When a Clearer Forecast Can Cloud an Expert’s Advice

A headquarters deciding how much capacity to build needs two things: a view of future demand and a regional manager’s assessment of local suppliers and crews. A strong demand forecast could help with the first question but complicate the second. If the manager benefits from a larger division, optimistic news may widen the gap between her preferred expansion and headquarters’ choice, making her account of local conditions harder to trust.

That is the starting point for Designing Public Information for Strategic Advice, a theoretical paper by Vasudha Jain of the Indian Institute of Technology Kanpur and Jaideep Roy of the University of Bath. The paper was submitted to arXiv on November 5, 2026; the landing page does not indicate journal acceptance. It develops a model rather than measuring how actual firms or governments communicate.

Designing the conversation

The authors model a decision maker, or principal, who commits to a study of future demand before an expert offers advice. The expert privately knows local conditions, which are assumed independent of demand. Both see the study’s report; the principal then chooses an irreversible action before demand is realized. The expert’s preferred action responds more strongly to demand than the principal’s, so a report changes their expected conflict even though it reveals nothing directly about local conditions.

In the baseline, the report is also the principal’s only demand information. More precise reporting can improve the forecast, but its effect on the expert’s incentives can make advice more or less useful. The authors use a model with a uniformly distributed local condition and quadratic losses for decisions away from each party’s preferred action. Their main calculations assume two possible demand-related conflict states.

The mathematical technique is information design: the principal chooses a rule for generating reports, subject to the constraint that the average of the beliefs those reports produce must equal the prior belief. The authors calculate the value of advice at each possible reported level of expected conflict, then identify the mix of reports that gives the principal the highest expected payoff. They also map outcomes numerically across model parameters.

Why partial disclosure can win

In the main case, advice is cheap talk: messages are costless and cannot be enforced, while the principal remains free to choose an action. As conflict increases, the number of distinctions the expert can credibly communicate falls at particular thresholds. The value of advice changes slope at those points, creating a reason to design reports that move the conversation away from a threshold rather than leave it there.

In one constructed example, fully revealing demand produces five distinguishable messages in the low-conflict state but only uninformative “babbling” in the high-conflict state. An optimal partially revealing study instead produces either four or two distinguishable messages, never three. It beats both full revelation and saying nothing: better advice more than compensates for the loss of forecasting precision. These numbers describe message categories in the model, not observed conversations.

The numerical exercise gives a sense of how parameter-dependent the result is. For one setting of the forecast-value parameter, the optimal study was neither fully revealing nor silent in about 38% of the grid-and-prior cases examined. Two interior reports occurred in about 7%, and skipping at least one intermediate communication regime in about 4%. The best two-interior designs reduced the principal’s loss by up to about 10% relative to the better of full disclosure and silence. These are shares of a numerical map, not frequencies in the economy.

Different rules, different reports

The disclosure prescription changes when the communication institution changes. With an optimally designed neutral mediator who receives the expert’s information and recommends an action, the model favors full revelation of conflict. With a principal able to commit in advance to a deterministic action rule, a cap can limit the expert’s discretion; the optimal study can reveal nothing, fully reveal the state, or sometimes certify high conflict while pooling other cases. In the main cheap-talk example, commitment makes silence optimal.

The authors also separate obtaining information from sharing it. If the principal keeps the raw data and commits in advance to a coarser public report, she can use the detailed forecast herself while influencing the expert’s incentives through what is disclosed. In this model, selective disclosure can improve advice without reducing the principal’s own forecasting precision.

The political analogy is a treasury setting a spending agency’s budget using the agency’s local knowledge and a public fiscal forecast. The model suggests that the precision of the forecast, the possibility of mediation and the existence of binding budget rules should be considered together. It does not establish that withholding detail from actual fiscal forecasts would improve budgets; transparency and other public costs of restricted disclosure are outside the modeled payoff.

The conclusions also depend on how the cheap-talk game is resolved. The authors select the most informative equilibrium at the reports used by the optimal study; if the expert instead sends uninformative messages there, their cheap-talk results fail. Independence between demand and local conditions, the assumed preferences and the binary benchmark matter too. The paper proves local robustness to small smooth changes in the expert’s type distribution, but leaves correlated states and broader departures for further research.