Monotone Comparative Statics without Lattices

This paper extends the theory of Monotone Comparative Statics to environments lacking a lattice structure by introducing the pseudo-lattice property, thereby enabling the analysis of mixed-strategy and trembling-hand perfect equilibria in multidimensional games.

The Gist

Imagine you are trying to predict how people will behave when the world around them changes. Maybe a tax goes up, a new technology arrives, or a competitor lowers their prices. Economists call this Comparative Statics. They want to know: If X changes, does Y go up or down?

For decades, the most powerful tool for answering this question was a mathematical framework called Monotone Comparative Statics (MCS). Think of this tool as a high-tech GPS for economic behavior. It's incredibly accurate, but it has a very strict rule: the map it uses must be a perfect Lattice.

The Problem: The "Lattice" Trap

In math, a "Lattice" is a very tidy, structured grid. Imagine a chessboard or a family tree where every two people have a clear "parent" (a highest common ancestor) and a "child" (a lowest common descendant).

The problem is that the real world is messy. Many important economic situations don't fit on a neat chessboard.

• Mixed Strategies: In games like Poker or Rock-Paper-Scissors, players often randomize their moves (e.g., "I'll play Rock 40% of the time"). The set of all possible random combinations is a "cloud" of possibilities, not a neat grid. It's like trying to arrange a cloud of smoke into a perfect chessboard.
• Risk and Uncertainty: When people choose between lotteries or uncertain futures, the math gets fuzzy.

Because the old GPS (MCS) required a perfect Lattice, it couldn't navigate these "cloudy" areas. Economists had to throw up their hands and say, "We can't predict how people will react to changes in these complex, random situations."

The Solution: The "Pseudo-Lattice"

The authors of this paper (Che, Kim, and Kojima) said, "Wait a minute. Do we really need a perfect chessboard? Or can we get by with something a little more flexible?"

They introduced a new concept called a Pseudo-Lattice.

The Analogy: The Hiking Trail vs. The Grid

• The Old Way (Lattice): Imagine you are hiking in a city with a perfect grid of streets. If you want to get from Point A to Point B, you can always find a "North-East" corner that is the exact meeting point of the two. The map is rigid and perfect.
• The New Way (Pseudo-Lattice): Imagine you are hiking in a forest. There are no grid lines. However, if you and a friend are at two different spots, you can still find a highest point you can both reach by going "up" (even if there are two different highest points, or a small ridge). You can also find a lowest point you can both reach by going "down."

The forest isn't a grid, but it has enough structure to guide you. You don't need a single, unique "North-East" corner; you just need some high ground and some low ground to anchor your direction.

What Did They Actually Do?

The authors rebuilt the entire GPS system using this "Forest Map" (Pseudo-Lattice) instead of the "City Grid" (Lattice).

1. They generalized the rules: They proved that even without a perfect grid, if you have these "highest" and "lowest" points, you can still predict that behavior will move in a specific direction (monotonically) when the environment changes.
2. They cracked the "Mixed Strategy" code: For the first time, they could apply these powerful prediction tools to games where players mix their strategies (like Poker). They showed that even in a "cloud" of random choices, there are "extreme" pure strategies that act as the boundaries. If the environment changes, the whole cloud of possibilities shifts up or down, bounded by these extremes.
3. They conquered "Perfect Equilibrium": This is the paper's biggest breakthrough. In game theory, a "Perfect Equilibrium" is a super-stable solution where players don't make silly mistakes (like "trembling hands"). Analyzing this usually requires looking at infinite layers of "what if" scenarios involving randomization. Because the math for randomization isn't a lattice, this was previously impossible to analyze with MCS. The authors used their new "Forest Map" to prove that these perfect equilibria do exist and do move predictably when conditions change.

Why Should You Care?

Think of this paper as upgrading the engine of a car that was stuck in the mud.

• Before: Economists could only analyze situations that were perfectly structured (like simple price wars or clear-cut production choices). If the situation involved risk, randomness, or complex strategies, they had to use weaker, less precise tools.
• After: They can now analyze anything that has a "highest" and "lowest" point, even if the middle is messy.

Real-world impact:

• Auctions: Predicting how bidders will change their random bidding strategies if the auction rules change.
• Finance: Understanding how investors shift their portfolios of risky assets when interest rates change.
• Policy: Designing better regulations for markets where companies use complex, randomized strategies to compete.

The Bottom Line

The authors took a rigid mathematical rule (the Lattice) that was blocking progress and replaced it with a flexible, robust rule (the Pseudo-Lattice). They showed that you don't need a perfect grid to navigate the economy; you just need enough structure to know which way is "up" and which way is "down."

This allows economists to finally apply their most powerful predictive tools to the messy, random, and complex realities of the real world.

Technical Summary

Here is a detailed technical summary of the paper "Monotone Comparative Statics without Lattices" by Yeon-Koo Che, Jinwoo Kim, and Fuhito Kojima.

1. Problem Statement

The theory of Monotone Comparative Statics (MCS) is a cornerstone of modern economic analysis, providing tools to predict how equilibrium outcomes change in response to exogenous parameters. Traditionally, this theory relies heavily on lattice theory (specifically, the assumption that the choice domain is a complete lattice).

• The Limitation: A lattice requires that any two elements have a unique least upper bound (join) and greatest lower bound (meet). This structure is essential for defining supermodularity and complementarity.
• The Gap: Many critical economic environments fail to satisfy the lattice property. The most prominent example is the space of mixed strategies (probability distributions over actions) ordered by stochastic dominance. Even if the underlying pure strategy space is a lattice, the space of lotteries is generally not a lattice because minimal upper bounds and maximal lower bounds may not be unique.
• Consequence: This limitation has prevented the application of MCS to mixed-strategy Nash equilibria and, crucially, to refinements like perfect equilibria (which require perturbing strategies to full support), leaving a significant gap in the theoretical toolkit for analyzing games with strategic complementarities in non-lattice domains.

2. Methodology and Core Concepts

The authors propose a generalization of the lattice framework by introducing a weaker structural condition called a Pseudo Lattice.

A. Pseudo Lattice Structure

• Definition: A partially ordered set $X$ is a pseudo lattice if for any $x, y \in X$, the set of minimal upper bounds ($x \vee y$) and the set of maximal lower bounds ($x \wedge y$) are non-empty. Unlike a standard lattice, these sets need not be singletons.
• Completeness: A complete pseudo lattice is chain-complete and possesses both a largest and a smallest element.
• Key Insight: The authors prove (Theorem 1) that any compact set with a largest and smallest element is a complete pseudo lattice. This includes the space of probability measures (mixed strategies) under stochastic dominance, provided the underlying space is compact and has extremal elements.

B. Generalized Order-Theoretic Tools

The paper redefines the necessary conditions for MCS and fixed-point theorems to operate on pseudo lattices:

1. Pseudo Quasi-Supermodularity: A generalization of the standard quasi-supermodularity condition. It ensures that the set of optimal choices maintains a specific order structure even when joins/meets are sets rather than points.
2. Set Orders: The authors utilize the Weak-Set (WS) order and the Pseudo Strong-Set (pSS) order to compare sets of optima. They establish that while the strong-set order is lost in non-lattice domains, the weak-set order (and a "sandwich" property) suffices to derive monotonicity.
3. Generalized Fixed-Point Theorem: They extend Tarski's Fixed Point Theorem and Zhou's Theorem. They prove that a monotonic correspondence on a complete pseudo lattice has a non-empty set of fixed points that forms a complete pseudo lattice (admitting maximal and minimal elements).

3. Key Contributions and Results

I. Individual Choice and Optimization

• Generalized Monotonicity Theorem: The authors characterize the monotonicity of the set of maximizers $M(t)$ with respect to a parameter $t$. They show that if the objective function is pseudo quasi-supermodular and satisfies single-crossing, the set of optimal choices shifts monotonically in the weak-set order.
• Structure of Maximizers: The set of maximizers inherits the complete pseudo lattice structure of the domain, ensuring the existence of extremal (largest and smallest) optimal choices.

II. Nash Equilibria in Games

• Pseudo Quasi-Supermodular Games: They define a class of games where strategy spaces are complete pseudo lattices and payoffs satisfy pseudo quasi-supermodularity.
• Existence and Structure: Using the generalized fixed-point theorem, they prove that the set of pure-strategy Nash equilibria is non-empty and forms a complete pseudo lattice.
• Comparative Statics: If the game parameters shift to strengthen incentives for higher actions, the entire set of Nash equilibria shifts upward in the weak-set order.
• Application to Bertrand Games: They demonstrate the framework's power by analyzing generalized Bertrand games with discontinuous payoffs (e.g., pure Bertrand competition), which standard MCS methods cannot handle due to the failure of single-crossing in certain regions.

III. Mixed-Strategy Nash Equilibria

• Breaking the Lattice Barrier: This is a primary contribution. The authors show that the space of mixed strategies $\Delta(S)$, while not a lattice, is a complete pseudo lattice.
• Extremal Pure Strategies: They prove that for pseudo supermodular games, the set of mixed-strategy equilibria is "sandwiched" between extremal pure-strategy equilibria.
• Result: The set of mixed-strategy Nash equilibria possesses largest and smallest elements (which are pure strategies) and exhibits monotone comparative statics in the weak-set order. This resolves a long-standing issue where MCS was previously inapplicable to mixed strategies.

IV. Perfect Equilibria (Trembling-Hand)

• Novel Contribution: This is the paper's most significant theoretical advance. They provide the first general monotone comparative statics analysis of perfect equilibria.
• Methodology: Perfect equilibria are defined as limits of Nash equilibria in a sequence of perturbed games where players are constrained to play full-support strategies.
• Mechanism:
  1. In each perturbed game, the constrained strategy space is a complete pseudo lattice (but not a lattice).
  2. The authors establish the existence of constrained-pure extremal equilibria for these perturbed games.
  3. They show these extremal equilibria shift monotonically with parameters.
  4. By taking the limit, they prove that perfect equilibria exist in pure strategies and that the set of perfect equilibria inherits the monotone comparative statics property.
• Significance: Previous literature (e.g., Simon and Stinchcombe, 1995) established existence but could not derive comparative statics for perfect equilibria due to the lack of a lattice structure in the limit.

4. Significance and Implications

• Theoretical Expansion: The paper fundamentally decouples the power of Monotone Comparative Statics from the rigid requirement of a lattice structure. It demonstrates that the core logic of complementarity and monotonicity relies only on the existence of extremal bounds, not their uniqueness.
• Unified Framework: It provides a unified approach to analyze pure strategies, mixed strategies, and equilibrium refinements (perfect equilibria) within a single mathematical framework.
• Practical Applications:
  • Game Theory: Enables rigorous analysis of strategic randomization and equilibrium refinements in supermodular games.
  • Information Design: The framework applies to spaces of information structures (ordered by mean-preserving spreads/convex order), which are pseudo lattices but not lattices.
  • Bayesian Games: Suggests a path for analyzing games with incomplete information where distributional strategies lack lattice properties.
• Robustness: The results hold under weaker continuity assumptions than previous literature, making the theory applicable to games with discontinuous payoffs (like Bertrand competition).

Conclusion

Che, Kim, and Kojima successfully generalize the foundations of Monotone Comparative Statics. By replacing the lattice assumption with the more general pseudo lattice property, they unlock the ability to perform comparative statics on mixed strategies and perfect equilibria—domains previously considered intractable for this methodology. This work bridges a critical gap between order-theoretic methods and the stochastic nature of modern game-theoretic analysis.