\)
And\(\forall \alpha >-1\),
This proves uniqueness and the proposition. □
Appendix C: Asymmetric Case
We can write the objective function as follows (notice \(\bar {{c}}_{i}\) instead of \(\bar {{c}},g_{i} (\cdot )\) instead of \(\bar {{c}},g({\cdot })\), \({\alpha _{i}}\) instead of \({\alpha }\), and \(\chi _{i}\) instead of \(\chi \), as costs for individual \(i)\):
Recall
Like before, with \(e_{i} =\exp \,(E_{i} )\) and
we get
Therefore, the F.O.C.s are:
That is,
Let \({\Phi } =\sum \limits _j {\exp \,(E_{j} )} =\sum \limits _j{e_{j}} .\) Then, we get
Since \(e_{i} =\exp \,(E_{i} )\), the F.O.C. becomes:
The first-order condition, simplifies as follows:
If \(\alpha _{i} = 0\forall i:\)We have \(\frac {\omega _{i}} {\chi _{i}} \frac {e_{i}} {{\Phi }^{2}}=\frac {\omega _{i}} {\Phi \chi _{i}} -1,\,or\,\,\frac {\omega _{i}} {\chi _{i}} ={\Phi } -{\Phi }^{2}\,\frac {\chi _{i}} {\omega _{i}} \)
Thus:
The following proposition summarizes our findings in the asymmetric case.
Proposition 1
Consider asymmetric but linear cost of effort, that is, cost of effort of individual i is given by\(\chi _{i}e_{i} ,\,\forall _{i}\).Let\(\omega _{i} =\left [{g_{i} \,(n-\bar {{n}})-\bar {{c}}_{i}} \right ]\,\frac {n-\bar {{n}}}{n-1}\). Then the Nash equilibrium level of effort of individual i is given by
Note that \(e_{r}^{\ast } >e_{s}^{\ast } \,\,\text {if}\,\,\frac {\chi _{s}} {\omega _{s}} >\frac {\chi _{r}} {\omega _{r}}\).
We can also compute equilibrium probabilities in this case to be as follows:
Notice that lower is the cost of exerting effort, \(\frac {\chi _{i}} {\omega _{i}} \), relative to \(\sum \limits _j {\frac {\chi _{j}} {\omega _{j}} } \), the higher is the probability of finding a seat (please refer to footnote 11 in the paper for an interpretation of heterogeneity of cost functions in our context).
Moreover, note that if all costs of effort \(\left ({\chi _{i}} \right )\) are multiplied by \(\tau \), then each level of effort is divided by \(\tau \), and the total cost of effort \(\sum \limits _j {\chi _{j} e_{j}} \) (for this case of αi = 0) remains the same. Taking effort multiplicatively has no impact of consumers’ surplus, so that the tax revenue \((\tau -1)\,\sum \limits _j {e_{j}}\) corresponds to the social benefit.
We can verify that in the symmetric case, we have:
as derived earlier.
Appendix D: Proof of Propsition 6
Proof
The planner minimizes the following objective function w.r.t. \(\,\bar {{n}}\):
Differentiatingw.r.t. \(\bar {{n}}\),the F.O.C. is:
This is a quadratic equation in\(\bar {{n}}\) and can be written inthe form \(A\bar {{n}}^{2}+B\bar {{n}}+C = 0\) where thecoefficients are as follows:
In general, it is difficult to get a closed-form solution to the above equation. Hence we let n and S be large (as isplausible), so that we get the following approximations:
Now substituting in the equation and solving, we get
So for any feasible solution, wemust have \(J\,(\pi +\bar {{c}})<1\). (This is plausiblegiven that \(\pi ,\,\bar {{c}}\) andJ are likely to be small and can be chosen appropriately). Hence we can solve\(\bar {{n}}^{\ast }\) from above tobe:
(Notice that\(Jn>S\), otherwise\(\bar {{n}}=n\) andthe problem of congestion would not be relevant.) We can check that the S.O.C. for minimizationalso holds.
However, notice that if n is very large, then\(\bar {{n}}^{\ast }\) will become negative,so that the optimal \(\bar {{n}}\) will be 0. Hence we can solve for the cut-off of n for feasible\(\bar {{n}}\), bysetting
This solves for
Call n in Eq. 27, \(\underline {n}.\) Hence
□
Appendix E: Proof of Proposition 7
Proof
Differentiating \(\bar {{n}}^{\ast } \) we get the following.
< 0,
This proves theproposition. □
Appendix F: Proof of Propsition Lemma 2: Derivation Of \(\kappa \)
Proof
Summing over all possible outcomes, we get
(There is no effortcounted for \(\varepsilon _{i} =-1.)\,\)Now\(\left ({\sum \nolimits _{{\begin {array}{*{20}c} {\bar {{\upsilon } }\in {\Omega }} \hfill \\ {\varepsilon _{i} (\bar {{\upsilon } })= 0} \hfill \end {array}} } 1} \right )\) counts the number ofvectors of length \(\left ({n-1} \right )\) (sincethe i th person’s outcome is known which is to get the low quality goods), where there are\(\bar {{n}}\) number of 1’s(since \(\bar {{n}}\) peopleare getting the higher quality good), and the rest can have any two possible outcomes,\(\varepsilon = 0,\,-1\), that isthey may have got the lower quality goods, or may not have got any goods at all. Hence we get thefollowing:
Similarly,\(\left ({\sum \nolimits _{{\begin {array}{*{20}c} {\bar {{\upsilon } }\in {\Omega }} \hfill \\ {\varepsilon _{i} (\bar {{\upsilon } })= 1} \hfill \end {array}} } 1} \right )\) counts the number ofvectors of length \(\left ({n-1} \right )\) with\(\left ({\bar {{n}}-1} \right )\) numberof 1’s (since the i th person has got the higher quality good), and is givenby
Substituting these in theabove expression, and letting \(\sum \nolimits _i {e_{ih}} =e_{H} \),and \(\sum \nolimits _i {e_{il}} =e_{L} \) we cansolve \(\kappa \) asfollowing:
□
Appendix G: Correlation
Let us see the correlation between the outcomes of two individuals, that is, say we are interested in the correlation of the outcomes players 1 and 2, Corr(ε1,ε2). Now
We know (from calculations before) that \(E\,(\varepsilon _{1} )=\Pr \,\{\varepsilon _{1} \}= 1\). Similarly, \(E\,(\varepsilon _{2} )=\Pr \,\{\varepsilon _{2} \}= 1.\)Also, \(E\,(\varepsilon _{1} \varepsilon _{2} )=\Pr \,\{\varepsilon _{1} \varepsilon _{2} \}= 1\) (since in all other possibilities, \(\varepsilon _{1} \)or \(\varepsilon _{2} \)or both are 0 and there is no contribution to the expectation). Now (restricting to the \(\lambda =\) 0 case), we have
(where K is as given in Eq. 4 with \(\lambda = 0)\)
Now
Substituting and simplifying, we get
Recall that
Hence we get
Also we can calculate that
Now by substituting all the expressions we can get \(Corr\,\left ({\varepsilon _{1} ,\varepsilon _{2}} \right )\). Now, in order to keep the calculations tractable, we make the simplifying assumption of letting n being large (so that \(\frac {1}{n}\approx 0\)). In this case, dividing the expression for \(Cov\,\left ({\varepsilon _{1} ,\varepsilon _{2}} \right )\) throughout by n and letting \(\frac {1}{n}\approx 0\), we get
Simplifying the above expression we get
Hence \(Corr\,\left ({\varepsilon _{1} ,\varepsilon _{2}} \right )<0\). In fact, we can calculate the expression for \(Corr\,\left ({\varepsilon _{1} ,\varepsilon _{2}} \right )\) more precisely in this case. Notice that when n is large
And variance can be calculated as
Hence substituting and simplifying, we get the expression for the correlation coefficient as below:
Notice that the correlation coefficient is generalizable to that between outcomes of any two individuals i andj after appropriately substituting for 1 and 2 in the expression on the R.H.S. Hence we see that, as expected, the correlation between outcomes of any two individuals is negative, that is, as one person’s chances of getting a better-quality good increases, that of the other falls. The following proposition summarizes the findings.
Proposition 2
The correlation coefficient between the outcomes of any two individuals, sayindividualiandindividual\(j, \quad i\ne j,\),is given as
Moreover, in case the number of better quality goods is small relative to the total number of consumers, so that\(\frac {\bar {{n}}}{n}\approx 0\), wecan further simplify the expression of correlation coefficient to arrive at the following corollary.
Corollary 1
If \(\frac {\bar {{n}}}{n}\approx 0\) then we get
Appendix H: Lottery
Consider a situation in which each person can buy more than one ticket for a limited number of prizes. Say there are k number of people, \(n \) the total number of tickets for \(\bar {{n}}\) number of prizes (seats), with \(n>\bar {{n}}\). Let \(e_{i} \) be the number of tickets purchased by player \(i \) (note this has nothing to do with effort as of now). Assume all available tickets are bought, that is \(\sum \nolimits _{i = 1}^k {e_{i}} =n_{\mathrm { } }\) and also \(e_{i} \,\in \,\left \{ {1,\,2,\mathellipsis ,n-k + 1} \right \}\). Hence the space of all outcomes \({\Omega } \), is as follows:
where \(\varepsilon _{ke_{k}} (\textbf {v})\) is the n-th component of the vector v.
The restriction \(\sum \limits _{i = 1}^k {\sum \limits _{j = 1}^{e_{i}} {\varepsilon _{ij} (\mathbf {v})} =\bar {{n}}}_{\mathrm { } }\) reflects the fact that in any outcome, all the prizes are won. Moreover, the first \(e_{1} \) outcomes in v reflect the outcomes for the tickets bought by player 1. Note that here \(\varepsilon _{ij} (\text {v})= 1\) means that in the outcome v the i-th person has won in the j th ticket and otherwise \(\varepsilon _{ij} (\text {v})= 0\). Also note that the number of outcomes in the space is given by
Now let us compute the probability that any person, say person 1, has won p prizes (where \(p\le e_{1} \), that is number of prizes won is less than the number of tickets bought by person 1 and \(p\le \bar {{n}}\), that is the number of prizes won is less than the total number of prizes). Now all outcomes are equally likely. Moreover person 1 wins p prizes with \(e_{1} \) number of tickets in \(\left ({\begin {array}{l} e_{1} \\ p \end {array}} \right )\) ways. For each way, the rest \(\bar {{n}}-p\) prizes can be won by \(n-e_{1} \) tickets in \(\left ({\begin {array}{l} n-e_{1} \\ \bar {{n}}-p \end {array}} \right )\). Hence the probability is given by:
Pr{Player 1 wins p prizes}\(={\begin {array}{*{20}c} {\sum \limits _{v\in {\Omega }}} \hfill \\ {\varepsilon _{11} (v)+\mathellipsis \varepsilon _{1e_{1}} (v)=p} \hfill \par \end {array}} \frac {1}{\left ({\begin {array}{l} n \\ \bar {{n}} \end {array}} \right )}\,\,\)
More generally, the probability that the \(l-\)th player will win p prizes is given by
Pr {Player \(\ell \) wins p prizes}\(=\frac {\left ({\begin {array}{l} e_{l} \\ p\,\, \end {array}} \right )\,\,\,\left ({\begin {array}{l} n-e_{l} \\ \bar {{n}}-p \end {array}} \right )}{\left ({\begin {array}{l} n \\ \bar {{n}} \end {array}} \right )}.\)
Now our model can be related to this lottery set-up in the following way: let each ticket, for example, to be a unit of effort exerted, so that more number of tickets will correspond to more effort exerted (of course, effort in our model is a continuous variable while number of tickets can only be discrete). If tickets have a (uniform) price, that might be interpreted as \(\chi \) in our model. Hence the total expenditure, of person i, to purchase \(e_{i}\) number of tickets, would be \(\chi e_{i} \), the total ‘effort cost’ of player i in our model (assuming \(\alpha = 0\), or linear cost of effort).
Interestingly, in the (usual) case with every person buying only one ticket \(e_{i} = 1,\,\,\forall {i}\) (or alternatively the symmetric effort case with everybody exerting the same effort), the probability that the ith person wins is given by
Pr {Player i wins p prizes}\(~=~\frac {\left ({\begin {array}{l} 1 \\ 1\,\, \end {array}} \right )\,\,\,\left ({\begin {array}{l} n-1 \\ \bar {{n}}-1 \end {array}} \right )}{\left ({\begin {array}{l} n \\ \bar {{n}} \end {array}} \right )}=\frac {\bar {{n}}}{n}\),
which is the usual random probability model that we had.
Rights and permissions
About this article
Cite this article
de Palma, A., Munshi, S. Multi-player, Multi-prize, Imperfectly Discriminating Contests. Methodol Comput Appl Probab 21, 593–632 (2019). https://doi.org/10.1007/s11009-018-9628-1
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11009-018-9628-1