Game-Theoretic Derivation of Discrete Distributions and Discrete Pricing Formulas

Akimichi Takemura, Taiji SuzukiJOURNAL 2007 · DOI

Abstract This paper explores the application of game-theoretic probability protocols to finite-horizon discrete games. We demonstrate that discrete distributions with finite support and standard financial pricing formulas, such as the Cox-Ross-Rubinstein (CRR) model, can be derived as logical consequences of game protocols rather than being assumed as a priori laws. By constructing replicating strategies for a "Skeptic" and neutral forecasting strategies, we show that the game-theoretic framework provides a robust alternative to measure-theoretic probability, eliminating the need for pre-existing probabilistic assumptions.

1. Introduction: The Foundations of Probability

In traditional statistics and mathematical finance, the standard approach is measure-theoretic. We typically begin by assuming a probability space and a specific distribution—for instance, assuming a stock price follows a geometric Brownian motion or a series of trials follows a Binomial distribution. However, the origin of these assumptions is often left unexamined.

This paper utilizes the game-theoretic probability framework, pioneered by Shafer and Vovk (2001), to bridge this gap. Instead of assuming randomness as a physical property of the world, we define a game with specific rules, or protocols. In this framework, probability emerges as a requirement for the game to be coherent and free of arbitrage. If a game is structured such that a "Skeptic" cannot become infinitely rich by betting against "Reality," then Reality is effectively forced to behave as if it were following a specific probability distribution.

2. The Game-Theoretic Protocol

The framework involves a multi-player game with a finite horizon $N$. The primary players are:

  • Skeptic: An investor who attempts to increase their capital by betting on the outcomes of Reality's moves.
  • Reality: The environment or market that determines the actual outcomes of the game.
  • Forecaster: An optional player who sets the "prices" or expected values for Reality's next move at the start of each round.

The game is played over a series of rounds. At each step, the Forecaster announces a price, the Skeptic decides how much to bet, and Reality decides the outcome. The goal is to determine the initial capital required by the Skeptic to reach a specific final payoff, regardless of Reality's moves.

3. Methodology: Replicating Strategies and Backward Induction

To derive specific distributions and pricing formulas, the paper employs several key mathematical techniques:

3.1 Replicating Strategies

Similar to the concept of "delta hedging" in finance, a replicating strategy is a betting sequence for the Skeptic that perfectly offsets risk. By constructing such a strategy, we can determine an "exact price" for a variable or a payoff function. If the Skeptic follows this strategy, their final capital will exactly match the target payoff, regardless of the path taken by Reality.

3.2 Backward Induction

To find the value of a "European option" (a function of the final state of the game), we use backward induction. We start at the final round $N$ and work backward to the beginning of the game. This process determines the initial capital necessary to guarantee a specific outcome, effectively deriving the "price" of the distribution's expectation.

3.3 Neutral Forecasting Strategies

By making the Forecaster’s prices dependent on the history of previous moves (a Markovian strategy), we can "force" Reality to follow complex distributions. This allows for the derivation of models involving sampling without replacement or contagion effects.

4. Deriving Classical Distributions

The paper demonstrates that various well-known discrete distributions are natural consequences of specific game protocols:

  • Binomial Distribution: Derived from a "Biased-Coin Game" protocol where the Forecaster sets a constant price for each round.
  • Urn Models (Hypergeometric and Polya): By specifying how the Forecaster updates prices based on the current "count" of outcomes, the authors derive distributions that represent sampling without replacement (Hypergeometric) or reinforcement/contagion (Polya).
  • Multivariate Extension: The framework is extended to $d$-dimensional vectors, allowing for the derivation of the Multinomial distribution.

5. Financial Applications: The CRR Formula

A significant result of this framework is the derivation of the Cox-Ross-Rubinstein (CRR) formula. In the standard binomial option pricing model, the CRR formula is used to price European options. In the game-theoretic context, this formula emerges naturally as the capital needed for a replicating strategy. The "risk-neutral probability" used in finance is not an assumption here, but a mathematical necessity to prevent the Skeptic from having a "sure win" (arbitrage).

6. Discussion and Universality

The authors prove a universality result: any arbitrary discrete distribution with finite support can be represented by a specific game tree. This implies that the game-theoretic approach is just as powerful as the standard measure-theoretic approach for discrete cases, while offering a more intuitive foundation based on the absence of arbitrage.

However, there are limitations to this expository work. The focus is strictly on finite-horizon games and distributions with finite support. While this covers many practical applications, it does not yet address continuous distributions or those with infinite support (such as the Poisson distribution) within this specific text. Furthermore, while the concept of "forcing" a distribution is used, the rigorous measure-theoretic proof of forcing typically requires infinite-horizon games, which are mentioned but not explored in depth here.

7. Conclusion

By shifting the focus from measure theory to game theory, we can derive the fundamental building blocks of statistics and finance from the simple requirement of game coherence. This framework provides a transparent look at why certain distributions appear in our models: they are the only outcomes consistent with a fair game between a Forecaster, a Skeptic, and Reality.

Summary updated

Abstract

In this expository paper, we illustrate the generality of the game-theoretic probability protocols of Shafer and Vovk (2001) in finite-horizon discrete games. By restricting ourselves to finite-horizon discrete games, we can explicitly describe how discrete distributions with finite support and discrete pricing formulas, such as the Cox-Ross-Rubinstein formula, are naturally derived from game-theoretic probability protocols. Corresponding to any discrete distribution with finite support, we construct a finite-horizon discrete game, a replicating strategy of Skeptic, and a neutral forecasting strategy of Forecaster, such that the discrete distribution is derived from the game. Construction of a replicating strategy is the same as in the standard arbitrage arguments of pricing European options in binomial tree models. However the game-theoretic framework is advantageous because it eliminates the need for any a priori probabilistic assumption.
BibTeX
@article{takemura2007game,
  author = {Akimichi Takemura and Taiji Suzuki},
  title = {{Game-Theoretic Derivation of Discrete Distributions and Discrete Pricing Formulas}},
  journal = {JOURNAL},
  year = {2007},
  doi = {10.14490/jjss.37.87},
}
Figure