Estimation, Fit & Counterfactuals (mksc-data-analysis)

When to trigger

• The model is specified and it is time to estimate and report
• Estimates exist but identification, fit, or counterfactuals are not yet convincing
• A reviewer says "the parameters are not credibly identified" or "the counterfactual is not validated"
• You need the replication package (data + estimation code) ready for acceptance

Estimate, then prove identification empirically

• Run the estimator matched to the model: GMM with the stated moment conditions (BLP), MLE/SMLE, simulated method of moments, or MCMC for hierarchical Bayes. Report standard errors that respect the estimation (e.g., GMM/sandwich, bootstrap, or posterior intervals) and the optimizer/convergence diagnostics.
• Demonstrate identification, not just assert it: show the identifying variation moves the relevant moments; report sensitivity of estimates to instruments; where feasible, a Monte Carlo recovering known parameters or a sensitivity-of-estimates-to-moments analysis strengthens the claim.
• First-stage/instrument strength for IV/GMM; relevance and exclusion discussed.

Assess model fit before trusting counterfactuals

• Report in-sample fit (predicted vs. actual shares/prices/moments) and, where possible, out-of-sample or holdout validation.
• Check economic plausibility of estimates (own-/cross-price elasticities, margins implied by supply FOCs, discount factors) against priors and institutional facts.
• For Bayesian models, report convergence (R-hat, effective sample size) and posterior predictive checks.

Counterfactuals are the payoff

• Re-solve the model under the policy/counterfactual, holding fixed only what theory says is fixed; recompute equilibrium prices/quantities where firms re-optimize.
• Report magnitudes with uncertainty (delta-method or simulation-based intervals), and decompose the mechanism driving the result.
• Discuss the scope and assumptions under which the counterfactual is valid.

Robustness

• Alternative specifications (functional form, instruments, heterogeneity), subsamples, and alternative normalizations.
• Show the headline result and key counterfactual survive the changes a referee will request.

Replication package (plan now, deposit on acceptance)

Per the Marketing Science Replication and Disclosure Policy, accepted papers submit data and estimation code sufficient for a peer to reproduce the essential content. For licensed data (NielsenIQ, Compustat, CRSP, Census), provide access instructions and the linking/build code rather than raw data. Keep a master script regenerating every table, figure, and counterfactual.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. Marketing Science is heavily structural/analytical; the chain below serves its reduced-form / field-experiment lane — structural demand and analytical modeling are outside this causal-inference toolchain.

• Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg — report the adjusted threshold.
• OVB sensitivity: oster_delta / sensemakr.
• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley; multilevel data → cluster at the right level.
• Re-fit off one handle: audit_result(result_id) lists the missing checks and the exact suggest_function for each.
• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Keep the decisive checks in the body and the exhaustive battery in the appendix. See the executed chain in the JF execution walkthrough.

Checklist

• Estimator run; appropriate SEs and convergence diagnostics reported
• Identification shown empirically (sensitivity/Monte Carlo/first stage)
• In-sample fit and (where possible) holdout/out-of-sample validation
• Estimates economically plausible (elasticities, margins, discount factor)
• Counterfactual re-solves equilibrium; magnitudes with uncertainty + mechanism
• Robustness to specification/instruments/normalization
• Replication package (code + data access/build) prepared

Anti-patterns

• Reporting point estimates with no identification or fit evidence.
• A counterfactual that holds firm behavior fixed when firms would re-optimize.
• Elasticities/margins that are economically implausible, unaddressed.
• "Code available on request" instead of a replication-ready package.

Evidence pass for Marketing Science

Use this as a second-pass capability check. First lock the demand/supply mechanism, fit evidence, and counterfactual decision margin; then test whether the manuscript addresses quantitative marketing reviewers who read the model through the managerial counterfactual it makes possible.

• Primary move: Audit unit, comparison, uncertainty, missingness, sensitivity, and reproducibility before making any prose or submission recommendation.
• Decision ledger: return claim / evidence / blocker / next edit rows so the next pass can patch the manuscript directly.
• Neighbor test: compare against Journal of Marketing Research for empirical marketing breadth, Management Science for wider OR/MS reach, Quantitative Marketing and Economics for specialist modeling; if the neighboring outlet has the stronger audience claim, recommend re-routing before polishing.
• Verification floor: before submission-ready advice, re-open resources/official-source-map.md for volatile rules and name the one unresolved fact that could change the recommendation.

Output format

【Estimator】GMM / MLE-SMLE / SMM / Bayes; SEs + convergence
【Identification evidence】sensitivity / Monte Carlo / first stage
【Fit】in-sample + holdout; economic plausibility of estimates
【Counterfactual】policy re-solved; magnitude ± uncertainty; mechanism
【Robustness】specs/instruments/normalizations
【Replication】data+code package status (licensed-data handling)
【Next step】mksc-contribution-framing