Lennart Brandt, Natalie Burr and Krisztian Gado
The Bank of England has a 2% annual inflation rate target in the ONS’ consumer prices index. But looking at its 700 item categories, we find that very few prices ever change by 2%. In fact, on a month-on-month basis, only about one fifth of prices change at all. Instead, we observe what economists call ‘sticky prices’: the price of an item will remain fixed for an extended amount of time and then adjust in one large step. We document the time-varying nature of stickiness by looking at the share of price changes and their distribution in the UK microdata. We find a visible discontinuity in price-setting in the first quarter of 2022, which has only partially unwound.
Theory of sticky prices and related literature
Understanding price-setting dynamics is essential for central banks. Most structural models in the literature use a form of time-dependent pricing, under which firms keep prices the same for fixed amounts of time (Taylor (1980)), or for random amounts of time such that there is uncertainty about the precise length of the price spell (Calvo (1983)). Another way of modelling sticky prices emphasises that firms will not just look at the time that has passed since they last adjusted its price, but also at how far their price is from some desired price level. This is called state-dependent pricing. Macroeconomic models don’t typically allow for time-variation in the degree of stickiness or switching between pricing strategies. Recently, however, firms in the Decision Maker Panel tell us that they have moved increasingly away from time-dependent towards state-dependent pricing. In this case, when there is a large shock affecting many firms, the shock leads to an increased frequency of price changes and so more immediate pass-through to overall inflation.
In order to better understand the pricing behaviour of firms in times of large inflationary shocks, we explore the pricing dynamics at the micro level using CPI microdata published by the ONS. We are of course not the only ones who have been interested in this type of data. Bank authors have been using this data set for a number of years. For example, Bunn and Ellis (2011) document stylised facts about pricing behaviour from the UK microdata and the August 2020 Monetary Policy Report used CPI microdata to inform policy. Elsewhere, Karadi et al (2020) use US microdata to analyse firms’ price-setting in response to changes in credit conditions and monetary policy. Nakamura et al (2018) analyse the societal cost of high inflation using microdata from the 1970s and 1980s, and Montag and Villar (2023) analyse the effect of more frequent price-changes on aggregate inflation during Covid. Relatedly, Davies (2021) finds that the difference between the share of price rises and price cuts in the UK microdata is related to aggregate inflation, focusing on price-setting during the pandemic. And finally, authors of the FT’s Alphaville blog have also been looking into these data (see here and here).
The data
The microdata spanning from 1996 until September 2023 is publicly available and updated monthly after each CPI release. It contains the monthly price quote data underpinning the ONS’ CPI series for over 700 items with identifiers at the shop, shop type, and region levels. We clean the data which works out to about 30 million observations. When identifying a price change in the data, which is ultimately what matters for inflation, we try to be as precise as possible with regards to the product and the timing of the change. To that end we only count the change if we find the same item, in the same region, in the same shop, in two exactly neighbouring months. For example, if a bag of potatoes cost £2 in January and £3 in March but was not recorded in February, rather than imputing a price we discard the observation since we cannot be sure in which month the change actually happened.
Stylised facts from the microdata
A brief look at the data lets us establish some stylised facts. Chart 1 shows a decomposition of these month-on-month price movements over all items in the data set. Four key observations stand out:
- Prices rise and fall all the time, but the vast majority of prices do not change between months. In any given month, on average since 1996, around 80% of prices remain unchanged relative to the previous month (blue line).
- The share of prices rising (in green) has increased notably since 2021 to an extent that has not happened in previous inflationary episodes in the sample (excluding VAT changes).
- The share of prices falling (in red) has fallen somewhat but remains stable since 2021, relative to historical average. The main margin of adjustment has been in the share of price increases.
- But, in recent months, while the share of prices rising has tapered off, it remains elevated relative to its historical average.
Chart 1: Decomposition of price movements month-on-month
Notes: The share of prices rising and falling reflect month-on-month changes. Shares are seasonally adjusted using the R package seasonal. Spikes in 2008, 2010 and 2011 are a consequence of UK VAT changes (17.5% to 15% in 2008, increase to 17.5% in 2010 and increase to 20% in 2011). The grey shaded area covers the time between March 2020 and July 2021 when the economy (and data collection) was most affected by the Covid pandemic. Dashed lines show the 2011–19 averages. Latest observation: September 2023.
Sources: ONS and authors’ calculations.
To be clear, this chart is not saying that 80% of products never change prices. If the price of an item remained constant between December and January, and rose between January and February, it would move from the blue into the green category during this period. Similarly, it would fall out of the green, into the blue or red, if from February to March the price again remained constant, or fell, respectively.
So, perhaps surprisingly, this chart shows that monthly price dynamics in the economy are driven by only a relatively small fraction of roughly 20% of all goods and services in the consumption basket. Also, we see that in the most recent episode, the shift into rising prices has been mostly out of the ‘no change’ category. Hence, fewer prices are staying fixed, and more are rising. It is worth noting that the recent up-tick in the shares of prices rising is only matched historically by those caused by VAT changes in 2008, 2010 and 2011, which however appear as one-off price spikes rather than a persistently higher share of price rises, as in 2022.
If it is a minority of total products whose price changes, it is important to take a closer look. Chart 2 shows the distribution of prices changes from 2019 by quarter (truncated at zero to exclude no-change observations). In line with the rise in the green line in Chart 1, we observe that over 2021 and 2022 a lot of mass moved into the right side of the distribution, that is the share of price increases, with the share of price decreases being relatively stable.
Chart 2: Evolution of the distribution of price changes by quarter 2019–23
Notes: The share of prices that did not change is excluded from these densities. The truncated densities are estimated in R via the Bounded Density Estimation package using the boundary kernel estimator. Darker colours correspond to quarters in which year-on-year CPI inflation was relatively high, lighter colours to quarters in which it was low. Each distribution represents month-on-month changes within the same quarter. Latest observation: 2023 Q3.
Sources: ONS and authors’ calculations.
A note on the chart: the distribution of price changes, when aggregate inflation is at or close to target, is roughly symmetric in logarithms. On this scale, a doubling (+100%) is equally far away from zero as a halving of the price (-50%). Due to sales, the doubling and halving of prices actually happens regularly in the data, which explains the bunching around these points. While these may be a source of seasonality in the data, which may obscure the underlying dynamics, we do not believe they are important for the overall shape of the distribution which we show here.
In Chart 3, we zoom in on a couple of these densities to better see differences in their shape. They are the densities corresponding to price changes in the third quarter of 2022 and 2023 alongside an average density over the pre-Covid period.
Chart 3: Comparison of densities from 2022 and 2023 against a pre-Covid average
Notes: The share of prices that did not change is excluded from these densities. The truncated densities are estimated in R via the Bounded Density Estimation package using the boundary kernel estimator. To compare densities across time, they are normalised to sum to the average share of prices falling and rising respectively within the quarter. The yellow line shows the pointwise average density over the third quarters of the years 2011–19.
Sources: ONS and authors’ calculations.
We can see how, compared to this historical average – which we use as a stand-in for pricing behaviour when inflation was close to the 2% inflation target – 2022 saw a huge number of prices increase while there was little change in the behaviour of the lower part of the distribution. In the latest data, this mass of increases has begun to subside, and, at the same time, there is a growing number of prices outright falling on the month. However, the modal price increase (that is, the most probable) is still elevated at about 6%, compared to roughly 3% on average during 2011–19).
Conclusion
To summarise, looking at the micro level of price changes, we find a visible discontinuity in price-setting in the first quarter of 2022. A variety of factors, such as the large rise in energy prices in early 2022, as well as supply-chain issues following Covid lockdowns, likely contributed to this significant change in price-setting dynamics in the UK (relative to any recent historical precedent at least). At the micro level, firms’ pricing decisions led to the emergence of a large rebalancing in the distribution of price changes. Suddenly, more prices for many different products were rising at the same time. Compared to the available history for these data, the recent period is unique. More research will be needed on the causes of this marked shift in the distribution of price changes, both at a micro and at a macro level.
In the very latest data, there is some evidence that the distribution of price changes has indeed begun to return in the direction of its historical average, though it is too soon to establish a trend.
Lennart Brandt and Natalie Burr work in the Bank’s External MPC Unit, and Krisztian Gado is a PhD candidate at Brandeis University.
If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.
A great post – fantastic to see new & insightful work in this area
Informative analysis, thanks.
I have long thought that Calvo pricing was unrealistic, so I am not surprised to read that the decision makers use and increasingly use, state-dependent price adjustment. I cannot imagine that retailers hold their prices while their shelves are emptied as fast as they can be replenished, simply because a price review is not due! I believe that COMPASS uses Calvo pricing, which may to some extent explain why the BoE forecasts underestimated quite how high inflation would reach in the recent peak, and perhaps the overestimation of the slowdown in real economic activity too. In short, prices adjusted faster than the models anticipated, so volumes adjusted less.
As a former BoE seasonal adjuster with the mental scars to show for it, I was amused to see that it can be done in R these days.