Normalisation: Cleaning the Base Year Before Forecasting

Normalisation is cleaning the base year so a forecast starts from a figure that would repeat. On Sankalp Industrial Systems Limited, invented, four checks are run over Year 0 and the Rs 2,88,00,00,000 of earnings before interest, tax, depreciation and amortisation (EBITDA) survives every one of them unchanged. Rs 1,00,00,00,000 of assets generate none of it, and those assets leave the model along with the cash. Both reappear at the bridge.

Normalisation matters because of what a forecast actually does with a base year. Every forward line in this model is a ratio lifted off Year 0 and then applied over and over: EBITDA at 24.0 per cent of revenue, depreciation and amortisation at 4.0 per cent, and net working capital pinned to 15.0 per cent of that same line. Lifting a ratio off a year that did not repeat produces a ratio that will not repeat either, applied five times over and then folded into a terminal valueOne lump figure standing in for every year of cash beyond the last one anybody troubled to forecast. that stands for every year after that. The base year is not one input among many, it is the shape the rest of the model gets grown into.

Which year is the base year, and why can it never be a forecast year?

The base year is Year 0, the last completed year. For Sankalp Industrial Systems Limited, invented, that means revenue of Rs 12,00,00,00,000, EBITDA of Rs 2,88,00,00,000, depreciation and amortisation of Rs 48,00,00,000 and therefore operating profit of Rs 2,40,00,00,000. The Year 0 figures are finished numbers. The year is shut, the ledgers are closed, and nobody is going to write another invoice into it.

Think about how a shopkeeper decides what to order for next month. He looks at what he actually sold last month, not at what he thinks he is selling right now with a week still to run. If he uses the running month he is working off part sales plus a hunch, and every order after that inherits the hunch. A base year taken from a year still in progress is a part year with somebody's estimate bolted onto the end, and the model then grows the estimate for five years and a perpetuity.

There are two more shapes the base year cannot take, and both look reasonable until they are said out loud. A forecast built on a forecast has no anchor in anything anybody counted, so the base year cannot be a forecast year. The base year also cannot be an average of several past years dressed up as one, unless the averaging is stated plainly. The moment several years are averaged, somebody has judged which years were ordinary, and that judgement then sits hidden inside a single number.

What is normalising actually trying to produce?

Normalising is trying to produce one number that would show up again if next year were unremarkable. Not the highest defensible number, not the number management would like on the record, and not the number that makes the model agree with what the shares happen to trade at. A repeatable one.

A repeatable number sounds like a soft target until the forecast's use of it becomes clear. The model does not carry Year 0 forward as a lump. The model strips three ratios out of Year 0 and applies them to every forecast year: the margin, the depreciation rate and the working capital intensity. Year 1 EBITDA of Rs 3,16,80,00,000 is not a view about Year 1 at all. Year 1 EBITDA is 24.0 per cent, the Year 0 margin, applied to Year 1 revenue. So is Year 5 at Rs 4,32,00,00,000. So, through the growth assumption, is every year past Year 5.

So the honest way to describe the exercise is this. The year is not being tidied for its own sake. The decision is which figures the model is allowed to assume repeat, and the four checks below make that decision in the open rather than inside a formula. The four are composition, one-offs, whose cash it is, and what the balance sheet contributed. The four run in that order for a reason: each one narrows what the next one has to look at.

Check one: do the parts add up to the whole being forecast?

Start with the least glamorous question in the exercise. Take the segment note, add the divisions together, and see whether they reach the consolidated figure. Sankalp Industrial Systems Limited, invented, runs three of them.

Revenue behaves. The three revenue lines come to Rs 12,00,00,00,000, matching the consolidated line to the rupee. Profit does not behave. Segment EBITDA reaches Rs 2,97,00,00,000 where the group line reads Rs 2,88,00,00,000, so Rs 9,00,00,000 is sitting between the two, and at a 24.75 per cent blended margin the divisions look better than the company does.

The gap has a name and the name is the finding. The gap is head office cost that has never been pushed down to any division: the group finance team, the chairman's office, the group insurance premium, the corporate audit fee. No division carries it because no division caused it on its own. Head office cost is real, it recurs, and nothing gets added back.

So what did the check buy, if the answer is that nothing changes? The check bought the knowledge that the segment note and the group account are two different objects. Suppose a modeller liked the look of those division margins, forecast the three of them separately, and added the results. The head office would have quietly disappeared from the model, and the group EBITDA would come out Rs 9,00,00,000 too high in Year 0 and too high by a similar margin every forecast year after it.

Check two: would this item be here again in an ordinary year?

Now the check everybody thinks normalisation actually is. An item comes out of the base year when it fails two questions, taken in order. Would it appear again in an ordinary year? And is it sitting on a line the forecast actually reads?

Most people run only the first question and then wonder why their adjustment did nothing. A set of accounts is full of items that are plainly one-offs and completely irrelevant to a cash flow forecast, so the second question is the one that does the work. An insurance recovery below operating profit. A revaluation reserve movement. A prior period tax refund. All genuinely non-recurring, none of them anywhere near the lines the model touches.

Sankalp Industrial Systems Limited, invented, gives a clean instance. Year 0 carries a special dividend of Rs 1.00 a share, Rs 20,00,00,000 across the 20,00,00,000 shares in issue, sitting on top of a regular dividend of Rs 2.60. The special dividend is unambiguously a one-off. The regular dividend has climbed in five even steps, 1.80, 2.00, 2.20, 2.40 and 2.60, and the special one was declared once and carried no promise about next year.

Run the second question and it stops dead. A dividend sits below every line this forecast uses. The model is built on operating profit, tax on that operating profit, depreciation, capital spending and working capital. The forecast never reaches the distribution line. The forecast never reads the line the Rs 20,00,00,000 sits on, so a textbook one-off moves the base year by nothing at all.

The discipline matters here more than the adjustment does. Before anything is touched, the question is which line the item sits on. Get that wrong and a whole afternoon can go on carefully cleaning a number the model was never going to look at, with the work feeling productive the entire time.

Check three: whose cash is sitting inside this figure?

Two companies sit inside Sankalp's Year 0 in two completely different ways, and telling them apart is the whole of this check.

Sankalp Coatings Private Limited, invented, is 75.0 per cent held and fully consolidated. ConsolidationA parent adds a subsidiary's whole revenue and whole cost into its own accounts, whatever share of that subsidiary it actually holds. means the whole of that subsidiary's revenue and the whole of its cost are already inside the group lines. Every rupee of its EBITDA is inside the Rs 2,88,00,00,000, including the quarter of it that belongs to somebody who is not a shareholder of the parent.

Aruna Tooling Private Limited, invented, is 26.0 per cent held and equity accountedA stake too small to consolidate shows up as one carrying value plus a slice of profit, so the investee's sales and costs never reach the parent's own lines.. Its revenue and its costs are nowhere in the group lines at all. The stake shows up as a carrying value of Rs 55,00,00,000 on the balance sheet and a share of profit far below the operating lines. None of its EBITDA is in the Rs 2,88,00,00,000, not a rupee of it.

Here is the trap on the consolidated side, and it costs money in both directions. Since the group holds only three quarters of the subsidiary, it is tempting to scale the base year EBITDA down to 75 per cent and forecast from there. Do not. The Rs 2,88,00,00,000 correctly states what the operating business generates, and ownership is settled once, later, at the bridge, where Rs 60,00,00,000 of minority interestWhatever slice of a fully consolidated subsidiary belongs to shareholders standing outside the group. is deducted. Cutting the EBITDA and then deducting the minority as well removes the same claim twice.

The trap on the equity accounted side runs the other way. Somebody reads about Aruna Tooling Private Limited in the annual report, decides its business is growing quickly, and lifts the forecast growth rate accordingly. Raising the growth rate that way forecasts growth in cash that was never in the base year to begin with. The associate can triple and the Rs 2,88,00,00,000 does not move.

Check four: which assets produced none of the profit being forecast?

The final check turns to the balance sheet and asks one question of every large item on it. Did this thing generate any part of the Rs 2,88,00,00,000 that is about to be grown?

Three items answer no. Sankalp Industrial Systems Limited, invented, holds cash and cash equivalents of Rs 1,20,00,00,000, split Rs 40,00,00,000 of operating float the business needs to trade through the month and Rs 80,00,00,000 of excess cashWhatever is left in the bank once the float needed to trade through an ordinary month has been set aside.. The company also holds a surplus land parcel carried at Rs 45,00,00,000 that nothing is built on. And it holds the Rs 55,00,00,000 stake in Aruna Tooling Private Limited. Check three has already established that the stake sits outside the operating lines. The land and the stake together are Rs 1,00,00,00,000 of non-operating assetSomething sitting on the balance sheet that earns nothing the forecast is counting..

None of the three produced a rupee of the EBITDA being forecast, so none of them belongs inside the forecast. All three leave the model. And here is the distinction readers lose more often than any other on this subject.

Leaving the model is not the same as being deducted. An item that leaves the operating forecast is not thrown away, it is valued somewhere else and by a different method. The land does not earn cash, so a cash flow model has nothing to say about it. Its value is what somebody would pay for the plot, and that goes onto the asset line of the bridge at Rs 45,00,00,000. Same for the associate stake at Rs 55,00,00,000. Same for the cash. The cash is not inside the model either, and it is added at the bridge in full.

Picture a household deciding what it can afford each month. The salary arrives every month, so the budget runs on the salary. The budget does not run on the gold in the locker. The gold is not worthless. The gold is not income, and it gets counted separately when the household totals up what it owns. Take the gold out of the monthly budget and also forget to count it in the total, and the household has been made poorer with a stroke of a pen.

Why does the working capital ratio get checked as well as the profit?

Profit is not the only ratio the model lifts off Year 0. The forecast pins net working capital to revenue at 15.0 per cent and leaves it there for every year it runs, so a base year cycle that happened to be unusually tight has just been assumed permanent, and each forecast year will quietly show cash the business is not going to produce.

So it gets measured. Receivables stand at Rs 2,16,00,00,000 against revenue of Rs 12,00,00,00,000. On a 365 day year that is 65.70 days of sales outstanding. Inventory of Rs 1,44,00,00,000 against cost of goods sold of Rs 7,20,00,00,000 is 73.00 days. Payables of Rs 1,80,00,00,000 against the same cost of goods sold is 91.25 days. Note which denominator goes with which line: receivable days run on revenue, inventory and payable days on cost of goods sold, and using revenue for all three is the commonest slip in the calculation.

Add them up the way the cash conversion cycleThe stretch of days between paying for stock and collecting from customers, once credit taken from suppliers has been netted off. is put together. Days of inventory plus days of receivables less days of payables gives 73.00 plus 65.70 less 91.25, or 47.45 days. And net working capital, being Rs 2,16,00,00,000 plus Rs 1,44,00,00,000 less Rs 1,80,00,00,000, comes to Rs 1,80,00,00,000. Net working capital of Rs 1,80,00,00,000 is exactly 15.0 per cent of revenue.

The finding is that the base year cycle is at an ordinary level, so the 15.0 per cent the forecast carries forward is the base year's own ratio and not an improvement smuggled in through the back door. Revenue adds a flat Rs 1,20,00,00,000 in each forecast year, and holding the ratio steady means Rs 18,00,00,000 of that goes straight into the cycle rather than to anybody. The Rs 18,00,00,000 is a direct consequence of a base year ratio nobody adjusted.

The printed figures otherwise look like an error, so one small thing is worth saying out loud. Printing the payable days as 91.3 and adding the three printed numbers gives 47.4 days. The 47.45 above is computed on the unrounded days. Both are right, and stating which of the two was used is the only way anybody can check the other one.

What did all four checks find on Sankalp's Year 0?

Run the four in order and the answer is the same every time. Nothing is added and nothing is taken away from the operating figure.

A net effect of nothing is a finding and not an absence of one. It says something specific about this set of accounts: the reported operating figure is already a repeatable figure, and the things that needed moving were sitting on the balance sheet rather than in the profit line. Manufacturing an adjustment here, so that the exercise had something to show for itself, would teach a habit that costs real money. The failure described below is exactly that habit.

What does an unclean base year cost by the time it reaches the answer?

Now the number this section exists to print. Suppose something had been missed. Suppose Rs 10,00,00,000 of EBITDA that will not repeat had been left inside Year 0. Rs 10,00,00,000 is a small amount against Rs 2,88,00,00,000, and against revenue of Rs 12,00,00,00,000 it is 0.83 of a margin point. In print it would round away to nothing.

Follow it through. The margin is 0.83 of a point higher, so every forecast year is 0.83 of a point higher on its own revenue. Depreciation, capital spending and working capital do not move, so the whole of that extra flows into operating profit after taxThe tax charge here is struck as though there were no borrowing at all, so what remains is trading profit before any lender has been paid. at this company's own assumed effective rate of 25.0 per cent. Year 1 gains Rs 8,25,00,000, rising in steps to Rs 11,25,00,000 in Year 5. Discount that at what this company's capital costs, a weighted average cost of capitalLenders and shareholders together are assumed to want a blended yearly return, and that blend is the rate a model discounts at. settled elsewhere at 12.00 per cent, and the five explicit years come out Rs 34,53,71,657 richer than they were.

Then the tail. Year 5 is Rs 11,25,00,000 richer, the terminal build grows it and takes its reinvestment out, and the terminal value rises by Rs 1,21,87,50,000. Discounted back five years that is Rs 69,15,51,480. Add the two halves together and Rs 10,00,00,000 left in the base year is worth Rs 1,03,69,23,137 of enterprise value.

A base year error is multiplied about tenfold on its way to the answer, and the exact multiplier on this model is 10.37 times. Against the model's Rs 21,28,13,79,094 of enterprise value that is 4.87 per cent, or Rs 5.18 a share. On an answer where the terminal figure is bearing 77.99 per cent of the weight, nothing below roughly a lakh in these totals carries any information at all.

Notice where the damage comes from. The tail stands in for an unbounded number of years and the inflated margin is baked into every one of them, so roughly one third of the damage lands in the five explicit years and two thirds in the tail. The split is the reason the multiplier is so much larger than intuition expects. Errors behave that way in an addition, so most people expect an error in a base year to stay roughly its own size. In a model built on ratios and extended by a perpetuity, an error compounds through the structure.

The relationship in that table is a straight line at 10.37 times. Every figure in it is printed here so that a reader who never touches the control below still has all of it.

Where does cleaning stop and construction begin?

Everything above assumed an honest attempt to get the base year right. The interesting failure is what happens when someone is trying to get it right and also, quietly, hoping it comes out high.

So the test that separates the two is not about the merit of any single adjustment. Each adjustment has an argument attached, and that argument gets won or lost on grounds that have nothing to do with the model, so judging them one at a time is precisely what fails. The test is directional: count the adjustments, then count how many of them lowered the base year, and if the answer is none, the exercise had its conclusion built into it whatever each individual line said.

Ordinary years contain both windfalls and setbacks, so a genuine cleaning pass on a genuinely messy year produces movement in both directions. A pass that produces four upward moves and nothing downward is not describing a company, it is describing whoever ran it.

What does an adjustment log look like, and why write it first?

The second test is simply whether there is a record. An adjustment written down with its reason before the forecast starts is something a colleague can argue with. An adjustment made inside a formula while the model is being built cannot even be found, let alone argued with, and six months later neither can the person who made it.

The artefact itself is unimpressive, and that is part of why so few people keep one. The log is three columns.

A log written afterwards describes what was already done rather than putting a decision open to challenge, so the log belongs before the first forecast cell is typed. And write it even when the net comes to nothing, as it does here. A log recording four checks and no change tells the next reader the checks were run, and that is a different and far more useful thing than silence.

Who actually runs these checks, and what do they do with the answer?

Three sorts of reader run this on the same set of accounts and want different things out of it.

Everything downstream is grown from what survives the four checks, so an equity analyst building a model runs them before a single forecast row exists. The output that matters is the margin: 24.0 per cent here, and the note that it is a base year ratio rather than a view about the future. The analyst who writes that down has made the single biggest assumption in the model visible to whoever reviews the work.

A credit officer at a lender runs a version of the same pass for a different reason. Lending decisions get sized against a repeatable earnings figure, so the composition check and the one-off check are the whole exercise. Head office cost that has not been allocated is exactly the sort of item that vanishes when a borrower presents division results, and the Rs 9,00,00,000 here would matter a great deal to somebody sizing a facility off EBITDA.

A corporate development team on the other side of a transaction runs the checks to build the figure the price will be quoted against. Here the whose-cash check earns its place: a multiple applied to an EBITDA that consolidates a subsidiary held at 75.0 per cent produces an enterprise value that includes a claim the buyer is not buying, and the Rs 60,00,00,000 of minority interest has to come out somewhere or the price is wrong by that amount.

All three share the second half of the discipline rather than the first. Running the checks is the easy part and takes an afternoon. Recording what the checks found, including when they found nothing, is what makes the resulting figure something another person can rely on rather than something they have to take on trust.