Markets Data and Sales Drivers

Summary

Chapter 2 covers the empirical raw material for market response models: data sources (scanner, warehouse, panels, mail), measurement of marketing variables (stock variables, relative indices, GRPs), aggregation choices (brand/category, store/market/national, weekly/monthly), and response measures (sales, market share, baseline).

Data Sources

Scanner Data

Scanner data are transaction records collected at point-of-sale (POS) terminals. Universal Product Codes (UPCs) track exact brand, size, price, and quantity sold. Scanner data exist at the store-week level and are aggregated by syndicators (IRI, Nielsen) to market-week panels.

Advantages: high frequency, objective, captures price and promotion exactly. Limitations: no household demographics, stockpiling behavior unobservable at aggregate level.

Source	Unit	Typical Use
Store scanner (IRI/Nielsen)	Store × week	Retail sales, price, promotion
Warehouse withdrawals	Warehouse × week	Manufacturer shipments
Consumer panels (BehaviorScan)	Household × trip	Purchase incidence, brand switching
Mail panels	Household × month	Attitudes, awareness
Advertising monitoring (BAR/LNA)	Brand × week × medium	GRPs, share of voice

Key Variable Types

Stock Variables

Stock Variable (Advertising Goodwill)

An advertising goodwill stock $G_t$ accumulates past advertising with geometric decay:

$$G_t=A_t+\lambda G_{t-1},0<\lambda <1$$

where $\lambda$ is the retention rate (carryover parameter) and $A_t$ is current advertising spend. The stock formulation is equivalent to the Koyck distributed lag — see Carryover Effects and Distributed Lags.

Relative Variables

Relative Index Variables

Three standard relative measures for competitive context:

• RIX (Relative Income Index): brand price relative to category average
• RCX (Relative Communication Index): brand advertising share relative to competitors
• RBX (Relative Brand Index): composite brand equity measure

Using relative variables reduces collinearity and captures competitive parity effects.

GRPs (Gross Rating Points)

GRPs

Gross Rating Points = reach × frequency. One GRP = 1% of the target audience exposed once. GRPs are the standard currency for advertising media planning and are used as the advertising input $A_t$ in many response models.

$$\text{GRP}=\text{Reach}\times \text{Average Frequency}$$

Baseline Volume

Baseline Sales

Baseline volume is the sales level that would occur without any promotional activity, typically estimated from a regression of sales on time trend and seasonal dummies with promotion periods excluded or treated as outliers. Incremental volume = total sales − baseline.

Baseline decomposition is essential for ROI calculation of trade promotions (which typically show large total lift but much of it is displacement of baseline volume).

Market Share

$$M{S}_t=\frac{Q_t}{Q_{T,t}}$$

where $Q_{T,t}$ = total category sales. Market share is bounded in $[0,1]$, motivating the logit/MCI share models in Market Share Models.

Aggregation

Temporal Aggregation

Aggregating weekly data to monthly or annual data biases carryover estimates. Clarke (1976) showed estimated retention rates can be 20-50× longer in annual vs. monthly data — see Carryover Effects and Distributed Lags for recovery procedures.

Cross-Sectional Aggregation

Level	Pros	Cons
Store	Most disaggregated, rich variation	Difficult to match to advertising
Market (DMA)	Standard AC Nielsen unit	Loses store-level heterogeneity
National	Matches national spending data	Masks regional variation

Brand vs. Category

MRMs can be estimated at brand level (selective demand) or category level (primary demand). The Schultz-Wittink decomposition separates a brand’s sales effect into:

• Primary demand (grows category)
• Selective demand (takes share from competitors)

Cross-Links

• Functional forms for response models: Functional Forms in Marketing
• Carryover and goodwill stock: Carryover Effects and Distributed Lags
• Market share models: Market Share Models
• Empirical advertising findings: Advertising and Promotion Effects
• Aggregation bias: Carryover Effects and Distributed Lags
• Bayesian MMM workflow (GRPs + adstock transform are the primary data inputs): Bayesian Media Mix Modeling - Overview