For most consumer brands, the fastest credible starting point is log-log regression on SKU-week sales data. If that estimate needs to hold up to a pricing decision with real money behind it, layer in an instrumental variables approach (IV/2SLS) or run a randomized price test. These methods, paired with the midpoint formula for quick sanity checks, feed directly into revenue and profit forecasting once you know how demand actually moves with price.
TL;DR:
- Use weekly SKU data with effective prices, promotion flags, and stockout periods removed; monthly totals hide promotional swings, while daily counts add noise.
- Ordinary regression understates sensitivity because prices respond to demand; use cost instruments and trust them only when the first stage F statistic exceeds 10.
- Use randomized price tests for high stakes decisions, but pilot a handful of representative SKUs first because experiments become costly across a large catalog.
- Compare revenue and contribution margin, calculate break even unit growth, and test low and high elasticity confidence bounds before making a permanent change.
- When absolute elasticity exceeds 1, demand is elastic and revenue rises as price falls; below 1, revenue rises as price increases.
Table of Contents
- At-a-glance guide to elasticity testing methods and when to use each
- Log-log regression explained: model, data prep, and reading the coefficient
- Identification and causality: endogeneity, instruments, and randomized tests
- Building a data-design checklist founders can actually run
- Turning an elasticity estimate into pricing and profit decisions
- What founders consistently get wrong about elasticity testing
- How we help founders turn elasticity data into pricing decisions
- FAQ
- Sources
At-a-glance guide to elasticity testing methods and when to use each
Every elasticity method trades off speed against rigor, and knowing where you sit on that line in any given quarter matters more than knowing the math.
- Midpoint formula: a back-of-envelope check using two price points and two quantities, useful for a quick gut check before committing analyst time.
- Log-log regression: the default workhorse for historical data, estimating a constant elasticity coefficient across your observed price range.
- Semi-log or linear variants: worth testing when your price range is wide enough that a constant elasticity assumption starts to strain.
- IV/2SLS: necessary when price itself is driven by demand or cost shocks you can’t observe, which is most of the time.
- Randomized price tests: the gold standard for causality, reserved for decisions big enough to justify the operational cost.
- Demand systems (AIDS or BLP): built for brands managing dozens of SKUs with meaningful substitution between them.
Start with log-log regression on clean historical data; add instruments or experiments if endogeneity is suspected; use demand systems for complex SKU portfolios
Log-log regression explained: model, data prep, and reading the coefficient
The standard model is ln(Q) = α + E·ln(P) + controls, where Q is quantity sold, P is price, and E is the coefficient you care about. Read E directly as the percent change in quantity for every 1% change in price, which is exactly how price elasticity of demand is formally defined: the ratio of percentage change in quantity demanded to percentage change in price.
Statistic callout: When the absolute value of elasticity exceeds 1, demand is elastic and revenue rises as price falls; when it’s below 1, demand is inelastic and revenue rises as price increases, according to MIT’s pricing coursework. That single threshold tells you more about your pricing power than almost any other number in your business.
Before you run the regression, three data decisions shape everything downstream:
- Use effective price, accounting for discounts and promotions, to accurately capture price impact.
- Default to SKU-week as your aggregation grain. It smooths out daily noise without hiding the within-month promotional swings that monthly data erase.
- Check your residuals and confidence interval after fitting the model, and compare a semi-log or linear specification when your price range is wide enough that constant elasticity feels like a stretch.
The remaining 20% of the analysis is where causality gets tricky.
Identification and causality: endogeneity, instruments, and randomized tests
Here’s the problem nobody mentions until it bites them: price is rarely set at random. You raise prices when demand is strong and discount when it’s weak, which means ordinary least squares regression tends to understate how sensitive your customers actually are. Research on scanner data confirms this directly: naive OLS regressions of log quantity on log price are biased toward zero because prices are endogenous, and recovering a causal elasticity requires instruments.
The fix is a variable that moves price without moving demand directly, as explained in detail by AI demand forecasting technologies. In practice, that means:
- Cost shifters, such as input or shipping cost changes that affect your price but have no direct bearing on customer demand.
- Hausman instruments, which use prices for the same product in other, unrelated markets as a proxy for cost-driven price movement.
- BLP-style instruments, more common in demand-system work, that exploit the structure of competing products in a category.
Whichever instrument you choose, check the first-stage F-statistic before trusting the result. A weak instrument amplifies bias rather than removing it, as the same scanner data research notes, so an F-statistic above 10 is the commonly cited threshold for calling an instrument strong enough to use.
Pro Tip: When naïve and IV elasticity estimates differ significantly, the IV estimate is more reliable. That divergence is itself a signal that endogeneity was distorting the naive estimate.
Randomized price tests sidestep the instrument question entirely by assigning prices at random, which is why field experiment research treats them as the most reliable source of causal price effects. The catch is that they scale poorly: testing every SKU in a large catalog gets expensive fast. A pilot experiment on a handful of representative SKUs can estimate how much variation (statisticians call this sparsity) exists across your catalog, which tells you whether a full experimental program is worth the spend or whether a model-based estimate will do.
Building a data-design checklist founders can actually run
Good elasticity estimates come from disciplined data habits, not fancier math. A few rules cover most of the ground.
- Default to SKU-week aggregation. Daily data is noisy; monthly data buries the promotional spikes that often explain your biggest demand swings.
- Look for at least a handful of genuinely distinct price points with a meaningful spread between them. A dataset where price barely moved will never produce a reliable elasticity.
- Track promotion flags, feature or display events, competitor pricing, and inventory status alongside every price point, and strip out stockout periods before you model anything.
- If SKU-level history is thin, aggregate only across SKUs that are genuinely comparable rather than lumping unrelated products together for the sake of sample size.
Statistic callout: Guidance on scanner-style demand analysis supports weekly aggregation as the practical default because it reduces noise relative to daily counts while still preserving the within-month variation that a monthly rollup would erase. That single choice of grain often matters more than which regression you ultimately run.
One more thing worth internalizing: elasticity isn’t always constant across your price range. Retail pricing research describes a ‘latitude of acceptance,’ a band within which customers barely notice a price change and outside of which sensitivity jumps. An estimate built entirely inside that band may understate how customers react once you push past it.
Turning an elasticity estimate into pricing and profit decisions
Once you have a credible E, the path to a pricing decision is mechanical.
- Map the expected percent change in quantity against the percent price change, then translate both into revenue to see the net direction.
- Run the same math against contribution margin, not just revenue, since a price cut that grows units can still shrink profit if margins are thin.
- Calculate the break-even volume change: the minimum unit shift that keeps profit flat, so you know how much cushion you actually have.
- Segment the estimate by channel or customer cohort where you can, since brand leaders and heavily advertised products tend to show lower price sensitivity than the category average, which changes the math for your best sellers.
Pro Tip: Run the profit calculation across the low and high ends of your elasticity confidence interval before locking in a permanent price change. A point estimate can look safe while the range around it isn’t.
For a deeper walkthrough of margin math specific to this revenue band, our piece on price increase modeling covers the break-even arithmetic in more detail, and our DTC Unit Economics Calculator turns these inputs into a margin picture in minutes.
What founders consistently get wrong about elasticity testing
The better habit is to treat elasticity testing as a recurring diagnostic tied to your SKU economics, not a one-time research project. Prioritize the handful of SKUs that actually move your revenue and margin, test those rigorously, and let the long tail ride on rougher estimates. Speed beats precision almost every time a founder is weighing a pricing call against a cash runway that doesn’t forgive delay.
How we help founders turn elasticity data into pricing decisions
Running a log-log regression is one thing. Knowing whether the resulting price move actually protects your margin, your cash position, and your growth plan is another, and that’s where we come in. Our DTC Operator Diagnostic gives you a fast read on where pricing fits into your broader operating picture, and a Founder Hour gets you direct, founder-to-founder input on a specific pricing decision before you commit to it.
For brands ready to go deeper, our Founder Advisory and Financial Health Assessment engagements translate elasticity findings into a full profitability plan. Start with our DTC Unit Economics Calculator or book a diagnostic to see where your pricing stands today.
FAQ
What is the price elasticity of demand formula?
Price elasticity of demand equals the percent change in quantity demanded divided by the percent change in price, written as Ed = %ΔQ / %ΔP. When the absolute value exceeds 1, demand is elastic and revenue rises as price falls; below 1, demand is inelastic.
How do I choose between log-log regression and a randomized price test?
Start with log-log regression when you have clean historical SKU-week data, since it’s faster and cheaper to run. Move to a randomized price test when the decision is big enough to justify the cost of a controlled experiment, because field experiments deliver the most reliable causal estimates but scale poorly across many products.
Why does my elasticity estimate look too low?
A common cause is endogeneity: prices often move for reasons tied to demand, which biases ordinary regression toward zero. Adding an instrument such as a cost shifter, or checking the first-stage F-statistic on your instrument, usually restores a more accurate estimate.
What data grain should I use for elasticity testing?
Weekly aggregation at the SKU level balances noise reduction with preservation of relevant price variation. Pairing this grain with effective price, calculated as revenue divided by units, keeps discounts from distorting your result.
How does elasticity testing fit into pricing and profit decisions?
An elasticity coefficient lets you translate a proposed price change into an expected change in units sold, and from there into revenue and margin impact. Calculating impact on contribution margins reveals whether a price change improves profitability.
Sources
- 15.818 Pricing (MIT): Lecture notes: Measuring Customer Reactions to Prices
- Field experiments and pricing (MIT paper)
- Demand elasticity with scanner data: MetricGate blog
- Consumer price sensitivity in the retail industry: latitude of acceptance with heterogeneous demand