Inflation and the Phillips curve · Lesson 11 of 12
The Phillips curve in the data: the US, and India
Ten lessons built the curve, moved it and priced it. This one looks at the numbers it was meant to describe. A scatter of inflation against unemployment, plotted for a real country, does not look like one downward line, and the reasons are the ideas of the earlier lessons. It is a reading lesson: it has no past-paper set, and it ends with a few checked exercises.
A question to hold on to
If the Phillips curve is true, a country that plots its inflation against its unemployment rate, year after year, should see a downward line. Take the United States from 1991 to 2024 and fit a line. Does it slope down, and is the slope the same in each decade?
A question from the data, not from a past paperSolved at the end ↓
The problem
A scatter is not the curve
The short-run Phillips curve of Lesson 7, π = πᵉ − β(u − uₙ), is one curve only while expected inflation, the natural rate and supply conditions stay put. Each year they do not. Expected inflation moves (Lesson 7), supply shocks shift the curve (Lesson 9) and the natural rate changes with the mark-up and bargaining (Lesson 8). A year’s dot lies on that year’s curve, and the curves differ. A scatter over many years mixes points from many curves, so a fitted line through it is not an estimate of one curve’s slope.
The answer to the opening question, from the table below: over all of 1991 to 2024 the fitted slope is −0.31, downward, but its standard error is 0.14 and R² is only 0.13, and the sign changes from decade to decade: 0.29 for the 1990s, −0.50 for 2001 to 2010, 0.14 for 2011 to 2019 and −0.90 for 2020 to 2024.
| Years | Points | Fitted slope | Standard error | R² |
|---|---|---|---|---|
| 1991 to 2000 | 10 | 0.29 | 0.18 | 0.24 |
| 2001 to 2010 | 10 | −0.50 | 0.15 | 0.59 |
| 2011 to 2019 | 9 | 0.14 | 0.16 | 0.09 |
| 2020 to 2024 | 5 | −0.90 | 0.57 | 0.46 |
| 1991 to 2024 | 34 | −0.31 | 0.14 | 0.13 |
United States, consumer price inflation on the modelled unemployment rate, one point a year. Source: World Bank, World Development Indicators, series FP.CPI.TOTL.ZG and SL.UEM.TOTL.ZS, read 2026-10-25. Computed from the stored series.
Interactive
Try it: the United States, decade by decade
Press through the periods. The thin line joins the dots in time order. Watch what the path does around 2009 and 2020 to 2022: it travels a long way from one cluster to another rather than sliding along one line.
One dot a year: consumer price inflation against the unemployment rate. The thin line joins the years in time order, and the red line is the straight line of best fit for the years you pick.
The fitted line
Inflation = 5.46 − 0.50 × unemployment, from 10 years. The slope has standard error 0.15 and R² is 0.59.
What the slope says
2001 to 2010: a downward slope, as the textbook curve says.
What to notice
Press through the periods. The fitted slope changes in size and in sign, and with only 10 dots it is often no larger than its standard error. A scatter of inflation against unemployment mixes points from different short-run curves: expected inflation, supply shocks and the natural rate all move between years. Source: World Bank, World Development Indicators, series FP.CPI.TOTL.ZG and SL.UEM.TOTL.ZS, read 2026-10-25.
Reading the loops
Why the slope jumps about
- Shifts of the curve. In 2009 unemployment reached 9.3% and inflation was −0.4%. In 2022 unemployment was 3.6% and inflation 8.0%. Low unemployment with high inflation and high unemployment with low inflation are both points on a curve, but not on the same one; the 2022 dot is consistent with a curve that has been shifted up, by a supply shock or by higher expected inflation, but a scatter cannot say which.
- Flat stretches. From 2011 to 2019 inflation stayed between 0.1% and 3.2% while unemployment fell from 8.9% to 3.7%. If inflation barely responds to unemployment, the slope β is small, or expectations were so firmly anchored that the curve moved little.
- Few dots. The 2020 to 2024 fit has five points. A standard error of 0.57 on a slope of −0.90 means five dots can show almost any slope.
The takeaway
The curve is a short-run relation between inflation and unemployment given expectations, the natural rate and supply conditions. To see it in data you must hold those fixed, or measure them, and a plain scatter does neither.
Pause and predict
In a scatter for one country over thirty years, the fitted slope of inflation on unemployment is close to zero. This shows
A second country
India: inflation moves, the series does not
Switch the lab to India. From 1991 to 2018 inflation in this data ranges from 3.3% to 13.9%, while the unemployment series stays between 7.5% and 7.7%. With an explanatory variable that hardly moves, a fitted slope has nothing to work with: look at its standard error in the lab. The series is a modelled estimate, so this says as much about the series as about India, and it should not be read as evidence for or against a trade-off.
What does move Indian inflation is covered earlier. Food is a large share of the consumer price basket (Lesson 1 gives the weights), and food prices are driven by harvests and supply conditions, which are supply shocks in the language of Lesson 9. The India Atlas decomposes a month’s inflation by item.
Limits
What this data cannot tell us
- The long history is missing. The unemployment series used here begins in 1991, so the 1960s trade-off, the 1970s breakdown and the early 1980s disinflation, the episodes the textbook curve is built on, are not in it. A source that allows its data to be stored would be needed to show them.
- Annual data hide the year. A quarter-by-quarter loop is more detailed than a dot a year.
- Inflation here is consumer prices only, not the GDP deflator or core inflation (Lesson 1).
- Expected inflation is not observed. The lab plots actual inflation, so it cannot separate a move along a curve from a shift of it.
Traps
What to remember when someone shows you a scatter
Watch out for
The mistakes examiners are fishing for
- One fitted line over many years is not one curve. It averages curves that shifted.
- Compare the slope with its standard error. A slope smaller than its own error cannot be told from zero.
- Few points, wide error. Five years can show almost any slope.
- A series that barely moves cannot show a response. The fault may lie in the series.
- Low inflation with high unemployment, and high inflation with low unemployment, are not a contradiction. They are different curves.
Revise
The lesson on one screen
The lesson on one screen
- A scatter
- Mixes points from different short-run curves, so its slope is not β.
- What shifts the curve
- Expected inflation (7), the natural rate (8), supply shocks (9).
- US data
- The fitted slope changes size and sign between decades, and is often within its standard error.
- India data
- Inflation swings widely while the modelled unemployment series barely moves: no curve to fit.
- Method
- Compare the slope with its standard error, count the points, and ask what was held fixed.
Practice
Check yourself
Quick questions
1A scatter of inflation against unemployment for one country over many years need not trace one Phillips curve because
2A fitted slope that is smaller than its standard error means
3Fitted over different decades of the US data, the slope of inflation on unemployment
4India's modelled unemployment series barely moves between 1991 and 2018. A fit of inflation on it
5The 1960s to 1980s history behind the textbook curve is not in this lesson's data because
Numerical problems from the data
A fitted slope
Use the lab’s US data for 2001 to 2010. To one decimal place, what is the fitted slope of inflation on unemployment (inflation points per point of unemployment)?
A fall in unemployment
US unemployment in the data was 9.633% in 2010 and 3.669% in 2019. By how many percentage points did it fall? (Two decimals.)
A fall in inflation
India’s consumer price inflation in the data was 11.989% in 2010 and 3.328% in 2017. By how many percentage points did it fall? (Two decimals.)
What next
Where to go next
This lesson has no past-paper set because the exams do not ask about data. The module’s practice sets are in the other lessons, and the last lesson, the Delhi School set, collects the hardest questions by technique.
Read next
All resources →MacroeconomicsInflation and the Phillips curve: the Delhi School problems on wage-setting, the natural rate and the NAIRUThe Delhi School entrance's seventeen wage-setting, price-setting and Phillips curve questions, worked by technique, with the Natural-Rate Workbench preloaded with each set-up.MacroeconomicsThe original Phillips curve: unemployment, wage and price inflation, and the trade-offWhat the curve shows and why it slopes down, the step from wage inflation to price inflation, the menu of choice, and the moves every question tests: along the curve, a shift, the intercept.MacroeconomicsInflation, unemployment and the Phillips curve: a module from price indices to the natural rateA module in twelve lessons: measuring inflation and unemployment, the causes and costs of inflation, the original and expectations-augmented Phillips curves, Okun's law, wage-setting and price-setting, supply shocks, disinflation, the curve in real data, and a lesson of Delhi School problems. With labs, worked questions and graded past papers.