Syntha Lab RU

Price elasticity of demand: how to estimate it from data without being fooled

How to obtain price elasticity of demand from sales history: log-log regression, point and arc elasticity, cross elasticity, the optimal price and the main traps (seasonality, promotions, cannibalization). The result decides how far a price can be changed, what volume response to expect, and when the estimate cannot be trusted.

Math and analytics Brand and retail Verified: 2026-11-13

Open interactively →

Why this document

The pricing lead and the commercial director read this before changing a price or running a promotion. It matters because prices are moved on how demand will react, and a naive estimate from history usually lies: a low price coincides with a weak season, so the relationship looks weaker than it is. Without controlling for season and promotion it is easy to underestimate sensitivity and cut a price for nothing, or raise one for nothing. An estimate with controls and an interval shows what can be relied on.

At a glance

What is this document and why is it needed?
How to obtain price elasticity of demand from sales history: log-log regression, point and arc elasticity, cross elasticity, the optimal price and the main traps (seasonality, promotions, cannibalization). The result decides how far a price can be changed, what volume response to expect, and when the estimate cannot be trusted.
Who owns the document and how often is it rebuilt?
Pricing or commercial analyst; estimated once a season by category and before every major price change; read by the buyer, the product director and the finance lead.
What mistakes are the most common?
Elasticity is computed from price alone without a season control → a biased estimate (understated in the example) → add the seasonal index and a promo flag. The estimate is extrapolated beyond the price range → an unrealistic forecast → apply only within the observed range, pilot for the rest. The post-promotion dip is ignored → the effect is overstated → use a window with the post-promotion trough.

For whom

Brand
A brand changes the recommended retail price and wholesale terms rarely and for a whole season: elasticity in own channels (D2C, marketplaces) is estimated from price and promotion history, and for wholesale from sell-out response at different price points. An error in elasticity costs a whole season of margin.
Retail
A retailer changes prices often (promotions, markdowns) and has many observations by store and week: elasticity is estimated by category and price band with controls for season and promotion, and cross elasticity is used to gauge cannibalization within a category.

What it consists of

What elasticity is and how to read it

  • Elasticity E = percentage change in volume per 1% change in price; with E = −2 a 10% price cut gives roughly a 20% volume rise. In absolute value |E| > 1 means elastic demand (revenue rises when price falls), |E| < 1 inelastic.
  • Elasticity is not a constant: it depends on the price range, season, item role (basic or fashion) and competitors; an estimate is valid only within the observed price range.

Estimation methods

  • Arc: E = [(q₂ − q₁)/((q₁ + q₂)/2)] ÷ [(p₂ − p₁)/((p₁ + p₂)/2)]; for price 4,000 → RUB 3,200 and volume 80 → 132 units it is -2.21, the point (log) estimate is -2.24. The two numbers are close for small changes; for deep changes arc describes an episode more reliably.
  • Log-log regression: ln q = a + E·ln p + controls + error; the coefficient on ln p is the elasticity (constant in the model). Controls: seasonal index, promotion and display flag, stock availability, competitor price.
  • Cross elasticity E_AB = (% change in volume of B) ÷ (% change in price of A): with a 20% price cut on A and a 6% drop in volume of B it is 0.3: the items substitute for each other and part of A growth is cannibalization of B.

How not to be fooled

  • The main trap is price coinciding with season: prices are cut when demand is falling anyway. In the example (52 weeks, true value 1.8) the naive one-variable regression gives 1.1 while with season and promotion controls it is 1.8 ± 0.36: the naive estimate understates sensitivity by 39%.
  • A promotion simultaneously cuts price and adds display, mailing and shelf space: without a separate promo flag the advertising effect is credited to price and elasticity is overstated.
  • Pull-forward: sales dip after a promotion; elasticity must be computed over a window that includes the post-promotion dip, otherwise the effect is overstated.
  • Brand: prices differ across partners and channels and are linked, so elasticity is computed by channel and flows between channels are checked. Retailer: prices are cut non-randomly (on leftovers), so stock and item age must be controlled.

The profit-maximizing price

  • With constant elasticity e (absolute, e > 1) and cost c the optimal price is p* = c·e/(e − 1): for c = RUB 2,000 and e = 2.2 that is RUB 3667. This is an upper bound of the reasoning: the real price is limited by positioning, competitors and the price ladder.
  • At |e| ≤ 1 the formula is undefined: profit rises with price and only competition and brand perception limit it.

Brand and retailer: elasticity under different conditions

  • Brand: price for a brand is the recommended retail and the wholesale price: elasticity in retail channels is estimated on own sales, and for wholesale the partner sell-out response to its price change is used.
  • Retailer: price for a retailer is the shelf price: it has many observations by store and week, so a panel regression with store, season and promotion controls is possible.
  • Brand: brand prices change rarely and in steps (season, price list review), so there is little variation and the estimate is backed by a price test on part of D2C traffic or on a marketplace.
  • Retailer: retailer prices change often but not randomly; to separate the price effect from the effect of leftovers and item age, age and stock are added to the model.
  • Brand: a brand computes cross elasticity between its lines and between channels: a price cut in D2C shifts partner sales and this flow enters the effect estimate.
  • Retailer: a retailer computes cross elasticity between substitutes within a category and between brands; cannibalization within the category shows whether the model brought growth or merely shifted sales.
  • Brand: a brand agrees markdown depth with partners: a too deep brand promotion in D2C devalues partner leftovers, so elasticity is complemented by an estimate of partner reactions.
  • Retailer: a retailer chooses markdown depth by elasticity and by the leftover liquidation price: elasticity together with the payoff threshold shows when depth stops working.
  • Brand: a brand decides on price a season ahead, so the elasticity interval is turned into revenue scenarios (low, middle, high elasticity).
  • Retailer: a retailer revisits the price decision weekly, so the elasticity estimate is updated over a rolling window and used for automatic markdown rules.

Where the data comes from

Data sourceFields
Weekly sales and pricesVolume, price excluding VAT (before and after discount), share of promotional sales by SKU, store and week; at least 40-60 observations per group.
Promotion and event calendarPromotion type, display, mailing, holidays, weather anomalies, stock availability.
Competitor pricesPrice monitoring of comparable items; without it elasticity mixes with competitor reactions.

How to build it step by step

  1. Step 1. Define the estimation group: a category and price band, not a single SKU (too few observations per SKU).
  2. Brand: Step 2. Take the series by channel (D2C, marketplace) separately and add partner prices of the same model to the controls.
  3. Retailer: Step 3. Take the series by week and store and add controls for item age and stock to separate a price cut from natural decay.
  4. Step 4. Extract the seasonal index (see demand forecasting) and a promo flag; reconstruct demand on out-of-stock days.
  5. Step 5. Plot ln q against ln p and look at the cloud: if price barely varies or varies only with season the estimate will be unreliable.
  6. Step 6. Estimate the regression of ln q on ln p without and with controls; compare coefficients and get the standard error and interval.
  7. Step 7. Check plausibility: negative sign, magnitude in reasonable bounds, interval excludes zero; estimates by category should not contradict common sense.
  8. Step 8. Estimate cross elasticity for pairs of substitutes and include cannibalization in the promotion forecast.
  9. Step 9. Confirm the estimate by an experiment where possible (a price test on a group of stores or part of traffic) and recompute.
  10. Step 10. Apply the estimate only within the observed price range; for going beyond it plan a pilot.

Formulas

MeasureFormulasNote
Point elasticityE = ln(q₂/q₁) / ln(p₂/p₁) (for small changes E ≈ (dq/q)/(dp/p))Symmetric with respect to the direction of the change.
Arc elasticityE = [(q₂ − q₁) / ((q₁ + q₂)/2)] / [(p₂ − p₁) / ((p₁ + p₂)/2)]Describes one episode; does not remove the effect of season.
Log-log modelln q_t = a + E·ln p_t + β₁·ln s_t + β₂·promo_t + ε_ts_t seasonal index; E is the constant elasticity; OLS.
Estimate intervalE ± 1.96·SE(E), SE = √( s²·[(XᵀX)⁻¹]_(22) ), s² = SSE/(n − k)Assumes independent residuals; with autocorrelation the interval is too narrow.
Cross elasticityE_AB = (Δq_B / q_B) / (Δp_A / p_A)Positive for substitutes, negative for complements.
Optimal pricep* = c · e / (e − 1), e = |E| > 1Only with constant elasticity; c is cost excluding VAT.
Volume change under a price moveq₂ / q₁ = (p₂ / p₁)^EFor E = −2 a 30% price cut gives a 2.04-fold volume rise.

Formulas in notation

OLS estimate
β̂ = (XᵀX)⁻¹ Xᵀ y; SE(β̂_j) = √( σ̂²·[(XᵀX)⁻¹]_{jj} )
Solution of the normal equations and the standard error.
Profit as a function of price
π(p) = (p − c)·q₀·(p/p₀)^{−e}
The profit curve for the chart.

Worked example

Elasticity estimated from 52 weeks: naive, with controls and by category

EstimateElasticity (absolute)Standard errorTrue model value
Naive (price only)1.10.191.8
With season and promo controls1.80.191.8
Apparel1.680.181.6
Footwear1.360.171.3
Accessories2.110.192.1
Kids1.040.191.1
Men1.360.181.4
Women1.660.21.9

Weekly series are generated in code from a known elasticity (demand = level × season × price^−e × promo effect × noise), prices are cut mostly in the weak season. The naive estimate 1.1 is understated; the estimate with controls 1.8 at R² 0.95 is closer to the truth. The data is illustrative: on real data the direction of the bias can be any.

Figures

Naive estimate: ln volume against ln price
Scatter plot: 52 points; regression line: y = 6.323 + −1.102·x66.26.46.66.877.2−0.4−0.3−0.2−0.1−0−0.356; 6.55−0.32; 6.55−0.274; 6.46−0.088; 6.03−0.109; 6.17−0.26; 6.53−0.175; 6.53−0.017; 6.11−0.331; 6.74−0.249; 6.580; 6.15−0.134; 6.48−0.013; 6.23−0.008; 6.34−0.233; 6.81−0.287; 6.93−0.008; 6.44−0.011; 6.37−0.3; 7.08−0.003; 6.39−0.303; 6.97−0.01; 6.48−0.011; 6.41−0.02; 6.54−0.008; 6.46−0.009; 6.47−0.172; 6.94−0.017; 6.41−0.013; 6.5−0.016; 6.42−0.018; 6.48−0.134; 6.68−0.019; 6.44−0.001; 6.43−0.01; 6.37−0.01; 6.38−0.013; 6.26−0.012; 6.4−0.015; 6.22−0.014; 6.2−0.005; 6.17−0.007; 6.13−0.287; 6.61−0.011; 6.06−0.323; 6.69−0.257; 6.53−0.32; 6.72−0.353; 6.63−0.335; 6.64−0.312; 6.53−0.306; 6.4−0.209; 6.2ln price (price index)↑ ln weekly volume
  • points
  • regression line: y = 6.323 − 1.102·x
Data as a table
ln price (price index)ln weekly volume
−0.366.55
−0.326.55
−0.276.46
−0.096.03
−0.116.17
−0.266.53
−0.186.53
−0.026.11
−0.336.74
−0.256.58
06.15
−0.136.48
−0.016.23
−0.016.34
−0.236.81
−0.296.93
−0.016.44
−0.016.37
−0.37.08
−06.39
−0.36.97
−0.016.48
−0.016.41
−0.026.54
−0.016.46
−0.016.47
−0.176.94
−0.026.41
−0.016.5
−0.026.42
−0.026.48
−0.136.68
−0.026.44
−06.43
−0.016.37
−0.016.38
−0.016.26
−0.016.4
−0.026.22
−0.016.2
−0.016.17
−0.016.13
−0.296.61
−0.016.06
−0.326.69
−0.266.53
−0.326.72
−0.356.63
−0.346.64
−0.316.53
−0.316.4
−0.216.2

Naive estimate: ln volume against ln price. The line through the cloud has a slope weaker than the true one: low prices coincide with the weak season.

After controlling for season and promotion
Scatter plot: 52 points; regression line: y = 6.181 + −1.801·x66.26.46.66.87−0.4−0.3−0.2−0.1−0−0.356; 6.83−0.32; 6.81−0.274; 6.7−0.088; 6.33−0.109; 6.43−0.26; 6.66−0.175; 6.61−0.017; 6.25−0.331; 6.75−0.249; 6.540; 6.18−0.134; 6.46−0.013; 6.18−0.008; 6.26−0.233; 6.61−0.287; 6.7−0.008; 6.28−0.011; 6.18−0.3; 6.78−0.003; 6.17−0.303; 6.64−0.01; 6.23−0.011; 6.15−0.02; 6.27−0.008; 6.19−0.009; 6.2−0.172; 6.57−0.017; 6.14−0.013; 6.23−0.016; 6.16−0.018; 6.23−0.134; 6.45−0.019; 6.22−0.001; 6.22−0.01; 6.19−0.01; 6.22−0.013; 6.12−0.012; 6.29−0.015; 6.14−0.014; 6.15−0.005; 6.15−0.007; 6.15−0.287; 6.58−0.011; 6.15−0.323; 6.74−0.257; 6.62−0.32; 6.85−0.353; 6.79−0.335; 6.84−0.312; 6.77−0.306; 6.67−0.209; 6.48ln price (price index)↑ ln volume net of season and promo
  • points
  • regression line: y = 6.181 − 1.801·x
Data as a table
ln price (price index)ln volume net of season and promo
−0.366.83
−0.326.81
−0.276.7
−0.096.33
−0.116.43
−0.266.66
−0.186.61
−0.026.25
−0.336.75
−0.256.54
06.18
−0.136.46
−0.016.18
−0.016.26
−0.236.61
−0.296.7
−0.016.28
−0.016.18
−0.36.78
−06.17
−0.36.64
−0.016.23
−0.016.15
−0.026.27
−0.016.19
−0.016.2
−0.176.57
−0.026.14
−0.016.23
−0.026.16
−0.026.23
−0.136.45
−0.026.22
−06.22
−0.016.19
−0.016.22
−0.016.12
−0.016.29
−0.026.14
−0.016.15
−0.016.15
−0.016.15
−0.296.58
−0.016.15
−0.326.74
−0.266.62
−0.326.85
−0.356.79
−0.346.84
−0.316.77
−0.316.67
−0.216.48

After controlling for season and promotion. The season and promotion contributions are removed from ln volume: the slope is close to the true value of 1.8.

Elasticity estimates and the true value
Bar chart: Naive 1.1, With controls 1.8, Lower 95% bound 1.44, Upper 95% bound 2.16, True value 1.800.511.522.5NaiveNaiveElasticity (absolute): 1.11.1With controlsWith controlsElasticity (absolute): 1.81.8Lower 95% boundLower 95% boundElasticity (absolute): 1.441.44Upper 95% boundUpper 95% boundElasticity (absolute): 2.162.16True valueTrue valueElasticity (absolute): 1.81.8absolute
Data as a table
MeasureElasticity (absolute)
Naive1.1
With controls1.8
Lower 95% bound1.44
Upper 95% bound2.16
True value1.8

Elasticity estimates and the true value. Controls bring the estimate back toward the truth; the interval shows the residual uncertainty.

Estimates for six slices and the set values
Bar chart: Apparel 1.68, Footwear 1.36, Accessories 2.11, Kids 1.04, Men 1.36, Women 1.6600.511.522.5ApparelApparelEstimate: 1.681.68Set value: 1.61.6FootwearFootwearEstimate: 1.361.36Set value: 1.31.3AccessoriesAccessoriesEstimate: 2.112.11Set value: 2.12.1KidsKidsEstimate: 1.041.04Set value: 1.11.1MenMenEstimate: 1.361.36Set value: 1.41.4WomenWomenEstimate: 1.661.66Set value: 1.91.9absolute
  • Estimate
  • Set value
Data as a table
MeasureEstimateSet value
Apparel1.681.6
Footwear1.361.3
Accessories2.112.1
Kids1.041.1
Men1.361.4
Women1.661.9

Estimates for six slices and the set values. The estimate with controls recovers the set values within error.

Profit as a function of price at e = 2.2
Line chart: Profit, RUB thousand: 141 → 1721401601802002202 5003 1003 6674 4005 2005 600p*Profit, RUB thousand, 2 500: 141Profit, RUB thousand, 2 800: 175Profit, RUB thousand, 3 100: 193Profit, RUB thousand, 3 400: 200Profit, RUB thousand, 3 667: 202Profit, RUB thousand, 4 000: 200Profit, RUB thousand, 4 400: 195Profit, RUB thousand, 4 800: 188Profit, RUB thousand, 5 200: 180Profit, RUB thousand, 5 600: 172RUB thousand
Data as a table
MeasureProfit, RUB thousand
2 500141
2 800175
3 100193
3 400200
3 667202
4 000200
4 400195
4 800188
5 200180
5 600172

Profit as a function of price at e = 2.2. The maximum is near RUB 3,670 (p* = c·e/(e−1)); towards higher prices profit falls more slowly than towards lower ones.

Specifics by category

Apparel
Elasticity is higher on seasonal and fashion items and lower on basics; compute by price band and role, not for the whole category; at season end elasticity falls with interest in the item.
Footwear
Purchases are pricier and rarer, choice depends on size and feel: price elasticity is moderate while size availability matters strongly; without a stock control it is understated.
Accessories
Gift purchases are less price-sensitive at peaks and more so outside them; estimate elasticity separately for the peak and the off-season.
Kids
Need-driven purchases (size, school) make demand less elastic on essential items; for fashionable kids items elasticity is closer to women apparel.
Men
Demand is steadier and less sensitive to short promotions but sensitive to price level and brand; cross effects among basics are weaker.
Women
A wide choice and strong substitutes give high cross elasticity and cannibalization within a category; a promotion on one model shifts sales from neighbors.

Common mistakes

  • Elasticity is computed from price alone without a season control → a biased estimate (understated in the example) → add the seasonal index and a promo flag.
  • The estimate is extrapolated beyond the price range → an unrealistic forecast → apply only within the observed range, pilot for the rest.
  • The post-promotion dip is ignored → the effect is overstated → use a window with the post-promotion trough.
  • Cannibalization is not counted → growth of the promoted model is taken for category growth → estimate cross elasticity and look at the whole category.
  • The optimal price from the formula is taken as final → the price ladder and positioning are broken → use p* as an upper guide, not a target.

Russian specifics

  • Price for elasticity is taken excluding VAT: the 22% rate does not change relative price change but a rate change inside the series creates a false shelf-price jump. Rule: Tax Code Art. 164: Tax Code, Art. 164 (VAT rates).
  • The recommended retail price of a brand for partners must not become price coordination between competitors; the exchange of price information with partners is limited by competition-protection requirements. Rule: Law 135-FZ of 26.07.2006: Federal Law of 26.07.2006 No. 135-FZ on protection of competition (rule check not completed).
  • Promotional prices are subject to advertising law requirements on conditions and terms; promotion terms come from the calendar that also feeds elasticity. Rule: Law 38-FZ of 13.03.2006: Federal Law of 13.03.2006 No. 38-FZ on advertising (rule check not completed).

Open interactively →