Use 5 to 10 years of history. A 3-year window can sit entirely inside one phase of a cycle (all boom or all bust) and it'll happily show you a "trend" that's just weather. Five years usually spans at least one demand wobble. Ten years spans a full capex or credit cycle and at least one real shock. In the last decade you've had demonetization, GST, COVID, the 2021-22 commodity spike and the 2022 rate reset. Your standing tool is the historical strip you built for the returns-on-capital mini-project, with one row per metric and one column per year. Here's how to read it.
Pull out four things, in order:
Level. The period average. A company that averaged 23% EBITDA margin over ten years is a 23% business until proven otherwise. It isn't the 28% of its best year, which its investor deck will quote, or the 18% of its worst, which the bear note will.
Advertisement
How do you describe a trend: percentage points and CAGR?
Trend. Direction and slope. Two mechanical conventions, drilled now and used forever:
- Margins and other ratios move in percentage points (pp), never "percent". A margin going from 12% to 15% rose 3 percentage points, or 25% relatively, and you should say which. Sloppy language here causes real errors of 10x.
- Growth compounds, so summarize it with CAGR =
(ending ÷ beginning)^(1/n) − 1, where n is the number of intervals. FY16 to FY25 is nine intervals, not ten. Know CAGR's two blind spots as well. It's path-blind, so 100, 40, 100 over four years prints the same 0% as 100, 100, 100. And it's endpoint-hostage: start the window at a trough, end it at a peak, and you've manufactured growth. For cyclicals, average three years at each end before computing, or simply plot every year and let your eye do the work.
Why does volatility matter as much as the average?
Volatility. The range and standard deviation of the series. Stability is information in its own right. A margin with a ±1pp band across ten years is telling you about pricing power, pass-through contracts or category dominance. A ±8pp band is telling you the company is a price taker. Here are two illustrative strips, both modeled on ~FY16 to FY25 scale. Pull the actuals from screener.in and verify.
| EBITDA margin, % | FY16 | FY17 | FY18 | FY19 | FY20 | FY21 | FY22 | FY23 | FY24 | FY25 | Mean | SD |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Tata Steel (illustrative) | 10 | 15 | 20 | 21 | 15 | 22 | 28 | 13 | 12 | 13 | ~17 | ~5.5pp |
| Nestlé India (illustrative) | 21 | 22 | 23 | 23 | 24 | 24 | 23 | 22 | 23 | 24 | ~23 | ~1pp |
The levels are only six points apart, but the volatility differs five-fold, and that difference drives everything downstream. It decides which year's margin you may extrapolate. For Nestlé that's roughly any year. For Tata Steel it's none, and you normalize to mid-cycle, a Phase 3 skill. It also decides which valuation tools apply and how much debt each company can carry. Two companies with the same average aren't the same business.
How do you spot and classify a break in the trend?
Inflections. A break in an established trend is the most valuable object in a time series, because something happened. Somewhere there's an MD&A for that year that explains what, or conspicuously doesn't. So circle the inflection year and read that year's MD&A and results commentary. Then classify the break. It's either (a) economics: demand, price, competition. Or it's (b) perimeter, where an acquisition, disposal or demerger changed what "the company" is. Or it's (c) accounting. The rules changed and the business didn't. Class (c) is the trap that catches professionals.
Telecom is the canonical Indian example. EBITDA margins jump sharply in FY20, and Bharti Airtel's reported margin steps up by several points that year. A large slice of the step is Ind AS 116 adoption. From FY20 rent left operating expenses and became depreciation plus interest, which raised EBITDA mechanically with no cash or economics attached, and the transition note in the FY20 annual report quantifies it. The real Airtel inflection, tariff repair after the price war, is in there too, but you can't size it until you strip out the accounting step. Time series need normalization just as much as peer tables do, and the hygiene layer shows you how. Keep a few similar Indian discontinuities on file. With the Ind AS transition itself, FY17 comparatives were restated. GST replaced excise in FY18 and changed reported revenue definitions for many manufacturers. And any demerger breaks a series. ITC's hotels business left the perimeter in January 2025, so every post-FY25 ITC series has to be spliced or restated.
Should trend analysis use real or nominal growth?
One more adjustment: inflation. Growth is nominal unless you say otherwise, and India runs hotter than the US. A 10% revenue CAGR is ~4% real at India's ~6% inflation and ~7.5% real at US ~2.5%. Those are long-run ballparks, so pull current CPI from mospi.gov.in and bls.gov. Comparing an Indian company's nominal growth with a US company's nominal growth flatters the Indian one by roughly the inflation differential. Whenever the comparison crosses the two economies, state growth in real terms, or at least footnote the differential.