How to Forecast Free Cash Flow: Five Lines That Must Agree

Forecasting free cash flow means forecasting five lines and then making them agree with each other. Revenue, margin, depreciation, capital expenditure and working capital each get their own rule, and the check is that the reinvestment and the profit growth imply the same rate. For Sankalp Industrial Systems Limited, invented, the fifth year ends at Rs 1,70,00,00,000.

Every line a free cash flow forecast contains is already sitting in a tiffin service, and the lines register faster over a lunch box than over a spreadsheet. So a tiffin service is the place to start. A woman cooks and delivers about two hundred lunch boxes a day to three office buildings. She wants to reach four hundred. Asked what has to happen, she will set it out in order, without having been taught any of this.

More boxes means more money coming in, and she can say almost exactly how many more boxes she can physically make: one more cook, one more delivery cycle, one more route. Boxes multiplied by price is the revenue line. Her food cost and her gas and her rent take a fairly steady share of what comes in, and she does not expect that share to improve just because she is bigger. The steady share of the takings is the margin line. Her steel containers and her two cycles wear out and get replaced every few years whether she grows or not. Wearing out and being replaced is depreciation. To actually double, she has to buy a second set of containers, a bigger stove and a third cycle, and that money leaves before a single new box is sold. Buying the second stove and the third cycle is capital expenditure. And she buys vegetables every morning while three of her offices pay her at the end of the month, so the bigger she gets, the more of her own money is permanently parked in tomorrow's lunch. Money parked in tomorrow's lunch is working capital.

Free cash flow is what is left of the money coming in after all five of those things have taken what they need, and forecasting it means writing down a rule for each of the five and then checking that the five rules describe the same business. The last step is the one almost nobody does. The five lines below are built for one invented company, checked four times, and stopped at the end of the fifth year. Present value, terminal value and company value are covered separately.

What does it mean for five lines to agree rather than just add up?

Any five columns of numbers can be made to add up. Adding up is arithmetic, and a spreadsheet does it unasked. The problem is precisely that. A sheet that foots gives the warm feeling of having checked something when nothing at all has been checked. Adding up is a property of the columns. Agreeing with itself is a property of the business the columns claim to describe, and it is a far stronger condition.

Here is the condition in one line. Growth is not free. A company that will earn more next year than it earns this year has to have put something into the ground to make that happen, and how much it has to put in depends on how productive new spending is. So the growth a forecast shows and the spending a forecast shows are not two independent facts. Growth and spending are the same fact, seen from two ends. If the forecast shows growth that its own spending could not have bought, the forecast is describing a company that cannot exist, however plausible each individual row looks in isolation.

Koller, Goedhart and Wessels put growth, the return on invested capital and value into a single expression, and that expression is what makes this check possible at all. The expression says that the growth rate is the share of profit put back into the business multiplied by what that new money earns. Two numbers that can be read off a forecast, multiplied together, must give a third number that can be read off the same forecast. When they do, the forecast is internally consistent. When they do not, one of the five lines is wrong and the sheet will never say which.

Depreciation is where half the confusion beginners have with this exercise lives. Look at the picture above and notice that depreciation appears twice. Depreciation was never money leaving the building, so it comes off revenue on the way down to profit and then comes back on the way through to cash. The line that reduces profit and the line that reduces cash are different lines, and depreciation is exactly the difference between them. Depreciation is why a forecast of profit is not a forecast of cash, and why doing the second one properly takes five rules rather than three.

Sankalp Industrial Systems Limited, an invented manufacturer, makes industrial valves and precision castings and sells the parts and service that go with them. The company has just finished a year with revenue of Rs 12,00,00,00,000, and everything that follows is a forecast of the five years after that, called Year 1 to Year 5. The years carry no calendar dates, and none are needed.

A rate or an amount?

A rate or an amount is the first decision and the most revealing one in the whole exercise. Most people make it without noticing they have made it. Something has to go into the revenue cell: either a percentage, or a number of rupees. A percentage and a rupee amount are not two ways of writing the same assumption; they are two different claims about what a business can physically do.

The tiffin service shows what is at stake. If she says she will grow twelve per cent a year, she is saying she will add twenty-four boxes next year, then twenty-seven, then thirty, then thirty-four, and by the fifth year she is adding forty-two boxes in a single year. Every year she must find more new customers than she found the year before, forever. If instead she says she will add fifty boxes a year, she is saying something much smaller and much easier to check: one more cook, one more route. Anyone can walk into her kitchen and see whether there is room for one more stove. Nobody can walk into her kitchen and see a percentage.

The same distinction decides a manufacturer's forecast, and for the same physical reason. A percentage rule quietly asks the company to add a bigger factory every year than it added the year before. An amount rule asks it to add roughly the same amount of capacity each year. A business with plant and machinery and a hiring plan usually does exactly that. On a company that makes physical things, an amount is nearly always the more defensible claim, and this forecast makes it.

The two rows of that picture are worth reading against each other. Year 1 is identical under both rules, and that is exactly why the choice is so easy to make carelessly. At the point where the assumption is typed, the two agree to the rupee. By Year 5 they have separated by Rs 1,32,61,20,000 of revenue, and more importantly they have separated in what they demand. The rupee rule asks for the same slab of new capacity every year. The percentage rule asks for Rs 1,75,69,20,000 of new sales in the fifth year alone, a demand for 46.41 per cent more new capacity than the first year needed. Nobody argued for that. The demand arrived on its own, inside a percentage sign.

How is the revenue line built, and why does its growth rate fall on its own?

The rule for Sankalp Industrial Systems Limited is one sentence long. Add Rs 1,20,00,00,000 of revenue every year to a base of Rs 12,00,00,00,000. One sentence is the whole revenue ruleThe stated basis on which each year's revenue is projected, written down in one line so somebody else can disagree with it., and everything else in the revenue line is arithmetic.

Now look at the third row. The falling growth rate there is the best thing this rule does, and nobody chose it. The growth rate falls every single year: 10.00, then 9.09, then 8.33, then 7.69, then 7.14 per cent. A declining growth rate is the single hardest thing to defend in a forecast when it has to be assumed, and here it is not assumed at all: it is a consequence of dividing a fixed number by a bigger and bigger base.

The free fade matters more than it sounds. In any forecast review, someone asks why growth is twelve per cent in Year 2 and eleven in Year 3 and ten in Year 4, and there is no good answer. The schedule of percentages was typed by a person who wanted the answer to look reasonable. Nobody can defend the second decimal of a fade. But everybody can defend one sentence about capacity, and the fade comes out of it for free. The arithmetic is doing the part that judgement is bad at.

The same trick runs through the rest of the forecast, and once it has been seen in the revenue line it is visible everywhere. Two other percentages in this guide fall over the five years without anybody having chosen a falling percentage, and both fall for the same reason: something underneath them is fixed in rupees while the base it is divided by keeps growing.

What does holding the margin flat commit the forecast to?

The margin rule here is even shorter than the revenue rule. Earnings before interest, tax, depreciation and amortisation (EBITDA) is a flat 24.0 per cent of revenue in every year, and depreciation is a flat 4.0 per cent of revenue in every year. Nothing improves. Nothing deteriorates. A flat marginHolding the EBITDA margin constant across the whole forecast, at 24.0 per cent of revenue here, so no value comes from the margin improving. is not a lazy assumption; it is a deliberately narrow one, and the narrowness is the point.

The value of a flat margin lies in what it rules out. Nothing in this forecast is earned by getting better at anything. There is no operating leverage, no procurement saving, no mix shift into the aftermarket business, no price rise ahead of cost. Disagreeing with the answer this forecast eventually produces means disagreeing with how much the company can sell, or with how much it must spend to sell it, or with the rate someone later applies to the cash. There is no margin assumption to attack, so nobody can say the margin assumption was heroic. A forecast whose margin offers nothing to attack is a very useful place to stand before an argument starts.

The household version is exact. A person who assumes their salary rises with inflation and their savings rate stays where it is has made a boring forecast that will mostly happen. A person who assumes their salary rises with inflation and that they will also start saving five percentage points more of it every year has smuggled a second, much softer promise into a plan that looks like one promise. The second plan is not more ambitious. The second plan is just harder to check.

Two things in that table are worth stopping on. The first is the tax line: 25.0 per cent is this invented company's own assumed effective rate, put into the model as an assumption, and it is not the tax rate of any country. The second is NOPATNet operating profit after tax: operating profit less tax, with no interest deducted, because it is the profit of the whole business before it is split between lenders and shareholders., the last row, and NOPAT is operating profit after tax with no interest taken out at all. Interest is not in this forecast anywhere. The lenders are dealt with somewhere else entirely, and putting them in here as well would count them twice.

Work the margins off the rupee rows rather than off any label. The habit of checking against the rupee rows is what catches errors. Rs 3,16,80,00,000 less Rs 52,80,00,000 is Rs 2,64,00,00,000, on revenue of Rs 13,20,00,00,000, or 20.0 per cent exactly. Three quarters of that is Rs 1,98,00,00,000, or 15.00 per cent of revenue exactly. Because both the margin and the tax rate are flat, NOPAT rises by exactly Rs 18,00,00,000 in every single year of the forecast. The rest of this guide turns on that number.

How are depreciation and capital expenditure kept consistent with each other?

Depreciation here is driven off revenue at 4.0 per cent, and that is a simplification worth naming out loud rather than hiding. Depreciation is properly a function of the asset base and the rate at which each class of asset is written down, not a function of sales. Driving it off revenue is a shortcut. The shortcut is defensible in this particular forecast because the asset base is being grown in step with revenue, so the two genuinely do move together. In a forecast where they did not, the same shortcut would be a mistake.

Capital expenditureMoney spent on fixed assets, both to replace what is wearing out and to add capacity the business does not yet have. gets an amount rule again, exactly like revenue. Capital expenditure runs Rs 1,34,80,00,000, Rs 1,39,60,00,000, Rs 1,44,40,00,000, Rs 1,49,20,00,000 and Rs 1,54,00,00,000, rising by Rs 4,80,00,000 each year. Expressed as a share of revenue that is 10.21, 9.69, 9.26, 8.88 and 8.56 per cent, falling steadily. Once again the falling percentage is a consequence and not a choice, and once again that is a feature: nobody has to sit in a room and defend why capital intensity improves by 43 basis points in Year 3.

Now the relationship that actually matters, and it is the fastest single sanity check anybody can run on somebody else's forecast. Put the capital expenditure line and the depreciation line next to each other and subtract. Year 1 spending of Rs 1,34,80,00,000 against depreciation of Rs 52,80,00,000 leaves Rs 82,00,00,000. The lower part is maintenance spendingThe part of capital expenditure that stands in for depreciation, replacing what is wearing out rather than adding anything new.: it is replacing what is wearing out, and it buys the company nothing it did not already have. The Rs 82,00,00,000 on top is growth capitalCapital expenditure beyond depreciation. It buys capacity the business did not have before, and it is what a growing revenue line has to be paid for with., and it is the entire physical reason the revenue line is allowed to go up.

Run the test the other way and it becomes a lie detector. A forecast where capital expenditure equals depreciation is a forecast of a business standing still, whatever its revenue row happens to say. There is no new capacity in it. If the revenue line in such a forecast is rising, the two lines are contradicting each other and one of them is fiction. A rising revenue line with no growth capital under it is the most common defect in a first model, and it takes eleven seconds to find.

The shape of this particular forecast is visible in the figure above. The growth block is Rs 82,00,00,000 in every one of the five years, identical. As the asset base gets larger, the maintenance block climbs from Rs 52,80,00,000 to Rs 72,00,00,000. The two blocks tell a coherent story. The company keeps adding the same slab of new capacity each year, and there is more plant to keep, so the cost of keeping it quietly rises. The revenue rule and the spending rule are telling the same story. A forecast should do exactly that, and usually does not.

How is the working capital movement forecast from days?

Working capital looks like a balance sheet detail and behaves like a direct cash deduction in every single year, so most beginners lose the thread there. So build it the way it is actually measured, from days, and let the ratio fall out at the end rather than asserting the ratio at the start.

At the end of Year 0 the company is holding receivables of Rs 2,16,00,00,000, inventory of Rs 1,44,00,00,000 and payables of Rs 1,80,00,00,000. Its cost of goods sold for the year was Rs 7,20,00,00,000, being 60.0 per cent of revenue. Using revenue for all three denominators is the single most common error in this calculation. Now convert each of the three into days, and mind which denominator each one takes. Days sales outstandingReceivables expressed as days of revenue: how long, on average, the company waits to be paid after it has sold something. is measured on revenue and comes to 65.70 days. Inventory is carried at cost of goods sold and suppliers invoice at it, so inventory days and payable days are measured on cost of goods sold. Inventory days come to 73.00 and payable days to 91.25.

The cash conversion cycleReceivable days plus inventory days less payable days: the number of days the business is funding its own operations out of its own money. is those three put together, 65.70 plus 73.00 less 91.25, or 47.45 days. Read that as the number of days the company is funding itself: it pays for material and labour, waits, and only gets its money back forty-seven and a half days later, and every rupee of growth has to carry that wait with it.

A rounding note goes with that figure and it is not pedantry. One decimal place prints payable days as 91.3, and adding the three printed figures then gives 47.4 days. The cycle above was computed on the unrounded day counts and is 47.45. Both are defensible; what is not defensible is quietly nudging one of the printed figures so the three printed numbers appear to add up. Every printed figure in this forecast is rounded for display, every identity is computed on the unrounded value, and where the two differ the text states which one was used.

Now the ratio, an output rather than an input. Receivables plus inventory less payables is Rs 2,16,00,00,000 plus Rs 1,44,00,00,000 less Rs 1,80,00,00,000, giving Rs 1,80,00,00,000 of net working capital against revenue of Rs 12,00,00,00,000. The ratio is exactly 15.0 per cent. The forecasting rule is then to hold the cycle where it is. Holding the cycle means holding 15.0 per cent of revenue, and the money the business has to find each year is 15.0 per cent of the revenue it adds. Fifteen per cent of Rs 1,20,00,00,000 is Rs 18,00,00,000, and that is the movement in every one of the five years.

A genuine trap sits in the next set of figures, and it is worth one warning. The stock of net working capital runs at Rs 1,98,00,00,000, Rs 2,16,00,00,000, Rs 2,34,00,00,000, Rs 2,52,00,00,000 and Rs 2,70,00,00,000 across the five years, the same set of rupee figures as NOPAT. The two rows are equal because both happen to be 15 per cent of the same revenue, and they measure entirely unrelated things. Two rows agreeing to the rupee for an uninteresting reason is exactly the shape of a coincidence that gets mistaken for a finding, and the only defence is to know why each row is what it is.

In what order do the five lines have to be built?

The order is not a matter of taste. Each line needs the answer to the line before it. Building them out of order forces the modeller to guess the same number twice and then reconcile the two guesses, and that is how a model acquires a circular reference and a bad afternoon.

Revenue is first because it is the only one of the five with nothing above it. The margin is second because a margin is a share of revenue and cannot be worked out before revenue exists. Depreciation is third, and it delivers operating profit, the tax charge and NOPAT in one step. The spending a company needs depends on the asset base the first three lines have implied, so capital expenditure is fourth. The working capital movement is a share of the revenue increment, and the increment is not known until revenue is done, so working capital is fifth. Free cash flow is not a sixth line to be forecast; it is what is left once all five exist, and a model that tries to forecast it directly has skipped the only part that was ever going to be checkable.

How do the five lines assemble into free cash flow?

There are two routes to the figure, and they give the same answer. The agreement is itself a useful check. The long route starts from NOPAT, adds depreciation back because it never left, then takes out what the company actually spent on assets and what it had to put into working capital. For Year 1: Rs 1,98,00,00,000 plus Rs 52,80,00,000 less Rs 1,34,80,00,000 less Rs 18,00,00,000 is Rs 98,00,00,000.

The short route collapses the three spending lines into one. Capital expenditure less depreciation plus the working capital movement is Rs 1,34,80,00,000 less Rs 52,80,00,000 plus Rs 18,00,00,000, or Rs 1,00,00,00,000. The Rs 1,00,00,00,000 is the net new invested capitalCapital expenditure less depreciation plus the movement in net working capital: the new money the business genuinely has to put in, over and above replacing what wore out., and it is the only thing standing between profit and cash. So free cash flow is simply NOPAT less that.

The fifth row is the one to stare at. Net new invested capital is Rs 1,00,00,00,000 in every single year, not approximately but exactly, so free cash flow to the firmThe cash the operating business produces for everybody who funded it, lenders and shareholders together, after tax and after everything it had to reinvest. is just NOPAT less Rs 1,00,00,00,000 all the way down. The constant is not a coincidence built into the case for decoration; it is what a company looks like when its spending rules and its revenue rule have been made to agree. Capital expenditure rises by Rs 4,80,00,000 a year, depreciation rises by Rs 4,80,00,000 a year, and the working capital movement never moves, so the three of them cancel to a constant.

And now the consequence, the reason a reader should care. NOPAT rises Rs 18,00,00,000 a year and the block taken out of it does not rise at all, so every rupee of the profit increase lands in cash. Free cash flow rises by exactly Rs 18,00,00,000 a year too: Rs 98,00,00,000, Rs 1,16,00,00,000, Rs 1,34,00,00,000, Rs 1,52,00,00,000 and Rs 1,70,00,00,000. Turned around, that is the cleanest statement there is of what growth costs. The company earns Rs 1,98,00,00,000 in Year 1 and its funders see Rs 98,00,00,000 of it. The other Rs 1,00,00,00,000 did not disappear and was not wasted. The Rs 1,00,00,00,000 bought next year's Rs 18,00,00,000.

What is the one check that proves the forecast agrees with itself?

Here it is, and it is one row on a spreadsheet. The growth in profit must equal the share of profit put back into the business multiplied by what that new money earns. Both of the inputs are already in the forecast, so nothing new has to be assumed to run it.

The reinvestment rateNet new invested capital expressed as a share of that year's NOPAT: the fraction of the profit that never reaches the funders because it went back into the business. is the Rs 1,00,00,00,000 over that year's NOPAT. Year 1 is Rs 1,00,00,00,000 over Rs 1,98,00,00,000, or 50.51 per cent. Then 46.30, 42.74, 39.68 and 37.04 per cent. The reinvestment rate falls every year for the same reason growth falls: the numerator is fixed in rupees and the denominator keeps growing.

The return on new invested capitalThe new profit divided by the new capital that produced it. Here it is 18.00 per cent, and it is an assumption of this forecast rather than something observed. is the other input. The company puts in Rs 1,00,00,00,000 and NOPAT rises Rs 18,00,00,000, so new capital earns 18.00 per cent. The 18.00 per cent is an assumption of this forecast and not a fact about anything. The return on new capital is in fact the single most load-bearing assumption in the whole model, and it is taken apart below rather than left as a number.

Run the check across all four year-on-year transitions and it holds exactly every time. Year 1 reinvestment of 50.51 per cent at 18.00 per cent implies 9.09 per cent of growth, and NOPAT does grow 9.09 per cent into Year 2. Then 46.30 gives 8.33, and it grows 8.33. Then 42.74 gives 7.69. Then 39.68 gives 7.14. Four checks, four exact agreements. A forecast that passes this check describes a company that could exist; one that fails it does not, however carefully each individual row was argued.

One caution about running the check on what is printed. The reinvestment rates above are rounded to two decimals for display, and 50.51 multiplied by 18.00 gives 9.0918 per cent rather than 9.09. The 9.0918 is not a defect in the forecast and it is not a rounding error to be fixed by adjusting a figure. The consistency checkConfirming that the reinvestment rate multiplied by the return on new capital equals the growth the forecast actually shows, in every transition. is computed on the unrounded values, where the Year 1 rate is Rs 1,00,00,00,000 over Rs 1,98,00,00,000 exactly, and that multiplied by 18.00 per cent is 9.0909 per cent, exactly the growth NOPAT actually shows. The practice is to print rounded, compute unrounded, and state which was done.

What is the 18.00 per cent assumption actually claiming?

Most forecasts leave the return on new capital as a number, and a number is very hard to argue with. Take it apart and it becomes a specific claim about the company that anybody on the shop floor can be asked about directly.

The margin is the place to start, and it is flat. NOPAT is 15.00 per cent of revenue everywhere in this forecast, on old capital and new capital alike. So the extra profit new spending produces is 15.00 per cent of the extra revenue it produces, and nothing else. The next question is how much revenue the new spending produces. Rs 1,00,00,00,000 of new capital is assumed to buy Rs 1,20,00,00,000 of new revenue, or 1.20 turns of capital, and 1.20 turns at a 15.00 per cent margin is Rs 18,00,00,000. Exactly the locked annual rise.

Now do the same for the business the company already has. The existing business carries Rs 12,00,00,00,000 of invested capital against Rs 12,00,00,00,000 of revenue, or 1.00 turn, and 1.00 turn at the same 15.00 per cent margin is a 15.00 per cent return. So the whole 18 against 15 assumption is an assumption about capital turnover and says nothing whatever about margin: new capacity is assumed to spin 1.20 times where the existing plant spins 1.00.

The restatement is worth the trouble because of what it does to the conversation. Saying new capital earns more is unfalsifiable and mildly flattering. Saying new machines will produce twenty per cent more output per rupee than the machines already on the floor is a claim a works manager can be asked about over tea, and either there is a reason for it or there is not. In this forecast there is no evidence recorded for it anywhere. The 1.20 turns is an assumption, it is stated as an assumption, and a reader who thinks new capacity would spin at the same 1.00 turn as the old should say so and watch what happens to the consistency check above.

What goes wrong when the check fails?

Three things are usually wrong, and each of them raises free cash flow rather than lowering it. A failing check therefore almost always flatters the answer rather than spoiling it.

The first is a revenue line growing faster than the capital expenditure line could support. The sales are in the forecast and the factory that makes them is not. The second is a margin that expands with no spending behind it, the same defect wearing better clothes. Getting better at something usually costs money, and a forecast that improves for free has left the cost out. The third is the quiet one: working capital held flat in rupees while revenue grows. Holding it flat silently releases cash the company never actually had.

The third defect hides well enough to deserve a sentence of its own. If revenue grows and the receivables line does not, the forecast is claiming that customers who are buying more will somehow be paying faster, and the cash that claim releases turns up in the free cash flow line looking like performance. On this company's numbers, holding working capital flat in rupees rather than at 15.0 per cent of revenue would add Rs 18,00,00,000 to every year, or Rs 90,00,00,000 across the five years, and nothing on the face of the sheet would say where it came from.

Where does the forecast stop, and why there?

At the end of Year 5, with free cash flow to the firm of Rs 1,70,00,00,000. Year 5 is where this guide ends, and it is a real ending rather than a pause for breath.

The company obviously does not stop trading on that day, and what happens afterwards is not ignored. A terminal value handles it, and a terminal value is a separate construction with its own assumptions and its own arguments, covered separately. The forecast produces a stream of cash and hands it on. The five lines stop short of any discount factor, present value, terminal value or enterprise value, and that restraint is deliberate: the quality of the five lines is a different question from what is done with them, and mixing the two is how a modeller ends up defending a valuation when the argument was really about capacity.

So the output of everything above is five numbers and a set of stated rules: Rs 98,00,00,000, Rs 1,16,00,00,000, Rs 1,34,00,00,000, Rs 1,52,00,00,000 and Rs 1,70,00,00,000, on a revenue rule of Rs 1,20,00,00,000 a year, a flat 24.0 per cent margin, depreciation at 4.0 per cent of revenue, an amount rule for capital expenditure, working capital at 15.0 per cent of revenue, and an assumed 18.00 per cent return on new capital that the four checks confirm the rest of the sheet is consistent with. Anybody who wants to disagree now has to name which of those they are disagreeing with. Naming one is the most useful thing a forecast can force somebody to do.