Regression-Based Multiples and Value Drivers

A junior analyst picks eight "comparable" companies, averages their EV/EBITDA multiples, and slaps that average on the target. The problem: not all eight peers are equally comparable. One grows twice as fast, one has half the margin, one carries much more leverage — and simply averaging pretends those differences don't exist. A multiple isn't a magic number attached to an industry; it's the market's compressed way of pricing growth, profitability, and risk, and different companies earn different multiples because they score differently on those exact drivers.

Think of pricing used cars by make and model alone, ignoring mileage and condition. Two "comparable" sedans of the same model year can be worth very different amounts once you account for how many miles are on the odometer. Regression-based multiples do for valuation what a mileage-adjusted pricing guide does for cars: instead of one average number for the whole peer group, you fit a line relating the multiple to the fundamentals that actually drive it, then read off where your specific company should sit on that line given its own numbers.

The regression

$$\text{EV/EBITDA}_i = \beta_0 + \beta_1 \times g_i + \beta_2 \times m_i + \varepsilon_i$$

Here $\text{EV/EBITDA}_i$ is company $i$'s observed multiple, $g_i$ is its expected growth rate, $m_i$ is its EBITDA margin, $\beta_0$ is the regression intercept (a baseline multiple), $\beta_1$ and $\beta_2$ are the estimated "prices" the market pays per unit of growth and per unit of margin, and $\varepsilon_i$ is the residual — the part of company $i$'s multiple the two drivers don't explain, which is the analyst's signal for over- or undervaluation relative to peers.

Worked example: fitting and reading the line

A peer group regression on six software companies yields $\beta_0 = 6.0$, $\beta_1 = 40$ (multiple points per 1.0 of growth rate, i.e., 100 percent growth), and $\beta_2 = 15$ (multiple points per 1.0 of EBITDA margin). A peer growing at 15 percent with a 25 percent margin has a fitted multiple of:

$$6.0 + 40 \times 0.15 + 15 \times 0.25 = 6.0 + 6.0 + 3.75 = 15.75\times$$

If that peer actually trades at 15.5×, it's priced almost exactly where its fundamentals say it should be — the residual is small.

Worked example: valuing the target off its own fundamentals

The target company grows at 22 percent with a 20 percent margin — faster but less profitable than the peer group's average. Plugging into the same fitted line:

$$6.0 + 40 \times 0.22 + 15 \times 0.20 = 6.0 + 8.8 + 3.0 = 17.8\times$$

If the peer group's simple average multiple was 14.0×, a naive comp analysis would apply 14.0× and undervalue the target, because it ignores that this company grows meaningfully faster than the average peer. The regression-implied 17.8× reflects the target's own growth and margin profile rather than the group average — a materially different, and better-justified, starting multiple.

What this means in practice

Banks and equity research desks run these regressions before assigning a multiple in a comp sheet, especially in sectors where the peer set is wide (software, biotech, retail) and growth rates vary enormously across names. The residual from the regression — how far a company sits above or below the fitted line — is itself a useful screen: names trading persistently below the line, after controlling for growth and margin, are flagged as statistically cheap, and vice versa.

Key terms

• Value driver — a fundamental (growth, margin, risk) that the market systematically prices into a multiple.
• Regression intercept ($\beta_0$) — the baseline multiple a company with zero growth and zero margin would notionally command.
• Residual — the gap between a company's actual multiple and the value predicted by its fundamentals; a screen for mispricing.
• Fitted multiple — the multiple implied by plugging a specific company's own drivers into the peer-group regression line.