Alpha is the return an investment earned beyond what you’d have expected given the risk it took. It’s the number every active fund manager is implicitly claiming to produce, and the number this blog’s mutual fund series has so far avoided using — because most of what gets called alpha is something else wearing the name.

There are two very different things people mean by it:

1. Raw excess return — the investment’s return minus the benchmark’s. Simple, and usually what a factsheet or a headline means.
2. Jensen’s alpha — the return minus what CAPM (the Capital Asset Pricing Model) says the investment should have earned, given its beta. This is the one that deserves the name.

The gap between the two is the whole lesson.

The formula

Raw excess return  =  R_investment − R_benchmark

Jensen's alpha     =  R_investment − [ R_f + β × (R_benchmark − R_f) ]

where  R_f  = risk-free rate
       β    = the investment's beta against that benchmark

The bracket is CAPM’s expected return: the risk-free rate, plus beta times the market’s excess return. A stock with a beta of 2 that returned 20% in a year the market returned 10% has no alpha — doubling the market is exactly what a beta of 2 is supposed to do.

Worked example: Britannia, two years

Daily closes from 1 April 2024 to 30 March 2026, the same series as the beta post (whose first daily return falls on the next trading day). Risk-free rate 6.5%, the illustrative G-Sec (Government of India bond) yield used throughout the WACC post. Historical data, for illustration only.

Same stock, same two years: the headline says Britannia beat the market by 5.56 points a year; the risk-adjusted version says 1.62. The difference is that the Nifty had a poor two years — the price index actually fell slightly — and a stock with a beta of 0.42 was expected to fall less. Most of the “outperformance” is just low beta doing what low beta does in a weak market. Reverse the market and the same low beta would have made Britannia look like a laggard.

One honest caveat on the inputs. Both returns here are price-only: Britannia’s share price and the Nifty 50 price index, with dividends left out, because a total return index series isn’t in this post’s data. That’s the very mistake the next section warns about. In the raw comparison the two missing dividend streams partly cancel. In the CAPM line they don’t: the risk-free rate is a full yield, but the market return it’s compared with is missing its dividends, so the market’s excess return is understated. Treat the alpha figure as rough — it illustrates the method, not a precise measurement.

Neither figure says anything about Britannia’s prospects. It’s two years of price history, and the beta post already showed that the beta itself moved from 0.29 to 0.55 across those two years — so the alpha number inherits every bit of that instability.

The fake alpha in every index fund factsheet

Here is a more instructive example. Take the UTI Nifty 50 Index Fund (regular plan) — a passive fund whose entire job is to match the index — and compute its calendar-year return minus the Nifty 50 price index:

Ten years, ten positive numbers, averaging +0.98 pp a year. Sources: fund NAV from AMFI via mfapi.in; index from Yahoo Finance. Data to 31 March 2026.

An index fund with a decade of positive “alpha” is obviously nonsense — and it is, because the benchmark is wrong. The price index (PRI) ignores the dividends the fifty companies pay; the fund receives and reinvests them. That one-point gap is roughly the Nifty’s dividend yield minus the fund’s expense ratio. Against the total return index (TRI), which includes dividends, the same fund would likely show a small negative number most years — the expense ratio and tracking error, which is what an index fund is supposed to show.

This is exactly why the benchmarks post insisted on comparing like with like, and why SEBI has required funds to benchmark against TRI since February 2018. Before that, a lot of Indian active-fund “alpha” was this dividend gap.

Common mistakes

• Calling excess return “alpha”. Excess return over a benchmark is a fact. Alpha is that fact with the risk taken to earn it subtracted out. A high-beta fund in a bull market has lots of the first and often none of the second.
• Benchmarking against a price index. It manufactures about a point a year of alpha out of dividends. Always check whether a comparison uses PRI or TRI.
• Reading alpha over a short window. Two years of Britannia produced an alpha estimate built on a beta that nearly doubled inside the window. Alpha needs a horizon long enough for the beta to mean something, and even then the standard error is usually larger than the estimate.
• Assuming past alpha persists. The consistency of the index fund’s “alpha” above is the exception — it’s a structural artefact. Genuine manager alpha is hard to find at all: S&P’s SPIVA India Year-End 2025 scorecard found about 76% of Indian active large-cap funds trailed their benchmark over the ten years to December 2025. Finding it and expecting it to continue is harder still.

Takeaway: alpha is what’s left of a return after the benchmark and the beta are accounted for. Britannia’s 5.56-point lead over a weak Nifty shrinks to 1.62 once its low beta is credited — and an index fund’s decade of “outperformance” over the price index is dividends, not skill.

The last post showed Britannia’s price as a simple line — one dot per day, joined up. It’s readable, and it throws away most of the information.

A line chart plots the closing price. But a trading day isn’t one number. The stock opened somewhere, traded up to some highest point, down to some lowest point, and closed somewhere else. A day that opened at ₹5,000, spiked to ₹5,300, collapsed to ₹4,900 and closed at ₹5,010 looks, on a line chart, almost identical to a day that drifted quietly from ₹5,000 to ₹5,010.

Those were not the same day. Candlesticks — developed by Japanese rice traders in the 1700s and popularised in the West by Steve Nison in the 1990s — show all four prices in a single bar.

The anatomy

Schematic illustration — not real price data.

Each candle has two parts:

And the colour tells you the direction:

• Green (or white/hollow) — the close was above the open. Buyers ended the session in control.
• Red (or black/filled) — the close was below the open. Sellers did.

That’s the whole notation. A long body means the open and close were far apart — a decisive session. A short body means they finished near each other, whatever happened in between. Long wicks mean the price went somewhere and came back.

Reading real candles

Here’s Britannia through the turn of 2026 — two months of real trading days:

Britannia (NSE: BRITANNIA), daily, mid-December 2025 to mid-February 2026. Source: Yahoo Finance. Historical data, for illustration only.

Some things you can read straight off it that a line chart would have hidden:

• Long upper wicks near the highs. Several sessions in early January pushed up and closed well below where they’d reached. Buyers got the price up there; it didn’t stay.
• Alternating colours in the middle. Long stretches of green-red-green-red with small bodies — sessions that opened and closed near each other. That’s a market with no settled view.
• The occasional long body. A few sessions ran decisively one way. Those are days when something happened.

The single-candle shapes worth knowing

A handful of individual candles have names. They describe the shape, and the usual interpretation attached to each is much softer than most sources admit.

And two-candle combinations:

Now the honest part. These shapes describe what happened. The predictive claims attached to them — “a hammer signals a reversal” — are much weaker than their confident names suggest. Hammers appear constantly in the middle of trends that carry straight on. In this dataset you can find hammers followed by rallies and hammers followed by further falls, and the candle looked identical both times.

Treat a named candle as a description of one session’s tug-of-war, not a forecast. That’s the level of claim the shape can actually support.

The dataset makes the point concretely. Applying a standard hammer definition to these 495 bars finds 25 of them. Ten trading days later, 15 had risen and 10 had fallen, with a median move of about +1%. That is close to a coin flip on a small sample — not a signal, and not a refutation either. It’s simply what one shape, on one stock, over two years, actually did.

Doing it in Python

Spotting shapes programmatically is mostly arithmetic on the four prices:

import pandas as pd

df = pd.read_csv("britannia-ohlcv-2024-04-to-2026-03.csv",
                 parse_dates=["date"]).set_index("date")

body = (df.close - df.open).abs()
rng = df.high - df.low
upper = df.high - df[["open", "close"]].max(axis=1)
lower = df[["open", "close"]].min(axis=1) - df.low

# a doji: body is tiny relative to the day's whole range
doji = body < 0.1 * rng
# a hammer: small body up top, lower wick at least twice the body
# (counts the shape only; textbook hammers also need a prior decline)
hammer = (lower > 2 * body) & (upper < body) & (rng > 0)

print(f"doji days: {doji.sum()}   hammer days: {hammer.sum()}")

Run it and you’ll get 59 doji days and 25 hammers out of 495 — which is the point. A shape that shows up dozens of times in two years on a single stock is not a rare omen.

Common mistakes

• Reading a candle without its context. A hammer at the bottom of a long decline and a hammer in the middle of a quiet range are the same shape and not the same information. The surrounding trend does most of the work.
• Trusting single candles. One session is a small sample of a continuous argument. Most practitioners who take candles seriously want confirmation from what follows, which by definition means waiting.
• Assuming green means good. A green candle means the close beat the open, nothing more. A stock can print a green candle on a day it fell 4% from the previous close, if it opened even lower.
• Forgetting the gap between sessions. The open often isn’t the previous close — overnight news moves it. Candles show the gap as empty space, and that space is real information the bodies don’t contain.
• Believing the more exotic names mean more. Three-candle formations with elaborate names are describing increasingly specific coincidences of shape. Specificity is not the same as reliability.

Takeaway: A candlestick packs open, high, low and close into one bar, so you can see not just where a session ended but how it got there — the body for the outcome, the wicks for the fight. The named shapes are a useful vocabulary for describing that fight, and a much weaker basis for predicting the next one than their confident names imply.

The capstone post ended the first module with a checklist and an honest limit: financial statements tell you what kind of business you’re looking at, not whether the price is right. This module is about the rest of what a company publishes — the 140 pages around those five pages of statements — and how to read it without being led by the nose.

Start with the uncomfortable fact. An annual report is the company’s account of its own year, written by management, approved by a board that management largely selected, and designed (photographs, typography, the chairman’s smile) to reassure. Most of it is truthful. Almost none of it is neutral. The skill is knowing which pages were written for you and which were written at you.

Desi Bites Foods, the fictional snacks company this blog has followed since its first post, listed on the main boards of the NSE (National Stock Exchange) and BSE in June 2025. Its first annual report as a listed company covers FY26 (year ended 31 March 2026), and it’s the example throughout this module. Everything about Desi Bites is invented; the structure of the report is what every Indian listed company files.

The map

Roughly 140 pages, in the order they appear — with the order you should actually read them in:

Two things jump out of that table.

First, the reading order is almost the reverse of the page order. The report opens with the most persuasive material and closes with the most informative. That isn’t an accident.

Second, the notes to accounts are half the report. Seventy pages of small type that most readers never open, sitting behind five pages of statements that this blog spent a whole module on. Every number in those five pages has its explanation in the seventy.

Step 1: the auditor’s report (six pages, read first)

The independent auditor’s report is the only part of the document written by someone the company didn’t employ to say nice things — though it did pay them, which is worth remembering. Three things to find in it:

The opinion. It comes in four grades. Unmodified (“true and fair view”) is the normal case. Qualified means “true and fair except for the following” — read the following. Adverse means the statements are materially wrong. Disclaimer means the auditor couldn’t get enough evidence to form a view at all. The last two are rare and are the loudest sound a document can make.

Key audit matters (KAMs). For listed companies, from audits of FY2018-19 onward (auditing standard SA 701), auditors must list the areas that took the most judgement. These are, by definition, where the numbers are softest. For Desi Bites the KAM is Revenue recognition around the year-end (cut-off) — a standard Key Audit Matter for a distributor-led business — which is exactly the area the next post shows being abused.

Emphasis-of-matter paragraphs. The auditor agrees with the statements but wants you to look at something — often a pending dispute, a significant uncertainty, or a change in accounting policy. It’s the auditor pointing. (A material doubt about going concern is louder still: under the revised SA 570 it gets its own separately headed section in the report.)

Step 2: the statements (five pages)

You know these. Balance sheet, income statement, cash flow, and the statement of changes in equity — which most readers skip and shouldn’t, because it’s where share issues, buybacks and dividends are laid out in one place.

Desi Bites’ FY26 headline figures, in ₹ lakh:

Revenue up 27.3%, PAT up 44.3%. Two new lines — other income and exceptional items — and two new balance sheet items, goodwill and a cash pile roughly five times last year’s. Every one of those is a question, and none of them is answered on these five pages.

Step 3: the MD&A (management’s version of why)

The management discussion and analysis (MD&A) is the one section where management is obliged to explain the year in words: volumes versus prices, input costs, new capacity, competition, what the risks are. It’s the only place the why is written down, which makes it indispensable — and it’s written by the people whose bonuses depend on the answer, which makes it something to read against the numbers rather than instead of them.

A practical test: for each claim in the MD&A, find the line in the statements that would be true if the claim were true. “Strong volume growth” should show up as revenue growing faster than any price increase mentioned. “Improved operational efficiency” should show up as opex growing slower than revenue. “Prudent working capital management” should show up in debtor days and inventory days. When the words and the lines disagree, believe the lines.

Desi Bites’ MD&A says revenue grew 27.3%. The notes say the existing business grew 18.0%; the other 9.3 points came from a company it bought in October. Both statements are true. Only one of them appears in the MD&A’s opening paragraph.

Step 4: the notes (seventy pages — but only seven matter first)

You don’t read all seventy. You read these, in this order:

The pattern: every note on that list is a place where a number on the statements could be true and misleading at the same time. Revenue can be real and pulled forward. A profit can be genuine and come from selling a building. A balance sheet can balance and omit a ₹120 lakh tax demand because it’s “contingent.” The notes are where the statements confess.

The next few posts in this module each take one of those notes and work through it on Desi Bites: the forensic checks, the quarterly rhythm, contingent liabilities and pledges, and goodwill.

Steps 5 to 8: the persuasive pages

Read the chairman’s letter after all of the above, and read it as a document about management rather than about the company. Does it mention the things you found in the notes? If the year had an exceptional item, a missed target, or a KAM about revenue cut-off, does the letter acknowledge it or talk about “headwinds” and “our journey”? A letter that engages with the bad news is worth more than one that doesn’t — and the difference is only visible if you already know what the bad news was.

The directors’ report and governance report are mostly statutory boilerplate, with two exceptions worth a minute each: the AOC-2 annexure (related-party contracts the board approved) and the list of independent directors who resigned during the year, with their stated reasons. “Personal reasons” three times in one year is information.

Common mistakes

• Reading front to back. The report is sequenced to persuade. Start at the auditor’s report and the notes; the letter can wait.
• Treating “unmodified opinion” as a clean bill of health. It means the statements fairly present what happened under the accounting rules. It does not mean the business is good, the profit is high quality, or the accounting choices were conservative. The key audit matters tell you where the auditor sweated.
• Skipping the notes because they’re long. Seven of them do most of the work. Start there.
• Believing the MD&A’s adjectives. “Robust,” “resilient” and “strategic” have no line on the P&L. Find the number each claim implies and check it.
• Ignoring the statement of changes in equity. It’s a one-page record of every share issued, every dividend paid and every reserve moved. For a company that just listed, it’s where dilution is spelled out.
• Reading one year. An annual report shows two years side by side. The useful unit is three to five reports, read for what changed — in policies as much as in numbers. A quiet change to the revenue recognition policy is a bigger event than a loud change in the margin.

Takeaway: An annual report is 140 pages written by the people it judges, sequenced to persuade: the reassuring letter up front, the seventy pages of notes at the back. Read it backwards — auditor’s report, statements, then the seven notes where a true number can still mislead — and treat the chairman’s letter as evidence about management, not about the company.

Every return needs a comparison before it means anything. A fund returning 14% in a year when its market returned 22% did badly. The same 14% in a year the market returned 4% was excellent.

A benchmark is that reference point — an index representing what the fund is trying to do. And the choice of benchmark quietly decides a great deal.

This post is mostly about one benchmark choice that used to flatter every equity fund in India, and how you can see it in the data.

The finding

Take the index fund this series has been using — a purely passive vehicle that holds the Nifty 50 and does nothing clever at all. Compare it to the Nifty 50 price index over the 18.5 years both exist:

Period 17 September 2007 to 31 March 2026. Fund NAV: AMFI via mfapi.in. Index: Yahoo Finance daily close, ^NSEI. Historical data, for illustration only.

The fund beat its own benchmark by 0.42 percentage points a year, while charging a fee, for 18.5 years.

That should look impossible. A fund tracking an index, after costs, ought to land slightly below it. Beating it consistently by a passive strategy isn’t skill — so what is it?

PRI versus TRI

The answer is dividends.

• A price index (PRI) measures only price movement of its constituents. When a company pays a dividend, its share price typically drops by roughly the dividend, and the price index falls with it. The dividend itself is never counted.
• A total return index (TRI) counts price movement and assumes dividends are reinvested.

A fund holding those fifty companies actually receives the dividends. They land in the fund and raise its NAV. So a fund measured against a price index gets credited with the entire dividend yield of the market as apparent outperformance — for doing nothing at all.

Roughly:

Fund return ≈ Index price return
            + dividend yield          ← the fund receives these
            − expense ratio           ← the fund pays this
            − other costs & cash drag ← part of the tracking difference

The 0.42 pp gap above is the Nifty’s dividend yield, net of this fund’s costs. It is an artefact of the comparison, not a result.

Why this mattered enormously

Until 2018, Indian mutual funds were generally permitted to benchmark against price indices. Which meant every actively managed equity fund got a free head start of roughly the market’s dividend yield — a percentage point or more a year — before any manager did anything.

Fund A claims it beat its benchmark by 1.5% a year. If most of that came from dividends the benchmark structurally ignored, the manager’s actual contribution was a fraction of the claim.

SEBI required benchmarking against total return indices from 1 February 2018 (circular SEBI/HO/IMD/DF3/CIR/P/2018/04, 4 January 2018). It’s an unglamorous rule change and one of the more consequential ones for retail investors, because it removed a systematic bias from every performance claim in the industry.

The practical check: when you see “the fund beat its benchmark,” find out whether the benchmark was TRI. For older track records spanning pre-2018, some of the claimed outperformance may simply be missing dividends.

Choosing a benchmark honestly

Three rules.

It must match what the fund actually does. A small-cap fund measured against the Nifty 50 is being measured against the wrong market. A fund that holds 30% debt shouldn’t be compared to a pure equity index.

It must be total return, and stated. Per above.

It must be fixed in advance. A benchmark chosen after the fact, from several candidates, is an argument rather than a measurement.

Category comparison, and its trap

The other common comparison is against the fund’s category — “top quartile among flexi-cap funds.” Useful, with one serious caveat.

Survivorship bias. Funds that perform badly get merged or wound up, and disappear from the category. Compare today’s surviving funds and you’re comparing against a group with the failures deleted. The surviving average is flattered by exactly the funds that no longer exist.

This applies to almost every “average category return” you’ll see, and it biases in one direction — always upward.

Two smaller cautions: category definitions changed materially with SEBI’s 2017–18 scheme rationalisation, so pre- and post-2018 category comparisons aren’t like-for-like. And within a category, funds can run very different risk levels — beating your category with far more volatility isn’t obviously winning, which is what Sharpe was for.

Doing it in Python

import pandas as pd

nav = pd.read_csv("uti-nifty50-index-fund-nav.csv",
                  parse_dates=["date"]).set_index("date")
idx = pd.read_csv("nifty50-price-index.csv",
                  parse_dates=["date"]).set_index("date").nifty50_pri_close
fund = nav.nav_regular_growth.dropna()

start = max(idx.index[0], fund.index[0])
end = pd.Timestamp("2026-03-31")
years = (end - start).days / 365.25

def cagr(s):
    return ((s.asof(end) / s.asof(start)) ** (1/years) - 1) * 100

print(f"index (PRI) {cagr(idx):.2f}%   fund {cagr(fund):.2f}%   "
      f"gap {cagr(fund) - cagr(idx):+.2f} pp")

Note both series must be compared over the same window — hence taking the later of the two start dates. Comparing a fund’s 20-year record to an index’s 18-year record is a common and entirely avoidable error.

Common mistakes

• Not checking whether the benchmark is TRI. As shown, a price index hands every fund the dividend yield as free outperformance.
• Accepting a benchmark that doesn’t match the mandate. Small-cap funds compared to large-cap indices, hybrid funds to pure equity.
• Forgetting survivorship bias in category averages. The failures were removed from the comparison set.
• Comparing over different periods. Same window, always.
• Treating benchmark outperformance as skill without checking risk. More return from more risk isn’t the same achievement.
• Assuming a benchmark-beating record persists. Past outperformance is a weak predictor of future outperformance — much weaker than most fund marketing implies.

Takeaway: A benchmark decides what a return means, and the choice does more work than it appears to. A purely passive index fund out-returned the Nifty 50 price index by 0.42 percentage points a year for 18.5 years purely because the price index ignores dividends the fund actually collects. Check that any benchmark is total-return, matches the mandate, and was fixed in advance, before reading anything into beating it.

Beta is a stock’s sensitivity to the overall market. A beta of 1.0 says the stock has tended to move one-for-one with the index; 1.5 says it has swung about half again as hard in both directions; 0.5 says it has moved about half as much. It is the number the WACC post plugged into CAPM (the Capital Asset Pricing Model) to get a cost of equity, and asked you to accept as “illustrative”. This post shows where it comes from and why “illustrative” was the honest word.

Two things beta is not. It isn’t a measure of how risky a business is in the sense of whether it might go bust — a high beta just means a company whose price swings with the market, whatever the business underneath looks like. (A share can be jumpy and still have a low beta, if its jumps don’t line up with the index’s.) And it isn’t stable: it’s estimated from past prices, and past prices change their minds.

The formula

Beta = Covariance(stock returns, market returns) / Variance(market returns)

     = Correlation(stock, market) × ( σ_stock / σ_market )

The second line is the useful one. Beta is correlation (how reliably the two move together) multiplied by relative volatility (how much bigger the stock’s swings are). A stock can have a low beta because it doesn’t follow the market, or because it follows it but barely moves. They’re different situations with the same number.

In practice you regress the stock’s daily (or weekly, or monthly) returns on the index’s returns; beta is the slope of that line.

Worked example: Britannia against the Nifty 50

Daily returns for Britannia (NSE: BRITANNIA) and the Nifty 50 price index, 2 April 2024 to 30 March 2026 (the first return uses the 1 April 2024 close). Source: NSE closing prices via Yahoo Finance (stock, index). Historical data, for illustration only.

Check the identity: 0.28 × (20.6 / 13.8) = 0.42. Same answer as the regression, which is the point — beta is that product.

Three things worth reading off this table.

Britannia is a low-beta stock, at 0.42 over the full window. That fits the textbook: a biscuit company’s sales don’t swing with the economic cycle the way a steel mill’s do, so its shares tend not to swing with the index either.

Beta is unstable. Split the same two years in half and you get 0.29 in one year and 0.55 in the next — the second is nearly double the first. Nothing about the business changed that much. The estimate did. Measured on weekly returns instead of daily, the full-window beta comes out at 0.54. Same stock, same two years, four defensible numbers ranging from 0.29 to 0.55.

The index explains very little. An R² of 0.08 means the Nifty’s daily moves explain only about 8% of the variation in Britannia’s daily returns. The other 92%-odd is the stock doing its own thing — results, input costs, a competitor’s price cut. For a large, liquid stock that is normal, and it is why beta is a statement about tendency, not a prediction for any given day.

The WACC post used a beta of 1.1 for Desi Bites, a fictional small-cap. That’s a reasonable convention for a small packaged-foods company; it is not something you could measure, because Desi Bites has no price history. Even for Britannia, which has plenty, the table above shows why any single beta deserves a range around it.

Common mistakes

• Treating beta as a fixed property of a company. It’s an estimate from a window of past prices. Change the window or the frequency and it changes — here, from 0.29 to 0.55 on adjacent years.
• Reading low beta as low risk. Beta only captures the market-wide part of the movement. A stock with a beta of 0.4 and an R² of 0.08 has most of its risk in the part beta doesn’t measure.
• Plugging a two-decimal beta into CAPM and reporting the cost of equity to two decimals. The input has one significant figure of confidence; the output can’t have more. This is why the sensitivity post exists.
• Comparing betas measured against different indices. A beta against the Nifty 50 and one against a small-cap index aren’t the same quantity.

Takeaway: beta is correlation times relative volatility — how much a stock has tended to move for a 1% move in the index. Britannia’s is 0.42 over two years, but 0.29 in one of those years and 0.55 in the other, which is roughly all you need to know about treating any beta as a precise number.

By now this blog has covered how to read three financial statements, roughly twenty ratios, and the machinery of valuation. Each one arrived in isolation. That’s the wrong way to actually use them — nobody computes a debtor-days figure and stops.

So this post, the capstone to the first module of this series, does one pass over one real company, in the order you’d sensibly do it, showing how the pieces build on each other. The company is Britannia Industries, the anchor used throughout this blog, and every figure comes from its audited consolidated FY25 results (year ended 31 March 2025, filed with the NSE (National Stock Exchange) and BSE on 8 May 2025), with the share price being the NSE close on 30 June 2025. All of it historical, and used here purely to illustrate the method.

One thing this post does not do is arrive at a verdict on the stock. There’s a section at the end on why that’s a deliberate stopping point rather than a cop-out.

Step 1: what does the business actually do?

Before a single ratio. Britannia makes biscuits, bread, cakes, rusk and dairy products, sells them through a distribution network reaching millions of Indian retail outlets, and owns brands — Good Day, Marie Gold, NutriChoice, Milk Bikis — that people ask for by name.

That paragraph already tells you what to expect from the numbers: modest gross margins (food inputs are commodities), heavy advertising spend, fast inventory turns (biscuits have shelf lives), and pricing power that shows up in the margin line rather than in volumes. If the ratios contradicted that picture, the interesting question would be why.

Skipping this step is how people end up computing a current ratio for a bank and concluding something silly.

Step 2: profitability

A 40.9% gross margin narrowing to a 12.1% net margin is the shape of a consumer-brands business: the gap is advertising, distribution and staff. Returns are high — a 49.5% ROCE means the business earns roughly half its capital employed back every year in operating profit.

Note the FY25 detail that a single year’s table hides. Revenue grew about 7% while EBITDA was almost flat, because input costs rose faster than prices — gross margin compressed year on year. Good years and bad years both need reading; this was a margin-pressure year for a business that usually doesn’t have them.

Step 3: why the returns are what they are

The DuPont decomposition turns the ROE from a score into an explanation:

Turnover and leverage are near-identical across the two companies. Nearly all of the ROE gap is margin, which is consistent with brand and scale. That’s a specific, checkable claim about where the value in this business sits, and it points you at the right things to monitor: pricing power and input costs, not asset utilisation.

Step 4: efficiency and working capital

That negative cash conversion cycle is the most interesting number in this entire post. Britannia collects from its customers in about 9.1 days while taking around 60.3 days to pay its own suppliers. Its suppliers are, in effect, financing its working capital.

That’s not an accounting trick — it’s what distribution power looks like in the accounts. Distributors pay quickly because they need the stock; suppliers accept long terms because the volume is worth having. Growth funds itself rather than consuming cash, which is why this business can grow without constantly raising money.

Step 5: is the balance sheet safe?

Here’s where reading ratios in isolation would mislead you badly. A current ratio of 1.08 and a quick ratio of 0.74 look, by textbook rules of thumb, like a liquidity problem.

They aren’t, and the other rows explain why. Interest is covered 20.7 times over. Net debt is negative — the company holds more cash and liquid investments than total borrowings. And the negative cash conversion cycle from the previous step means large trade payables are a structural feature of how this business runs, not a sign of trouble paying bills. Those payables inflate current liabilities, which is exactly what drags the current ratio down.

The lesson generalises: a ratio that looks alarming in isolation often has its explanation two ratios away. Rules of thumb are a prompt to investigate, not a finding.

Step 6: is the profit real?

An OCF/PAT ratio above 1 means reported profit is converting into actual cash — the single most useful check against accounting that flatters the income statement. Britannia’s 1.14x is healthy.

But apply the scepticism the FCF post built in. Free cash flow rose from FY24 to FY25 mainly because capex fell from 3.3% to 2.1% of revenue — operating cash flow actually declined slightly. A rising FCF driven by a capex pause is a different fact from a rising FCF driven by better operations, and only one of them is repeatable.

Step 7: what is the market paying?

At the 30 June 2025 closing price of ₹5,851:

Every one of these needs the context the earlier posts supplied. Take the P/B of 32.4x. P/B = P/E × ROE, so a business earning 52.5% on its book value mechanically produces a high P/B at any given P/E. Part of the reason ROE is that high is that the assets doing the work (brands, distribution relationships, shelf position) were built through the P&L over decades and appear nowhere on the balance sheet. That explains why the number is large; whether 32.4x is justified is exactly the question this post doesn’t answer.

The PEG of ~35.9 is the one to be most careful with, and the PEG post covered why: it divides a high P/E by a single weak year’s growth. That makes it unstable — a different base year would give a wildly different figure — so it tells you little either way here.

Step 8: what this exercise cannot tell you

The natural next step would be a discounted cash flow model, a value per share, and a comparison against the ₹5,851 price. This blog stops here, on purpose, and it’s worth being straight about why.

The compliance reason. Wealth Primer is educational. Publishing an intrinsic value for a specific listed stock is functionally a price target, and price targets are the work of SEBI-registered research analysts. That registration exists for good reasons and this blog doesn’t hold it. So the DCF machinery in this series was built on a fictional company, where the method can be shown in full without the output being mistaken for a call.

The intellectual reason, which matters more. Everything above is backward-looking. FY25 is history. The ratios describe a year that has already happened, and a valuation depends almost entirely on what happens next — which of these numbers persist, which mean-revert, which are about to be disrupted by something not in the accounts at all.

Financial statement analysis is superb at telling you what kind of business you’re looking at and which questions to ask next. It is close to silent on whether the price is right. The things it can’t see are exactly the things that usually decide the outcome:

• Management quality, capital allocation instincts, and integrity
• Competitive dynamics and whether the moat is widening or eroding
• Regulatory and input-cost shifts still ahead
• What the market has already priced in

The toolkit gets you to an informed question. It doesn’t get you to an answer, and any framework claiming otherwise is selling something.

How to actually run this on a company

Condensed to a checklist:

1. Understand the business first. What it sells, to whom, and why they buy it. Then predict roughly what the ratios should look like.
2. Read three years, not one. Trends carry more information than levels, and one year is mostly noise.
3. Profitability, then efficiency, then leverage, then cash. In that order — each layer explains the previous one.
4. Run DuPont on the ROE. It converts a score into a reason.
5. Check profit converts to cash. OCF/PAT above 1, sustained.
6. Never read a ratio alone. Britannia’s current ratio of 1.08 would have misled you completely without the four numbers around it.
7. Compare against peers and against its own history, not against textbook thresholds.
8. Write down what would change your mind. If nothing in the accounts could, you weren’t analysing — you were justifying.

Takeaway: No single ratio tells you anything; the toolkit works because each number explains the last one, and a figure that looks alarming alone usually has its answer two ratios away. Used well, financial statement analysis tells you what kind of business you’re looking at and which questions to ask next. It stops well short of telling you whether the price is right, and that limit is worth respecting.

Everything on this blog so far has asked one question: what is this business worth? Read the statements, compute the ratios, forecast the cash flows, discount them back. The capstone closed that arc.

Technical analysis (TA) asks something else: what is this price doing? It studies the record of what buyers and sellers actually did — price and volume — and largely ignores what the company sells, what it earns, and who runs it. To a technical analyst, all of that information has already been expressed by people putting money down, and the chart is the record of it.

That’s a genuinely different discipline, not a rival version of the same one. It’s also the part of investing most surrounded by nonsense, so this series is going to be careful about separating what the technique is from what people claim it can do.

The three assumptions underneath everything

Technical analysis rests on three claims, usually traced to Charles Dow’s writing around 1900. Every indicator in this series inherits them, so they’re worth stating plainly and holding onto sceptically.

1. The price discounts everything. Whatever is knowable about a company — earnings, management, industry shifts, and whatever the well-informed know before anyone else — is already reflected in the price. So studying the price is enough.

2. Prices move in trends. A price in motion tends to stay in motion until something changes it. This is the load-bearing assumption: without it, no trend line or moving average means anything.

3. History tends to repeat. Chart formations recur because they’re produced by human behaviour under greed and fear, and human behaviour is fairly stable across decades.

Notice that none of these is a law. They’re empirical claims — they may hold strongly, weakly, or not at all, and they can hold in one market and fail in another. Assumption 1 in particular is a strong form of market efficiency that sits oddly beside the whole enterprise: if the price already reflects everything, it’s not obvious why studying its past shape should help. That tension is real, it’s unresolved, and the limitations post near the end of this series returns to it properly.

What a chart actually is

Strip away the jargon and a price chart is a record of transactions. Each point says: at this moment, a buyer and a seller disagreed about the future enough to trade, and agreed on this price.

Here’s the dataset this entire series uses — daily closing prices for Britannia Industries, the same company the rest of this blog has been analysing:

Britannia (NSE: BRITANNIA), closing prices, daily bars, 1 April 2024 to 30 March 2026. Source: Yahoo Finance. Historical data, for illustration only.

Two years, 495 trading days. The stock closed at a peak of ₹6,446.05 on 1 October 2024, then fell 29.0% to a closing low of ₹4,575.2 on 4 March 2025, then spent a year recovering. (Measured from the intraday high to the intraday low, the fall was a little bigger — the next few posts use those extremes.)

A fundamental analyst looks at that and asks what changed about the business. (Something did — the gross margin post covered the input-cost squeeze.) A technical analyst looks at the same picture and asks about the shape: where did the falling stop, how many times was a level tested, is the recovery still intact.

Both are looking at Britannia. They’re not looking at the same thing.

About the data in this series

Every chart in this series is built from one file, and you can download it:

Two deliberate choices worth explaining.

It’s real data, and it’s old. The series ends 30 March 2026, more than 6 months before this post publishes. That lag is a rule this blog follows for anything used as a worked example, and it has a useful side effect: nothing here can be read as a comment on where the price is going now.

It’s one stock, over two years. That is nowhere near enough to prove anything about whether an indicator works. When a signal in this series succeeds or fails, that’s an illustration of the mechanism, never evidence about the technique. Anyone claiming an indicator “works” needs thousands of instances across many markets, not one flattering chart — and this series will keep saying so.

Following along in Python

Every calculation in this series is short enough to run yourself. The setup:

import pandas as pd

df = pd.read_csv("britannia-ohlcv-2024-04-to-2026-03.csv",
                 parse_dates=["date"]).set_index("date")
print(df.head())
print(f"{len(df)} bars, {df.index[0].date()} to {df.index[-1].date()}")

Five columns — open, high, low, close, volume — and every indicator in the rest of this series is built from those five numbers and nothing else. That’s worth sitting with. The entire apparatus of technical analysis, all the indicators with impressive names, comes out of five columns of arithmetic.

Where this series is going

The last four are a second module, added after the first ten. The limitations post isn’t a disclaimer bolted on at the end. Technical analysis has real, well-documented weaknesses — hindsight bias, ambiguity about what counts as a signal, and the awkward fact that many published tests of indicators look much weaker once data snooping and trading costs are accounted for (Park and Irwin’s 2007 survey in the Journal of Economic Surveys is the standard reference). A series that showed you eight indicators and skipped the reckoning would be selling something.

Common mistakes

• Treating TA and fundamental analysis as opponents. They answer different questions. Plenty of people use fundamentals to decide what interests them and charts to think about when — and plenty of people use neither well.
• Believing a pattern predicts the future. At its strongest, a chart describes what has happened and where prices have previously reacted. It assigns no probabilities, and it does not know what’s coming.
• Assuming an indicator works because someone showed you a chart where it did. Any indicator can be made to look brilliant with a well-chosen example. That’s a statement about the example.
• Using TA on something that doesn’t trade much. These techniques assume a liquid market with continuous two-way trading. On a thinly traded small-cap, a “pattern” may be three trades by two people.
• Skipping the question of whether the company is solvent. A chart cannot tell you a company is a fraud, or that its debt is about to be restructured. The ratio toolkit exists for the questions a price series structurally cannot answer.

Takeaway: Technical analysis studies price and volume rather than the business behind them, resting on three assumptions — that price reflects everything, that trends persist, and that behaviour repeats — none of which is a law. Held that way it’s a genuine lens on what buyers and sellers have actually done. Held as prophecy, it’s astrology with better charts.

The last post ended with a value of ₹397.72 a share for Desi Bites. Two decimal places. It looks like a measurement.

It isn’t. It’s the output of a chain of estimates — a growth rate that fades on a schedule someone chose, a margin path someone assumed, a beta borrowed from a sector, an equity risk premium that reasonable people put anywhere in a range several percentage points wide, and a perpetual growth rate for a company that will outlive everyone reading this. Every one of those is arguable. The spreadsheet reports ₹397.72 because spreadsheets always report something.

This post is about the two habits that keep a discounted cash flow (DCF) model honest: showing how much the answer moves when the inputs move, and refusing to act on the answer unless there’s room to be wrong.

Sensitivity: show the range, not the point

A sensitivity table re-runs the model across a grid of assumptions and prints every answer. The two inputs worth gridding are almost always WACC and terminal growth, because those two do the most damage.

Value per share (₹) for Desi Bites, at varying discount rates and perpetual growth rates:

The base case sits in the middle at ₹397.72. But move WACC one percentage point in each direction and terminal growth two points — the edges of the grid, and a range no analyst would call unreasonable — and the answer runs from about ₹334 to ₹525. The high end is 57% above the low end. Even at just one point on each input, it’s ₹349 to ₹474.

That’s the honest output of this model. Not ₹397.72, but “somewhere in the high 300s to low 400s if my central assumptions hold, and plausibly ₹334 to ₹525 across assumptions I can’t rule out.”

A range is less satisfying than a number. It’s also true, which is the better property for something you’re about to risk money on.

Which assumptions actually matter

Not every input deserves equal worry. Rough sense of what moves the Desi Bites valuation, in order:

The pattern is worth noticing: the assumptions with the most influence are also the hardest to defend. Tax rates are knowable. Perpetual growth rates are not. Precision and importance run in opposite directions here, which is the fundamental awkwardness of the whole exercise.

Margin of safety

If a valuation is a range rather than a number, acting only when the price sits at the bottom of the plausible range gives you room to be wrong. That buffer is the margin of safety — an idea from Benjamin Graham, and probably the single most durable concept in fundamental analysis.

Margin of Safety (%) = (Intrinsic Value − Price) / Intrinsic Value × 100

Run it on Desi Bites at its IPO price. With a DCF value of about ₹397.72 and a price of ₹640, the margin of safety is (₹397.72 − ₹640) / ₹397.72 ≈ −61%. Negative: the price sits well above the model’s value, so there’s no buffer at all. For a 30% margin against this model, the price would need to be around ₹278. That’s arithmetic on a fictional company, not a price anyone should wait for — and the rest of this post is about how little the ₹397.72 itself deserves to be trusted.

The logic is defensive rather than clever. Your model will be wrong; the question is only by how much and in which direction. A price well below your estimated value means you can be substantially mistaken and still not lose money. A price at or above it means every one of your assumptions has to come good just to break even.

How large a buffer? There’s no formula, and anyone offering one is overselling. The sensible principle is that the buffer should scale with your uncertainty:

Desi Bites hits three of those four. It’s a small-cap with three years of audited accounts, 74.2% of its value sits in the terminal value, and the sensitivity range spans ₹334 to ₹525. That combination argues for a wide buffer, whatever number you’d normally use.

Where margin of safety gets misused

Two failure modes, both common.

The first is treating the buffer as a substitute for understanding the business. A 50% discount to a valuation built on assumptions you can’t defend isn’t safety, it’s a bigger bet on a worse model. Graham’s buffer was meant to absorb ordinary estimation error, not to compensate for not knowing what a company does.

The second is assuming a gap between price and value must be an opportunity. Sometimes the market knows something the model doesn’t — a customer concentration risk, a promoter dispute, a regulatory change working its way through. When the market disagrees with your model, “I’m right and they’re wrong” is one explanation, and it’s rarely the first one to reach for. The reverse DCF from the last post is the better instinct: work out what the price is assuming, then go and check whether that assumption is defensible.

Common mistakes

• Reporting a DCF as a single number. If you’ve built the model you’ve already got the sensitivity grid — it’s the same code in a loop. Publishing a point estimate while sitting on a range is a choice to look more certain than you are.
• Tuning assumptions until the model agrees with the price. The most seductive error in valuation, because the result feels like confirmation. It’s the opposite: you’ve fitted the answer to the data. If you catch yourself nudging terminal growth to close a gap, stop.
• Using a wide margin of safety to justify a business you don’t understand. The buffer covers estimation error, not ignorance.
• Applying the same buffer to every company. A predictable business with twenty years of stable cash flows and a loss-making three-year-old startup do not warrant the same discount.
• Forgetting the model can be wrong in the good direction. Sensitivity cuts both ways, and a business that outperforms your fade assumption is worth more than your model says. Margin of safety is about surviving errors, not about assuming the worst case is the true case.
• Believing precision equals accuracy. ₹397.72 is precise. Whether it’s accurate depends entirely on assumptions no spreadsheet can check.

Takeaway: A DCF produces a number, but what it actually supports is a range — so run the sensitivity grid and quote the range, because that’s the honest output. Then insist on a gap between price and value big enough to absorb the fact that you’ll be wrong about something. The margin of safety isn’t a way of being cleverer than the market; it’s an admission that you won’t be.

A SIP — systematic investment plan — invests a fixed amount at fixed intervals, usually monthly. It’s the default recommendation for Indian retail investors, and the reasoning behind it is genuinely sound: it automates the habit, it removes the need to decide when to invest, and buying a fixed rupee amount means you get more units when prices are low and fewer when they’re high.

That last effect is called rupee cost averaging, and it’s real.

It is also surrounded by claims that the data does not support. So this post does two things: shows you how to measure SIP returns properly, and then tests the claims against twenty years.

Why you can’t use CAGR

With a lump sum, CAGR works — one amount, one start date, one end date.

A SIP breaks that completely. Each instalment has been invested for a different length of time. The first has compounded for twenty years; last month’s has compounded for a month. There is no single “holding period.”

The measure that handles this is XIRR — extended internal rate of return: the single annual rate that makes the present value of all your cash flows equal zero.

Find r such that:

    Σ  CFₜ / (1 + r)^(daysₜ / 365.25)  =  0

where CFₜ = each instalment (negative, money out)
            plus the final value (positive, money in)

There’s no closed-form solution — it’s found numerically. Spreadsheets have XIRR(); Python needs a root-finder. (Excel’s XIRR() divides by 365, not 365.25, so its answer can differ from the code below in the second decimal.)

Twenty years of ₹10,000 a month

UTI Nifty 50 Index Fund, Regular Plan - Growth (AMFI scheme code 100822). Source: AMFI via mfapi.in. Historical data, for illustration only.

The full twenty-year row is the one that gets quoted: ₹23.9 lakh invested becomes ₹77.8 lakh. That’s a real result and a good advertisement for the habit.

Now read the fourth row.

Five years of discipline, and nothing to show

From January 2007 to January 2012, someone invested ₹10,000 every single month without fail — 61 instalments, ₹6,10,000 — straight through the worst crash in modern Indian market history, never missing, never panicking.

Their XIRR was −0.02%.

Five years of doing everything the marketing tells you to do, and the money came back essentially unchanged. Not a disaster. Not a gain either.

This is the fact that “SIPs protect you from market crashes” cannot survive. And here’s the part that surprises people: rupee cost averaging didn’t even come out ahead. The same ₹6,10,000 put in as a single lump sum in January 2007 was worth about ₹6,92,146 by January 2012 — roughly 2.56% a year. The lump sum bought near the top and still did better, because the SIP kept buying all the way through the 2010–11 highs too. Averaging changes your entry prices. It doesn’t guarantee better ones.

That makes the lesson sharper, not weaker. A SIP buys more units when prices fall, which lowers your average cost compared with the prices you paid. It does not make a falling market rise, and it doesn’t promise you’ll beat the money you could have invested on day one.

And the other side of the ledger

Being fair to SIPs, because the honest picture cuts both ways.

Across the full twenty-year run, the portfolio’s value sat below the total amount invested in only 12 of 239 months. The longest continuous stretch was 8 months, from October 2008 to May 2009, and the worst it ever looked was a deficit of about ₹1.09 lakh.

So: through a 60% crash, a disciplined SIP was underwater roughly 5% of the time. That’s a genuinely strong argument for the mechanism.

Both facts are true. A SIP was rarely underwater across twenty years, and a five-year SIP starting at the wrong moment returned nothing. Any presentation giving you only one of those is selling something.

The claims, tested

“SIP returns are guaranteed.” No. The fourth row returned −0.02%.

“SIPs beat lump sum investing.” Not reliably. In a rising market, lump sum wins, because money invested earlier compounds longer. SIPs win when markets fall early in the period. On this fund, a lump sum beat a five-year SIP of the same total in 170 of the 179 monthly start dates we could test — markets rose more often than they fell, so money invested earlier usually had longer to compound. SIPs are chosen mainly because most people receive money monthly, and because they remove the decision entirely. Even the Jan 2007 row, where the crash came early, went to the lump sum.

“Stop your SIP when markets are high.” This is timing, wearing a SIP costume. It requires knowing what “high” means in advance. The Jan 2007 row cuts both ways here. In hindsight, pausing through the 2010–11 highs would have helped this SIP. But you’d have needed to know in 2010 that those were highs, and not the start of another leg up — and the same rule would have had you pause in 2007 and miss the cheapest units of 2008–09. What counted as “high” is only obvious afterwards.

“SIP averaging means you can’t lose.” Averaging lowers your average purchase price. It cannot make a five-year decline profitable.

“Longer SIPs always work.” The 20-year record here is good. It is one market, one period. The rolling returns post made the same caution about any historical distribution.

Doing it in Python

import pandas as pd
from scipy.optimize import brentq

nav = pd.read_csv("uti-nifty50-index-fund-nav.csv",
                  parse_dates=["date"]).set_index("date")
r = nav.nav_regular_growth.dropna()

def xirr(flows):                       # flows: list of (date, amount)
    t0 = flows[0][0]
    npv = lambda rate: sum(cf / (1 + rate) ** ((t - t0).days / 365.25)
                           for t, cf in flows)
    return brentq(npv, -0.99, 10)

def sip(series, start, end, amount=10000):
    units, flows = 0.0, []
    for d in pd.date_range(start, end, freq="MS"):
        price = series.asof(d)
        if pd.isna(price):             # NAV not published yet — skip
            continue
        units += amount / price
        flows.append((d, -amount))
    value = units * series.asof(pd.Timestamp(end))
    flows.append((pd.Timestamp(end), value))
    return len(flows) - 1, value, xirr(flows) * 100

n, value, rate = sip(r, "2007-01-01", "2012-01-01")
print(f"{n} instalments, value Rs {value:,.0f}, XIRR {rate:.2f}%")

That if pd.isna(price): continue is not decoration. The first scheduled instalment of the twenty-year run falls on 1 April 2006, before the fund’s first published NAV on 3 April — which is why the table says 239 instalments and not 240. Getting this wrong silently overstates what you invested.

Common mistakes

• Using CAGR on a SIP. Different instalments have different holding periods. XIRR exists for exactly this.
• Comparing a SIP’s XIRR to a lump sum’s CAGR. They measure different things over different exposure profiles.
• Believing averaging removes risk. It improves your average price. It cannot turn a falling market into a rising one.
• Stopping a SIP when markets fall. This inverts the mechanism — the cheap units are the entire benefit.
• Pausing a SIP when markets look “high.” That’s market timing, and it needs the same impossible foresight as any other timing decision.
• Judging a SIP over a period shorter than the asset’s drawdown recovery. The 2008 recovery took almost six years. A three-year SIP horizon in equity is a bet on not meeting one of those.

Takeaway: XIRR is the right way to measure a SIP, because each instalment has been invested for a different length of time. Measured properly, a twenty-year SIP into this fund turned ₹23.9 lakh into ₹77.8 lakh — and a five-year SIP starting January 2007 returned −0.02%. The habit is excellent; the guarantee it’s usually sold with does not exist.

The last post forecast Desi Bites’ free cash flow to the firm (FCFF) through FY30. But the company doesn’t dissolve on 31 March 2030. It carries on generating cash for decades, and all of that has value too.

Forecasting it year by year isn’t the answer — nobody can model FY47 with a straight face. Instead, everything past the forecast horizon gets collapsed into a single figure: the terminal value, the worth of all remaining cash flows as of the final forecast year.

It sounds like a tidying-up exercise. It is usually most of the answer. In the model below it’s 74.2% of the enterprise value — the five years we carefully forecast account for barely a quarter. That ratio is typical, and it should make you handle this input more carefully than any other in the model, not less.

The formula

The standard approach is the Gordon growth model, which values a cash flow stream growing at a constant rate forever:

                FCFF_final × (1 + g)
Terminal Value = ─────────────────────
                      (WACC − g)

where  g = perpetual growth rate, forever

Then discount that back like any other future amount:

PV of Terminal Value = Terminal Value / (1 + WACC)ⁿ

Note it’s discounted over n years — five here, not six. The terminal value is expressed as of the end of FY30, so it travels back the same distance as the FY30 cash flow.

Choosing g, and the constraint that governs it

There’s one rule about the perpetual growth rate, and it isn’t a convention — it’s arithmetic. g must be less than WACC. Look at the denominator: as g approaches WACC, the terminal value approaches infinity, and past it the formula returns a negative number that means nothing.

That isn’t a quirk of the formula. It’s the formula refusing to model something incoherent: a company growing faster than its cost of capital, forever, would eventually be worth more than everything else in existence.

In practice g should sit at or below the long-run nominal growth rate of the economy the company operates in — roughly 10–11% nominal for India, being real growth plus inflation. This series uses 5%. A common alternative is to use the expected long-run inflation rate, on the reasoning that a mature company grows with prices and no faster.

Anything approaching that nominal rate — call it 10% — is a claim that the business will keep pace with the entire Indian economy in perpetuity. Anything above it is a claim that the business will outgrow the economy forever, and eventually become a bigger and bigger slice of it. Either is a big thing to assert in a spreadsheet cell.

Worked example: assembling the whole discounted cash flow model

Everything from the earlier DCF posts, in one place. All figures ₹ lakh, as of 1 April 2025, discounted at 13.91% (carried as 13.914% in the arithmetic, so the figures below reproduce on a calculator).

Step 1 — discount the forecast cash flows.

Step 2 — compute and discount the terminal value.

TV = 421.6 × (1 + 0.05) / (0.13914 − 0.05)
   = 421.6 × 1.05 / 0.08914
   = 4966.1

PV of TV = 4966.1 / 1.13914⁵      (factor 0.5213, unrounded 0.52133)
         = 2589.0

Step 3 — add them for enterprise value.

Step 4 — bridge from enterprise value to equity value.

This is the step people skip. Enterprise value is what the whole business is worth, to lenders and shareholders together. Shareholders own what’s left after the lenders are paid, so subtract net debt:

Desi Bites holds far more cash than debt after its IPO — net debt of −₹1,480 lakh, i.e. net cash of ₹1,480 lakh — so this step adds value rather than subtracting it. Subtracting a negative is a reliable place to fumble a sign; the sanity check is that a company with spare cash must be worth more than the same company without it.

Step 5 — divide by shares.

Desi Bites Foods — the fictional case study. The bar shows how much of the valuation rests on the terminal value assumption. Illustration of a method only; this is not a real company and no listed stock is being valued here.

The model disagrees with the market

Desi Bites listed at ₹640. This DCF says ₹397.72 — some 38% below the IPO price.

The tempting move here is to go back and adjust assumptions until the model agrees with the price. Resist it, thoroughly. A model tuned to match a price you already knew has told you nothing you didn’t already know.

The useful move is to run the logic backwards and ask: what would have to be true for ₹640 to be right? Holding the same forecast and the same WACC, that price implies a perpetual growth rate of 9.62% — a claim that Desi Bites keeps pace with the entire Indian economy, forever. That’s right up against the ceiling from the previous section.

Or, keeping terminal growth at 5% and pushing on the operating assumptions instead:

Now the ₹640 price says something specific and testable: it’s priced for sustained 25–30% revenue growth with meaningful margin expansion. Whether that’s optimistic or reasonable is a judgement about the business — but at least it’s a judgement about something concrete, rather than a squabble about whether a stock “looks expensive.”

This is a reverse DCF, and for most people it’s the more useful direction of travel. Forward DCF asks you to produce assumptions and hands you a number. Reverse DCF takes the market’s number and hands you the assumptions hiding inside it — and those assumptions are far easier to argue with.

The calculator

The full model as runnable Python — change the assumptions at the top and everything downstream re-computes:

FY25_REVENUE, TAX, DEP_PCT, NWC_PCT = 2592.0, 0.25, 0.045, 0.0868
WACC, TERMINAL_G = 0.13914, 0.05
NET_DEBT, SHARES = -1480.0, 12.5

growth    = [0.18, 0.16, 0.14, 0.12, 0.10]
margin    = [0.172, 0.174, 0.175, 0.175, 0.175]
capex_pct = [0.08, 0.08, 0.06, 0.05, 0.05]

flows, revenue = [], FY25_REVENUE
for g, m, cx in zip(growth, margin, capex_pct):
    prev, revenue = revenue, revenue * (1 + g)
    dep = revenue * DEP_PCT
    ebit = revenue * m - dep
    flows.append(ebit * (1 - TAX) + dep - revenue * cx - (revenue - prev) * NWC_PCT)

pv_flows = sum(cf / (1 + WACC) ** n for n, cf in enumerate(flows, start=1))
tv = flows[-1] * (1 + TERMINAL_G) / (WACC - TERMINAL_G)
pv_tv = tv / (1 + WACC) ** len(flows)

ev = pv_flows + pv_tv
equity = ev - NET_DEBT
print(f"PV of forecast   {pv_flows:8.1f}")
print(f"PV of terminal   {pv_tv:8.1f}  ({pv_tv / ev:.0%} of EV)")
print(f"Enterprise value {ev:8.1f}")
print(f"Value per share  {equity / SHARES:8.2f}")

Common mistakes

• Not checking what share of the value is terminal. If terminal value is 74.2% of your enterprise value, you have not really valued five years of forecasts — you’ve valued a growth rate in perpetuity, with a five-year preamble. Always compute this percentage. If it’s above about 80%, the forecast horizon is probably too short.
• Setting g too close to WACC. The gap between them is the denominator. Narrow it and the terminal value explodes on arithmetic alone, not on anything you learned about the business.
• Discounting the terminal value by the wrong number of years. It’s n years for an n-year forecast, not n + 1. Off by one here and you cut the terminal value’s contribution by roughly the WACC.
• Getting the net debt sign wrong. Subtract net debt from enterprise value. When the company holds net cash, net debt is negative and you’re subtracting a negative — which adds. Sanity-check the direction every time.
• Using the wrong share count. Use the diluted, post-issue count, as the EPS post covered. Dividing by the pre-IPO count here would have produced ₹497 a share instead of ₹397.72 — a 25% error from one wrong cell.
• Presenting the output as a precise number. ₹397.72 is the arithmetic consequence of a stack of estimates. It is not what the share is worth to two decimal places, and the next post is entirely about why.

Takeaway: Terminal value collapses everything past the forecast horizon into one number, and that number is usually most of the valuation — so the perpetual growth rate deserves more scrutiny than any line in the forecast. And when the model disagrees with the market price, the productive question is never “which is right” but “what is the market assuming that I’m not.”