marketing-science-academic-writing

# Marketing Science Academic Writing > Skill by [ara.so](https://ara.so) — Marketing Skills collection. A comprehensive skill for writing quantitative marketing science papers targeting flagship journals (Marketing Science, JMR, JM, JCR, JAMS, IJRM, QME, Marketing Letters). Guides you through the complete 0-to-draft pipeline: topic positioning, consumer utility modeling, identification design, structural estimation, counterfactual simulations, and manuscript assembly. ## What This Skill Does This skill transforms your research idea into a complete academic paper draft by: 1. **Topic Positioning**: Gap analysis, journal selection, contribution framing 2. **Consumer Utility Modeling**: Micro-founded demand systems, notation, utility specification 3. **Identification & Estimation**: Causal inference strategies (DID/RDD/IV), structural estimation (BLP/GMM) 4. **Counterfactuals & Experiments**: Simulation design, conjoint analysis, field experiments 5. **Full Draft Assembly**: Complete LaTeX manuscript with INFORMS formatting **Key differentiator**: Domain-specific conventions for marketing journals that generic writing skills don't cover — utility model structure, identification justification, structural estimation workflows, and journal-specific reviewer expectations. ## Installation The project is a skill repository that you reference, not a package you install. The skill files provide structured guidance and reference materials. ### For AI Coding Agents **OpenCode**: ```bash cp -r marketing-science-writing ~/.config/opencode/skills/ ``` **Claude Code**: ```bash cp -r marketing-science-writing ~/.claude/skills/ ``` **Codex**: ```bash cp -r marketing-science-writing ~/.agents/skills/ ``` **Cursor / Windsurf**: ```bash cp -r marketing-science-writing ~/.cursor/skills/ # or cp -r marketing-science-writing ~/.windsurf/skills/ ``` ### Manual Integration Reference `SKILL.md` and the `references/` directory in your agent's custom skills path, or concatenate the markdown files into your system prompt. ## Project Structure ``` marketing-science-writing/ ├── SKILL.md # Main skill orchestration ├── references/ │ ├── journal-characteristics.md # 8 journals: profiles & expectations │ ├── modeling-conventions.md # Utility models, demand systems │ ├── identification-guide.md # DID, RDD, IV strategies │ ├── estimation-guide.md # BLP, GMM, MLE, Bayesian │ ├── counterfactual-guide.md # Simulation design │ ├── conjoint-analysis.md # Conjoint experiment design │ ├── field-experiments.md # Field experiment protocols │ ├── marketing-implications.md # Actionable implications framework │ ├── reviewer-expectations.md # Reviewer checklist │ └── writing-patterns.md # Reusable writing templates ├── assets/ │ └── demand-model-reference.md # Generic demand model structures ├── examples/ │ ├── manuscript_template.tex # INFORMS LaTeX template │ └── blp_estimation_example.py # BLP estimation starter code └── README.md ``` ## Core Usage Patterns ### Pattern 1: Full 0-to-Draft Pipeline **Trigger**: "Take me through the full marketing science pipeline for a paper on [topic]" **Workflow**: 1. **Stage 1**: Topic positioning → gap table, journal selection, contribution list 2. **Stage 2**: Consumer utility model → notation, utility specification, demand derivation 3. **Stage 3**: Identification & estimation → DID/RDD/IV design, estimator choice 4. **Stage 4**: Counterfactuals & experiments → simulation design, experiment protocol 5. **Stage 5**: Full draft → LaTeX manuscript assembly **Example**: ``` User: "Write a Marketing Science paper on BLP demand estimation for the smartphone market with merger counterfactuals." Agent: [Executes Stage 1] → Gap table: 8 closest papers (Nair 2007, Berry et al. 1995, etc.) → Target journal: Marketing Science (structural + IO focus) → Contributions: (1) First BLP on smartphone market, (2) New demand-side micro moments using clickstream data, (3) Merger simulation with regulatory implications [Stage 2] → Utility specification with random coefficients on price, screen size, brand → Nested logit within operating system → Supply-side: Bertrand-Nash pricing game [Stage 3] → Identification: Cost shifters as IVs (component prices from Counterpoint) → Micro moments: Match cross-price elasticities from clickstream → GMM estimation with optimal weighting matrix [Stage 4] → Counterfactual: Simulate Apple-Samsung merger → Equilibrium: New Nash equilibrium prices post-merger → Welfare: Consumer surplus loss, producer profit gain [Stage 5] → Complete LaTeX manuscript with INFORMS template ``` ### Pattern 2: Stage-Specific Deep Dive **Trigger**: "Help me design the consumer utility model for my paper" **Focus**: Stage 2 only — utility specification, notation system, demand derivation. **Example**: ``` User: "I'm studying ride-sharing demand. Help me specify the utility model." Agent: # Consumer Utility Model ## Notation - $i$: consumer index - $j$: ride option (Uber, Lyft, taxi, outside option) - $t$: time period - $p_{jt}$: price per mile - $x_{jt}$: observable characteristics (wait time, vehicle quality) - $\xi_{jt}$: unobserved quality shock ## Indirect Utility Consumer $i$ obtains utility from option $j$ at time $t$: $$u_{ijt} = \alpha_i p_{jt} + x_{jt}'\beta_i + \xi_{jt} + \epsilon_{ijt}$$ where: - $\alpha_i \sim N(\alpha, \sigma_\alpha^2)$: random price coefficient (negative) - $\beta_i \sim N(\beta, \Sigma_\beta)$: random coefficients on characteristics - $\epsilon_{ijt}$: i.i.d. Type-I extreme value ## Choice Probability (Random Coefficients Logit) The probability consumer $i$ chooses option $j$: $$s_{ij}(\theta) = \int \frac{\exp(u_{ijt})}{\sum_{k=0}^J \exp(u_{ikt})} dF(\alpha_i, \beta_i)$$ Aggregate market share: $s_j = \int s_{ij}(\theta) dF(i)$ ## Demand Inversion For estimation, invert share equation: $$\delta_j = \ln(s_j) - \ln(s_0)$$ where $\delta_j = \alpha p_j + x_j'\beta + \xi_j$ (mean utility). ``` ### Pattern 3: Identification Strategy Design **Trigger**: "I need to design the identification strategy for my marketing paper" **Focus**: Stage 3 — causal inference or structural identification. **Example for DID**: ``` User: "Amazon introduced same-day delivery in select cities. I want to estimate the causal effect on sales." Agent: # Identification Strategy: Difference-in-Differences ## Treatment & Control - **Treatment**: Cities with same-day delivery (rollout: 2014-2016, staggered) - **Control**: Cities without same-day delivery (matched on demographics) ## Specification $$\log(sales_{ct}) = \beta_0 + \beta_1 SameDay_{ct} + \gamma_c + \lambda_t + \epsilon_{ct}$$ where: - $SameDay_{ct}$: indicator for city $c$ having same-day delivery at time $t$ - $\gamma_c$: city fixed effects - $\lambda_t$: time fixed effects ## Parallel Trends Test Event study specification: $$\log(sales_{ct}) = \sum_{k=-K}^{-2} \delta_k D_{ct}^k + \sum_{k=0}^{K} \delta_k D_{ct}^k + \gamma_c + \lambda_t + \epsilon_{ct}$$ where $D_{ct}^k = 1$ if city $c$ is $k$ periods from treatment. **Test**: $H_0: \delta_{-K} = \cdots = \delta_{-2} = 0$ (pre-trends) ## Robustness Checks 1. Callaway-Sant'Anna estimator (staggered treatment) 2. Synthetic control for treated cities 3. Triple-differences: same-day × product category × time ``` **Example for IV (Structural)**: ``` User: "Price is endogenous in my demand model. How do I construct instruments?" Agent: # Instrumental Variables for Demand Estimation ## Endogeneity Problem Price $p_j$ is correlated with unobserved quality $\xi_j$ due to: 1. Simultaneous pricing (supply-side response to demand shocks) 2. Omitted product characteristics ## IV Strategy: BLP Instruments ### 1. Cost Shifters **Instruments**: Input prices, wage indices, exchange rates **Exclusion restriction**: Affect marginal cost → price, but not consumer utility directly **Example**: Steel prices for automotive demand estimation ### 2. Hausman Instruments **Instruments**: Prices of same product in other markets **Assumption**: Market-specific demand shocks are uncorrelated **Example**: Price of iPhone 13 in New York as IV for Boston demand ### 3. Differentiation Instruments **Instruments**: Number of competing products, sum of rival characteristics **Intuition**: More competition → lower markups → lower prices **Construction**: ```python # Sum of rival characteristics df['IV_rival_features'] = df.groupby('market')['feature_score'].transform( lambda x: x.sum() - x ) # Number of rivals in same nest df['IV_nest_count'] = df.groupby(['market', 'nest']).transform('size') - 1 ``` ### First-Stage Test $$p_j = \pi_0 + \pi_1 Z_j + \pi_2 x_j + \nu_j$$ **Requirement**: $F$-statistic > 10 (Stock-Yogo weak IV test) ``` ### Pattern 4: BLP Estimation Implementation **Trigger**: "Set up BLP estimation code for my demand model" **Output**: Working Python code using `pyblp` library. **Example**: ```python import pyblp import numpy as np import pandas as pd # Load data # Expected columns: market_ids, product_ids, shares, prices, characteristics, instruments data = pd.read_csv('smartphone_data.csv') # Define product formulation # Random coefficients on: price, screen_size, brand_apple, brand_samsung product_formulations = ( pyblp.Formulation('1 + prices + screen_size + C(brand)'), # linear pyblp.Formulation('1 + prices + screen_size') # random coefficients ) # Define agent formulation (demographics for random coefficients) agent_formulation = pyblp.Formulation('0 + income + age') # Problem setup problem = pyblp.Problem( product_formulations=product_formulations, product_data=data, agent_formulation=agent_formulation, integration=pyblp.Integration('monte_carlo', size=1000, seed=0) ) # Initial parameters initial_sigma = np.diag([0.5, 0.3, 0.2]) # std dev for price, screen, brand coefs initial_pi = np.zeros((3, 2)) # interactions with demographics # Optimization results = problem.solve( sigma=initial_sigma, pi=initial_pi, optimization=pyblp.Optimization('l-bfgs-b', {'maxiter': 1000}), iteration=pyblp.Iteration('simple', {'atol': 1e-12}) ) # Results summary print(results) # Elasticities elasticities = results.compute_elasticities() print("\nOwn-price elasticities:") print(np.diag(elasticities).mean()) # Micro moments (optional) # Match observed cross-price elasticity from clickstream data observed_cross_elasticity = 0.15 micro_moments = pyblp.MicroMoment( name='cross_price_elasticity', value=observed_cross_elasticity, compute=lambda data, results: results.compute_elasticities()[0, 1] ) # Re-estimate with micro moments results_micro = problem.solve( sigma=initial_sigma, pi=initial_pi, micro_moments=[micro_moments], optimization=pyblp.Optimization('l-bfgs-b') ) # Counterfactual: Merger simulation # Merge firm 1 (Apple) and firm 2 (Samsung) changed_ownership = data['firm_ids'].copy() changed_ownership[data['firm_ids'] == 2] = 1 # Samsung products now owned by Apple merger_results = results.solve_merger( changed_ownership=changed_ownership ) print("\nMerger Effects:") print(f"Price change: {merger_results.price_changes.mean():.2%}") print(f"Consumer surplus change: ${merger_results.consumer_surplus_change.sum()/1e6:.1f}M") ``` **Key Configuration**: - **Integration method**: Monte Carlo (1000 draws) for random coefficients - **Optimization**: L-BFGS-B (handles bounds on sigma) - **Iteration**: Simple fixed-point iteration for contraction mapping - **Micro moments**: Optional — improve identification with additional data moments **Environment variables** (if using cloud compute): ```bash export PYBLP_CACHE_DIR=/path/to/cache # Cache intermediate results export OMP_NUM_THREADS=8 # Parallel computation ``` ### Pattern 5: Counterfactual Simulation **Trigger**: "Design counterfactual simulations for my structural model" **Example**: ``` User: "I estimated a demand model. Now simulate a price discrimination ban." Agent: # Counterfactual: Uniform Pricing Mandate ## Baseline Equilibrium Current state: Firms set different prices across markets (3rd-degree price discrimination) ## Counterfactual Scenario Policy: Firms must charge the same price in all markets ## Equilibrium Computation ### Step 1: Compute Baseline Already have: $p^*, q^*, \pi^*$ from estimation ### Step 2: Constrained Optimization Firm $f$ solves: $$\max_{p_f} \sum_{m=1}^M (p_f - mc_{fm}) \cdot q_{fm}(p_f, p_{-f})$$ subject to: $p_f$ is constant across markets $m$ ### Step 3: Nash Equilibrium Iterate best responses until convergence: ```python def compute_uniform_pricing_equilibrium(results, mc, markets, max_iter=100, tol=1e-6): """ Compute Nash equilibrium under uniform pricing constraint. Parameters: - results: pyblp estimation results