Research & Evidence

Search for a method

Search titles, questions, territories and MSC identifiers.

    ← All methods
    METHOD DOSSIERMSC-P-022Pricing scienceVerified scientific dossier

    How do you estimate price elasticity and its uncertainty?

    In a log-log model, the coefficient on log price is a constant elasticity conditional on included variables. Without a credible design it remains associational and may be biased by price endogeneity.

    Scientific editorial team: Marketing Science Center

    Direct answer

    Estimate conditional sensitivity with interval and diagnostics.

    In a log-log model, the coefficient on log price is a constant elasticity conditional on included variables. Without a credible design it remains associational and may be biased by price endogeneity.

    Tellis, 1988

    01

    Direct answer

    In a log-log regression, the log-price coefficient estimates a constant elasticity conditional on included variables. On MSC-001, the season-controlled model gives −2.4586; this is an association, not a causal price effect.

    02

    Scientific question

    For the observed product, market and period, what relative quantity change is associated with a 1% price change, with what uncertainty and over what support?

    03

    Population, unit and horizon

    The unit is a product-market week. The data contain 24 weeks, prices from €10 to €11 and no promotions. Simulation horizons must remain close to this support.

    04

    Statistical target

    The default target is β in E[ln(Q)|ln(P),S]=α+βln(P)+γS. Counterfactual demand Q(p) separately requires a credible identification design.

    05

    Required data

    Use strictly positive quantity and price, time identifier, promotions, seasonality, competitor prices, availability, distribution and common drivers of price and demand. Declare each variable as pre- or post-treatment.

    06

    Assumptions

    Assume an adequate log-log form, errors compatible with the chosen inference, no unmodelled break and, for causal reading, no residual confounding or justified exogenous variation. R² does not prove these assumptions.

    07

    Model

    The simple model is ln(Q)=α+βln(P)+ε. The reference model adds season index S. Promotion is constantly zero and must be excluded to avoid a no-variation column.

    08

    Reproducible calculation

    Validate 24 rows, log-transform price and quantity, estimate both OLS specifications, calculate residual standard error, 95% CI and exact change (1.05)^β−1. The script fails when expected support is absent.

    09

    Results

    Simple model: β=−2.4107, SE=0.1198. With seasonality: β=−2.4586, SE=0.0251, 95% CI [−2.5078, −2.4093]. Under constant elasticity, +5% price corresponds to −11.3039% conditional volume.

    MSC-001 · log-log OLS
    SpecificationβSE95% CI
    ln(P)−2.4106830.119794—
    ln(P) + S−2.4585720.025131[−2.507828; −2.409317]

    10

    Uncertainty

    The conventional CI is conditional on the model and 24 observations. It includes neither functional-form error, endogeneity nor regime-change uncertainty. A real time series requires variance compatible with serial correlation.

    11

    Price endogeneity

    Managers may change price in response to unobserved expected demand. Price and error then correlate; OLS is associational and potentially biased. MSC-P-020 addresses instrumental-variable identification.

    12

    Diagnostics

    Plot residuals and fits, compare specifications, inspect leverage and influence, test temporal stability, serial correlation and collinearity, document scenario support and investigate the price-setting rule.

    13

    Common errors

    Common errors are calling β causal without a design, reading −2.46 as −2.46 units, applying β to a large change without exponentiation, controlling a post-price variable, ignoring zeros or extrapolating beyond support.

    14

    Interpretation

    The estimate describes highly price-sensitive demand in this small constructed dataset. Its narrow precision reflects artificial regularity; it is neither an industry benchmark nor a guarantee of real response.

    15

    Supported decision

    Use the associational estimate to compare nearby scenarios, show conditional volume sensitivity and decide what additional data or experiments are needed.

    16

    Unsupported decision

    Do not set an optimal price, promise a sales change, transport β to another market or causally attribute volume decline to price without identification and economic constraints.

    17

    Implementation

    The CC0 CSV and MIT Python script run dependency-free OLS, fail on nonpositive price or quantity and report both specifications, CI and the +5% scenario.

    18

    Expected deliverable

    Include question, unit, support, associational/causal status, specification, variables, coefficient, SE, CI, diagnostics, sensitivity, bounded scenario, supported/unsupported decision and reproducible files.

    19

    Scientific sources

    Tellis defines elasticity, links the log-log coefficient to elasticity and documents specification bias. Villas-Boas and Winer show why price endogeneity can bias parameters. Neither source validates MSC-001 as a real market.

    1. Tellis (1988) ↗Full text verified

    Dataset · Tool

    Dataset · MSC-001Synthetic price elasticity→Tool · MSC-T04Price-volume impact→

    Method connections

    Parent territoryPricing science: connecting price, demand and contribution

    Read next

    MSC-P-018Predictive or causal regression: what are you trying to estimate?→MSC-P-019How do you diagnose a marketing regression before interpreting it?→MSC-P-003How should uncertainty in a marketing result be expressed?→