marketing-science-writing

Follow the five-stage pipeline: position the topic with a gap table, specify a consumer utility model such as random-coefficients logit, design an identification strategy, run counterfactual simulations, and assemble the LaTeX manuscript. The skill encodes Marketing Science conventions for model presentation and identification justification.

The skill provides a BLPModel class using numpy, pandas, and scipy that computes predicted shares via simulation, inverts shares to mean utilities with contraction mapping, and estimates random coefficient standard deviations with two-step GMM. Instruments combine BLP-style sums of rival characteristics with cost shifters.

Structural models with counterfactuals fit Marketing Science or QME, field experiments with causal effects fit JMR or JM, and consumer psychology work fits JCR. The skill's journal-characteristics reference details scope, length, and reviewer expectations for eight journals.

Non-convergence usually comes from poor starting values or weak instruments. Use starting sigmas near zero, check first-stage F-statistics exceed 10, and add damping such as delta_new = 0.7 * delta + 0.3 * update to stabilize iterations.

Yes, it generates code that solves Bertrand-Nash equilibrium prices under pre- and post-merger ownership matrices, then computes changes in consumer surplus, profits, and total welfare. Multi-start strategies and adaptive damping handle equilibrium non-convergence.

It targets quantitative marketing journals and structural or experimental methods, so it does not cover qualitative research or non-marketing fields. Users must supply their own data and descriptive statistics, and estimation code requires Python with scipy and pandas installed.