Discounted Cash Flow: A Model That Is Mostly Terminal Value

The claim that a DCF is the rigorous method

Valuation methods divide roughly into two families. The relative family prices a company against something else: against the multiple its peers trade on, against its own history, against a sector average. The absolute family attempts to derive a figure from the business itself, by forecasting the cash it will produce and deciding what that cash is worth today. A discounted cash flow model is the standard member of the second family, and its reputation rests on that distinction. It is supposed to be the method that does not depend on the crowd.

The appeal is genuine. A multiple is silent about why it is the level it is. Our explainer on what a price to earnings ratio actually measures makes the case that the number only becomes informative in comparison, against a peer set or against the company's own past, and that a bare multiple with no comparison attached is close to uninterpretable. A DCF appears to escape that problem entirely. It has forecasts, a discount rate, an explicit arithmetic chain from operating performance to a figure per share. Nothing in it refers to what anyone else is paying.

Here is the difficulty. That chain has an end, and the end is a perpetuity. No forecast runs forever, so every DCF stops explicitly forecasting at some year and replaces all subsequent years with a single number. That number is usually produced by taking the last forecast year's cash flow, growing it by an assumed rate forever, and dividing by the difference between the discount rate and that growth rate. Two inputs, both chosen by the analyst, neither observable, and a division that becomes unstable as they approach each other.

So the question worth answering is not whether the method is sound in principle. It is arithmetic, and the arithmetic is correct. The question is how much of the answer that arithmetic produces actually rests on the forecast the analyst did the work on, and how much rests on the two numbers typed into the terminal value formula. That is a measurable quantity, and measuring it on a model you can inspect line by line is the point of the rest of this page.

None of what follows argues that a DCF is useless. It argues that its output has been widely misread. A model whose answer is dominated by two chosen inputs is an excellent instrument for asking what those inputs would have to be, and a poor instrument for producing a figure to act on. That distinction between a sensitivity instrument and a price oracle is the whole of the argument, and everything below is an attempt to demonstrate it rather than assert it.

The entity, and every assumption written down so it can be disputed

The company below is constructed. It is called Entity M, it is an illustrative manufacturer, and no part of it corresponds to any real listed business. That is a deliberate choice rather than a caution. A constructed entity lets every driver be stated openly, lets the arithmetic be reproduced or challenged, and removes any possibility that a worked example is read as a view on a security someone owns. Every rupee figure on this page is illustrative and comes from the model described here.

The inputs a real model would need come from the filed statements, and knowing where each one lives is half the work. Revenue, operating margin and depreciation come off the profit and loss account; capital expenditure and the movement in working capital come off the cash flow statement; net debt and the share count come off the balance sheet and the shareholding disclosures. Our guide to reading an annual report like an analyst walks through where each of those sits and how the three statements tie to each other, which is worth having in front of you before you attempt to populate a model from a real filing.

One input deserves more care than the rest, and it is the one the whole model is built on. Free cash flow is not a reported line. It is assembled from the cash flow statement, and it inherits every judgement inside that statement, including how aggressively working capital has been managed and what has been classified as operating rather than investing. Our companion piece on reading the cash flow statement covers that document in its own right, and a model built on a cash flow figure the analyst has not interrogated is a precise calculation on an unexamined input.

For Entity M the drivers are these. Base year revenue is 4,200 crore rupees, illustrative. Revenue grows 12.0 percent in the first forecast year and the growth rate fades linearly to 5.0 percent by year ten, which produces a compound rate of 8.48 percent across the explicit period. The operating margin starts at 14.0 percent and fades to 13.2 percent, on the ordinary assumption that competition erodes an advantage rather than leaving it untouched for a decade. Depreciation runs at 4.5 percent of revenue and capital expenditure at 5.5 percent, so the business reinvests slightly more than it consumes in wear. Incremental working capital absorbs 12 percent of every rupee of revenue growth. Net debt is 950 crore rupees and there are 46 crore shares outstanding, both illustrative.

Free cash flow to the firm is then operating profit after tax, plus depreciation, less capital expenditure, less the increase in working capital. Tax is applied at 25.17 percent, which is the effective rate for a domestic company that has elected the concessional regime under section 115BAA of the Income-tax Act 1961: a 22 percent base rate, a 10 percent surcharge and a 4 percent cess, which multiply to 25.168 percent. That is one of the few inputs on this page that is a fact rather than a judgement, and it is worth noting how few of them there are.

Building the discount rate, and being honest about what is known

The discount rate is where a DCF quietly decides most of its answer, and it is almost always presented as a single number with no working shown. Built up properly it has four or five components, and separating them makes visible which are observable and which are assumptions wearing the clothes of measurements.

The risk free rate is the one genuinely verifiable input. The Reserve Bank of India publishes a weekly series on its National Summary Data Page, and for the week ended 7 August 2026 the ten year government security par yield compiled by Financial Benchmarks India stood at 6.79 percent, on a page dated 14 August 2026. The policy repo rate in the same table was 5.25 percent. A ten year yield is the conventional anchor for a model with a ten year explicit forecast, and unlike everything that follows it can be looked up and checked.

The equity risk premium cannot. This is the single most important point in this section and it is routinely glossed over. There is no observable equity risk premium for India or for anywhere else, because it is the compensation investors are currently demanding for holding equity rather than government paper, and nobody publishes that. It has to be estimated, and the estimation methods disagree with each other. A historical average depends entirely on the window chosen and on whether the market being averaged had an unusual few decades. A forward looking implied premium depends on the growth assumptions embedded in the index level, which is the same circularity this page is about. A country risk adjusted construction depends on the adjustment. These approaches can differ by several percentage points for the same market on the same day.

This page therefore uses 6.0 percent as a stated assumption and labels it as one. It is not a fact, it is a placeholder chosen to be within the territory a practitioner might occupy, and the correct response to disagreeing with it is not to argue but to read the sensitivity grid further down, which shows exactly what a different choice does. The same holds for the relative risk measure, set here at 1.05, and for the pre-tax cost of debt at 8.25 percent, and for the assumed capital structure of 80 percent equity and 20 percent debt at market values. Each is defensible and none is measured.

Read that table as a scorecard on the method rather than as a set of instructions. Of the inputs that determine an 11.71 percent discount rate, exactly one can be checked against a primary source. There is a circularity worth naming too: the weights are supposed to be market values, so the equity weight depends on the market capitalisation, which is the price the model is being used to form an opinion about. That does not invalidate anything, but it does mean the phrase independent of what other people are paying is doing less work than it appears to.

The forecast, line by line

With the drivers stated, the explicit forecast is mechanical. Each year's revenue follows from the growth path, the operating margin follows from the margin path, and free cash flow falls out of the four adjustments. The only step that involves any judgement at this stage has already been made, which is choosing the drivers themselves.

The pattern in that figure is the first thing worth noticing, and it is a property of discounting rather than of this company. Free cash flow rises steadily from 385 crore rupees in year one to 787 crore in year ten, more than doubling. The present value of that cash does the opposite: it peaks early, at about 345 crore in year one, and declines every year afterwards to 260 crore by year ten. By year ten the discount factor is 0.331, so a rupee arriving then is worth about thirty three paise today. The model rewards the analyst for forecasting a decade and then discards most of the later years' contribution.

That is the arithmetic setting up the problem. The years you can say most about, the near ones, contribute the least in absolute terms because they are the smallest cash flows. The years you can say least about contribute more per rupee of effort but are discounted hardest. And everything after the forecast, about which you can say nothing specific at all, arrives as a single lump.

Ten years of forecast cash discounts to 3,138.5 crore rupees of present value. The terminal value, computed as the tenth year's free cash flow grown at 4.5 percent and divided by the gap between the discount rate and that growth rate, comes to 11,413.1 crore rupees, which discounts back to 3,772.4 crore. Adding the two gives an enterprise value of 6,910.8 crore rupees. Subtracting 950 crore of net debt gives an equity value of 5,960.8 crore, and dividing by 46 crore shares gives 129.58 rupees a share. All illustrative.

That figure, 129.58 rupees, has two decimal places and looks like the product of careful work. Most of the rest of this page is about why those two decimal places are the least defensible thing on the page.

Taking the answer apart

The decomposition is the central computation here, and it is trivially easy to run and almost never published. Every DCF already contains it: the present value of the explicit forecast and the present value of the terminal value are two numbers that get added together, so the split is available for free. Reporting it changes how the output reads.

In this model the terminal value contributes 54.6 percent of the enterprise value. The ten years of forecast contribute 45.4 percent, and the first five of those years, the only ones about which anybody has a defensible opinion, contribute 24.5 percent. Slightly less than a quarter of the answer rests on the part of the forecast a careful analyst could actually defend. The majority rests on a single number produced by a growth assumption and a discount rate applied to infinity.

Now the complication, because the honest version of this argument has one and leaving it out would be the same sin the page is warning about. The terminal share is not a fixed property of DCF models. Take the identical company, on the identical growth and margin path, with the identical discount rate, and change nothing except the year at which the explicit forecast stops. Cut it at five years and the terminal value is 73.7 percent of the answer. Cut it at ten and it is 54.6 percent. At fifteen it is 39.6 percent, and at twenty it is 28.8 percent.

So an analyst who wanted a model that looked less dependent on its terminal value could simply forecast for longer, and the share would fall without a single assumption improving. This is where a lazy version of the argument collapses, and the collapse is instructive. The terminal share on its own is not a measure of model quality, because it is partly a reporting choice.

What makes the finding survive is the second column of that comparison. Across those four horizons the value per share moved from about 119 rupees to about 133 rupees, a spread of roughly 11 percent, while the terminal share moved from 74 percent to 29 percent. Extending the forecast did not change what the model thought the company was worth. It moved value across the boundary between two labels. The reason is that the extra explicit years were themselves forecast on the fading path that converges on the perpetuity assumption, so lengthening the forecast just spells out in slow motion what the perpetuity was already assuming.

That reframes the point more precisely, and more damningly. The problem is not that the terminal value is a large share of the answer. The problem is that the perpetual growth assumption is doing most of the work regardless of how you label it, and a short forecast merely makes the dependence visible. An analyst who extends the forecast to twenty years and reports a terminal share of 29 percent has not reduced the model's exposure to that assumption at all. They have hidden it inside years eleven to twenty, where it is harder to see and looks like analysis.

There is a cross-check worth running alongside the decomposition, and it takes one line of arithmetic. Divide the undiscounted terminal value by the final forecast year's EBITDA and you get the multiple the model is implicitly paying for the business at the end of the forecast. Here that is 6.80 times, which is an unremarkable figure for a mature manufacturer and therefore reassuring. When that implied multiple comes out at a number no buyer has ever paid for a comparable business, the perpetuity assumption is telling you something the per share figure is hiding.

The grid that is the reason this page exists

Everything so far has used one discount rate and one perpetual growth rate. Both are assumptions. The natural question is what happens to the answer if a different reasonable person had chosen differently, and the natural way to answer it is to compute every combination rather than to describe the effect in words.

The ranges below are deliberately modest. Discount rates run from 9.5 to 13.5 percent, which brackets the built up 11.71 percent by about two points either side and is easily produced by a different view on the equity risk premium alone. Perpetual growth runs from 3.0 to 6.0 percent, all of which sit at or below a plausible long run nominal growth rate for the economy, which is the standard ceiling on the assumption. No cell in the grid is a straw man.

The lowest cell is 89.48 rupees a share and the highest is 264.33. The highest is 2.95 times the lowest. That single ratio is the most honest summary of what a DCF tells you, and it is the number that ought to appear next to any per share figure a model produces.

Read the grid as a shape rather than as a set of cells. The values change gently along the middle and steeply toward the top left corner, where a low discount rate meets a high growth rate. That asymmetry is not noise. It is the perpetuity formula: the gap between the two inputs is the denominator, so the same absolute change in either input has a much larger effect when the gap is already small. A one percentage point change in the discount rate is not a one percentage point change in the answer, and the size of its effect depends on where you already are.

The same grid tells you something about the decomposition too. At the bottom right corner, a 13.5 percent discount rate against 3.0 percent growth, the terminal value is 43.0 percent of the enterprise value. At the top left, 9.5 percent against 6.0 percent, it is 73.4 percent. So the fraction of the answer that rests on the perpetuity is itself a function of the two assumptions being tested, and both quantities move together in the same direction. The inputs that raise the answer are the same inputs that raise the share of the answer resting on them.

There is a common defence of DCF that this grid should retire. It runs: yes the model is sensitive, but you sanity check the output against what the market is paying, and if they are far apart you revisit your assumptions. Look at what that procedure actually does. The grid spans 89 to 264 rupees. Almost any quoted price in that range can be reached by an adjustment to two inputs that no one could call unreasonable. A model that can be tuned to agree with the price, by a process that feels like diligence rather than fitting, is not providing an independent check on the price. It is providing a narrative for it, and the narrative is more persuasive than the price alone precisely because it arrives with a spreadsheet attached.

The failure mode that produces nonsense

The perpetuity formula divides by the difference between the discount rate and the perpetual growth rate. As that difference narrows the quotient grows without limit, and at the point where the growth rate equals the discount rate the expression is undefined. This is not an exotic edge case. It is the most common way a spreadsheet produces a valuation that nobody sanity checks, because the arithmetic never complains.

The base case here has a comfortable gap of 7.21 percentage points. The question worth computing is how far that gap has to close before the output stops describing a business.

The answers are more alarming than the shape of the curve suggests, because the damage starts long before anything looks broken. Close the gap from 7.21 points to 2.91 points and the value per share doubles, to about 259 rupees. A 2.91 point gap is not absurd. It corresponds to an 11.71 percent discount rate against 8.80 percent perpetual growth, which is a number an optimistic analyst might type without pausing. It is worth pausing, because a perpetual rate is a claim that the company keeps growing at that pace after every competitor, every product cycle and every technology has had its turn, and if the figure exceeds the long run nominal growth of the surrounding economy it is also a claim that the company eventually becomes that economy.

At a gap of 1.72 points the value reaches about 409 rupees, and the cross-check from the previous section fires: the terminal value now implies paying 30 times the tenth year's EBITDA for the business in perpetuity. That is the point at which the model has stopped describing an asset anyone would transact in. At a gap of 1.02 points the value is about 662 rupees and the terminal value is 90 percent of the answer, so the ten years of forecasting have become decoration. Push to 0.10 points and the model returns 6,359 rupees a share for the same company with the same cash flows, an implied exit multiple of 524 times, and a terminal value that is 98.9 percent of the total.

Two practical defences follow directly. The first is a hard rule that the perpetual growth rate cannot exceed the long run nominal growth rate of the economy the company operates in, because a business growing faster than its economy forever would eventually be larger than that economy. This is widely stated and widely ignored, usually by half a point at a time. The second is the implied exit multiple, computed above, which catches the failure from the other direction and is the more useful of the two because it converts an abstract assumption into a transaction price you can judge.

It is worth being clear about which direction this bias runs. The divergence is one sided. Widening the gap compresses the value toward a floor; narrowing it sends the value up without limit. So the errors that a careless assumption produces are overwhelmingly errors that make a company look more valuable, and an analyst who is already inclined toward a conclusion will find the input that supports it sitting right there in a cell that looks like every other cell.

Running the model backwards

Everything up to this point has been an argument that a DCF cannot produce a trustworthy figure per share. There is one way to use the same machinery that does not depend on it producing one, and most treatments of DCF leave it out.

Take the market price as given. Hold the discount rate, the perpetual growth rate and every other driver exactly where they are. Then solve for the one input the price is really an opinion about: the growth rate the business would have to deliver for that price to be the right price. This is a reverse DCF, and it converts an unanswerable question into an answerable one. Instead of asking what is this business worth, which requires you to be right about six assumptions, you ask what does this price require, and then you consider whether the requirement is plausible for a business of this kind.

Run on Entity M, the base case forecast of 129.58 rupees a share is equivalent to a constant revenue growth rate of 9.33 percent a year for ten years. Take an illustrative quoted price of 158 rupees, about 22 percent above the model's own output, and the required growth rate is 12.22 percent a year for ten years. So the gap between the model and the price is not a mysterious 22 percent discrepancy to be argued about. It is a specific, checkable requirement of 2.89 percentage points a year of additional revenue growth, sustained for a decade.

That reformulation is the entire value of the exercise, because the second question can be investigated and the first cannot. Whether a business is worth 158 rupees is unanswerable without settling every assumption on this page. Whether a manufacturer of this type can compound revenue at 12.22 percent for ten years is a question about addressable market, capacity, competitive position and management record. It has an evidence base. It can be discussed with someone who disagrees. It can be revisited when the next set of results arrives, because you know precisely what the price needed and can check whether the year delivered it.

The reverse calculation also inherits the model's sensitivity, and the figure is honest about that rather than hiding it. The gold band on the chart is the same question answered at a 10.0 percent discount rate and at a 13.0 percent one. At the illustrative price of 158 rupees, the required growth runs from 7.54 percent to 15.42 percent depending on which discount rate you accept. Change the perpetual growth assumption instead, from 3.0 percent to 5.5 percent, and the requirement moves from 14.19 percent to 10.66 percent. The reverse DCF does not escape the assumption problem. What it does is make the assumption and its consequence sit next to each other, in units a reader can argue with.

Used this way the model becomes a device for interrogating a price rather than replacing it. It stops producing a figure that invites action and starts producing a sentence of the form: at this price, the buyer is assuming the following. That sentence is falsifiable in a way that a per share number never is, and it is available for any company whose statements you can read.

What the model can and cannot support

The conclusion is narrower than either the enthusiasts or the dismissers would like, so it is worth stating without softening.

A DCF cannot support a figure per share presented as a valuation. The grid on this page produced 63 of them, spanning a 2.95 times range, from assumptions no one could call unreasonable. Quoting one of those cells to two decimal places, which is what a model naturally outputs and what a spreadsheet naturally displays, communicates a precision that the same spreadsheet has already disproved on the adjacent tab. The precision is not merely unhelpful, it is actively misleading, because it is the most persuasive part of the output and the least defensible.

A DCF cannot escape the crowd either, which is the claim that gives it its reputation. The capital structure weights are supposed to be market values. The relative risk measure is estimated against a market index. Any forward looking equity risk premium is derived from the level of that index. Three of the discount rate's components are downstream of the same prices the model is supposed to be independent of, and that dependence is inside the machinery rather than declared at the front.

What a DCF can support is a much shorter and much more useful list. It can tell you how much of an answer rests on a perpetuity, which is the decomposition, and which almost no published model shows. It can tell you how wide the honest range is, which is the sensitivity grid, and that range is the correct output of the exercise rather than a footnote to it. It can tell you when an assumption has quietly stopped describing a business, through the implied exit multiple and the gap between growth and discount rate. And it can tell you what a given price is assuming, which is the reverse calculation, and which is the only output on this page that can be checked against the world as the years arrive.

There is a boundary worth marking around all of it. Valuation sits downstream of a question a model like this cannot answer, which is whether the business deserves to be owned at all. Our guide to fundamental analysis for Indian retail investors sets out the screen that answers that earlier question, and it stops deliberately short of valuation for the reasons this page has spent five thousand words demonstrating. A model that runs a beautiful discounted cash flow on a business with deteriorating earnings quality has produced a precise answer to the wrong question, and the precision will make the error harder to see rather than easier.

The habit worth taking from this, if only one survives, is to stop reporting the number and start reporting the range and the requirement. What is the spread across defensible inputs. What fraction of the answer is the perpetuity. What growth would this price need. Those three outputs are computable from the same spreadsheet in an afternoon, they resist the false precision that the single figure invites, and learning to work in that register rather than in the register of confident single answers is a substantial part of what serious valuation training consists of, and is exactly what the method we teach is built around.

Every figure comes from a single deterministic model with no random component, so it reproduces identically on each run. Entity M is constructed and its drivers are stated in full in the body of the page: base year revenue, a fading revenue growth path, a fading operating margin path, depreciation and capital expenditure as fixed shares of revenue, and incremental working capital as a fixed share of revenue growth. Free cash flow to the firm is operating profit after tax plus depreciation less capital expenditure less the increase in working capital. The terminal value uses the perpetuity growth method on the final forecast year. The sensitivity grid re-values the identical cash flow forecast at every combination of nine discount rates and seven perpetual growth rates. The reverse calculation solves for a constant revenue growth rate by bisection, to a tolerance far finer than the two decimal places reported.

Two inputs are drawn from primary sources and are dated. The ten year government security par yield of 6.79 percent and the policy repo rate of 5.25 percent are from the Reserve Bank of India National Summary Data Page dated 14 August 2026, for the week ended 7 August 2026. The 25.17 percent effective tax rate follows from section 115BAA of the Income-tax Act 1961 and its associated surcharge and cess. Every other input, including the equity risk premium, the relative risk measure, the cost of debt, the capital structure weights and the perpetual growth rate, is a stated illustrative assumption and is labelled as such in the build-up table. Verify any rate at source before relying on it.

All results are illustrative and computed on a constructed entity. They are not a valuation of any company, they are not a track record, they are not a forecast, and no figure on this page is a target price or a view on any security. Nothing here is investment advice. The purpose of the exercise is to demonstrate properties of the valuation method itself, which hold regardless of which company the method is pointed at.

• The five ratio screen that decides what is worth owning →
• Reading the cash flow statement →
• What a price to earnings ratio actually measures →
• Accounting red flags in Indian filings →