Price Elasticity Modeling

The Price Elasticity of Demand (PED) is the core metric, calculated as the percentage change in quantity demanded divided by the percentage change in price. A coefficient of -2.0 indicates that a 1% price increase leads to a 2% drop in demand.

• Elastic (|PED| > 1): Demand is highly sensitive to price changes. Common for luxury goods or items with many substitutes.
• Inelastic (|PED| < 1): Demand is relatively insensitive to price. Typical for necessities or products with few alternatives.
• Unitary Elastic (|PED| = 1): Revenue remains constant as price changes.

The most common statistical implementation uses a log-log ordinary least squares (OLS) regression. By taking the natural logarithm of both price and quantity, the coefficient directly represents the constant elasticity.

• Model Form: log(Q) = α + β * log(P) + γ * Controls
• Interpretation: The coefficient β is the price elasticity.
• Advantage: Assumes a constant elasticity curve, which is a robust baseline before exploring more complex non-linear models like semi-log or polynomial forms.

A critical modeling challenge is endogeneity, where price is correlated with unobserved demand shocks (e.g., a price cut coinciding with a viral trend). This biases elasticity estimates.

• Solution: Use Instrumental Variables (IV) like competitor prices, supply-side cost shocks, or Hausman-style instruments from other markets.
• Two-Stage Least Squares (2SLS): First, predict price using the instrument; second, regress quantity on the predicted price to isolate the causal effect.

Aggregate elasticity masks significant variation across segments. Modern models estimate individual-level treatment effects to enable personalized pricing.

• Causal Forests: A non-parametric method that identifies how elasticity varies by customer features (e.g., loyalty status, basket size).
• Hierarchical Bayesian Models: Shrink individual segment estimates toward the population mean, preventing overfitting for small segments.
• Application: Identifying a loyal, high-income segment with an inelastic response versus a deal-seeking segment with high elasticity.

Measures the responsiveness of demand for Product A when the price of Product B changes. This is essential for modeling cannibalization and halo effects in a catalog.

• Substitutes (Positive Elasticity): A price increase for Coke leads to increased demand for Pepsi.
• Complements (Negative Elasticity): A price cut for printers increases demand for ink cartridges.
• Modeling: Requires a multi-equation demand system like the Almost Ideal Demand System (AIDS) to capture these interdependencies simultaneously.

Consumer sensitivity is not static; it shifts due to seasonality, product lifecycle stage, and macroeconomic trends. A robust model must detect concept drift.

• Rolling Window Regression: Estimates elasticity using only recent data, discarding obsolete historical patterns.
• Kalman Filters: A dynamic linear model where the elasticity coefficient is treated as a hidden state that evolves over time via a random walk.
• Use Case: Detecting a sudden shift to elastic behavior during a recession or a competitor's aggressive promotion.

Measures how demand for Product A changes when the price of Product B shifts. Critical for modeling cannibalization and halo effects in multi-product catalogs.

• Substitutes: Positive cross-elasticity (butter vs. margarine)
• Complements: Negative cross-elasticity (printers vs. ink cartridges)
• Formula: %ΔQ_A / %ΔP_B

Retailers use this to avoid self-cannibalization when discounting one SKU inadvertently kills sales of a higher-margin alternative.

Determines the maximum price a consumer accepts before abandoning a purchase. Elasticity modeling provides aggregate demand curves; WTP maps individual thresholds.

• Gabor-Granger method: Directly surveys price points
• Van Westendorp model: Identifies acceptable price ranges
• Conjoint analysis: Derives WTP from feature trade-offs

Combining elasticity curves with WTP distributions enables price discrimination engines to segment users by sensitivity.

Extends static elasticity models into continuous learning agents that adapt pricing policies through market interaction. Unlike regression-based elasticity, RL handles non-stationary environments.

• Q-Learning: Learns state-action value functions for price points
• Contextual Bandits: Balances exploration of new prices with exploitation of known elasticities
• Markov Decision Processes: Models sequential pricing decisions with delayed rewards

Elasticity estimates often serve as the initial policy before RL fine-tunes through live feedback.

Distinguishes correlation from causation when measuring price sensitivity. Raw elasticity calculations often conflate price effects with seasonality, competitor moves, or promotional overlap.

• Difference-in-Differences: Compares treated vs. control markets
• Instrumental Variables: Uses exogenous shocks to isolate price effects
• Propensity Score Matching: Controls for confounding variables

Without causal identification, elasticity estimates risk being biased and non-actionable for revenue optimization.

The degradation of elasticity model accuracy as market conditions evolve. Consumer sensitivity shifts due to economic cycles, competitor entry, or trend saturation.

• Sudden drift: Shock events like supply chain disruptions
• Gradual drift: Slow changes in brand perception or demographics
• Recurring drift: Seasonal elasticity patterns

Detection requires online model monitoring with statistical tests (ADWIN, Kolmogorov-Smirnov) to trigger retraining before stale elasticities cause revenue leakage.

Estimates the incremental causal effect of a price change on individual customers, identifying four segments:

• Sure Things: Buy regardless of price
• Lost Causes: Won't buy at any price
• Persuadables: Convert only with discount
• Sleeping Dogs: Buy only at full price

While elasticity models aggregate demand response, uplift modeling pinpoints which customers to target, preventing margin destruction on price-insensitive segments.