Manufacturing and the PLI Scheme in India

A production-linked incentive pays a producer in proportion to what it actually makes, instead of paying up front for a plant to be built. Paying against output rather than against a plant is the whole instrument. The design moves the risk of a plant that never produces from the public purse to the producer, and if nothing is made, nothing at all is paid.

Why a country makes what it makes, and why what a place gives up matters more than what it happens to be good at, is comparative costThe rule that a place should make whatever costs it the least in things given up, rather than whatever it can make in the largest quantity. Taught in full at the start of this reading order and only used here. and it is settled. A tariffA charge collected at the border when goods come in. Who ends up paying it is a separate question from who hands over the money, and both are covered separately. splits its cost between buyer and seller rather than landing wholly on the foreign seller, and that is settled too. The price of a country's exports against the price of its imports fixes what the country can afford, and that is settled as well. All three are borrowed here and taught in full elsewhere.

Governments announce support for manufacturing, and those announcements have a shape: a headline number, a mechanism named in a phrase or two, and a great deal left unsaid. The unsaid things decide whether the headline means anything at all, and a reader who closes the question early never finds out which ones did.

India runs a Production Linked Incentive (PLI) Scheme. The Ministry of Commerce and Industry and the Ministry of Finance hold its outlay, its list of sectors, its targets, its dates, its eligibility rules and any assessment of whether it has worked. Those two ministries are where the scheme as it actually stands is set out. A live scheme moves, so any fixed record of its figures is wrong from the day those figures change.

All the arithmetic runs on the Republic of Sankhya, an invented country that trades nowhere, and on Nirvaha Machine Works, an invented producer inside it. Sankhya makes onions cheaply and machines expensively. One worker-year applied there yields either 100 quintalsA quintal is one hundred kilograms. Bulk farm output in India is counted and priced this way rather than by the tonne or the kilogram. of onions or 2 machines, so one machine costs 50 quintals of onions surrendered. In Marut, its invented trading partner, the same machine costs only 10. Sankhya therefore buys machines: 2,50,000 of them at Rs 50,000/- each, or Rs 1,250 crore, against onion exports of Rs 1,000 crore. Onions and machines are one product pair rather than the whole external trade of the economy, and on that pair the balance is minus Rs 250 crore.

Hold the two prices against each other for a moment. Almost everybody skips that check. Onions leave Sankhya at Rs 2,000/- a quintal and machines land at Rs 50,000/-, so in goods rather than money a machine costs 25 quintals of onions. A price of 25 quintals sits between the 10 quintals Marut gives up to build one and the 50 Sankhya gives up. A price has to sit inside that band for both sides to want the exchange at all. Money prices landing outside that band would be describing a trade Sankhya would simply refuse, and no amount of tidy rupee arithmetic would rescue them.

Whether a country should build manufacturing deliberately, whether any scheme should exist and whether any scheme has worked are live questions about real policy. Each of them turns on evidence gathered years after an announcement, so the questions come first and the answers come much later.

Why would a country try to build an industry it does not already make cheaply?

Begin at the objection, one a careful reader will already be holding. Comparative cost says Sankhya should grow onions and buy its machines from Marut, a place that gives up a fifth as much to make one. Nirvaha Machine Works wants to make machines in Sankhya. Has somebody stopped reading halfway through the rule?

No, and the reason sits in a single word that the rule never says out loud: today. A cost table is a photograph. The photograph records what each place gives up to make each thing at the moment it was taken, and it is completely correct about that moment. A cost table says nothing whatsoever about what those numbers would be after ten years of somebody actually doing the work.

The gap between today's cost and tomorrow's is the entire claim behind every deliberate attempt to build an industry. Some costs fall with practice. The fortieth machine takes less time than the first because the mistakes have already been made, the supplier down the road now stocks the right steel, and three people in the workshop know things nobody wrote down. Volume brings economies of scaleThe fall in cost per unit that comes from making more of something, because fixed costs are spread wider and bigger runs waste less. A general idea from the fundamentals, used here rather than taught here. on top of that, so cost per machine can fall for two separate reasons at once.

The same thing happens in a kitchen. The first time somebody in the house attempts a dish it takes ninety minutes and a third of it goes in the bin. By the fortieth attempt it takes twenty five minutes and none of it is wasted. Nobody looks at the first attempt, works out the cost per plate, and concludes that the household should buy that dish from outside forever. Everybody understands instinctively that the first attempt is not the standing cost.

The whole case for deliberate industrial support rests on the gap between what a cost table records and what a cost table cannot record, and a reader who does not see that gap will think the idea contradicts comparative cost. The idea does not contradict comparative cost. The dispute is narrower and more interesting, and it is about whether today's costs are fixed.

Be honest about what kind of statement that is. The claim that costs fall with practice is a claim, not a fact. The same claim may be true for one product in one place and false for another product in the same place. The learning may be real and still be smaller than what it costs to buy the time in which it happens. And notice the limit of the claim. Nothing in it says that the learning will happen. The claim says only that learning can happen, so everything after this depends on conditions rather than on hope.

There is a second route to the same goal, mentioned here only so it can be set aside. A country can make imported goods dearer instead of making home production cheaper. A barrier at the border does that. Its cost does not land where an announcement usually implies it lands, either: on the Sankhya machine, a 20 per cent charge on a landing price of Rs 50,000/- puts the buyer's price anywhere between Rs 50,000/- and Rs 60,000/-, depending on how much of the charge the foreign seller absorbs rather than passes on. At 60 per cent passed on, the buyer pays Rs 56,000/- and the exporter absorbs Rs 4,000/-. Who carries a barrier is covered separately and is not reopened here. An incentive works from the other side: it lowers the cost of making rather than raising the cost of buying.

What kind of instrument is a production-linked incentive?

A production-linked incentive is a payment linked to output. The definition is short enough to carry around: the producer makes a thing and sells it, and the state pays a stated share of what was made and sold. Make more, receive more. Make nothing, receive nothing.

Put Sankhya numbers on it, all of them invented. Nirvaha Machine Works proposes to make 10,000 machines over a four year window, at the same Rs 50,000/- a machine that Marut charges. Meeting that promise in full would be Rs 50,00,00,000/- of production. Sankhya offers an incentive of 5 per cent of the value Nirvaha actually sells, up to a ceiling of Rs 2,50,00,000/-. Work that through at full delivery: 5 per cent of Rs 50,00,00,000/- is Rs 2,50,00,000/-, exactly the ceiling. The instrument was sized so that meeting the promise in full uses the whole of it and not a rupee more.

Notice what travels with each machine. Five per cent of Rs 50,000/- is Rs 2,500/-, so every machine that leaves the yard and finds a buyer carries Rs 2,500/- of public payment attached to it, and no machine that fails to exist carries anything. Because the payment is a rate applied to a thing that has already happened, the instrument cannot pay for a plant that was never finished.

The ceiling matters more than it looks. Above the promise the payment simply stops rising. If Nirvaha made 13,000 machines instead of 10,000, the rate on its own would give Rs 3,25,00,000/-, and the ceiling holds the payment at Rs 2,50,00,000/-, withholding Rs 75,00,000/- of what the rate would otherwise have produced. A reader who forgets the ceiling will think a wildly successful producer keeps collecting at the same rate forever, and the announcement almost never says this out loud.

The size is worth keeping honest. Ten thousand machines is 4 per cent of the 2,50,000 that Sankhya buys from abroad. Even a promise met perfectly replaces a small slice of the import bill, and any announcement that sounds like it replaces the whole of it is being read too generously. There is also a gap between the two events in the middle of all this: the announcement, and the disbursementThe moment money actually leaves the paying authority and reaches the receiver, as against the moment a payment is announced or approved. The two can sit years apart. that eventually follows it once output has been counted and checked.

How is paying for output different from paying for a plant?

The other way to support a producer is to pay for the input. An up-front grant hands over money against something going in: land bought, a shed built, machines installed. The paperwork is checked, the plant is inspected, the money leaves. All of that can be completed before a single unit of output exists.

Both instruments can be sized to cost the same. Sankhya could offer Nirvaha a grant of Rs 2,50,00,000/- towards the plant, or the output-linked payment worked through above with its ceiling of Rs 2,50,00,000/-. In the budget line where the cost is written, they are the same number. In every other respect they are opposites.

The grant puts the risk of a plant that never produces on the public purse, and the output-linked payment puts that same risk on the producer, and that is the strongest argument anybody makes for the second form. Under the grant, a plant that is built and then never runs well has already been paid for. Under the incentive, the same plant collects nothing, and the money spent building it was the producer's own.

Two ways to pay a tailor make the same point in an afternoon. Half the money can be handed over in advance so cloth and thread can be bought, or payment can be made per finished shirt. The first gets the work started for somebody who has nothing to start with. The second means that if no shirt is ever finished, the customer is not out of pocket. Neither arrangement is the clever one. The two arrangements allocate one specific risk differently, and everybody who has ever paid a deposit already knows which risk it is.

The output-linked form has a cost of its own. The plant still has to be built with money that arrives before any output does, and the incentive cannot help with that. A producer who can raise that money is helped a great deal by a payment that arrives later. A producer who cannot raise it is not helped at all, and may not apply. The instrument therefore selects who can use it, and that selection is part of what it does, not a side effect of it.

What do the two instruments cost when the producer delivers, and when it does not?

Four cells settle it. Take Nirvaha under both instruments, and run each through two outcomes: the promise met in full at 10,000 machines, and a bad four years at 2,000 machines, a fifth of what was promised. Every figure below is worked from the same two rules already stated, the flat grant of Rs 2,50,00,000/- and the 5 per cent rate with its ceiling.

Read the left column first and then the right one. In the left column the two instruments are indistinguishable, and that is exactly why an announcement can describe either and sound the same. In the right column they are not remotely the same thing. The grant has paid Rs 2,50,00,000/- for 2,000 machines, or Rs 12,500/- of public money per machine, a quarter of what the machine itself sells for. The incentive has paid Rs 50,00,000/- for the same 2,000 machines, and its cost per machine has not moved at all. A rate of Rs 2,500/- a machine is the whole instrument, so it has nothing to move. Five rupees of public money under the grant for every one under the incentive, on identical output.

The two instruments cost the same on paper and differ entirely in what happens when things go wrong, and that difference is the only comparison between them worth making. Anyone comparing them on the headline alone is comparing the one number on which they were built to agree.

What has to be true for an instrument like this to work?

Three things, and they are conditions rather than predictions. Stating them as conditions is the point: an instrument can then be tested without forecasting anything, and a design question does not turn into an argument about which side anybody is on.

First, the learning has to be real. Cost per machine has to actually fall because the work is being done, and not merely appear to fall because a payment arrived and was counted as income. The difference between a cost that really fell and a payment counted as income is testable, and it is the one most often skipped.

Second, the support has to end. An instrument with a stated end is making a claim about itself: that the thing being built will stand up on the far side of it. An instrument that quietly renews every time the end approaches has stopped being a way of building an industry and has become a way of running one.

Third, what was built has to survive the ending. Output that holds after the payments stop is evidence that something was built. Output that collapses the month the payments stop was rented rather than built, and the rent has now come due.

An instrument with no end date tests none of the three, so the exit is a design feature and not an afterthought. Without an end, condition two is unmet by construction, condition three can never be run, and condition one becomes almost impossible to separate from the payment itself. Notice that all three conditions can be checked long after the announcement and none of them can be checked from it. Being unable to settle any of the three on announcement day is the honest position for a reader to be in.

What goes wrong with industrial support, wherever it is tried?

Three failure modes are named often enough, in enough places, that any reader of an announcement should have them ready. None of them points at anything in particular, and the reason to learn them is that they are the questions the announcement will not raise about itself.

The first is capture. The payment goes to producers who were going to build the plant anyway, so public money buys output that already existed in somebody's plan. Nothing changes except who holds the cash. Capture is genuinely hard to detect. Proving it requires a counterfactualWhat would have happened if the thing under examination had not happened. It cannot be observed directly, only estimated, which is what makes claims about it arguable. and nobody can observe one.

The second is holding up what should close. Support keeps a producer running that would otherwise have stopped, and the workers, the machines and the money stay where they are instead of moving to something that can stand on its own. The cost here is not the payment. The cost is everything those resources would have done elsewhere, and it never appears in any account of the scheme.

The third is assembly without capability. Parts arrive finished, get fastened together locally, and the counting says the thing was made here. Output rises on paper and the ability to make the difficult part never moves, and moving it was the entire point of trying. Assembly without capability is especially awkward because the early stage of genuinely building capability looks identical to it, and only time separates the two.

The three failure modes are the standing risks of the instrument type, in the same way that a covenant has standing risks, and naming a standing risk is not an accusation against whoever carries it. A reader who knows all three, and knows that none of them can be settled from an announcement, is reading well.

How is one assessed without reaching a verdict?

There is a method, and running the method is not the same as announcing a result. Three questions, each of which needs evidence gathered years after the announcement, and each of which can come back either way.

Did output rise by more than the payment bought? If a scheme paid for a certain quantity and that exact quantity appeared, the payment and the output are one event described twice. Something happened only if the rise outran what was purchased.

Did the thing survive the support ending? The survival question cannot be asked until the support has actually ended. Patience is therefore part of the method, and a verdict delivered in year one is not a verdict.

Did capability move, or only assembly? Look for where the difficult part is made now against where it was made before. The capability question is the slowest of the three and the one that decides whether the exercise did what it set out to do.

Assessing any real scheme is a live question about real policy, and a live question of that kind is settled by evidence rather than by argument. The three questions above, and the standard of evidence each one needs, are durable, and they do not go stale the way a verdict would.

Where is the scheme itself set out?

At the issuer, and nowhere else. Everything specific about a real scheme lives in documents that are published, amended and republished: which goods it covers, what rate applies to what, who qualifies, how long the window runs, and what has been paid so far. Every one of those can change, and several of them usually have by the time any explanation of them is a year old.

What does an analyst check about a company that receives support of any kind?

One thing before everything else: what the margin looks like without it. A business that works only while the support runs is a different business from one that used the support to get somewhere, and the accounts of the two can look identical for as long as the payments continue.

Work it on Nirvaha, with invented figures. Suppose the promise is met: revenue of Rs 50,00,00,000/-, operating costs of Rs 48,50,00,000/-, so Rs 1,50,00,000/- of operating profit before any support, or an operating marginOperating profit as a share of revenue, before financing costs and tax. It is the standard way to compare how much of each rupee of sales survives the running of the business. of 3.00 per cent. Add the incentive of Rs 2,50,00,000/- and operating profit becomes Rs 4,00,00,000/-, a margin of 8.00 per cent. The year the support ends, on unchanged revenue and unchanged costs, that margin falls back to 3.00 per cent and the business earns Rs 2,50,00,000/- less than it did the year before while doing exactly the same amount of work.

The unsupported margin is what remains, so a lender or an analyst reads the supported margin and the unsupported margin as two different businesses and prices the second one. The practical checks follow from that: find the support in the accounts and see whether it sits in revenue or below it, find out when it ends, ask what the cost per unit has done over the period rather than what the profit has done, and work out whether the debt taken on to build the plant is serviceable at the unsupported margin. A household knows this shape already. A salary that includes a temporary allowance is not a salary, and nobody sensible takes a twenty year loan against the allowance.

Why is an announced ceiling not a forecast of spending?

There is one reading error worth naming here, and it is made by careful people reading quickly.

There is a second version of the same error, and it is larger. A reader sees the payment figure and takes it for the size of what will be built. On the Sankhya numbers, a payment of Rs 2,50,00,000/- sits against production of Rs 50,00,00,000/-, and 5 per cent of something is not the something. Reading the payment as the size of the plant is out by a factor of twenty here, and by whatever the rate happens to be everywhere else.

The Republic of Sankhya, Marut and Nirvaha Machine Works are invented.
Educational material. Not advice on any investment, tax, budget or market position.