Elasticity from a price and sales table: log-log regression with an interval
From a table of prices and sales over equal periods estimates the price elasticity of demand by least squares in logarithms, shows the confidence interval, the goodness of fit, the forecast of sales and gross profit at a new price and the profit-maximizing price.
At a glance
- What question does it answer?
- How sales followed the price, how sure the estimate is and what it promises at a new price.
- What data is needed?
- A table of periods: the average price excl. VAT and sales in units.
- How to read the result?
- Example on the default values (10 periods, illustrative data): elasticity −1.73, the interval from −1.92 to −1.54, R² 98.2%; when the price falls from 4,500 to RUB 4,300 sales rise by 8.2%, gross profit per period changes by −RUB 99,838. Example on the default values: Demand elasticity: −1.7 (−1.73); Interval: lower bound: −1.9 (−1.92); Interval: upper bound: −1.5 (−1.54); Standard error of elasticity: 0.1 (0.083). The figures are illustrative: enter your own to see your result.
Why and how to reason
Elasticity shows by how many percent sales change if the price changes by one percent. To estimate it, take real periods with different prices and look at how sales changed. The relationship is non-linear: a rise of 100 from 3,600 affects demand more than from 5,400. So both measures are taken in logarithms, and the slope of the line in logarithms is the elasticity. For ten periods it came out at minus 1.73: a 1% rise in price lowers sales by 1.73%. The interval from minus 1.92 to minus 1.54 shows how reliable the estimate is. R² of 98% says price explains almost all the variability. Cutting the price from 4,500 to 4,300 (−4.4%) gives +8.2% sales, but gross profit falls by 100 thousand: margin per unit fell from 2,300 to 2,100.
When to use it
This is the calculation for the case when you have a history of prices and sales for comparable periods without other strong influences (promotions, stock-outs, season) and need to decide whether to change the price. Do not use it if price changed together with other factors: the estimate will carry the influence of others onto price. Assumptions: elasticity is constant over the price range, periods of equal length, no prices outside the observed range.
How to read the result
First check the interval: it is narrow (−1.92 to −1.54), so the estimate is stable, and both bounds exceed one in absolute value. With elasticity above one in absolute value a price cut raises revenue but not necessarily profit: here +8.2% sales do not cover the loss of margin. The price of maximum gross profit (5,226) is above the present one: the model advises raising the price, but do not go beyond the data range (the maximum observed is 5,400). Check R² and the number of periods: at R² below 60% or fewer than six periods the estimate is unreliable. It is worth keeping the interval in your notes: at the next recalculation you will see whether customer sensitivity to price has changed.
Common mistakes
Typical trap: taking periods with and without promotions, not entering cost, extrapolating beyond the price range, treating a causal link as proven by correlation alone, mixing SKUs of different categories. For reliability confirm the estimate with a small price test on part of the traffic. Be especially careful with short series: ten points give only a guide, and an error in one point shifts the whole result.
What data is needed?
- A table of periods: the average price excl. VAT and sales in units.
- The current and the new price.
- Unit cost for profit.
Inputs
| Field | Unit | Value in the example | Meaning |
|---|---|---|---|
| Current price | ₽ | 4,500 | The price excl. VAT the change is counted from. The example value is illustrative. |
| New price | ₽ | 4,300 | The price excl. VAT you test. Outside the table range the forecast is unreliable. |
| Unit cost | ₽ | 2,200 | For gross profit and the profit-maximizing price. May be left empty. |
How it is calculated
| Step | Formula |
|---|---|
| Logarithms | x = ln(price), y = ln(sales) |
| Elasticity | e = Σ(x − x̄)(y − ȳ) ÷ Σ(x − x̄)²; a = ȳ − e × x̄ |
| Interval | e ± t(0.975; n − 2) × s ÷ √Σ(x − x̄)² |
| Forecast | sales = exp(a + e × ln(price)) |
| Maximising price | p* = cost × e ÷ (1 + e) when e < −1 |
Worked numbers
Default values from the app; in the interactive version you enter your own.
- Demand elasticity
- −1.7 (−1.73)By how many percent sales change when the price rises 1%.
- Interval: lower bound
- −1.9 (−1.92)
- Interval: upper bound
- −1.5 (−1.54)
- Standard error of elasticity
- 0.1 (0.083)
- R²
- 98.2 %What share of the scatter in sales the price explains.
- Periods in the estimate
- 10
- Sales at the current price (by the model)
- 3,505 units
- Sales at the new price (by the model)
- 3,791 units
- Change in sales
- +8.2 %
- Change in gross profit
- −99,838 ₽Per period, by the model.
- Gross profit maximising price
- 5,226.42 ₽Only within the data range and at constant elasticity.
- points
- regression line: y = 22.689 − 1.727·x
Data as a table
| ln price | ln sales | |
|---|---|---|
| 8.19 | 8.57 | |
| 8.24 | 8.42 | |
| 8.29 | 8.4 | |
| 8.34 | 8.24 | |
| 8.39 | 8.22 | |
| 8.43 | 8.1 | |
| 8.48 | 8.1 | |
| 8.52 | 7.95 | |
| 8.56 | 7.92 | |
| 8.59 | 7.84 |
- points
- regression line
Data as a table
| Price | Sales, units | |
|---|---|---|
| 3,600 | 5,270 | |
| 3,800 | 4,525 | |
| 4,000 | 4,445 | |
| 4,200 | 3,780 | |
| 4,400 | 3,710 | |
| 4,600 | 3,305 | |
| 4,800 | 3,295 | |
| 5,000 | 2,840 | |
| 5,200 | 2,765 | |
| 5,400 | 2,540 |
The table used in the estimate
| Price | Sales | By the model | Deviation |
|---|---|---|---|
| 3,600 ₽ | 5,270 | 5,153 | 2.3 % |
| 3,800 ₽ | 4,525 | 4,694 | −3.6 % |
| 4,000 ₽ | 4,445 | 4,296 | 3.5 % |
| 4,200 ₽ | 3,780 | 3,949 | −4.3 % |
| 4,400 ₽ | 3,710 | 3,644 | 1.8 % |
| 4,600 ₽ | 3,305 | 3,375 | −2.1 % |
| 4,800 ₽ | 3,295 | 3,135 | 5.1 % |
| 5,000 ₽ | 2,840 | 2,922 | −2.8 % |
| 5,200 ₽ | 2,765 | 2,731 | 1.3 % |
| 5,400 ₽ | 2,540 | 2,558 | −0.7 % |
Elasticity from a table answers “how sales followed the price”, not “what will happen”. Prices moved together with the season, advertising and size availability, so check it with a test. For footwear and children’s count by size and the core: elasticity of the end sizes differs.
Nuances and limits
- Reading the estimate as causal when prices moved with the season and advertising.
- Forecasting a price outside the table range.
- Taking out-of-stock periods: sales in them are limited by stock and not demand.