What’s the question?
Most earnings forecasts start with revenue. An analyst builds a sales line, applies a margin, and reads off profit. The margin is usually carried forward from recent history, because it is the number hardest to defend and easiest to leave alone.
That workflow contains a testable claim: revenue growth is what separates one company’s profit record from another’s, and margin is a second-order adjustment. The claim can be checked directly, because the relationship is an identity rather than a model. Profit equals revenue multiplied by net margin, which is net income divided by revenue. Take logarithms and profit growth splits into two additive pieces: growth in revenue, and change in margin. Nothing is estimated and no residual is left over.
Which piece carries the spread across companies? Two firms measured five years apart can report profit growth 60 percentage points apart. That gap traces either to their sales, or to everything between the top line and the bottom line: pricing, input costs, interest, tax.
The approach
- Take the S&P 500 roster as it stood on 31 December 2019, keyed to entity identifiers rather than tickers. Today’s list is a list of survivors, and a ticker is only a lease; either would tilt the answer before the arithmetic starts.
- Hold out banks, property trusts and regulated utilities. Revenue means something different for a lender, and a utility earns a return set in a rate case rather than in a market.
- Pull annual revenue and net income for fiscal 2017 through fiscal 2024, where fiscal year Y covers period ends from June of Y to May of Y+1. That places a January year end in the year it describes.
- Compare three-year totals rather than single years: fiscal 2017-2019 against fiscal 2022-2024, since one year of impairments can set a base that a growth rate then amplifies. Companies without all six fiscal years, or without positive totals in both windows, leave the sample.
- Split each company’s profit growth into its two terms, then decompose the variance of profit growth across companies into the variance of each term plus twice their covariance.
Code
import numpy as np
import pandas as pd
import xfinlink as xfl
from scipy import stats
xfl.set_api_key("YOUR_API_KEY") # free at https://xfinlink.com/signup
BASE, RECENT = [2017, 2018, 2019], [2022, 2023, 2024]
HELD_OUT = {"Financials", "Real Estate", "Utilities"}
members = xfl.index("sp500", as_of="2019-12-31").dropna(subset=["entity_id"])
ids = sorted(set(members["entity_id"].astype(int)))
frames = [xfl.fundamentals(entity_id=ids[i:i + 100], period_type="annual",
start="2017-06-01", end="2025-05-31",
fields=["revenue", "net_income"], max_rows=100000)
for i in range(0, len(ids), 100)]
f = pd.concat([x for x in frames if len(x)], ignore_index=True)
d = pd.to_datetime(f["period_end"])
f["fy"] = np.where(d.dt.month < 6, d.dt.year - 1, d.dt.year)
f = f[~f["gics_sector"].isin(HELD_OUT)].dropna(subset=["revenue", "net_income"])
f = f.sort_values("period_end").groupby(["entity_id", "fy"], as_index=False).tail(1)
rows = []
for eid, g in f.groupby("entity_id"):
g = g.set_index("fy")
if not all(y in g.index for y in BASE + RECENT):
continue
a, b = g.loc[BASE], g.loc[RECENT]
r1, r2 = a["revenue"].sum(), b["revenue"].sum()
p1, p2 = a["net_income"].sum(), b["net_income"].sum()
if min(r1, r2, p1, p2) <= 0:
continue
rows.append({"g_rev": np.log(r2 / r1),
"g_mar": np.log((p2 / r2) / (p1 / r1)),
"g_prof": np.log(p2 / p1)})
w = pd.DataFrame(rows)
var_p = w["g_prof"].var()
cov2 = 2 * w[["g_rev", "g_mar"]].cov().iloc[0, 1]
print("revenue %.1f%% margin %.1f%% covariance %.1f%%"
% (100 * w["g_rev"].var() / var_p, 100 * w["g_mar"].var() / var_p, 100 * cov2 / var_p))
print("R-squared on revenue growth alone: %.3f"
% stats.linregress(w["g_rev"], w["g_prof"]).rvalue ** 2)Full script with formatting and visualisation: revenue-growth-vs-margin-profit-decomposition-python.py
Output
Five-year profit growth decomposition, S&P 500 members outside banks, property trusts and regulated utilities Roster point-in-time at 2019-12-31; fiscal 2017-2019 totals against fiscal 2022-2024 totals 500 on the roster, 361 after the sector hold-out, 296 in the sample Median five-year change (each line a separate median, so the two terms need not multiply out to the profit figure) net profit +55.9% revenue term +35.0% net margin term +8.6% median net margin 10.4% -> 11.4% Share of the cross-sectional variance in profit growth revenue growth 14.4% net margin change 79.7% covariance of the two 5.9% margin share, 1st/99th trim 71.2% (n = 290) R-squared, profit growth on revenue growth 0.210 revenue up and profit down 18.2% of companies revenue down and profit up 4.1% of companies By revenue-growth quartile n revenue margin profit profit fell Q1 slowest 74 +3.2% -3.0% +3.0% 49% Q2 74 +26.0% -4.4% +20.7% 30% Q3 74 +43.5% +12.8% +63.3% 15% Q4 fastest 74 +87.2% +17.7% +117.6% 5% By sector n revenue margin profit net margin Energy 18 +63.9% +64.8% +169.7% 7.6% -> 12.0% Information Technology 47 +39.5% +15.1% +73.2% 14.3% -> 17.9% Health Care 49 +47.5% -9.3% +56.4% 13.2% -> 12.5% Materials 24 +31.3% +7.6% +48.6% 10.1% -> 10.5% Industrials 66 +33.3% +8.0% +47.6% 10.0% -> 11.5% Consumer Discretionary 45 +31.2% +13.5% +44.0% 7.9% -> 9.3% Consumer Staples 32 +26.0% -7.8% +12.8% 11.0% -> 9.3% Communication Services 15 +36.2% -11.4% +11.6% 13.0% -> 9.4%
What this tells us
For the typical company, revenue did most of the work. Median profit growth was 55.9%, with the revenue term supplying 35.0% and the margin term 8.6%. Net margin at the median company widened from 10.4% to 11.4% across five years containing a pandemic and an inflation shock.
The spread tells a different story. Variance in profit growth across the 296 companies splits into 14.4% from revenue growth, 79.7% from margin change, and 5.9% from the covariance between them. A regression of profit growth on revenue growth alone returns an R-squared of 0.210, and that figure is generous, because it credits revenue with the shared covariance too. Four fifths of the difference between one company’s five-year profit record and another’s sits in what happened to margin.
Some of that is definitional, since the margin term equals profit growth minus revenue growth and collects whatever revenue does not explain. The substance is how tightly revenue growth clusters. The middle half of the sample grew revenue between 17.5% and 60.7%; the same middle half grew profit between 1.3% and 110.0%, a band two and a half times wider. Large American companies differ far less in how fast they sell than in what they keep.
The quartile table shows how loose the company-level link is. Among the 74 slowest revenue growers, 36 ended the period with lower profit; among the 74 fastest, 4 did. Across the sample, 18.2% grew revenue and earned less at the end of it.
Sector medians locate the moves: energy gained 64.8% of margin off a depressed base, health care gave back 9.3%, consumer staples 7.8%. Trimming the extreme 1% at each end of profit growth leaves the margin share at 71.2%, so a few violent recoveries are not carrying the result.
So what?
A revenue forecast settles a minority of the earnings question. Effort spent refining a sales projection past its second decimal place buys less accuracy than an hour on the margin path: pricing power against input costs, mix shift, the interest bill on refinanced debt, the tax rate, and how much of reported profit is one-off.
Two habits follow. When screening, rank candidates on margin trajectory rather than sales trajectory; sales growth in a large-cap universe is compressed into a narrow band, and ranking on a narrow band mostly ranks noise. When building a model, treat a flat-margin assumption as an assumption about four fifths of the answer and write it down as such.
Built with xfinlink — free financial data API for Python.