How to Read CPI Inflation Data: Headline and Drivers
Read a price release in a fixed order: the vintage, the base, the headline, then the split between the volatile items and the rest, then the weights behind that split. The split comes before any conclusion. A headline is an average, and an average can be moved by a small corner of what it averages.
The Republic of Sankhya, an invented economy, publishes the price release read throughout. Every level, weight and rate below re-runs to the total printed beside it.
Build the headline out of the release's own weights and index levels.
Nine figures come off a price release, three for each group: the weight, the index level for the month being reported, and the index level for the same month one year earlier. Type those nine in and the panel builds the headline the way the release built it, group by group, and shows both totals that have to hold. The fields open on the invented Sankhya first print.
The nine defaults are the Sankhya first print, and they are worth having in plain text as well as in the fields. Food carries a 40.00 per cent weight and its index moved from 112.50 a year earlier to 126.00, a rate of 12.00 per cent and a contribution of 4.80 points. Energy carries 10.00 per cent and moved from 120.00 to 130.80, a rate of 9.00 per cent and a contribution of 0.90 points. The rest of the basket carries 50.00 per cent and moved from 150.00 to 153.00, a rate of 2.00 per cent and a contribution of 1.00 point. The weights add to 100.00 per cent, the contributions add to 6.70 points, and 6.70 points of contribution is a headline of 6.70 per cent. Nothing in that run is a judgement; it is nine figures read off a document and one multiplication each.
A consumer price index (CPI) and the basketThe fixed list of items an index prices, and the share of household spending each of those items is taken to represent. it prices are settled subjects, built earlier in this library. The order in which a release built on them gets read is not settled at all. A price release carries one figure that everybody quotes and a long table that almost nobody opens, and the difference between a careful reader and a careless one is hardly ever knowledge of inflation. The difference is which of those two things gets opened first. Every step below produces something written down, and the next step may use only what the step before it produced.
What can this panel be driven into?
Two readings go wrong on price releases often enough to have names, and both of them can be produced in the fields above rather than merely described. The first is the base effect. Leave all three of the current month's levels exactly where they are, at 126.00, 130.80 and 153.00, and change nothing but food's level for the same month a year earlier. At 112.50, the published base, the headline is 6.70 per cent. Move that one field to 120.00, an ordinary month in which nothing collapsed, and the headline is 3.90 per cent. Move it to 140.00, a month in which food prices had spiked, and the headline is minus 2.10 per cent. Not one of this month's prices moved, and the published rate travelled 8.80 points.
The second misreading is the one the failure block below is built around, and the last two buttons produce it. With food's levels at 112.50 and 135.00, energy's at 120.00 and 118.80, and the rest at 150.00 and 150.30, the headline is 8.00 per cent: food alone hands over 8.00 points, energy takes 0.10 of that back, and the other half of the basket adds 0.10, so the volatile pair carries 7.90 of the eight. The last button moves every group 8.00 per cent. The headline is 8.00 per cent again, and this time the volatile pair and the rest hand over 4.00 points each. Two runs, one headline of 8.00 per cent, and a 3.90 point gap in what the volatile pair did. Energy's rate in the first of those runs is minus 1.00 per cent: a group whose prices fell runs to the left of the zero line and its figure carries the word minus, and it is a fall rather than a fault.
India, for the institutions and nothing else
The National Statistical OfficeThe government body that compiles and publishes India's official statistics. Naming it here says what kind of body it is, nothing about what it has published. is the body that compiles India's official statistical releases, and the Index of Industrial Production and the Purchasing Managers' Index are two releases named elsewhere in this reading order for what they measure. A level, a base period, a release schedule or a target for any of them goes stale the moment it is typed. Read each one at its own source.
What does this order rest on?
Three things are already sitting on the table before step one begins. The first is the vintage, covered separately: the same release can carry three different values for the same quantity and none of them is wrong. The second is the reading order itself, built on an output release and reused here, unchanged in shape, on a price release. The third is the price index and its basket, covered earlier still.
Three settled subjects leave a genuinely narrow remainder, and the narrowness is the point. Other subjects are covered elsewhere: what inflation is, what an aggregateAny figure built by adding or averaging many smaller ones, such as total output or an economy wide price level, rather than measuring one thing directly. measures, or why prices move at all. Seven actions remain, in one order, each of which hands the next one a written output and nothing else. A reader who runs them out of order produces a sentence instead of a sheet, and the sentence is usually wrong in a way nobody can trace.
Step one: what vintage is this figure?
Write down which release the numbers come from, and attach that label to every figure copied off it afterwards.
The line on the release saying which print this one is: an early print, a revised one, or something later still, each carrying its own vintageThe label saying which release of a number this particular one is, since the same quantity gets published more than once. Why the versions differ is covered separately.. Nothing else on the document is read at this step. Step one is too small to get wrong.
A rate written into a note with no label beside it. Two months on there is no way left to tell which of two or three correct values it was, and the label cannot be reconstructed after the fact.
The vintage of the release, written beside every figure.
Why the versions differ at all is covered separately, and step one does not need it. Step one needs only the size of the thing being labelled. On the Sankhya output series the same quantity reads 6.20 per cent at first print, 6.50 per cent at first revision and 6.80 per cent at a later vintage, a spread of 0.60 points, and not one of the three is an error by anybody. Price series revise less than output series, and they still revise, so the label does not get to lapse just because the reading has moved from output to prices.
Step one is finished. What is written on the sheet?
Step two: what base is this rate sitting on?
Name the period this rate is measured against, and put one word beside it. Ordinary, or odd.
The base month's own index levels, in the column the panel above calls the same month a year earlier, and whatever the release says about conditions in that month.
The word left off, or chosen after the headline has been looked at. On the Sankhya release the base is the month in which food prices had collapsed. Food's base level sits at 112.50 in the panel above rather than near 120.00, so the word is odd. A sheet that says nothing lets everybody downstream treat the far end as a quiet month.
The base period named, plus that one word.
Why an unusual base bends the rate sitting on top of it is a separate subject, covered on its own. Step two needs only the size of it, and the panel above has already shown that size: this month's three levels held exactly still, food's base month moved, and the headline travelled from 6.70 per cent to minus 2.10 per cent. A base effectThe part of a growth rate that comes from what the earlier period did rather than from what the current one did. Covered separately in this reading order. caught at step two is a note; the same effect caught after a conclusion has been written is a retraction.
Step three: what is the headline, and in which unit?
Take one rate, state it in per cent, and write the base period beside it. Take nothing else.
The all-groups line of the release, the year-on-year column. On the invented Sankhya release that line reads 6.70 per cent, against the base flagged odd at step two.
Two rates written down instead of one. The moment there are two there is a distance between them, and the sheet has acquired a second quantity in a second unit without saying so.
One rate, in per cent, with the base period named beside it.
The unit slips exactly here, so the refusal to take a second rate is worth one worked line. Suppose the comparable earlier Sankhya headline, at the same vintage, was 5.20 per cent. The distance between 6.70 and 5.20 is 1.50 points, and the figure below shows what else the same pair of rates can be made to say. Per cent is the unit of a rate, and points is the unit of the distance between two rates, and mixing them changes the size of every claim built on them. The index levelThe raw number an index prints, such as one hundred and twelve, before anybody turns a pair of them into a rate of change. Levels are built elsewhere in this library. the rates were computed from stays in the panel above and never enters this step.
Step three hands over the headline. In what unit?
Step four: how does the headline split?
Cut the one rate from step three in two: what the volatile items put into it, and what everything else put into it. State both in points.
The group rows of the release, and the volatile grouping the basket already carries. On the Sankhya run those come out at 5.70 points and 1.00 point, exactly what the panel above prints for the volatile pair and the rest.
Two contributions that do not add back to the headline, or two contributions labelled per cent. A share of a rate is in points. Calling the 5.70 a per cent figure makes it look like a rate of its own, and it is not one.
Two contributions, each in points, adding to the headline exactly.
Which items count as volatile is not decided here. The volatile grouping was built when the basket was built, and this step takes it as given and takes nothing else from it. The step is a cut, not an argument. The headline is an average, and an average says nothing about which of its parts moved, so the cut happens before any conclusion and not after. A household knows this instinctively. If the monthly grocery bill jumps, nobody says the bill went up and stops there. Somebody says the onions went up, and then the conversation is about onions rather than about groceries. Step four is that sentence, written down in points.
Look at what the cut reveals on this release. Food and energy hold half the Sankhya basket between them and account for 5.70 of the 6.70 points. The other half, the coreThe part of the basket left over once the items that jump around most are set aside. The contents of the core, and the use it is put to, are built earlier in this library. of it, accounts for 1.00 point. Half the basket did close to six times as much work as the other half, and the headline said none of that.
Step four hands over two contributions. In what unit are they stated?
Why does the split come before any conclusion rather than after one?
Step five: what weights sit behind those contributions?
Print the weight each group carries beside its contribution. Anybody holding the sheet can then rebuild the contribution from scratch.
The weighting diagram issued with the release, the same document the panel above takes its three weight fields from. One extra column on the sheet, no more.
Contributions printed with no weights beside them. A large move on a small weight and a small move on a large weight arrive at the sheet looking identical, so nobody can audit step four.
The weight beside each contribution, in per cent.
A contribution is a move times a weight, and that single fact is what makes the weight column worth a step of its own. It shows plainly on the Sankhya release. Energy carries a 10.00 per cent weight and moved 9.00 per cent, so the contribution is 0.90 points. The rest of the basket carries a 50.00 per cent weight and moved 2.00 per cent, so the contribution is 1.00 point. The two moves are 7.00 points of move apart. The two contributions are 0.10 points apart, and the smaller move has the larger contribution. Had energy moved 10.00 per cent instead of 9.00, the two contributions would have been identical at 1.00 point each, from moves five times apart. The everyday version is a household budget: the tomato price doubling is loud and small, the rent creeping up is quiet and large, and the monthly total cannot say which of them happened.
A contribution is a move times what?
Step six: are these two headlines even comparable?
Mark every comparison already on the sheet like for like, or strike it out. Do not qualify one and leave it standing.
Two things, and the second is the one that slips through. The vintage of each figure, already recorded at step one, and the basket each figure was built on.
A footnote saying the baskets differ, with the two numbers still sitting side by side. Side by side is how they get quoted, and the footnote does not travel with them.
Every comparison on the sheet marked, or refused outright.
The weights that turned the price moves into a headline are not the same weights after a basket revision, so two headlines are incomparable even at the same vintage. Hold every price move on the Sankhya release fixed and change nothing but the weights, and the headline reads 6.10 per cent instead of 6.70 per cent. The published figure moves 0.60 points with not one price behaving differently. The output of this step is blunt for that reason, rather than careful.
Two headlines are at the same vintage but were built on different baskets. Why does step six refuse the comparison?
Step seven: what would change this reading?
Name two or three developments that would make this sheet read differently, and write them down before any of them happens.
The outputs of the six steps above and nothing outside them. Each condition names a figure already on the sheet.
A condition phrased as a mood rather than a figure, or one written after the event it describes. Conditions written in advance are a test; the same conditions written afterwards are an excuse.
Two or three named conditions, and with them the sheet is finished.
On the Sankhya release there are three. First: a revised headline with the old contributions under it adds to nothing, so if the vintage changes the split is recomputed from the new figures rather than carried forward. Second: if the volatile contribution of 5.70 points drops towards the 1.00 point the rest of the basket produced, the two halves have swapped roles, and a sheet describing a food and energy event no longer describes this one. Third: if the weights are revised, every comparison mark on the sheet is erased and step six runs again from the beginning. Three lines, and they are what make the sheet re-readable by somebody who was not there when it was built, including the person who built it.
Where does the reading stop?
The reading stops when the seven outputs exist, and the figure below is the whole product. The seven steps never collect the material a statement about whether prices are rising too fast, or about what happens next, would need. No eighth step supplies it either.
The reading is complete once the outputs are written down, and not once a view has arrived. This feels unfinished the first few times, and that feeling is the order working. A sheet that stops at seven can be handed to somebody who disagrees with it, and that reader can find the exact line where the two accounts part company. A sheet that ends in a view cannot be taken apart that way. The view has already absorbed the parts.
How does an analyst run this in practice?
Somebody who reads price releases for a living does something that looks like cheating: they open the contributions before the headline. Opening the contributions first is not a shortcut past the order. Every step still runs and every output still gets written, and what changes is only where the attention goes. The headline is the one figure everybody in the room already has, so it distinguishes nothing; the contributions are what separate two readings that share it.
Watch how that plays out. Three people walk in holding the same 6.70 per cent. The first has only the headline and can add nothing to what is already on every screen. The second has the split, and can say that 5.70 of those points came out of half the basket while the other half produced 1.00 point. The third has the split, the weights and a comparison mark, and can say which earlier figure this one may be set against and which it may not. The consensusThe average of what forecasters expected a release to show, gathered before it lands. Measuring a print against it is covered separately in this reading order. everyone carried beforehand was built against the headline alone, so it cannot settle any of that.
A household, a lender and an analyst are each standing in one of those three places. The household reading only the headline hears that prices rose 6.70 per cent and cannot plan against it. The household holding the split knows that 5.70 of those points came from food and energy. One bill has been pushed up by a corner of the basket, the other by all of it. The lender pricing a loan and the analyst writing a note need that same line on a longer horizon. The decision each of them then takes sits outside the seven steps.
What does the whole order look like on one release?
Here is the Sankhya release run end to end. Step one: the vintage is the first print. Step two: the base is the period in which food prices had collapsed, and the word beside it is odd. Step three: the headline is 6.70 per cent against that base. Steps four and five are the table below, and every weight is printed so that any total in it can be rebuilt rather than trusted.
| Group | Weight, per cent | Move, per cent | Contribution, points |
|---|---|---|---|
| Food | 40.00 | 12.00 | 4.80 |
| Energy | 10.00 | 9.00 | 0.90 |
| The volatile items together | 50.00 | a move each, not an average | 5.70 |
| The rest of the basket | 50.00 | 2.00 | 1.00 |
| Everything else together | 50.00 | a move each, not an average | 1.00 |
| The headline, in per cent | 100.00 | 6.70 |
Step six then marks the comparisons. The earlier 5.20 per cent Sankhya headline is at the same vintage and on the same basket, so it is marked like for like, and the gap between them is 1.50 points. The 6.10 per cent figure computed on the revised weights sits on a different basket, so it is struck out. Step seven writes the three conditions. The sheet is finished, and it contains no view.
The same 6.70 per cent can be reached a second way. Hold the basket at 40.00, 10.00 and 50.00 per cent and change only the moves, to 1.00, 3.00 and 12.00 per cent. The contributions become 0.40, 0.30 and 6.00 points, so the volatile pair hands over 0.70 points against 6.00 from the rest, and the headline is 6.70 per cent again. Two sheets, one headline, and exactly 5.00 points of difference in what the volatile items did. Everything past step three disagrees.
Set every move and every weight, and watch the contributions rebuild the headline.
The second panel takes moves rather than levels, and taking moves is what lets the weights be pulled about as well. The sliders open on the same published run. The second button reaches 6.70 per cent from a different picture. The third holds every price move still and revises the weights instead, and that is step six in motion. Pulling food's move down and pushing the remaining half of the basket up shows how far the headline can stay still while the picture underneath it turns inside out.
Two readings both report a headline of 6.70 per cent. What can still differ between them?
What goes wrong when the headline is read alone?
A number reported, a picture invented
The failure has one move in it. A reader takes the 6.70 per cent off the release, writes that prices are rising faster than before, and never opens the table. Nothing in that sentence is arithmetically wrong. The trouble is that it is equally true of two releases that have almost nothing in common.
In the first, the volatile items contribute 2.60 of the 6.70 points while sitting on just 15.00 per cent of the basket. Vegetables alone moved 22.00 per cent. In the second, the volatile pair contributes 0.75 points, and the three larger groups all move 7.00 per cent, contributing 5.95 points between them. The first is a narrow, loud move on a small slice. The second is the whole basket moving together. A reader with only the headline cannot tell them apart, and neither can anybody they pass the headline to.
The fix is not more caution, it is step four: split the headline into contributions and print the weights before writing a single sentence about it. The split takes one line of arithmetic per item and it is the difference between reporting a number and describing what happened.
The fix has a limit. The fix is not deciding that the volatile items should be excluded, or that the second picture is the more serious one. Exclusion and seriousness are conclusions, and the order does not reach them. The order only insists that the two pictures stop looking identical.
A reader takes the 6.70 per cent headline and writes that prices are rising faster than before. What has gone wrong?
What must never be a step in this order?
An order is defined as much by what it refuses to contain as by the seven things it does. Three moves look like steps, get inserted by readers under time pressure, and are not steps here at any point.
| Never | Why it is not a step | What the order puts there instead |
|---|---|---|
| Pick the base month that flatters the reading | The base month is not a choice. It is the same month one year earlier, fixed by the release before anybody opened it. Choosing among candidate base months until the rate reads well is choosing the answer, and the panel above shows the size of the prize: the same three current levels give 6.70, 3.90 or minus 2.10 per cent depending only on which month sits at the far end. | Step two names the base the release used and flags it in one word, and the flag travels with the rate whether it flatters or not. |
| Quote a rate without the base it is measured from | A rate with no far end is not a smaller claim than a rate with one. It is a different quantity, and nobody downstream can tell which. Once the base is gone it cannot be recovered from the rate. | Step three writes the base beside the rate on the same line, so the two travel together or neither travels. |
| Carry forward a comparison step six struck out | A struck comparison that reappears further down the sheet, softened into a phrase such as broadly higher than last year, is the same incomparable pair with the mark removed. | Step six strikes it out rather than qualifying it, and step seven names the revision that would let it be recomputed. |
All three cost nothing at the time and are unrecoverable afterwards, and prohibition rather than counsel is the only form that catches them. Each of them produces a sheet that looks finished, and none of them can be detected by a reader who was handed only the result.
What lies outside this reading order?
Inflation itself, and how a price index and its basket are constructed, are built earlier in this library. Why figures revise is covered separately, and how an unusual base bends the rate sitting on top of it is covered separately as well.
Every rate, weight, contribution and comparison above belongs to the Sankhya release and is not a figure for any actual economy. The order stops at seven outputs. Whether prices are rising too fast, whether any reading is good or bad, and what the next release will show all lie outside it.
References and where to read the real thing
The two bodies below are named for what they publish.
| Source | Document | Site |
|---|---|---|
| Ministry of Statistics and Programme Implementation | Consumer Price Index release material, including its note on how item weights are set | mospi.gov.in |
| Reserve Bank of India | Database on the Indian Economy, the price statistics tables | rbi.org.in |
The Republic of Sankhya and its statistical office are invented.
Educational material. Not advice on any investment, tax, budget or market position.