Forward vs Historical Financials: Which One a Model Uses

A discounted cash flow discounts forward figures only. The historical record enters at exactly two points: it supplies the base year every forward ratio is copied from, and it is the evidence the forecast has to be argued against. On Sankalp Industrial Systems Limited, invented, profit compounded at 8.4938 per cent looking back and the forecast compounds at 8.4472 per cent looking forward.

Underneath that answer sits a division most readers collapse without noticing. A historical figure is a measurement of something that happened. A forward figure is a claim about something that has not. No amount of care, no spreadsheet discipline and no number of decimal places turns the second kind into the first. The useful question is never which of the two is better, but which job each one is doing in the sentence being written. Asking which job each figure is doing dissolves most of the confusion, and what is left is a small set of cleaning problems with a specific repair each.

What exactly is a historical figure, and what exactly is a forward one?

The easy one comes first. A historical figure describes a period that has already finished. Somebody counted the stock, added up the invoices, agreed the closing balances and closed the books. Whether a particular cost was classified correctly is arguable, and people do argue about it, but whether the period happened is not. Sankalp Industrial Systems Limited reports Year 0 revenue of Rs 12,00,00,00,000 and earnings before interest, tax, depreciation and amortisation (EBITDA)Strip out interest, tax and the charge for wearing out plant, and what is left is this line. The line moves with trading rather than with how the business was funded. of Rs 2,88,00,00,000. Revenue and EBITDA there are readings taken off a completed year.

A forward figure works the other way round. There is nothing yet to count, so nobody counted anything. Somebody decided what will happen and wrote it down. Sankalp's Year 5 EBITDA of Rs 4,32,00,00,000 is not a small figure and it is not a careless one, but it exists because a person assumed revenue reaches Rs 18,00,00,00,000 and assumed the margin holds at 24.0 per cent. Remove either assumption and the figure vanishes. A forward figure is a claim with arithmetic wrapped around it, and the arithmetic is the easy part.

Consider a household's last month against its next month. Last month's salary landed in the account and the exact amount can be read off. Next month's salary is what the household expects, and for most people the expectation is very good, but the two are still different kinds of statement. If the employer runs into trouble, one of the two numbers changes and the other one cannot. The difference between a counted figure and an expected one is the whole distinction, and it survives being moved from a household to a manufacturer of industrial valves.

What does a discounted cash flow actually discount?

Forward figures, and nothing else. A discounted cash flow looks backward-facing at first meeting: it is built on a company with a history, it is checked against a history, and its ratios come from a history. None of that changes what is in the discounting. Sankalp's model discounts five annual free cash flow to the firmOperations generate a figure, tax comes out of it, the spending needed to keep the plant running comes out of it, and what remains is measured before anybody funding the business has taken a rupee. figures and then one terminal valueEverything after the last forecast year, squeezed into a single figure so a model does not have to list years without end.. Every one of the six is a forecast.

Not one historical rupee is discounted anywhere in the model. The terminal value claims a company will go on operating indefinitely, so it is the most forward figure of them all, and on Sankalp it is the largest single item in the answer. Any impression that a valuation somehow blends past and future cash can be put down. The past supplies no cash to a discounted cash flow at all. The record supplies shape and evidence instead, and the two enter at different points.

Entry point one: how does the base year reach a model that discounts nothing historical?

Through the ratios. Year 0 is historical: revenue of Rs 12,00,00,00,000, EBITDA of Rs 2,88,00,00,000, a depreciation charge of Rs 48,00,00,000 and net working capitalStock sitting on the shelf plus what customers still owe, less what suppliers are still owed. Growth ties up more of it and shrinkage releases it. of Rs 1,80,00,00,000. Read as percentages of that year's revenue, those become 24.0 per cent, 4.0 per cent and 15.0 per cent. Every forecast year then applies exactly those three percentages to its own revenue line. The model is entirely forward in its cash and entirely historical in its shape, and both halves of that sentence do real work.

Working one forward year through shows it happening. Year 3 revenue is assumed at Rs 15,60,00,00,000. Applying the historical 24.0 per cent makes EBITDA Rs 3,74,40,00,000. Applying the historical 4.0 per cent makes depreciation Rs 62,40,00,000. Taking that charge out leaves operating profit of Rs 3,12,00,00,000. Measured against Year 3 revenue that comes to 20.0 per cent, the same share Year 0 produced. Nobody typed any of those figures. The figures fell out of two percentages read off a year that has already closed. A change to Year 0 moves all five forecast years together. A base year that carries a distortion therefore carries it into every year of the answer.

Which forward lines are not base year ratios at all?

Three of them, and noticing which is a good habit. Capital expenditure is the clearest case. Capital expenditure is not a percentage of anything historical. Its share of revenue drops in each forecast year in turn, and a schedule behaving like that was built line by line rather than carried across from a base year.

Two other forward lines behave the same way. The 25.0 per cent tax rate is the company's own assumed effective rate, and an assumed rate is not a reading. A weighted average cost of capitalEach provider of money wants a different return, and blending those returns in the proportions they supply gives the single hurdle a business has to clear. set at 12.00 per cent for discounting looks forward by construction and takes nothing from Year 0.

So a forecast is not one thing: some lines are the base year wearing a different date, and some lines are separate decisions that need arguing separately. A reader who treats the whole forecast as a single object misses this, and then cannot say which part of a disagreement is about the base year and which is about a judgement somebody made on top of it. The two kinds need different questions, so ask of every forward line whether it is a ratio carried across or a schedule built by hand.

Entry point two: if the past is never discounted, what is it for?

The past is the evidence. A forecast on its own is a set of assertions in a column, and the only material anybody has for deciding whether to believe those assertions is what this business has actually done. Evidence is the second and last place the historical record enters, and unlike the first it stays out of the arithmetic. The record enters the argument instead.

The everyday version runs like this. A student announces a score of 82 in the next examination. The examination has not happened, so there is no way to verify the announcement. Anybody can look at the four papers already sat. If those came in at 79, 80, 81 and 82, the claim is a continuation and doubting it would need a reason. If they came in at 54, 58, 61 and 63, the claim is a departure and the student now has to name what changed. Neither pattern proves anything about the next paper. Both change who has to do the explaining, and that is exactly what the historical record does in a valuation.

How is that check run on Sankalp Industrial Systems Limited?

By putting the two rates side by side and doing the compounding honestly. The historical side comes from five reported years of profit attributable to owners, set out in full just below. Five figures give four growth periods, and those four come out at 8.3333, 7.6923, 8.4337 and 9.5238 per cent.

Now the forward side. Revenue is assumed to reach Rs 18,00,00,00,000 by Year 5, up from Rs 12,00,00,00,000, and net operating profit after tax (NOPAT)Tax on the operating profit has already been deducted at this line, and no interest has. Lenders and shareholders share whatever the line comes to. is assumed to go from Rs 1,80,00,00,000 to Rs 2,70,00,00,000 across the same five. Both are exactly one and a half times, and one and a half times over five years compounds to 8.4472 per cent a year. So the record says 8.4938 per cent looking back and the forecast says 8.4472 per cent looking forward. The two rates sit 4.66 basis pointsPercentage points divide into a hundred of these. Two rates that separate only in the second decimal are much easier to discuss in them. apart.

Two notes on that 4.66 matter more than the coincidence itself. Subtracting the rounded figures, 8.49 less 8.45, gives four basis points and that is the wrong way to do it. The gap has to be computed on the unrounded rates, and that gives 4.66. Subtracting two printed figures is one of the commonest quiet errors in this whole subject. Second, the two rates are not measured over the same number of periods: four looking back and five looking forward, and that difference has to be stated rather than the two being presented as a clean pair.

What did the check find, and does a close match make the forecast right?

The check found a continuation. A continuation is a real finding and worth stating plainly, and it is also not evidence that the forecast is correct. A forecast in line with a company's own record is not verified by that record; it is simply not a departure from it. The consequence is about who has to argue rather than about what is true. A forecast running four hundred basis points above its own history needs somebody to name what is changing. Sankalp's forecast sits close enough to its history that a reader who wants to reject it has to name what is changing instead. Both are useful positions to be standing in, and neither settles anything.

The figure above shows why the coincidence should not be leaned on. The four historical periods range across 1.83 percentage points, from a slow 7.6923 to a quick 9.5238. Against that spread, a gap of 4.66 basis points between two compound rates is noise. If the record had happened to end one year earlier or later, the compound would have moved by far more than 4.66 basis points. The match is a fact about these particular five figures rather than a property of the business.

There is one more honesty point that the arithmetic hides. The historical series measures profit attributable to owners, struck after interest and after the share belonging to the minority in the subsidiary. The forecast rate measures operating profit after tax, struck before both. The two are different lines on the same company, so the comparison is indicative rather than exact. Presenting 8.4938 against 8.4472 without saying so quietly implies a precision the comparison does not have.

Cleaning problem one: what does a single one-off do to a series?

A single one-off breaks the comparison while leaving every figure correct, and that is the hardest kind of defect to see. Sankalp's Year 0 payout carries a regular dividend of Rs 2.60 a share and a special dividendA distribution declared once, deliberately labelled so that nobody reads it as a commitment to do the same thing again next year. of Rs 1.00 a share on top. Multiply each by the 20,00,00,000 shares outstanding and the regular half comes to Rs 52,00,00,000 while the special half comes to Rs 20,00,00,000, so Year 0 distributed Rs 72,00,00,000 in all, being Rs 3.60 a share.

Now put a share price of Rs 90.00 underneath both numerators. Count only what the company committed to and the yield is 2.8889 per cent. Count everything a holder actually received and the yield is exactly 4.0000 per cent. Both divisions are correct and both answers are real. The trouble starts the moment either figure is printed without saying which dividend produced it. Set 4.0000 per cent next to four earlier years of regular-only yields and the chart shows a jump that nobody at the company decided on, because the earlier years contain no special dividend to include.

Cleaning means putting a series on one basis, and it is narrower than people expect. No error has been made, so nobody is correcting one. The item is right and the series is misleading, and the fix is to decide which basis the series is on and hold to it. Take the special dividend out of Year 0 and the five years become comparable. Leave it in and add the equivalent items to the earlier years, if there are any, and they also become comparable. The one forbidden version is a series where one year answers a different question from the other four.

Cleaning problem two: what does a moving share count do to a per share series?

A moving share count breaks the series silently, and again every individual figure stays correct. Sankalp reported earnings per share of Rs 4.80 in the earliest year, then Rs 5.20, then Rs 5.60, then Rs 6.30, and Rs 6.90 in Year 0. Each of those five is the right answer to the right division. The problem is that the divisor moves. A buybackCash a company spends purchasing its own shares back from holders. The purchase cancels those shares and leaves a smaller number outstanding afterwards. of 75,00,000 shares at Rs 80.00 each, costing Rs 60,00,00,000, was executed at the end of year minus 2. So the first three figures divide by 20,75,00,000 shares and the last two divide by 20,00,00,000.

A growth rate run across that break measures two things at once, without any way to say how much of the rise is which. The reported series rises 43.75 per cent from Rs 4.80 to Rs 6.90 over the four periods. Profit attributable to owners rises 38.55 per cent over the same four periods, from Rs 99,60,00,000 to Rs 1,38,00,00,000. The 5.20 percentage points between those two figures is entirely the denominator.

The repair is mechanical. Divide every year's profit by one constant share count, conveniently the current 20,00,00,000, and the restated series becomes Rs 4.98, Rs 5.395, Rs 5.81, Rs 6.30 and Rs 6.90. The restated series rises by exactly 38.55 per cent, the same as profit. Dividing every term of a series by the same constant cannot change its growth rate. The restatement is a comparison device, not a picture of what earnings per share would have been without the buyback. The Rs 60,00,00,000 was spent, and the cash would otherwise have been earning something. Valuing that lost earning is a separate calculation.

What happens to a multiple when only the year underneath it changes?

Everything changes, and the first sight of it is startling. Sankalp's traded enterprise valuePut the value of the debt together with the value of the shares, and the total is what the operating business itself is worth to everybody funding it. is Rs 22,40,00,00,000. Rs 22,40,00,00,000 is one figure on one day. Divided by each year's EBITDA in turn, it gives six different multiples, none of which is wrong.

The same company, the same day, the same enterprise value, and six multiples running from 7.78 times down to 5.19 times. Nothing about the business changed while that column was being written. Only the size of the denominator did. A multiple quoted without its year is not a fact about the company. The multiple is a fact about the company and one particular year, with the year dropped.

Why does a forward multiple always look lower for a growing company?

Because the numerator is held still while the denominator is allowed to grow, and that is arithmetic rather than information. On Sankalp the fall is exactly measurable. EBITDA over the five forecast years rises by exactly one half, from Rs 2,88,00,00,000 to Rs 4,32,00,00,000. Holding the enterprise value fixed, the multiple therefore falls by exactly one third, from 7.78 times to 5.19 times. A multiple falls by exactly one third whenever its denominator rises by exactly one half, and no fact about the business is needed to produce that.

The direction is always the same for a growing company, and it never carries a message. If EBITDA is rising, every further year out gives a smaller multiple. If EBITDA were falling, every further year out would give a larger one. A reader who takes a low forward multiple as a sign of anything is reading the growth assumption back to themselves, dressed as an observation. The multiple has told them what somebody already assumed about the denominator.

When does a historical record stop being a useful guide?

In three specific situations, and each of them is checkable rather than a matter of feel. The first is a business that has changed shape. A company that has bought a division or sold one has a record describing a different company from the one being forecast, and stitching the two halves together produces a series about nothing in particular. The second is a record too short to carry a rate. The third is a period containing something plainly out of the ordinary, where a strike, a fire, a shutdown or a one-time contract makes a year unrepresentative of ordinary trading.

Run all three on Sankalp. The record names no acquisition and no disposal, so the first comes back clean. Profit in each of the five years is ordinary trading profit with nothing exceptional inside it, so the third comes back clean too. The second one bites, and it bites precisely because four growth periods still print as a rate. The 8.4938 per cent computed earlier came from four observations on one company, and quoting it as though it were a property of the business calls for saying how few numbers went into it.

What does the discipline look like in an actual sentence?

The discipline looks like naming the basis in the same sentence as the number, every single time, without exception and without apology for the extra words. A figure and its basis are one object, and separating them is what makes every failure described here possible. Not a footnote, not a column header three rows up, not a convention the reader is assumed to share. The same sentence.

The right hand column is longer and it reads worse, and it cannot be misquoted. A figure travelling with its basis attached survives being pasted into somebody else's note, cropped into a slide or quoted back six months later. A figure travelling alone acquires whatever basis the next reader happens to assume, and the next reader almost always assumes the one that suits them.

Who actually does this, and what do they do with the answer?

Three people, and their uses are different enough to be worth separating. Start with the lender. Inside a lending team the completed years get read before the forecast, and that order is deliberate: a facility comes back out of cash this business itself throws off, and the only evidence that it throws off any sits in years already finished. The forecast supplies management's intention. The record supplies management's track. Where the two disagree, the facility gets sized off the record and the difference becomes a question.

A valuation is built from the forecast and the record is the check on it, so an equity analyst reads the two the other way round. An analyst who computes Sankalp's 8.4938 per cent verifies nothing. The analyst establishes where the burden of argument now sits, and the note that follows either explains a departure or says plainly that there is none. An analyst who skips the historical check has written a forecast nobody can interrogate.

A person buying a small business does both at once and usually with worse information. Somebody buying a shop reads two years of till rolls and one sheet of projections. The till rolls are the base year and the evidence. The projections are the model. The instinct that makes people cautious about a stranger's projections and comfortable with their own till rolls is exactly right, and a valuation model is that instinct written out carefully enough to survive a spreadsheet.

Sankalp Industrial Systems Limited is invented.
Educational material. Not advice on any investment, tax, budget or market position.