--- name: econ-design description: Develop an economics question or model, design identification and inference, or select among research methods and designs, including structural IO, dynamic choice, equilibrium counterfactuals and welfare. --- # Economic design ## Purpose and input [Task boundaries and completion](../../runtime/CORE.md) and [operation ownership](../../registry/operations.json) govern this specialist. Use [PROSE](../../runtime/PROSE.md) when explaining the result. Input: Requested economic question, model or empirical design and supplied premises. Use relevant literature when contribution or source investigation is in scope, and documented source fields when empirical measurement or data feasibility is in scope. For an explicit assessment of a specified claim or artifact, econ-review owns the assessment and this skill returns its technical design judgment. Before design reasoning, apply [CORE task entry](../../runtime/CORE.md#task-entry). Read the applicable method section and its required dependencies in full; retrieve knowledge IDs and their true prerequisites only when the design needs them. A registry entry, summary, heading or knowledge lookup does not replace these bodies or this specialist's judgment. ## Procedure When the pending operation needs a library definition, derivation, model condition or worked example, activate [econ-knowledge](../econ-knowledge/SKILL.md) and select the needed IDs and required prerequisites. Retain this specialist’s decision and execution ownership, adopted economic specification, official package interfaces and applicable rules. ### Identify the requested design work When this design requires both an analytical result and source findings, or method applicability or selection remains unresolved, use [DECIDE](../../runtime/DECIDE.md) for problem formulation, decision selection and independent source integration. Select the requested output: a research question, derivation, empirical design or recommendation. For a research question, identify the behavior, incentive, information constraint, allocation or equilibrium it would explain; use econ-literature to compare that contribution with the closest papers. For an economic choice, apply [DECIDE's action-selection procedure](../../runtime/DECIDE.md#select-a-feasible-action), expressing incremental benefits and resource costs for the same population, period and units and including existing uses of the same capacity. An assessment support call returns the requested technical judgment to econ-review. When conflicting model results motivate a research question, express their choices and outcomes in a common model and vary the condition that can reverse the result. If aligned definitions resolve the difference, narrow the contribution accordingly. If an economic condition explains it, formulate the question around when that condition holds and select settings or measurements that distinguish its regimes. When proposing a behavioral hypothesis, identify its driver, the quantity to measure independently, and the comparison where competing hypotheses predict different outcomes. Specify how cases differ in that driver while holding established determinants of the choice fixed, and how the measure would explain choices beyond established explanations. Connect a diagnostic design to the condition preventing the requested outcome. Distinguish the observed shortfall, its candidate mechanism, the model margin an intervention changes and the resulting economic choice. Derive the comparison at that margin before specifying data collection or computation. Return its implication for the parent research claim or action, including displacement and participant responses that change the total objective. Keep established specifications attached to each delegated derivation or empirical comparison through [TEAM](../../runtime/TEAM.md). For a design-only request, complete the requested economic object and the assumptions and operations needed to establish it. An empirical design specifies its question, identifying comparison, measurement targets, estimand and inference; a theoretical design specifies primitives, feasible choices, solution concept and the proposition or comparative static sought. Include testable hypotheses when they are part of the requested design. When the request includes empirical measurement, calculate the comparison on the adopted sample and use its results to retain or revise the hypothesis. A completed design does not itself start measurement or estimation. ### For an empirical specification Start with the [estimand-to-method sequence](references/designs.md#choose-the-estimand-and-method): requested use, target population and contrast, observation and assignment units, outcome scale, identifying variation, and inference. Resolve these choices before selecting a package. For an adopted specification, reuse those choices and start at the requested change. Use the count, DID, heterogeneous-effect or structural branch that matches the economic object. Locate the unresolved link that can invalidate the requested conclusion: the measured outcome, the population or response margin identified, or the mapping from that effect to the proposed action. Resolve that link before downstream estimation variants and exhibits that depend on it. For a policy expansion, compare the people and choices moved by the new policy with those moved by the identifying variation. For a finite intervention, establish the response over its path. Return the decisive comparison and update only its affected calculations; retain the adopted specification and matching results. Use [design elements](references/designs.md#design-elements) to define the parameter, comparison, assignment, controls and inference units. Translate the intervention into the information, feasible actions or payoffs it changes and the eligibility rule that includes actors. For staged implementation, locate the changed gate and retain eligible actors who do not proceed through it. Use timing gaps between information, contracting and implementation to find a comparison where one choice is fixed while another can respond. When explanations predict different responses, choose the population, timing or participation boundary where their predictions separate, then assess the explanations against that comparison. When a proposed rule uses observed behavior to set eligibility, scores or prices, identify who can change that behavior after learning the rule. Compare fixed-input predictions with responses to the announced rule, including the cost of changing the input. Use the announcement-to-implementation interval and differences in the gain or cost of qualifying to distinguish anticipation from implementation effects; revise treatment timing, input assumptions or the policy comparison when those responses change the answer. Use the [design-specific section selector](references/designs.md#select-the-needed-section) to match estimation and inference to the parameter, treatment path, heterogeneity, dependence and assignment units. Interpret each requested diagnostic through its null, precision and sampling design. For implementation, pass the adopted estimator, sample, timing, weight meaning and inference requirements to [econ-data's engine selection](../econ-data/references/implementation.md#estimator-engine-and-algorithm-selection). Require the selected function to implement that parameter and the requested post-estimation comparison. A faster representation must preserve that contract; identify a proposed change in the economic specification and its affected comparisons as a design decision governed by [CORE](../../runtime/CORE.md). An analysis request continues through the package call, saved estimates and requested economic comparison. ### For a theoretical model Choose the requested object: a payoff comparison, optimal choice, equilibrium, dynamic path or optimized value. Use the matching [model-construction operations](references/designs.md#model-construction). Retain structure whose omission changes that object, including waiting, nonparticipation, switching or actor composition. A baseline representing an existing institution preserves its objectives, rights and priority rules; a change to those features is a separate intervention to evaluate. A payoff comparison requires payoffs; deriving choices additionally requires a behavioral rule. For interventions, include displacement into feasible alternative places, times or actions. If competitors or trading partners respond, include their outcomes where the requested aggregation requires them. When constructing an empirical counterfactual, reassess their eligibility as untreated comparisons. When connecting a model to observed data or interventions, use [observation mapping and interventions](references/designs.md#connect-observed-choices-and-outcomes-to-the-model-and-interventions) to map observed choices and outcomes to model parameters; use [restrictions and set-valued predictions](references/designs.md#construct-restrictions-and-set-valued-predictions) for deviations, multiple equilibria and compatible counterfactual sets. ### For behavioral responses and group comparisons Distinguish a behavioral mechanism's total or fixed cost from its cost of an additional unit. Use the actor's decision rule and initial choice to derive the response to a cost change. Establish which choices cross the participation or supply threshold, or which feasible alternative becomes the chosen treatment. Express changes in net gains and changes in actions as separate economic objects. An individual's monotone comparative static gives the direction of that actor's response. To compare group quantity changes, convert choices crossing the threshold into the comparison's quantity unit and apply population weights. When a change in the number participating can arise from entry or exit, compare those transitions and the quantities each contributes. Establish comparability of initial choice distributions, production opportunities and response margins at the affected thresholds. Choose the observation window from the response's adjustment and duration; a narrow window can precede implementation, while a long average can dilute a short-lived response. ### Establish feasibility For an end-to-end research request, keep discovery, design and execution in the same adopted specification. Connect each candidate's closest literature answer, added economic mechanism, target contrast and identifying restrictions to documented source fields, observation grain, period, linkage keys and the variation that identifies that contrast. Use the [public discovery connector](../econ-literature/scripts/research_sources.py) or the current connected sources to inspect those actual fields. Use the [design assessment](scripts/research_design.py) to bind the closest-answer passages and the owner's contribution judgment to actual CSV grain, missingness, period, variation and linkage observations. Adopt the candidate order from its economic contribution and decisive comparison under DECIDE; execute the first feasible candidate's recorded next operation through the [research execution path](../../harness/research.md). The host continues from a returned calculation to its economic interpretation, next determining comparison and final paper through the five existing owners. After each returned result, compare the selected mechanism's prediction with the observed quantity and its consequential rival. Retain the candidate when that comparison supports its next operation; revise the affected premise when new source evidence or the calculation changes it. Generate the next authorized retrieval, specification or artifact task in the existing record, execute it, and integrate its output. Stop the research cycle when the requested economic answer, artifacts and actions satisfy their original completion conditions. Freeze a specification-search protocol before its trials, including estimand, population, inference, evaluation criterion and partition roles. A change in the target or identifying restrictions receives its own adopted comparison and result identity. For each proposed empirical design, map required measures and comparisons to documented fields and source records within the task's cost and access limits. For a theoretical or computational design, establish compatibility of assumptions and feasibility of the required derivation or solution within those limits. Preserve supplied candidate IDs and scope. Use econ-data for field or linkage analysis. Repair input or implementation defects within the authorized design; record missing fields or permissions needed for the next operation and continue independent work. ### Choose among research designs When asked to select a design, compare feasible candidates on the requested economic object, assumptions needed to establish it, and data, computation, implementation and delay costs. For empirical designs, include identification and precision; for theoretical designs, compare the propositions and mechanisms their assumptions can establish. Evaluate numerical or statistical precision against a stated economic threshold or target contrast, deriving it from the task's objective when available. If that criterion is missing, request it when its value changes the design choice. Use [DECIDE](../../runtime/DECIDE.md#select-a-feasible-action) to rank feasible designs, establish switch conditions and stop information gathering when it cannot change the action; return the selected design or its selecting conditions. Keep adoption decisions and project status in operational records; research artifacts state the economic comparison, assumptions and results. ## Output The requested question, derivation, recommendation or design assessment, with the economic comparison and conditions that determine it. For a delegated design, return the requested economic object, applicable population, timing, assumptions, identifying comparison or equilibrium, the determining evidence and the precise next operation to its owner. Match each returned calculation and source to the adopted estimand, model and units before using it. After interruption, handoff or loss of context, restore the original request, authority, fixed specification, completed results and mandatory rule bodies that are unread, changed or no longer available in the active context through [TEAM's resumption contract](../../runtime/TEAM.md#resumption); reuse same-version reads and valid results. Continue the dependent work until the requested design or authorized analysis meets its own completion conditions. ## Specialized executable designs Select the registered condition that matches the adopted model: `method_extended_causal`, `method_extended_markets`, `method_extended_dynamic`, `method_extended_macro`, `method_agent_simulation` or `method_extended_numerics`. Their [causal](references/causal-models.md), [market](references/market-models.md), [dynamic](references/dynamic-models.md), [general-equilibrium](references/macro-models.md), [stochastic](references/dsge-models.md), [agent](references/agent-models.md) and [numerical](references/numerical-models.md) references give the independent equations, identifying observations, normalization, exact supported specification, calculation entry and scientific checks. Carry those same objects into estimation, counterfactuals and prose. Select `research_adaptive_experiments` for [source comparisons and adaptive experiments](references/adaptive-research.md). The proposal generator compares all five economic dimensions and actual data feasibility. Sol adopts the research target and authorized experiment axes in the existing governing context. The runner then generates new parameter candidates from completed results under the fixed evaluation and preserves the final held-out sample for the selected candidate. ## Registered operations Generated from registry/operations.json; the request determines the actual target and authority. - `design`: Choose a research question, model, identification strategy, study design or policy action.