Bang-Bang Correlation Sensitivity: A Local-to-Global Dichotomy for Correlation Preference
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Bang-Bang Correlation Sensitivity".
Warren: Here's the gist of the paper. Binary choice under risk can depend on how alternative outcomes are jointly distributed, and this paper shows that continuity, completeness, and betweenness impose a local-to-global restriction on this dependence The resulting framework distinguishes correlation effects from departures from classical independence, and yields testable restrictions linking choices between identically distributed lotteries to correlated Allais behavior Local-to-Global Dichotomy The first result establishes a dichotomy for correlation sensitivity, showing that under continuity, completeness, and betweenness, exactly one of two alternatives holds: either the preference is correlation insensitive at every p in Delta or it is correlation sensitive at every p in Delta This means that if the decision maker is indifferent between two copies of one full-support lottery under every possible coupling, then the pairing of received and forgone outcomes is irrelevant to choice throughout the domain The second theorem focuses on the correlation-sensitive branch, proving that continuity, completeness, and betweenness are necessary and sufficient for a nondegenerate correlation weighted utility (CWU) representation on the entire joint-distribution simplex Characterization of Utility Representations The paper introduces several representations to characterize preference structures based on the axioms (i) Correlation Weighted Utility (CWU): A correlation preference admits a CWU representation if there are skew-symmetric matrices ϕ and ψ such that π ∈ Π ⇐⇒ U(π):= ⟨ϕ, π⟩ + πT1ψπ2 ≥ 0 (ii) Correlation Expected Utility (CEU): Strong independence yields CEU below, which includes the relevant binary ST and RT specifications of Lanzani (2022) (iii) Skew-Symmetric Bilinear Utility (SSB): Imposing both strong independence and correlation insensitivity yields expected utility (EU), the common specialization of CEU and SSB Empirical Predictions and Testing The framework yields testable links between experimental research on choice under risk, examining preferences between identically distributed alternatives and the role of correlation in Allais behavior (i) SML-Allais Prediction: For a CWU preference DM with Fz above, the probabilities of the fanning-out choices (BL, AH) are both weakly decreasing (resp.
Ruth: increasing, constant) in the level of correlation λ if the decision maker exhibits ISOD (resp.
Warren: DSOD, CSOD) in the SML choice (ii) HILO Design Test: The joint SML–HILO test suggests that coupling changes at fixed marginals isolate correlation contrasts, while on the three-outcome support the SML cycle identifies the sign of ∆ϕ Extension to General Outcome Spaces The analysis is extended to a general metrizable outcome space by retaining only finitely supported lotteries as in Lanzani (2022) A support-lattice dichotomy is introduced, stating that for every nonempty finite S ⊆ X, exactly one of two branches holds regarding correlation insensitivity on the finite face J(S) This extension confirms that insensitivity is relative to S and does not rule out sensitivity if the outcome support is enlarged Distinguishing Utility Specifications The paper distinguishes between the four groups of preferences: EU, SSB, CEU, and CWU based on specific conditions imposed on the kernels ϕ and ψ For instance, sensitive CEU requires ϕ ≠ 0 while SSB has ϕ = 0 The joint SML–HILO test provides a two-by-two categorization of these groups The paper concludes that correlation insensitivity is either void or global under the axioms, and the framework provides a uniform representation connecting CWU, CEU, SSB, and EU The local-to-global implication extends to choice among multiple alternatives by showing that indifference among all decisions under every coupling of one common full-support marginal makes choice invariant to correlation across all marginal profiles This establishes a link between local SML response and cross-coupling Allais patterns The joint SML–HILO test provides distinct sources of identification for these behavioral types The analysis is further extended to a context-dependent counterfactual utility model, showing that indifference among all decisions under every coupling of one common full-support marginal makes choice invariant to correlation across all marginal profiles This extension distinguishes sensitivity to correlation from dependence on the marginal context The discussion concludes by showing that the SML-CI condition at some full-support marginal is equivalent to global CI in a multi-marginal setting References Allais, M., Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école américaine Bell, D.
Ruth: E., Regret in decision making under uncertainty Bohren, J.
Title and authors: Ruth: So we're looking at this paper today, "Bang-Bang Correlation Sensitivity: A Local-to-Global Dichotomy for Correlation Preference." It basically says that whether your choice depends on the correlation between outcomes is either always true across everything or never true across everything, given some basic rules about how we think.
Warren: Right, so it’s drawing a line between preferences that are sensitive to correlation and those that aren't. That sounds like a neat way to organize the different kinds of choice models we already have in mind.
Iris: The authors are focusing on how continuity, completeness, and betweenness—which are just technical ways of describing the structure of our choices—impose this restriction on the joint distribution of outcomes for binary choices under risk.
Ruth: Exactly. They show that if you have a decision maker who is indifferent between two copies of one full-support lottery under every possible way those outcomes can be coupled, then the correlation between them doesn't matter for their choice anywhere in the whole domain.
Warren: That’s a big deal because it suggests that if we find someone who is indifferent to these paired lotteries everywhere, we don't need to worry about how correlated the actual results are when they make a choice.
Iris: But the interesting part is when you look at the correlation-sensitive side. The three axioms mentioned—continuity, completeness, and betweenness—they aren't just nice conditions; they actually characterize a specific kind of utility representation called correlation weighted utility.
Ruth: So if someone has that kind of preference structure, they can be described using this correlation weighted utility model which mixes comparing outcomes at the same time with a simple bilinear function involving the marginal lotteries.
Warren: And in the opposite case, when it's correlation insensitive, another axiom comes into play that leads to Fishburn’s skew-symmetric bilinear representation. It’s like having two different mathematical tools for describing those two very different kinds of preferences depending on which condition holds true.
Iris: That distinction helps us separate the effects of correlation from just moving away from classical independence assumptions, which is a key goal of this work in characterizing utility representations.
Ruth: And they extend this characterization to finitely supported lotteries on general metrizable outcome spaces, meaning it doesn't just apply to simple cases but more general ones where the outcomes might be spread out in some way.
Title and authors: Warren: That extension makes the framework more robust, showing that the insensitivity we talk about is relative to the specific set of outcomes you are looking at, not a universal property across every possible outcome space.
Iris: The paper also distinguishes between four types of preferences: expected utility, skew-symmetric bilinear utility, correlation expected utility, and correlation weighted utility based on conditions on those kernels phi and psi.
Ruth: And they link this to other concepts we've discussed before like strong independence leading to correlation expected utility which includes some of the relevant binary specifications from Lanzani.
Warren: It connects the dots between these different mathematical descriptions, showing how they all relate under specific conditions on those underlying structures.
Iris: The empirical predictions are quite concrete as well, suggesting testable links between choice experiments and Allais behavior when we look at preferences between identically distributed alternatives.
Ruth: Specifically, for a correlation weighted utility preference with a certain condition called Fz above, the probabilities of the fanning-out choices, like BL or AH, will either be weakly decreasing or increasing depending on whether the decision maker exhibits ISOD, DSOD, or CSOD in their same-marginal lotteries.
Warren: So if you see those specific patterns in how people distribute their fanning choices as correlation changes, it points directly to that underlying correlation weighted utility structure.
Iris: And they also have a joint SML HILO test that suggests coupling changes at fixed marginals isolate the correlation contrasts, while on the three-outcome support, the SML cycle helps identify the sign of phi.
Ruth: That test sounds like it could be very useful for experimentalists because it gives them a way to disentangle whether they are seeing a correlation effect or just some artifact of how we set up the comparison.
Warren: It allows researchers to use both designs—the same marginal design and the joint SML HILO test—to get a two-by-two categorization of these preference groups, which is pretty powerful for experimental validation.
Iris: The conclusion boils down to saying that correlation insensitivity is either void or global under the three main axioms, and their framework provides a uniform way to connect correlation weighted utility with expected utility and skew-symmetric bilinear utility.
Title and authors: Ruth: It establishes that indifference among all decisions under every coupling of one common full-support marginal makes the choice invariant to correlation across all marginal profiles, which is a strong link between local same-marginal response and cross-coupling patterns.
Warren: That means you can use local SML response as a starting point to predict how choices behave when you consider multiple alternatives together under different correlations.
Iris: And they extend this even further into a context-dependent counterfactual utility model, showing that indifference among all decisions under every coupling of one common full-support marginal makes choice invariant to correlation across all marginal profiles.
Ruth: That final extension is interesting because it distinguishes sensitivity to correlation from dependence on the specific marginal context you are working with.
Warren: It's a useful distinction for modeling, showing that the same pattern of indifference can mean different things depending on whether you're looking at one decision or multiple decisions at once.
Iris: So, in short, this paper provides a local-to-global restriction on how correlation dependence works by linking three structural axioms to specific mathematical representations of utility.
Ruth: It gives us a clear framework for separating the effects of correlation from other departures from classical independence assumptions in choice under risk.
Warren: The empirical predictions are testable through those SML Allais and HILO designs, which offer distinct ways to identify these preference structures in action.
Iris: The main limitation they mention is that the analysis on a general metrizable outcome space only retains finitely supported lotteries, meaning it stops working if we have infinitely many possible outcomes.
Ruth: So while the theory is powerful for finite settings, we need to be careful when applying it to scenarios with an infinite number of possibilities.
Warren: It sets up a clear boundary for where this framework is most applicable in practice, which is good because it tells us exactly where we can expect these results to hold true.
Iris: Overall, the paper establishes that correlation insensitivity is either void or global under those conditions, unifying several important utility specifications into one coherent structure.
Ruth: And by connecting local same-marginal responses to cross-coupling Allais patterns, they’ve provided a new way to test these theories in choice under risk.
Warren: It connects the mathematical structure of preferences to actual behavioral patterns seen in experiments, which is always a good sign for quantitative modeling.
The paper's summary: Ruth: So, to recap, this paper is trying to sort out whether someone's choice depends on how correlated the outcomes are between two different options, and they find that this dependency is either global or not at all because of some basic rules about how we structure our preferences.
Warren: Yeah, it’s basically saying that correlation insensitivity has to be either a universal truth across everything or it doesn't exist at all under these axioms. That’s a pretty strong statement on the structure of choice.
Iris: What’s interesting is how they build this framework, using things like skew-symmetric matrices and expected utility representations to map out the different types of preferences—like correlation weighted utility versus expected utility—based on those structural conditions.
Ruth: Exactly, it’s not just one model; they’re creating a classification system for these preference structures based on how sensitive they are to that correlation. You get this distinction between something called correlation insensitive and something sensitive in every possible scenario.
Warren: That leads into the empirical side where they suggest we can actually test this in the real world, linking stuff like Allais behavior to how much correlation is present. They have these specific predictions about how choices—like fanning out—should change as the correlation level shifts.
Iris: And they’re trying to be rigorous about this identification, using tests like the joint same-marginal and HILO test to see if a single mathematical structure can explain both local responses and cross-coupling patterns. That’s where you check if your chosen identification method actually works.
Ruth: The big implication for us is that it gives us a way to separate the effect of correlation itself from just moving away from classical independence assumptions, which is super helpful when we’re trying to model complex decision-making.
Warren: And they extend this idea beyond just two choices; they show you can use a more general version of expected utility that predicts choice invariance across all marginal profiles, which helps us distinguish between being sensitive to correlation and just being sensitive to the specific context of the marginal lotteries we’re looking at.
Iris: The limitation is that they stick to finitely supported lotteries on general metrizable outcome spaces, so if you’re dealing with an infinite number of possible outcomes, this specific framework doesn't cover that whole space.
Ruth: Still, it provides a very tight connection between those local same-marginal responses and the broader patterns seen in cross-coupling Allais behavior, which is something we can use to test these utility models empirically.
The paper's improvements: Tom: So, to wrap up, the authors suggest several ways to improve this framework, mainly by making it more flexible for different types of data structures and by allowing AI to estimate these preference types more directly.
Ruth: Right, they’re looking at three main improvements: first, they want to extend the analysis beyond simple finite sets of outcomes to a general metrizable outcome space, which makes the results apply in a wider variety of real-world scenarios where outcomes aren't just discrete points.
Warren: And second, they’re introducing this new testable link combining the same-marginal design with four different HILO situations, which should let researchers figure out if one single correlation kernel can explain both local responses and those cross-coupling Allais patterns we talked about earlier.
Iris: They’re also proposing a stochastic specification where the AI can estimate the type weights jointly with those kernel parameters, which means we could potentially estimate whether a preference is expected utility, skew-symmetric bilinear utility, or correlation weighted utility all at once.
Ruth: That’s significant because it moves us from just classifying preferences to actually estimating *which* preference structure someone has based on their choices under risk. It lets the AI figure out the underlying mathematical model itself.
Warren: From a quantitative side, this is useful because it gives us a way to predict behavioral movements based on whether someone shows local increasing or decreasing sensitivity as the dependence between outcomes changes from maximal positive dependence toward independence.
Iris: That directional prediction is really valuable for testing; if we see those specific patterns in fanning-out choices, we can have a hypothesis about the underlying utility structure before we even look at the correlation data.
Ruth: So, by improving the identification and allowing AI to estimate parameters within this unified framework, they’re building something that connects local decision-making directly to global patterns across different marginals.
Warren: It really suggests that we can use these local SML responses as a starting point to predict how choices behave when you consider multiple alternatives together under different correlations, which is a powerful way to see the whole picture.
Iris: The main caveat they flag is still that this extension to general spaces still relies on keeping only finitely supported lotteries, so we’re still limited in what it can perfectly model about infinite outcome possibilities.
Conclusion: Tom: So to wrap up, this paper on "Bang-Bang Correlation Sensitivity: A Local-to-Global Dichotomy for Correlation Preference" establishes that correlation insensitivity is either global or nonexistent under those basic axioms, and it provides a unified way to connect several different utility models.
Ruth: It really lays out how continuity, completeness, and betweenness act as these restrictions on the joint distribution of outcomes when we’re dealing with binary choices under risk. That’s the core mechanism they’re using.
Warren: And what I like is that it gives us a clear way to separate correlation effects from just departures from classical independence assumptions in choice under risk. It moves beyond just looking at one type of utility function and tries to organize them all together.
Iris: I think the main contribution is the link they draw between local same-marginal responses and those cross-coupling Allais patterns, which gives us a testable way to connect what people do when they make one choice versus how they behave when comparing multiple choices together.
Ruth: Exactly, so we’ve got this uniform representation that connects correlation weighted utility with expected utility and skew-symmetric bilinear utility based on conditions on those kernels phi and psi. It’s a really neat way to structure the theory of preference.
Warren: And for the practical side, it suggests that if you see specific patterns in how choices distribute themselves as correlation changes, it points directly to that underlying correlation weighted utility structure we discussed earlier.
Iris: The extension they made to finitely supported lotteries is important because it shows this framework holds true even when the outcome space isn't perfectly continuous across the whole range.
Ruth: So ultimately, this paper gives us a robust tool for distinguishing between different types of correlation dependence in decision-making, moving us closer to a more precise model of how people make choices when risk is involved.
Warren: It’s solid work for understanding the structural constraints on preferences, which is something we need as we try to build models that can actually predict risky behavior without just relying on historical averages.
Iris: We’ll see how this local-to-global structure plays out when we look at more complex scenarios involving multiple alternatives and different marginal contexts in our next discussion.
Yi-Chun Chen, Soo Hong Chew, Xinhan Zhang
Department of Economics and Risk Management Institute, National University of Singapore · Center for Intelligence Economic Science, Southwestern University of Finance and Economics
econ.TH
Submitted: 2026-11-03
Updated: 2026-11-03
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
Importance score: 99/100
The gist: Binary choice under risk can depend on how alternative outcomes are jointly distributed, and this paper shows that continuity, completeness…
Key concepts
- Local-to-Global Dichotomy
- This result shows that under continuity, completeness, and betweenness axioms, correlation insensitivity must be either true everywhere or false everywhere across the entire decision space. This means a preference is either insensitive to correlation at every possible point or sensitive to it at every possible point.
- Correlation Weighted Utility (CWU)
- This utility representation allows for modeling preferences that depend on how outcomes are jointly distributed. It uses skew-symmetric matrices ($\phi$ and $\psi$) to define preference, distinguishing correlation effects from standard expected utility by capturing the joint distribution structure.
- SML-Allais Prediction
- This prediction links choice under risk (SML) with Allais behavior. For a CWU preference, it predicts that the probabilities of 'fanning-out' choices change in a predictable way (weakly decreasing or increasing) based on the level of correlation ($\lambda$), depending on whether the decision maker exhibits ISOD, DSOD, or CSOD behavior.
Terminology
Summary
Binary choice under risk can depend on how alternative outcomes are jointly distributed, and this paper shows that continuity, completeness, and betweenness impose a local-to-global restriction on this dependence The resulting framework distinguishes correlation effects from departures from classical independence, and yields testable restrictions linking choices between identically distributed lotteries to correlated Allais behavior
Local-to-Global Dichotomy
The first result establishes a dichotomy for correlation sensitivity, showing that under continuity, completeness, and betweenness, exactly one of two alternatives holds: either the preference is correlation insensitive at every p in Delta or it is correlation sensitive at every p in Delta This means that if the decision maker is indifferent between two copies of one full-support lottery under every possible coupling, then the pairing of received and forgone outcomes is irrelevant to choice throughout the domain The second theorem focuses on the correlation-sensitive branch, proving that continuity, completeness, and betweenness are necessary and sufficient for a nondegenerate correlation weighted utility (CWU) representation on the entire joint-distribution simplex
Characterization of Utility Representations
The paper introduces several representations to characterize preference structures based on the axioms
(i) Correlation Weighted Utility (CWU):
A correlation preference admits a CWU representation if there are skew-symmetric matrices ϕ and ψ such that π ∈ Π ⇐⇒ U(π):= ⟨ϕ, π⟩ + πT1ψπ2 ≥ 0
(ii) Correlation Expected Utility (CEU):
Strong independence yields CEU below, which includes the relevant binary ST and RT specifications of Lanzani (2022)
(iii) Skew-Symmetric Bilinear Utility (SSB):
Imposing both strong independence and correlation insensitivity yields expected utility (EU), the common specialization of CEU and SSB
Empirical Predictions and Testing
The framework yields testable links between experimental research on choice under risk, examining preferences between identically distributed alternatives and the role of correlation in Allais behavior
(i) SML-Allais Prediction:
For a CWU preference DM with Fz above, the probabilities of the fanning-out choices (BL, AH) are both weakly decreasing (resp. increasing, constant) in the level of correlation λ if the decision maker exhibits ISOD (resp. p. 2 If the decision maker is indifferent between two copies of one
full-support lottery under every possible coupling, then the pairing of received and
forgone outcomes is irrelevant to choice throughout the domain. Such indifference is
innocuous in the classical domain of marginal lotteries, where correlation insensitivity
is implicit and reflexivity makes the decision maker indifferent between two copies
of the same lottery. If the decision maker is indifferent between two copies of one
full-support lottery under every possible coupling, then the pairing of received and
forgone outcomes is irrelevant to choice throughout the domain. Such indifference is
innocuous in the classical domain of marginal lotteries, where correlation insensitivity
is implicit and reflexivity makes the decision maker indifferent between two copies
of the same lottery.Where the paper says it
Where the paper says it
(ii) HILO Design Test:
The joint SML–HILO test suggests that coupling changes at fixed marginals isolate correlation contrasts, while on the three-outcome support the SML cycle identifies the sign of ∆ϕ
Extension to General Outcome Spaces
The analysis is extended to a general metrizable outcome space by retaining only finitely supported lotteries as in Lanzani (2022) A support-lattice dichotomy is introduced, stating that for every nonempty finite S ⊆ X, exactly one of two branches holds regarding correlation insensitivity on the finite face J(S) This extension confirms that insensitivity is relative to S and does not rule out sensitivity if the outcome support is enlarged
Distinguishing Utility Specifications
The paper distinguishes between the four groups of preferences: EU, SSB, CEU, and CWU based on specific conditions imposed on the kernels ϕ and ψ For instance, sensitive CEU requires ϕ ≠ 0 while SSB has ϕ = 0 The joint SML–HILO test provides a two-by-two categorization of these groups
Conclusion
The paper concludes that correlation insensitivity is either void or global under the axioms, and the framework provides a uniform representation connecting CWU, CEU, SSB, and EU The local-to-global implication extends to choice among multiple alternatives by showing that indifference among all decisions under every coupling of one common full-support marginal makes choice invariant to correlation across all marginal profiles This establishes a link between local SML response and cross-coupling Allais patterns The joint SML–HILO test provides distinct sources of identification for these behavioral types The analysis is further extended to a context-dependent counterfactual utility model, showing that indifference among all decisions under every coupling of one common full-support marginal makes choice invariant to correlation across all marginal profiles This extension distinguishes sensitivity to correlation from dependence on the marginal context The discussion concludes by showing that the SML-CI condition at some full-support marginal is equivalent to global CI in a multi-marginal setting
References
Allais, M., Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école américaine
Bell, D. E., Regret in decision making under uncertainty
Bohren, J. A., Mental models of information: The role of attention and memory
Bordalo, P., Salience theory of choice under risk
Bruhin, A., Risk and rationality: The relative importance of probability weighting and choice set dependence
Chen, Y.-C., Correlation preference
Chew, S. H., A generalization of the quasilinear mean with applications to the measurement of income inequality and decision theory resolving the Allais paradox
Chew, S. H., The projective independence axiom
Chew, S. H., A generalization of the quasilinear mean with applications to the measurement of income inequality and decision theory resolving the Allais paradox
Chew, S. H., Empirical tests of weighted utility theory
Fishburn, P. C., Separation theorems and expected utilities
Fishburn, P. C., Nontransitive measurable utility
Fishburn, P. C., Transitive measurable utility
Fishburn, P. C., Nonlinear Preference and Utility Theory
Frydman, C., The role of salience in choice under risk: An experimental investigation
Galichon, A., Dual theory of choice with multivariate risks
Humphrey, S. J., Who accepts Savage’s axiom now? p. 7 This is the usual topological regularity condition, now imposed on joint distributions
rather than on pairs of marginal lotteries. Axiom 2 (Completeness).Where the paper says it
Koch, B., An axiomatic foundation for decisions with counterfactual utility
Kolotilin, A., Persuasion and matching: Optimal productive transport
Lanzani, G., Correlation made simple: Applications to salience and regret theory
Leland, J. W., Similarity judgments in choice under uncertainty: A reinterpretation of the predictions of regret theory
Loewenfeld, M., Salience or event-splitting? p. 0 The resulting
framework distinguishes correlation effects from departures from classical independence, and yields testable restrictions linking choices between identically
distributed lotteries to correlated Allais behavior. Keywords: correlation preference, regret theory, salience theory, optimal
transport, correlation weighted utility, correlation expected utility.Where the paper says it
Loewenfeld, M., Uncovering correlation sensitivity in decision making under risk
Loomes, G., Are preferences monotonic? p. 3 Our framework establishes a testable link between two strands of experimental
research on choice under risk. The first examines preferences between identically
distributed alternatives, from early tests of regret theory and similarity judgments
(Loomes et al.Where the paper says it
Loomes, G., Regret theory: An alternative theory of rational choice under uncertainty
Loewes, M., A generalization of the quasilinear mean with applications to the measurement of income inequality and decision theory resolving the Allais paradox
Machina, M. J., A more robust definition of subjective probability
Nakamura, Y., Bilinear utility and a threshold structure for nontransitive preferences
Nelsen, R. B., An introduction to copulas
Ostermair, C., An experimental investigation of the Allais paradox with subjective probabilities and correlated outcomes
Samuelson, P. A., Probability, utility, and the independence axiom
Starmer, C., Testing new theories of choice under uncertainty using the common consequence effect
Vohra, R. V., Mechanism Design: A Linear Programming Approach, volume 47
Yaari, M. E., The dual theory of choice under risk
--- Page 30 ---
The paper is titled Bang-Bang Correlation Sensitivity: A Local-to-Global Dichotomy for Correlation Preference
and it matters because it establishes a local-to-global dichotomy for correlation preference, showing that correlation insensitivity is either void or global under the axioms
The framework yields testable links between experimental research on choice under risk, examining preferences between identically distributed alternatives and the role of correlation in Allais behavior
For a CWU preference DM with Fz above, the probabilities of the fanning-out choices (BL, AH) are both weakly decreasing (resp. increasing, constant) in the level of correlation λ if the decision maker exhibits ISOD (resp.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Correlation preference modeling can be refined to distinguish between correlation effects and departures from classical independence by using a framework that organizes familiar models along two dimensions:
the strength of independence and the role of correlation.
This allows AI systems to classify choice behaviors based on whether a preference iscorrelation sensitive
orcorrelation insensitive.
-
A new testable link between choices can be established by combining the same-marginal design of Loewenfeld and Zheng with the four HILO situations. This joint test allows researchers to determine if a single correlation kernel accounts for both local same-marginal responses and cross-coupling Allais patterns, separating these effects from the marginal correlation term.
-
AI systems can perform type estimation by using a stochastic specification where
type k the choice probability Pr(A π, k) = Λ(σkUk(π))
to estimate type weights jointly with kernel parameters. This allows for the estimation of EU, SSB, CEU, and CWU groups within a single framework. -
The model can predict behavioral movements based on whether a decision-maker exhibits
local ISOD (increasing sensitivity to outcome differences), local CSOD (constant sensitivity), or local DSOD (decreasing sensitivity)
in the SML choice. This allows for directional predictions regarding the movement of choices as dependence moves from maximal positive dependence toward independence. -
The framework extends to choice among multiple alternatives using a generalization of CEU that compares multiple correlated lotteries, allowing for
choice invariant to correlation across all marginal profiles.
This distinguishes sensitivity to correlation from dependence on the marginal context in multi-alternative decision-making scenarios.
Abstract
Binary choice under risk can depend on how the alternative outcomes are jointly distributed. We show that in a finite outcome space, continuity, completeness, and betweenness impose a local-to-global restriction on this dependence: indifference between two alternatives with one common marginal distribution, however their outcomes are jointly distributed, implies correlation insensitivity for all lotteries supported on the same outcomes. In the correlation-sensitive case, these three axioms alone characterize correlation weighted utility, which combines statewise comparisons with a bilinear function of the marginal lotteries. In the correlation-insensitive case, an additional projective independence axiom yields Fishburn's (1982) skew-symmetric bilinear representation. Strong independence reduces correlation weighted utility to Lanzani's (2022) correlation expected utility representation. We extend the correlation-sensitive characterization to finitely supported lotteries on general metrizable outcome spaces. The resulting framework distinguishes correlation effects from departures from classical independence, and yields testable restrictions linking choices between identically distributed lotteries to correlated Allais behavior.