Demand Measurement and Price Elasticity

Introduction

Retail pricing decisions are fundamentally demand decisions. A retailer may know its costs, competitors’ prices and desired margin, yet still price poorly if it cannot estimate how customers respond to price changes. Demand measurement converts observed customer and market behaviour into evidence about quantity sold, revenue, margin and customer response. Price elasticity then expresses the responsiveness of demand to changes in price or related commercial variables.

For retail managers, elasticity is not merely a theoretical ratio. It supports markdown planning, promotional design, assortment decisions, private-label strategy, price architecture and revenue forecasting. However, elasticity estimates can be misleading when they confuse correlation with causation, ignore stock-outs, fail to control for promotions, or use data that do not represent the target market.

This chapter explains own-price, cross-price and promotional elasticity; introduces transaction, panel, survey and market-research data; distinguishes association from causal impact; and shows how retailers can combine quantitative models with managerial judgement.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain demand, elasticity and the principal determinants of retail demand.
2. Calculate and interpret own-price, cross-price and promotional elasticity.
3. Distinguish correlation from causation in retail data.
4. Evaluate transaction, panel, survey and market-research data for forecasting.
5. Specify and interpret basic demand models.
6. Design a practical elasticity study and recognise its limitations.
7. Apply elasticity evidence to pricing, promotions and retail decision-making.

1. Demand in a Retail Context

Demand is the quantity of a product that consumers are willing and able to purchase at specified conditions during a specified period. In retailing, observed sales are not always equal to demand. If a product sells out, recorded sales reflect available inventory rather than the quantity customers would have purchased. Consequently, demand measurement must consider price, availability and the selling environment.

A useful general representation is:

Here, Q_i is quantity demanded for product i; P_i is its own price; P_j represents prices of substitutes or complements; I is consumer income or purchasing power; A represents advertising and promotion; X includes product, customer and store characteristics; S represents seasonality and calendar effects; T captures trend; and ε_i is an unobserved component. The function is not a complete causal model until the variables, timing and identification assumptions are specified.

Retail demand is shaped by several forces:

• Price and perceived value: Consumers compare the monetary price with expected benefits and alternatives.
• Income and budget constraints: Essential and discretionary categories respond differently to changes in purchasing power.
• Substitutes and complements: Products compete with close alternatives and may also stimulate demand for associated products.
• Availability and convenience: Location, delivery speed, shelf placement and stock availability influence purchase probability.
• Promotion and communication: Displays, coupons, loyalty rewards and advertising can alter both awareness and price perception.
• Time and context: Weather, festivals, pay cycles, holidays and local events affect demand.
• Customer heterogeneity: Different segments may exhibit substantially different elasticities.

A critical distinction is between market demand and observed sales. Market demand concerns the quantity consumers would purchase under stated conditions. Observed sales are the transactions actually recorded and may be constrained by inventory, store hours, website performance, assortment and execution quality.

2. Elasticity: Meaning and Interpretation

Elasticity measures the percentage change in one variable resulting from a one-percent change in another variable. The percentage formulation makes comparisons possible across products with different units and price levels.

The general elasticity of demand with respect to a variable X is:

For sufficiently small changes, the point elasticity is:

The derivative ∂Q/∂X represents the local change in quantity associated with a change in X. The ratio is dimensionless, provided both percentage changes use the same basis. Elasticity is local: it may differ across customer segments, price ranges, stores and time periods.

For discrete before-and-after observations, the midpoint or arc elasticity is preferred because it avoids making the result depend on which observation is treated as the base:

The midpoint method is most useful when changes are not infinitesimally small. It assumes the two observations are sufficiently comparable and that other influential conditions have been controlled or are unchanged.

2.1 Own-Price Elasticity

Own-price elasticity measures the responsiveness of a product’s quantity demanded to its own price:

For ordinary downward-sloping demand, E_p is negative. Managers often discuss its absolute value, |E_p|, to classify responsiveness:

These categories must be interpreted carefully. They describe quantity response, not automatically profit response. A price increase may raise revenue when demand is inelastic, but profitability also depends on unit cost, volume, fixed costs, inventory and competitive reactions.

Worked Example 1: Own-Price Elasticity

A retailer raises the price of a product from £10 to £11. Quantity falls from 1,000 to 920 units. Using arc elasticity:

The demand is inelastic over this interval because |E_arc| ≈ 0.875 < 1. Revenue before the change is £10,000; after the change it is £10,120. The example illustrates why revenue may rise when demand is inelastic. It does not prove that profit rises, because cost information is absent.

2.2 Cross-Price Elasticity

Cross-price elasticity measures the response of demand for product i to the price of product j:

A positive value often indicates substitutes: if the price of branded coffee increases, demand for a private-label alternative may rise. A negative value often indicates complements: if the price of game consoles rises, demand for compatible games may fall. A value near zero may indicate weak economic relationship, although measurement error can also produce a low estimate.

Cross-price elasticity is especially important in assortment pricing. Changing the price of one SKU may shift demand among sizes, brands or pack formats rather than changing total category demand. A retailer should therefore examine both SKU-level and category-level effects.

2.3 Promotional Elasticity

Promotional elasticity measures the responsiveness of quantity to a promotional variable. Depending on the design, the variable may be discount depth, promotional price, display exposure, coupon availability or advertising expenditure:

A promotional elasticity estimate is meaningful only when the promotional measure is clearly defined. “Promotion” may combine several actions, such as a temporary price reduction, end-cap display, digital advertising and loyalty points. If these elements are not separated, the estimate represents a bundle effect rather than the effect of one instrument.

Promotional sales also include baseline demand, incremental demand and forward buying. Baseline demand is expected sales without the promotion. Incremental demand is additional sales caused by the promotion. Forward buying occurs when customers purchase earlier or in larger quantities, potentially reducing later sales. A high sales spike therefore does not necessarily mean that a promotion generated a proportionately high increase in total consumption.

3. From Elasticity to Revenue and Profit

If price is P and quantity is Q, revenue is:

For a small price change, the approximate percentage change in revenue is:

Since %∆Q ≈ E_p × %∆P, it follows that:

This approximation assumes a sufficiently small change and a stable elasticity over the interval. With negative own-price elasticity, a price increase tends to raise revenue when E_p > −1 and reduce revenue when E_p < −1. This is a revenue result, not a profit rule.

With unit variable cost c, contribution profit before fixed costs is:

Under a differentiable demand function and a single product, the profit-maximising interior price satisfies the Lerner condition:

The condition assumes a firm with market power, known or stable marginal cost, no binding capacity constraint, a correctly specified demand relationship and a meaningful point elasticity. Retail managers should not apply it mechanically where competition, inventory liquidation, multi-product substitution or strategic objectives dominate.

4. Measuring Demand with Retail Data

4.1 Transaction and Point-of-Sale Data

Transaction data record purchases at SKU, basket, store and time levels. Typical fields include product identifier, quantity, paid price, discount, timestamp, store or channel, customer identifier where permitted, payment method and promotion code.

Strengths include large sample size, actual purchase behaviour, detailed timing and the ability to estimate heterogeneous responses. Limitations include stock-out bias, missing competitor prices, incomplete customer information, return transactions, price changes triggered by anticipated demand, and inability to observe non-buyers.

A transaction dataset should be audited for:

• duplicate or cancelled transactions;
• returns and exchanges;
• tax-inclusive versus tax-exclusive prices;
• coupon attribution and loyalty redemptions;
• stock availability and replenishment;
• product substitutions and pack-size changes;
• store closures and channel migration.

4.2 Household and Consumer Panels

Panel data follow the same households or individuals over time. They can reveal purchase frequency, brand switching, category penetration and customer-level heterogeneity. Panel attrition, under-reporting, recruitment bias and privacy restrictions must be considered. A panel can show what participating households purchased, but it may not represent non-participants or all retail channels.

4.3 Surveys and Stated-Preference Data

Surveys can measure attitudes, willingness to pay, perceived value, purchase intentions and reactions to hypothetical prices. Choice experiments can present respondents with alternative products and prices, enabling estimation of stated preferences.

Stated data are useful when a product is new or historical variation is limited. However, hypothetical bias is a central limitation: people may claim that they would buy at a price that they would reject in an actual store. Survey design should use realistic choice tasks, randomisation, comprehension checks and, where possible, calibration against observed behaviour.

4.4 Market-Research and Competitive Data

Market-research sources include syndicated category data, competitor price audits, store traffic measures, web-scraped prices, macroeconomic indicators, search trends and distribution measures. These data help estimate market share, category growth and competitive response. Their value depends on coverage, frequency, methodological transparency and comparability of product definitions.

5. Correlation, Causation and Identification

Correlation describes statistical association; causation concerns what would happen to demand if a specific factor were changed while relevant alternatives were held constant. A retailer may observe that products with displays sell more, but displays may have been assigned to products already expected to sell well. The observed association then overstates the causal display effect.

Common sources of bias include:

• Confounding: A third factor affects both price and sales, such as seasonality.
• Reverse causality: Expected high demand causes a retailer to raise price.
• Selection bias: Promotions are targeted at stores or customers unlike the comparison group.
• Simultaneity: Competitors and the retailer change prices in response to each other.
• Stock-out censoring: Low recorded sales reflect limited supply rather than weak demand.
• Post-treatment bias: A variable affected by the promotion is incorrectly controlled for.

5.1 Randomised Experiments

A/B tests or geographically controlled experiments can provide strong causal evidence. Stores, customers or online visitors are randomly assigned to treatment and control conditions. Randomisation balances observed and unobserved factors in expectation.

A simple average treatment effect is:

The notation means the difference between expected quantity under treatment and expected quantity under control. Valid interpretation requires appropriate randomisation, stable treatment exposure, no material interference between units and accurate measurement. In retail, interference may occur when customers travel between stores or when online prices are visible across regions.

5.2 Quasi-Experimental Approaches

When randomisation is impractical, analysts may use difference-in-differences, instrumental variables, regression discontinuity or matched comparisons. Difference-in-differences compares changes over time in a treated group with changes in a control group:

The key assumption is parallel trends: absent treatment, the treated and control groups would have followed similar trends. This assumption should be examined using pre-treatment data; it cannot be guaranteed merely by applying the formula.

6. Demand Models for Estimation and Forecasting

6.1 Linear Demand Model

A basic model is:

Here, Q_t is quantity in period t; P_t is own price; A_t is promotional exposure; P_j,t is a related product’s price; S_t represents seasonality; T_t is trend; β terms are parameters; and u_t is an error term. A negative β₁ is consistent with ordinary demand, but the coefficient is not itself elasticity. At a given observation, own-price elasticity is approximately:

This model is easy to interpret but may predict negative quantities and imposes constant unit effects.

6.2 Log-Log Demand Model

A commonly used alternative is:

In this specification, β₁ is the own-price elasticity, β₂ is cross-price elasticity and β₃ is promotional elasticity, subject to correct specification and interpretation of the variables. Z_t is a vector of control variables. Log transformations require positive observations and careful treatment of zero sales or zero promotion. Adding an arbitrary constant to zeros changes interpretation and may bias results.

6.3 Panel and Hierarchical Models

Store-by-week or customer-by-period data permit fixed-effects models that control for stable differences among stores or customers. A typical model includes store effects, time effects and interactions between price and customer characteristics. Hierarchical or mixed-effects models can partially pool information across stores, categories or segments, which is useful when some units have limited observations. These models require attention to clustering, shrinkage assumptions and out-of-sample validation.

6.4 Forecast Evaluation

Forecast quality should be evaluated on data not used to estimate the model. Common metrics include mean absolute error:

and root mean squared error:

Here, Q_t is actual demand, Q̂_t is forecast demand and n is the number of observations. MAE is expressed in the units of quantity; RMSE penalises large errors more heavily. Forecast accuracy alone is insufficient: a model may forecast well but provide biased causal elasticity estimates.

7. A Practical Elasticity-Estimation Workflow

A robust retail study generally follows these stages:

1. Define the decision: price increase, markdown, promotion, assortment change or category strategy.
2. Define the outcome: units, revenue, contribution, repeat purchase, category spend or market share.
3. Specify the unit of analysis: SKU-store-day, customer-week, category-region or another defensible level.
4. Construct variables: paid price, regular price, discount depth, display, availability, competitor price, seasonality and customer characteristics.
5. Diagnose the data: inspect outliers, missingness, returns, stock-outs and structural breaks.
6. Choose a design: experiment where feasible; otherwise use a credible quasi-experimental or observational design.
7. Estimate the model: include economically relevant controls and appropriate fixed effects or clustering.
8. Test robustness: compare specifications, samples, time windows and customer segments.
9. Validate forecasts: use holdout periods and assess errors by store, product and demand level.
10. Translate results into decisions: simulate revenue, margin, inventory and competitor responses.
11. Monitor after implementation: compare predicted and realised outcomes and update the model.

Elasticity should be reported with confidence intervals or other uncertainty measures. A point estimate of −1.2 without uncertainty may create false precision, particularly when price variation is limited or observations are correlated.

8. Illustrative Case: FreshBite Supermarkets

Background

FreshBite, a fictional regional supermarket chain, sells a 500-gram private-label coffee pack alongside a leading national brand. Management is considering a 5% price increase on the private-label product and wants to understand own-price and cross-price effects.

Problem

Historical data show that the private-label price was higher during holiday periods, when coffee demand was also strong. A simple comparison would therefore risk concluding that higher prices increase demand. In addition, the national brand frequently ran promotions, and some stores experienced stock-outs.

Analysis

FreshBite assembled 52 weeks of store-SKU data, including units, net paid price, regular price, national-brand price, display status, promotional leaflet exposure, inventory availability, store fixed effects, week fixed effects and local competitor prices. Weeks with severe stock-outs were flagged rather than treated as ordinary zero-demand observations.

The team first plotted prices and quantities and examined pre-existing trends. It then estimated a log-log panel model with store and week fixed effects. The private-label own-price coefficient was −1.35, with a confidence interval that indicated meaningful uncertainty. The cross-price coefficient on the national brand was +0.42, suggesting substitution: when the national brand became more expensive, private-label demand tended to rise. A display variable was positive, but because displays were allocated partly by category managers, the team did not interpret its coefficient as fully causal.

To improve identification, FreshBite conducted a six-week staggered test. Comparable stores were randomly assigned to maintain the existing price or implement the proposed increase, while promotional activity was held constant. The experimental estimate was smaller in magnitude than the historical estimate, indicating that the observational model had likely captured some confounding variation.

Approach and Solution

Management simulated three scenarios: no change, a 5% price increase and a 5% increase combined with additional display support. The simulation incorporated estimated own-price response, expected substitution from the national brand, contribution margin and inventory constraints. The price increase was implemented only in stores with reliable availability, while a monitoring rule was established for unit volume, category revenue, contribution and customer switching.

Outcome

The price increase improved private-label unit margin but reduced unit volume. Category contribution increased modestly in the test stores, although the effect varied by local competition and customer segment. The chain decided not to apply one national price blindly. Instead, it adopted a segmented pricing policy and continued experimentation.

Learning Points

1. Historical price-sales relationships may be confounded by seasonality and managerial targeting.
2. SKU elasticity and category elasticity are different decision measures.
3. Cross-price effects can make a product-level price increase beneficial or harmful at category level.
4. Stock availability must be incorporated into demand analysis.
5. Experiments and observational models are complementary rather than interchangeable.
6. Estimates should be converted into contribution and inventory scenarios, not used as isolated statistics.

9. Managerial, Ethical and Governance Considerations

Pricing analytics can improve value creation, but it also creates governance responsibilities. Customer-level data should be collected and processed lawfully, transparently and proportionately. Personally identifiable information should be minimised, access controlled and retained only as necessary. Models should be reviewed for discriminatory effects, particularly when prices, offers or credit-related conditions vary across customer groups.

Dynamic pricing requires clear communication. A price that changes frequently or differs across channels may be commercially lawful in some settings but still damage trust if customers perceive it as unfair or deceptive. Retailers must comply with consumer-protection, competition, pricing-display, tax and privacy requirements in relevant jurisdictions.

AI and automated pricing systems can identify nonlinear patterns and personalise forecasts, but they do not remove the need for causal design. Automated systems may learn historical discrimination, respond to spurious correlations or amplify competitor reactions. Human oversight, audit trails, approval thresholds, drift monitoring and the ability to explain material decisions are essential controls.

10. Limitations and Contemporary Developments

Elasticity is not a permanent product attribute. It changes with competition, inflation, income, brand positioning, channel, inventory, consumer expectations and time horizon. Short-run elasticity may differ from long-run elasticity because consumers need time to switch brands or change habits. Digital retail also creates rapid price comparison, personalised promotions and large volumes of clickstream data, but browsing is not identical to purchasing.

Current practice increasingly combines causal experiments, machine-learning forecasting and structured economic models. Machine learning can improve predictive accuracy through nonlinearities and interactions, whereas econometric models often provide clearer causal interpretation. The strongest systems separate these goals: forecasting models predict demand, while causal models estimate the effect of an intervention. Both should be validated against business outcomes and monitored for changes in the data-generating process.

Best-Practice Checklist

• Define demand and the decision before selecting a method.
• Use net paid prices and distinguish regular price from promotional price.
• Record stock availability and identify stock-out periods.
• Control for seasonality, trends, holidays and competitor actions.
• Prefer randomised tests when operationally and ethically feasible.
• Do not interpret association as causal impact without a credible design.
• Estimate both SKU-level and category-level effects.
• Segment cautiously and avoid unstable estimates from very small samples.
• Report uncertainty, sensitivity and forecast performance.
• Translate elasticity into revenue, contribution, inventory and customer outcomes.
• Protect customer data and review automated pricing for fairness and compliance.
• Monitor realised results and update estimates as market conditions change.

Chapter Summary

Demand measurement links retail observations to pricing decisions, but observed sales are not identical to unconstrained demand. Own-price elasticity measures the response of a product’s quantity to its own price; cross-price elasticity measures substitution or complementarity; and promotional elasticity measures response to a defined promotional variable. Arc elasticity is appropriate for finite changes, while point elasticity describes local responsiveness.

Elasticity assists revenue and contribution analysis, but it must be interpreted within a complete demand system. Transaction, panel, survey and market-research data each provide valuable but imperfect evidence. Correlation does not establish causation: confounding, reverse causality, selection, simultaneity and stock-outs can bias estimates. Experiments provide strong evidence when properly designed, while quasi-experimental and panel methods can support credible inference when randomisation is unavailable.

A professional elasticity workflow defines the decision, constructs reliable variables, selects an appropriate design, estimates and validates models, reports uncertainty, and converts results into operational scenarios. Ethical data governance, consumer transparency and human oversight are essential when pricing systems use customer-level data or AI.

Key Terms

Arc elasticity: Elasticity calculated from finite changes using midpoint percentage changes.

Baseline demand: Expected demand in the absence of a specified intervention.

Causal effect: The change in an outcome attributable to an intervention under a defined counterfactual.

Complement: A product whose demand is generally related negatively to the price of another product.

Confounding: A situation in which another factor influences both the explanatory variable and the outcome.

Cross-price elasticity: Responsiveness of demand for one product to the price of another.

Demand: Quantity consumers are willing and able to purchase under specified conditions.

Difference-in-differences: A quasi-experimental estimator comparing changes across treatment and control groups.

Elastic demand: Demand for which the absolute value of elasticity exceeds one.

Forward buying: Purchasing earlier or in greater quantity because of a temporary promotion.

Incremental demand: Additional demand attributable to an intervention relative to baseline.

Inelastic demand: Demand for which the absolute value of elasticity is below one.

Own-price elasticity: Responsiveness of quantity demanded to the product’s own price.

Panel data: Repeated observations on the same households, customers, stores or other units.

Point elasticity: Local elasticity evaluated at a particular point on a demand relationship.

Promotional elasticity: Responsiveness of demand to a specified promotional variable.

Stock-out bias: Distortion caused when observed sales are limited by unavailable inventory.

Substitute: A product whose demand tends to increase when the price of another product increases.

Treatment effect: Difference in outcomes caused by exposure to an intervention.

Review Questions

Short Answer Questions

1. Why are observed retail sales not always equal to demand?
2. Define own-price elasticity and explain the meaning of a negative value.
3. What is the difference between a substitute and a complement?
4. Why can a promotion create forward buying?
5. State one important assumption of difference-in-differences analysis.

Long Answer Questions

1. Explain the differences among transaction, panel, survey and market-research data, including the strengths and limitations of each.
2. Discuss why correlation between price and sales does not necessarily establish a causal price effect.
3. Explain how own-price elasticity can inform revenue and contribution decisions.
4. Compare linear and log-log demand models and identify situations in which each may be useful.
5. Develop a complete workflow for estimating promotional elasticity in a supermarket category.

Analytical / Application Questions

1. A product’s price rises from £20 to £22 and quantity falls from 500 to 460 units. Calculate arc price elasticity and classify demand as elastic or inelastic. Explain the revenue implication.
2. A retailer observes higher sales in stores receiving end-cap displays. Design an empirical strategy to determine whether the displays caused the increase.
3. A private-label product has a positive cross-price elasticity with a national brand. Explain how this result should influence a proposed price increase, and identify additional data required before implementation.

Multiple Choice Questions with Answers

Question 1

Which expression correctly defines own-price elasticity of demand?

A. Percentage change in price divided by percentage change in quantity

B. Percentage change in quantity divided by percentage change in own price

C. Change in revenue divided by change in cost

D. Change in market share divided by change in income

Correct Answer: B. Percentage change in quantity divided by percentage change in own price

Question 2

If own-price elasticity is −0.6, demand is best classified as:

A. Elastic

B. Unit elastic

C. Inelastic

D. Perfectly elastic

Correct Answer: C. Inelastic

Question 3

A positive cross-price elasticity most commonly indicates that two products are:

A. Complements

B. Unrelated necessities

C. Joint products

D. Substitutes

Correct Answer: D. Substitutes

Question 4

A retailer raises price from £10 to £11 and quantity falls from 1,000 to 920 units. The approximate arc elasticity is:

A. −0.875

B. −1.250

C. 0.875

D. 1.143

Correct Answer: A. −0.875

Question 5

Which problem occurs when a product sells out and recorded sales are treated as demand?

A. Selection neutrality

B. Stock-out bias

C. Perfect identification

D. Measurement invariance

Correct Answer: B. Stock-out bias

Question 6

In a log-log demand model, the coefficient on log own price is commonly interpreted as:

A. The fixed cost of the product

B. The unit margin

C. Own-price elasticity

D. The average basket size

Correct Answer: C. Own-price elasticity

Question 7

Which condition is central to a difference-in-differences design?

A. Prices must be identical in every period

B. Treatment and control groups must have equal sample sizes

C. The outcome must never vary before treatment

D. Treated and control groups would have followed parallel trends without treatment

Correct Answer: D. Treated and control groups would have followed parallel trends without treatment

Question 8

Why might a positive association between displays and sales fail to prove that displays caused higher sales?

A. Displays may be assigned to products already expected to sell well

B. Sales can never be measured in units

C. Displays automatically reduce availability

D. Price has no relationship with demand

Correct Answer: A. Displays may be assigned to products already expected to sell well

Question 9

Which metric is expressed in the same units as quantity?

A. Mean absolute error

B. Cross-price elasticity

C. Promotional elasticity

D. Percentage margin

Correct Answer: A. Mean absolute error

Question 10

A promotion produces a large sales spike, but sales fall below normal in later weeks. Which concept may explain part of this pattern?

A. Parallel trends

B. Forward buying

C. Perfect competition

D. Constant marginal cost

Correct Answer: B. Forward buying