Wber Theory Model

Theory and Model Craft (wber-theory-model)

When to trigger

• A reduced-form result needs a model to interpret a mechanism or run a counterfactual
• A referee asks "what is the model that rationalizes this?" or "what does welfare do?"
• You want to extrapolate beyond the estimated policy to an un-tried policy (requires structure)
• The paper has a heavy model but it is decorative — it does not discipline the empirics or the counterfactual
• You are unsure whether WBER expects a formal model here at all

Does WBER even want a model here?

WBER publishes both theoretical and empirical development research, but most accepted papers are empirical, and a formal model is a means, not a merit badge. Add a model only when it earns its place:

• To interpret — convert a reduced-form coefficient into a structural parameter (an elasticity, a friction) policymakers can reason about.
• To extrapolate — answer a counterfactual the data alone cannot (an un-tried transfer size, a national roll-out, a price change).
• To aggregate — move from a partial-equilibrium treatment effect to a general-equilibrium or welfare statement.

If none of these apply, a clean reduced-form evaluation with a clearly stated conceptual framework is the better WBER paper. A decorative model that the empirics ignore is a liability — it invites referee attacks for no payoff.

Disciplining a development model

• Tie every parameter to data. Name what in the developing-country data identifies each parameter (a moment, an elasticity, an experimental treatment effect). "We calibrate to the literature" is weak; "the experimental LATE pins the take-up elasticity" is strong.
• Match the friction to the setting. Development models live or die on the right friction: credit/insurance constraints, missing markets, information asymmetries, enforcement/state-capacity limits, search frictions in informal labor markets. Use the one the data support, not a textbook default.
• Validate out of sample. Show the model reproduces a moment it was not fit to — ideally a treatment effect from the very experiment/reform that motivates the paper.
• Make the counterfactual honest. State which parameters you assume are policy-invariant (Lucas critique) and why; bound the GE channels you cannot fully model.

Reduced-form ↔ structure handoff

Referee pushback mapped to the model fix

• "What rationalizes this reduced-form pattern?" → Add the minimal model that generates it; do not over-build. Show the model's comparative static matches the sign and rough magnitude you estimate.
• "Your parameters are calibration in disguise." → Tie each parameter to a data moment and report which moment moves which parameter; validate on an untargeted moment.
• "The counterfactual assumes invariance you never defend." → State explicitly which behavioral parameters you treat as policy-invariant and argue why the Lucas critique does not bite here.
• "This is a rich-world model bolted onto a poor-country setting." → Replace the frictionless core with the binding development friction (credit, insurance, information, enforcement) the data reveal.
• "The model adds nothing the regressions don't." → Either give it a job (a counterfactual the data cannot answer) or cut it to a conceptual framework.

Checklist

• The model's job is named: interpret, extrapolate, or aggregate (not decoration)
• Each parameter is tied to identifying variation in the developing-country data
• The friction matches the setting (credit/insurance/information/enforcement/search)
• An untargeted moment or out-of-sample treatment effect validates the model
• Counterfactual states policy-invariance assumptions and bounds GE channels
• Model assumptions are kept separate from policy interpretation in the text
• If no model is warranted, a clear conceptual framework replaces it