Delegation without Priors

arXiv:2609.35156 · econ.TH · Submitted 2026-09-28 · Read on arXiv

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: We're looking at the paper: "Delegation without Priors".

Warren: Here's the gist of the paper. A principal relies on information held by a biased agent in decision-making to design a delegation rule that minimizes worst-case regret without a prior over the circumstances that may arise <ref:2609.35156#pg2>.

Ruth: The optimal delegation combines a default action, a fixed gap above it, and a higher discretionary interval with random boundaries to hedge exposure across potential worst-case states <ref:2609.35156#pg2>.

Warren: The Core Problem and Setup The central design problem is how much discretion to grant when an agent has private information that may lead them to steer the decision toward their own interests when preferences are imperfectly aligned <ref:2609.35156#pg3>.

Ruth: The principal's decision is modeled as a real number whose desirability depends on a one-dimensional payoff-relevant state which lies in a known interval but for which the principal has no prior <ref:2609.35156#pg3>.

Title and authors: Ruth: So we’re looking at this paper titled Delegation without Priors, and the authors are Wei He, Fei Li, and Jiangtao Li. It sounds like it tackles the problem of when you let someone who has private information make decisions for you without knowing what kind of trouble might happen.

Warren: Yeah, that’s right. The core idea is designing a delegation rule that keeps you from taking too much risk when your decision depends on something the agent knows better than you do. It’s about minimizing the worst-case regret without needing a specific guess about what the future looks like.

Iris: Exactly. The title itself sets up this ambiguity, this lack of prior information, which is a big deal in real-world delegation where preferences aren't perfectly lined up between the principal and the agent.

Ruth: Right, so instead of trying to guess what’s going to happen—like when a market might crash or boom—this paper focuses on building a rule that works no matter what happens. It suggests a specific structure for this rule.

Warren: And that structure is this default action, plus a fixed gap above it, and then a wider discretionary interval with random boundaries. It’s designed to handle the uncertainty by separating whether you even leave the default and how much room the agent has once they do leave it.

Iris: That separation is key. It means the default and that gap act as a hurdle, limiting big adjustments when there isn't much reason for them, while that higher interval gives flexibility when you actually need to make a significant change.

Ruth: So, the paper proposes this specific structure to handle the uncertainty without needing a prior about future circumstances. This sounds like a practical tool for any situation where you’re delegating something important.

The paper's summary: Warren: The summary of Delegation without Priors explains that the optimal way to deal with this is not just one simple rule, but a mechanism that matches the agent’s conditional mean and variance schedules with your own. It’s about finding the best action lottery at every state while keeping things incentive compatible across all possible state distributions.

Iris: That sounds like a very technical way of saying they are optimizing the delegation strategy to be robust against any distribution of future states, which is what ambiguity means in this context.

Ruth: I see. So it’s not just about finding one perfect rule for one specific future scenario, but finding a rule that works well across all possible futures simultaneously. It’s a kind of hedging mechanism built right into the design itself.

Warren: Right, and they find that this approach helps solve the central problem: how much discretion to grant when preferences are imperfectly aligned. The default and gap structure exploits the partial alignment of preferences, creating a hurdle for leaving the baseline action that only kicks in when it makes sense relative to how much the agent wants to move up as you move up.

Iris: So it’s showing that even with imperfect alignment, you can design a mechanism where the incentive logic naturally produces this default-gap-interval structure, and then they randomize the boundaries of that higher interval to hedge your exposure.

Ruth: It seems like they are using randomization in a very specific way—randomizing the floor and ceiling of that discretionary interval. That’s how you hedge against those potential worst-case states the principal might face.

The paper's improvements: Iris: One thing I noticed is that the authors show how this default and gap structure also offers a broader view of familiar delegation arrangements, suggesting it covers many existing ways people delegate decisions. That’s a nice practical point for applying the theory.

Ruth: It’s not just about creating a new mechanism; it’s about understanding why existing structures work, even if they aren't perfectly optimal under this model. The way the incentive logic exploits that partial alignment is what drives the whole design forward.

Warren: And they show how this structure helps separate two different things: it separates whether you are going to depart from the default and it separates how much discretion you get once you’ve decided to depart, which is a nice way to look at the trade-off.

Iris: Exactly. And then they show that by using randomization in this specific manner, they can reproduce the responsiveness and dispersion of the unrestricted optimum, even though they are dealing with a constrained setting.

Ruth: That means you get the benefits of being able to adjust your decision when things get really big, while still having a safe baseline when things are small or uncertain. It’s about getting both responsiveness and safety at once.

Conclusion: Warren: So to wrap up, the main implication of Delegation without Priors is that the optimal rule is this default-gap-interval structure using a higher interval with random boundaries. It separates access to discretion from the scope of discretion, which is a really clean way to think about it.

Ruth: And they show that this structure arises because incentive compatibility requires responsiveness to be reflected in action dispersion, and randomization implements both the optimal mean and variance for that kind of robust design. It’s not just a one-size-fits-all answer.

Iris: I think what stands out is how this model provides a microfoundation for frictions where incentive considerations in delegated capital allocation can make intermediate investment scales unavailable, which is something you see in real economic life.

Warren: Right, and they also mention that the worst-case regret actually increases as the bias gets stronger, and when the bias is near one, the stochastic value approaches a centralized guarantee of one/four.

Ruth: So it’s not a perfect solution everywhere—the gain from randomization doesn't vanish at both ends of that bias range, but it still gives you a robust framework for making decisions under ambiguity. That’s what we have for Delegation without Priors.

The Chinese University of Hong Kong · University of North Carolina at Chapel Hill

econ.TH

Submitted: 2026-09-28

Updated: 2026-11-08

License: http://creativecommons.org/publicdomain/zero/1.0/

Importance score: 99/100

The gist: A principal relies on information held by a biased agent in decision-making to design a delegation rule that minimizes worst-case regret without a prior over the circumstances that may arise…

Key concepts

Robust Decision-Making Under Prior Uncertainty
This concept focuses on designing delegation rules that perform well even when the principal has no specific knowledge about future conditions or states of the world. The goal is to minimize the worst possible loss (regret) across all possibilities, ensuring a reliable outcome regardless of what happens next.
Default-Gap-Interval Structure
This is the specific structure of an optimal delegation rule. It separates decisions into three parts: a common default action, a fixed gap above it that limits small deviations, and a higher discretionary interval with random boundaries for larger adjustments. This structure balances stability with necessary flexibility.
Optimal Stochastic Mechanism Design
This involves finding the best way to design an agent's decision-making process (a mechanism) when the agent has private information. The optimal method matches the conditional mean and variance of the agent's choices at every state, ensuring incentive compatibility while achieving a robust guarantee.
Lumpy Resource Allocation Modeling
The paper shows how this delegation framework can model real-world frictions in capital allocation. Because of the constraints imposed by incentive considerations, intermediate investment scales might become unavailable. This structure helps explain why investments tend to be in discrete, lumpy sizes rather than continuous values.

Terminology

Summary

A principal relies on information held by a biased agent in decision-making to design a delegation rule that minimizes worst-case regret without a prior over the circumstances that may arise <ref:2609. p. 0

Where the paper says it

Delegation without Priors∗ Wei He Fei Li Jiangtao Li Hanqing Ye September 28, 2026 Abstract A principal relies on information held by a biased agent in decision-making. Without a prior over the circumstances that may arise, she designs a delegation rule to minimize the worst-case regret.

35156#pg2>. The optimal delegation combines a default action, a fixed gap above it, and a higher discretionary interval with random boundaries to hedge exposure across potential worst-case states <ref:2609. p. 0
Where the paper says it

Without a prior over the circumstances that may arise, she designs a delegation rule to minimize the worst-case regret. Optimal delegation combines a default action, a fixed gap above it, and a higher discretionary interval with random boundaries.

35156#pg2>.

The Core Problem and Setup

The central design problem is how much discretion to grant when an agent has private information that may lead them to steer the decision toward their own interests when preferences are imperfectly aligned <ref:2609.35156#pg3>. The principal's decision is modeled as a real number whose desirability depends on a one-dimensional payoff-relevant state which lies in a known interval but for which the principal has no prior <ref:2609.35156#pg3>. The study focuses on a canonical quadratic-loss environment where both parties prefer higher actions as the state increases, while the agent’s preferred action is shifted upward relative to the principal's <ref:2609.35156#pg3>.

The Optimal Rule Structure

The optimal robust delegation rule has a simple structure consisting of a common default action, a gap above it, and a discretionary interval with random boundaries <ref:2609.35156#pg4>. This structure separates two margins: whether the decision departs from the default and how much discretion the agent receives conditional on doing so <ref:2609. p. 0

Where the paper says it

Optimal delegation combines a default action, a fixed gap above it, and a higher discretionary interval with random boundaries. This structure separates two margins: whether the decision departs from the default and how much discretion the agent receives conditional on doing so.

35156#pg4>. The default and gap limit excessive adjustment when little adaptation is warranted, while the higher interval preserves flexibility when larger adjustments are needed <ref:2609. p. 0
Where the paper says it

This structure separates two margins: whether the decision departs from the default and how much discretion the agent receives conditional on doing so. The default and gap limit excessive adjustment when little adaptation is warranted, while the higher interval preserves flexibility when larger adjustments are needed.

35156#pg4>. Randomizing the discretionary interval’s floor and ceiling hedges the principal’s exposure across potential worst-case states <ref:2609. p. 0
Where the paper says it

The default and gap limit excessive adjustment when little adaptation is warranted, while the higher interval preserves flexibility when larger adjustments are needed. Randomizing the discretionary interval’s floor and ceiling hedges the principal’s exposure across potential worst-case states.

35156#pg4>.

Mechanism Design Insights

The incentive logic exploits the partial alignment of preferences, where the default and gap create a hurdle for departing from the baseline <ref:2609.35156#pg5>. The optimal robust mechanism uses increasing dispersion whenever it remains responsive to the agent’s information, stopping when further responsiveness would require too much additional dispersion <ref:2609. p. 2

Where the paper says it

Under worst-case evaluation, doing so leaves the principal fully exposed to the preference conflict in any state, which is undesirable. The optimal robust mechanism therefore uses increasing dispersion whenever it remains responsive to the agent’s information.

35156#pg5>. The deterministic outcome at the lowest state generates the common default and gap, while randomizing the boundaries of the higher interval reproduces the responsiveness and dispersion of the unrestricted optimum <ref:2609. p. 3
Where the paper says it

A random delegation rule implements the optimal mechanism. The deterministic outcome at the lowest state generates the common default and gap, while randomizing the boundaries of the higher interval reproduces the responsiveness and dispersion of the unrestricted optimum.

35156#pg5>.

Implementation via Randomization

The optimal rule is implemented by a random delegation rule, which is a special stochastic mechanism where conditional on each realized set D, the induced action aD (theta) must be the agent’s optimal choice from that set at every state <ref:2609.35156#pg18>. This construction shows that despite the stronger ex post incentive requirement imposed by delegation, the principal loses nothing relative to unrestricted stochastic mechanism design <ref:2609. p. 3

Where the paper says it

Thus, despite the stronger ex post incentive requirement imposed by delegation, the principal loses nothing relative to unrestricted stochastic mechanism design. The discussion above shows that both randomization and the gap in the delegation set are essential for implementing the optimal robust mechanism.

35156#pg5>. The rule preserves both players’ expected payoffs at every state and therefore attains the unrestricted robust guarantee <ref:2609. p. 1
Where the paper says it

The criterion instead provides a prior-free guarantee: a rule that performs well under it does so regardless of which state arises. Our main result characterizes the optimal robust delegation rule and shows that it attains the value of the unrestricted stochastic mechanism-design problem.

35156#pg19>.

Key Findings on Optimality

The optimal random delegation rule has a common default action a0 = b(1 − 1/2 e(1−1/b)) and excludes all actions in (a0, 2b − a0) almost surely <ref:2609.35156#pg10>. This default-and-gap structure has a simple observable implication: investment is lumpy in scale <ref:2609. p. 8

Where the paper says it

Every optimal random delegation rule has the same common default 𝑎0 = 𝑏(1 − 1 2 𝑒 1−1/𝑏 ) and excludes all actions in (𝑎0, 2𝑏 − 𝑎0) almost surely. The default-and-gap structure has a simple observable implication: investment is lumpy in scale.

35156#pg9>. The quantitative macro investment literature commonly captures such behavior through nonconvex adjustment costs <ref:2609. p. 8
Where the paper says it

This is consistent with the well-documented tendency of capital expenditure to cluster at very low and relatively high levels, with comparatively little activity at intermediate scales (Doms and Dunne, 1998; Cooper and Haltiwanger, 2006). The quantitative macro investment literature commonly captures such behavior through nonconvex adjustment costs.

35156#pg9>. The minimax regret achieved by this rule is Vd = a 2 0 < min(b squared, 1/4) which is the optimum of the unrestricted stochastic mechanism-design problem <ref:2609. p. 8
Where the paper says it

Moreover, the corresponding minimax regret is 𝑉𝐷 = 𝑎 2 0 < min{𝑏 2 , 1/4}, which is the optimum of the unrestricted stochastic mechanism-design problem. The optimal rule has three components: a common default action, a gap above the default, and a discretionary interval with random boundaries.

35156#pg10>. The structure separates access to discretion from the scope of discretion <ref:2609. p. 4
Where the paper says it

design margin: partial alignment allows a common default and a gap to govern whether the agent departs from the baseline, while a separate discretionary interval governs how much latitude he receives once he does. Robust delegation therefore separates access to discretion from the scope of discretion.

35156#pg4>.

Conclusion and Implications

The paper concludes that the optimal rule separately governs whether the decision leaves a common default and how much discretion follows, using a gap and a higher interval with random boundaries <ref:2609.35156#pg20>. The mechanism-design relaxation explains this structure because incentive compatibility makes responsiveness require action dispersion, and random delegation implements the optimal mean and variance <ref:2609. p. 5

Where the paper says it

Our analysis enriches this literature by studying a delegation environment without transfers, where incentive compatibility creates an endogenous cost of responsiveness: making the decision more responsive to the agent’s information requires additional action dispersion. Solving the unrestricted robust mechanism problem reveals how this tradeoff shapes the optimal stochastic allocation and provides the benchmark that guides our characterization of optimal delegation.

35156#pg20>. Deterministic delegation forces an all-or-nothing choice between full discretion and a fixed decision <ref:2609. p. 7
Where the paper says it

The minimax constant action is 1/2, yielding worst-case regret 1/4. Robust deterministic delegation is therefore all-or-nothing: full delegation or no delegation.

35156#pg9>. The paper provides a principal–agent microfoundation for frictions where incentive considerations in delegated capital allocation can endogenously make intermediate investment scales unavailable <ref:2609. p. 9
Where the paper says it

Our mechanism provides a principal–agent microfoundation for such frictions: incentive considerations in delegated capital allocation can endogenously make intermediate investment scales unavailable. This mechanism has a direct institutional counterpart in firms’ internal capital-budgeting processes.

35156#pg9>. The worst-case regret is increasing in the bias, and near one, the stochastic value approaches the centralized guarantee 1/4. p. 7
Where the paper says it

Any deterministic rule that remains responsive to the agent’s information must expose the principal to the agent’s full bias somewhere, generating worst-case regret of at least 𝑏 2 . Further restrictions on the feasible set can only introduce additional distortions across states, so full delegation 𝐷 = [0, 1] attains the best robust guarantee among responsive deterministic rules.

The gain from randomization vanishes at both extremes of the bias range. p. 7
Where the paper says it

𝐷2 has the opposite profile: it exposes the principal to the agent’s bias at low states, but preserves the flexibility needed when the state is high. Randomization hedges these opposing statewise exposures.

The optimal rule separates access to discretion from the scope of discretion <ref:2609. p. 4
Where the paper says it

Robust delegation therefore separates access to discretion from the scope of discretion. Second, our saddle-point argument endogenously identifies a least-favorable distribution against which the robust rule is optimal.

35156#pg4>.

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

  1. Robust decision-making under prior uncertainty: The improved system can design delegation rules that minimize worst-case regret without needing a specific prior over future circumstances, by using a default-gap-interval structure to separate the threshold for departure from the scope of discretion.

  2. Optimal mechanism design under ambiguity: The system can characterize optimal stochastic mechanisms by matching conditional mean and variance schedules, allowing it to select the best action lottery at every state while ensuring incentive compatibility across all possible state distributions.

  3. Lumpy resource allocation modeling: The AI can model frictions in capital allocation or project evaluation where intermediate investment scales are unavailable, mirroring how incentive considerations in delegated capital allocation can endogenously make intermediate investment scales unavailable.

Abstract

A principal makes decisions relying on the private information held by a biased agent. Lacking a prior over the circumstances that may arise, she seeks a delegation rule that minimizes worst-case regret relative to the complete-information benchmark. The optimal delegation mechanism governs two distinct margins: whether to depart from a common default action, and how much discretion to grant conditional on departure. This rule pairs the default action with an action gap, followed by a discretionary interval with randomized boundaries designed to balance regret across states. This structure is globally optimal within the class of all stochastic, incentive-compatible mechanisms.

Sources

Related papers