Regression-based multiples: what growth, ROIC and margin justify

A peer median treats every company in a group as if it deserved the same multiple. A regression-based multiple does something smarter: it looks at how multiples vary across many companies, finds how much of that variation lines up with growth, returns on capital and margins, and then asks what multiple this particular company's fundamentals would justify. It is a way to make a peer comparison fairer.

Why fundamentals should drive multiples

Valuation theory says a multiple is not arbitrary. In a simple growth model, the price-to-earnings ratio depends on the payout of earnings, the growth rate of those earnings and the required return:

P/E = Payout ratio x (1 + g) / (r - g)

Here g is the long-run growth rate and r is the cost of equity. Higher growth raises the multiple, and higher risk (a higher r) lowers it. Payout depends on how much a company must reinvest to grow, which in turn depends on its return on invested capital: a business that earns high returns needs to reinvest less to grow, so it can pay out more. Higher margins matter for the same reason, since they signal a stronger business. That is why growth, ROIC and margin are the natural explanatory variables. See return on invested capital and gross, operating and net margin.

How the method works

1. Gather a cross-section. Collect the multiple (such as EV/EBIT or P/E) and the fundamentals for many companies, often a whole sector or the whole market.
2. Run the regression. A regression fits a line (or plane) through the data, so that the multiple is explained as a base level plus a contribution from each variable.
3. Plug in the company. Enter the company's own growth, ROIC and margin to get the model's predicted multiple.
4. Compare. Set the predicted multiple against the actual one. The gap, called the residual, is the part the fundamentals do not explain.

Predicted multiple = a + b1 x growth + b2 x ROIC + b3 x margin
Residual           = Actual multiple - Predicted multiple

Example (illustrative numbers)

Suppose a regression across many companies in a sector produced this invented equation for the P/E ratio, with growth, ROIC and margin all in percent:

Predicted P/E = 8 + 0.6 x growth + 0.25 x ROIC + 0.3 x margin

Two companies in that sector:

Company A trades below what its fundamentals would typically justify, and Company B above it. If both sat at the sector median of, say, 20, a simple peer comparison would call them equal, while the regression separates them. Note what the regression does not say: it does not say A will rise or B will fall. It says that, relative to the pattern in the data, A is priced lower and B higher than their fundamentals alone would suggest.

How to read it

Rules of thumb:

• A large negative residual (cheap versus fundamentals) is a prompt to investigate, not a signal to buy. Look for risks the three inputs cannot capture, such as litigation, customer concentration or a fading competitive position.
• A positive residual may reflect qualities the model does not see, or optimism.
• The size of the residual matters relative to the noise. Small gaps are well inside normal error.

Common mistakes and limits

• Low explanatory power. Even good models explain only part of the variation in multiples. The leftover variation is large, so predicted multiples are rough estimates.
• The market's mood is in the data. The regression describes how the market is pricing fundamentals today. If investors overpay for growth across the board, the fitted line inherits that.
• Inputs are estimates. Growth is especially uncertain. Using one year's growth can mislead, and forecasts carry their own error.
• Correlated variables. Growth, margin and ROIC often move together, which makes individual coefficients unstable even when the overall fit is acceptable.
• Outliers and small samples. A handful of extreme companies can tilt the line. Sectors with few members give weak estimates.
• Different periods and sectors. Coefficients change over time and between industries, so a regression fitted in one context should not be reused in another.
• Not a substitute for judgment. The model compares; it does not value. The article on relative valuation with peer multiples covers the simpler approach it refines.

Where to see it on StocksXray

One of the three models in StocksXray's fair value is a fundamentals-adjusted multiple, a regression on growth, ROIC and margin. It is blended with equal weight alongside the peer multiple model and a 10-year DCF. Because all three carry error, the site treats the result as a range, a point explained in why fair value is a range. The methodology page has the full description.

Key takeaways

• A regression-based multiple estimates the valuation ratio that growth, ROIC and margin typically justify.
• It improves on the peer median by adjusting for differences in quality and growth.
• The gap between actual and predicted multiples is a question to investigate, not a trade signal.
• Expect noisy output: low explanatory power, shifting coefficients and uncertain inputs all limit precision.