Josh Martin and Julian Reynolds

How much have higher import prices increased consumer prices in the UK and euro area? This post explores this question using a framework grounded in some fundamental economic and national accounting concepts. Starting with the GDP price, we adjust for relative import and export prices to arrive at a consumer prices measure – this gives us a sense of the impact of import prices and the terms of trade shock on consumer price inflation. For the euro area, aggregating imports across member countries, which includes trade between members, risks overstating total imports and thus the effect on inflation. Using supplementary data to resolve this issue, we find that the euro area terms of trade shock has been larger than the UK’s.

Why does the terms of trade matter for inflation?

Much commentary has attributed the recent inflationary impulse to the impact of the ‘terms of trade’ shock (eg Lane (2022)). This is defined as the ratio of a country’s export prices to its import prices. If the price of imports increases relative to exports, this makes it harder for an economy to pay for its imports through its exports revenue, and its terms of trade deteriorate. As net fuel importers, the UK and euro area have experienced an adverse terms of trade shock due to the surge in global energy prices, following Russia’s invasion of Ukraine (Dhingra (2023)).

How does this shock affect inflation? The National Accounts framework can help.

Our variable of interest – consumer prices – can be thought of as a weighted average of domestically produced products for domestic consumption, and imported products for domestic consumption. Clearly import prices are relevant for consumer prices, as some of domestic consumption is from imports. If import prices go up by more than the price of domestically-produced products for domestic consumption, then it will push up on consumer price inflation.

The price of all domestically produced products is the GDP deflator, but not all of GDP is for domestic consumption. To get from the GDP price to consumer prices, we need to subtract the effects of things that are in GDP but are not part of the consumer price basket, especially export prices. By subtracting the effect of export prices on the GDP price, just the relevant bit of the GDP price is left.

More formally, we can relate consumer prices (Pc) and the GDP deflator (Pv) using the methodology outlined by Haskel (2023):

• We start with the usual GDP identity. GDP (V) equals the sum of consumption (C), investment (I), government expenditure (G) and net trade (X-M). For sake of argument, we can ignore I and G (they make relatively little difference here).
• It follows that: Pv=aPc+bPx-cPm, where Pm is the price of imports, and Px is the price of exports, and where a, b and c are the shares of consumption, exports and imports in GDP respectively.
• Rearranging and dividing through by a, gives: Pc=Pv+(c/a)(Pm-Pv)-(b/a)(Px-Pv), where (c/a) and (b/a) are the shares of imports and exports in consumption, rather than in GDP.

In words, consumer prices are GDP prices, plus share-weighted import prices relative to the GDP price, minus share-weighted export prices relative to the GDP price. We’ll also account for the small difference between the National Accounts consumption price and the CPI (HICP for the euro area).

What has happened to import prices in the UK and euro area?

Using the approach outlined above, we can quantify the impact of the terms of trade shock on UK and euro area inflation, starting with import prices. The aggregate price of imports is measured by import price indices from the National Accounts, compiled by the ONS for the UK, and by each country’s national statistical institute for euro-area countries, collated by Eurostat.

The annual change in import prices in the UK peaked in 2022 Q3 at 22.2%, while in the euro area it peaked at 20.5% in 2022 Q2. In total, between the end of 2019 and the end of 2022, import prices rose by 26.0% in the UK and 25.0% in the euro area. These are not very different.

How important are imports in the UK and euro area?

What does appear different, from official data directly, is the importance of imports relative to GDP. In the euro area, imports apparently account for about 50% of GDP typically, and are about 90% of the size of total household consumption expenditure. By contrast in the UK, imports are about 30% the size of GDP, and 50% the size of consumption. This suggests the euro area is almost twice as import intensive as the UK. That doesn’t chime with expectations for an open economy like the UK.

Indeed, the data are misleading us here. When Eurostat aggregate the national accounts data for the euro area, they simply add up the values for each country. That means they add up Germany’s GDP, imports and exports, with France’s GDP, imports and exports, and so on. But of course, Germany’s imports could be France’s exports, and Germany’s exports could be France’s imports. Viewed as individual countries this makes sense. But as an aggregate, including within-block trade becomes less useful.

Using additional data from Eurostat, we can net out this intra-euro-area trade. Since every intra-euro-area trade flow is one member country’s import and another’s export, total imports and exports are reduced by the same amount (GDP is unaffected).

What does that do for the euro-area import share? Chart 1 shows the import share of GDP and consumption for the UK and euro area, with and without adjustment for the intra-euro-area trade. After adjustment, these shares look much like the UK – about 30% of GDP and 50% of consumption.

Chart 1: Import share of GDP and consumption, UK and euro area, 2019

Sources: Eurostat, ONS and authors’ calculations.

What has happened to export prices?

Starting from the GDP price, we need to add share-weighted relative import prices, and deduct share-weighted relative export prices, to get to a measure relevant of consumer prices. Higher export prices do not directly affect consumer prices. But if export prices increase by more than the GDP price, then less of the GDP price inflation is relevant for domestic consumer prices.

UK export prices increased 22.0% in total between the end of 2019 and the end of 2022. The euro area also saw an increase in export prices, although by somewhat less than the UK. One explanation for this difference could be that the UK exports some energy products, such as North Sea oil, and has therefore benefited somewhat from higher energy (export) prices. Such export revenue would only directly benefit some firms and workers, and so the typical UK citizen might be experiencing a larger terms of trade shock than the aggregate figures suggest.

Bringing it all together

Table A shows a decomposition of consumer price inflation into the GDP deflator, and the terms of trade effect, for the UK and euro area, following the method in Haskel (2023). Charts 2 and 3 show a time series of these effects. Consumer prices have increased by more than the GDP price in both the UK and euro area, reflecting higher relative import prices. The size of this effect is the price of imports relative to the change in the GDP deflator, weighted by the imports share of consumption (as in Chart 1). This pushed up UK CPI inflation by 6.3 percentage points, and euro-area HICP inflation by 7.7 percentage points.

However, since relative export prices also increased, we need to adjust for that. This offsets 4.0 percentage points of the GDP price inflation in the UK and 3.5 percentage points in the euro area, in terms of the contribution to consumer price inflation. The terms of trade effect is the difference between the import and export price adjustments, and adds to inflation for both economies.

Table A: Contributions to consumer price inflation, cumulative from 2019 Q4 to 2022 Q4

Notes: Terms of trade = import price adjustment plus export price adjustment. Consumption price inflation (CPI in UK, HICP in euro area) = GDP deflator plus export price adjustment plus import price adjustment plus other. Other accounts for difference between National Accounts consumption deflator and CPI/HICP, and other small factors – see Haskel (2023) for details.

Sources: Eurostat, ONS and authors’ calculations.

It turns out that the intra-euro-area trade adjustment is neutral for the terms of trade effect. Since every intra-euro-area import is also an intra-euro-area export, then conceptually these are entirely netted off and there is no change in the net trade position. In practice, there can be differences due to inconsistencies in reporting imports and exports across euro-area countries, but these are typically small. This is shown by the hashed bars in Chart 2, which are roughly offsetting.

Chart 2: Trade price contribution to annual euro-area HICP inflation, 2019 Q4 to 2022 Q4

Notes: Method as in Haskel (2023), and described in this blog.

Sources: Eurostat and authors’ calculations.

Chart 3: Trade price contribution to annual UK CPI inflation, 2019 Q4 to 2022 Q4

Notes: Method as in Haskel (2023), and described in this blog.

Sources: ONS and authors’ calculations.

Conclusion

The approach set out in our blog helps us understand the impact of higher relative import prices on UK and euro-area inflation. Looking at import prices alone is inadequate because the inclusion of intra-euro-area trade flows in national accounts data distorts upwards the contribution of import prices to inflation in the euro area. Working in the national accounts framework, we also need to account for the effect of export prices. Adjusting for these factors, we still find that the import price shock has been larger in the euro area than the UK, but our estimate of the magnitude of the shock is more in line.

Josh Martin works in the Bank’s External MPC Unit and Julian Reynolds works in the Bank’s Global Analysis Division.

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.

Yuliya Baranova, Eleanor Holbrook, David MacDonald, William Rawstorne, Nicholas Vause and Georgia Waddington

The functioning of major government bond and related repo markets has deteriorated on several occasions in recent years as trading demand has overwhelmed dealers’ intermediation capacity. Seeking a remedy, Duffie (2020) proposes a study of the costs and benefits of a clearing mandate in these markets. Such a policy could boost dealers’ intermediation capacity by allowing more of their trades to be netted, thereby reducing their balance sheet exposures and capital requirements. In a recent staff working paper, we estimate the effects of comprehensive central clearing of cash gilt and gilt repo trades on UK dealer balance sheets during one particular stress episode. This post summarises those quantitative results and discusses qualitatively other costs and benefits of broader central clearing.

Background

Trading demand has overwhelmed dealer intermediation capacity in recent episodes such as the September 2019 strains in the US Treasury repo market, the March 2020 global ‘dash for cash’ (DFC) and the September-October 2022 stress in the UK gilt market. On each occasion, markets became highly illiquid. For example, Table A shows the extent to which liquidity deteriorated in UK and US government bond markets during the DFC episode.

Table A: Measures of deterioration in government bond market liquidity in March 2020

Moreover, with government bond markets growing faster than dealer balance sheets, there is a risk that such episodes could recur or even become more frequent in the future. Hence, policymakers are responding on three fronts. First, they are developing measures to dampen surges in demand for market liquidity. Second, they are considering policies to enhance market intermediation capacity. Third, they are evaluating the most effective ways for central banks to provide backstop support to core markets in extreme stresses.

Central clearing

Consideration of broader central clearing falls into the second set of policy responses. When a contract is centrally cleared, a central counterparty (CCP) becomes both the seller to the buyer and the buyer to the seller. As a result, the original buyer and seller no longer face each other as counterparties, but rather face a CCP, which is by design an extremely robust counterparty. Moreover, as illustrated in Chart 1, comprehensive central clearing would mean that instead of having exposures to multiple trading counterparties, dealers would only have exposure to a single counterparty: the CCP. Hence, buy and sell trades pending settlement or lending and borrowing via repo could be netted, reducing balance sheet exposures and capital requirements.

Chart 1: Effect of clearing arrangements on counterparty exposures

As shown in Table B, rates of central clearing vary among major government bond and related repo markets at present. In the United Kingdom, there is essentially no central clearing of cash gilts, while around one third of gilt repo trades are centrally cleared. Clearing rates are somewhat similar in the United States and Germany, while Japan sets a precedent for very high rates of central clearing in a major advanced economy.

Table B: Estimated share of centrally cleared trades in government bond markets

Netting benefits for gilt repo trades

Repo intermediation involves dealers borrowing cash against collateral from one counterparty and lending it against other collateral to other counterparties. Unless these trades can be netted, however, they increase dealers’ total exposures and hence capital requirements, as measured by the Basel III Leverage Exposure Measure (LEM) and Leverage Ratio (LR) respectively. Potentially, these capital requirements could constrain the volume of repo intermediation that dealers are able to provide.

The Leverage Ratio framework allows dealers to net exposures arising from repo and reverse repo trades if they: (i) have the same counterparty, (ii) have the same maturity date and (iii) are governed by a common legally enforceable netting agreement. Trades meeting these conditions would not increase LEMs (and hence LR capital requirements). Thus, to estimate the contribution of gilt repo and reverse repo trades to UK dealers’ exposures, we gather data on these trades from the Bank of England’s Sterling Money Markets Daily (SMMD) data set and net outstanding trades if conditions (i) and (ii) are satisfied, assuming condition (iii) is always satisfied. This gives us estimates of exposures under the status quo.

Then, to estimate counterfactual exposures under comprehensive central clearing, we substitute the recorded counterparties in the data set for a single CCP and repeat our netting calculations. We also estimate exposures in a second counterfactual in which repo maturity dates (apart from overnight repos) are additionally standardised to a common day of the week, making condition (ii) more likely to apply. We construct these estimates around the DFC period. The results are shown in Chart 2.

Chart 2: Contribution of gilt repo trades to total exposures of UK dealers during the DFC period

As shown by the total size of the orange bars, gilt repo activity accounted for about 200 basis points of UK dealers’ LEM in aggregate. Comprehensive central clearing would have reduced those exposures by around 40% (top parts of orange bars). If, in addition, repo maturity dates had been standardised, this reduction would have increased to around 55% (sum of top two parts of orange bars). Within the aggregate, however, there is significant variation: for some dealers, comprehensive central clearing and maturity standardisation would have eliminated the majority of exposures, while for others it would have eliminated only a small minority (sum of top two parts of aqua bars).

While the proportionate reduction in aggregate gilt-repo exposures from comprehensive central clearing and maturity standardisation is significant, it would only have boosted the leverage ratios (Tier 1 capital divided by LEMs) of UK dealers by 6 basis points on average. For comparison, the same dealers had an average of around 190 basis points of leverage-ratio headroom over regulatory requirements before the DFC. That suggests that clearing and maturity standardisation policies may not have alleviated any firm-wide leverage-ratio constraints during the DFC. However, they may have eased internal constraints on repo desks, especially on the assumption that dealers maintained their internal capital allocations during our counterfactual experiment.

Netting benefits for cash gilt trades

For cash trades, the Basel III leverage ratio has been harmonised since the beginning of 2023 to allow exposures relating to sales and purchases of financial assets to be netted across counterparties for all dealers regardless of their accounting practice. As exposures can now always be netted regardless of counterparty, this means the leverage ratio already allows for netting equivalent to that under comprehensive central clearing. So expanding central clearing would not bring any further netting benefits in cash markets.

At the time of the DFC, however, dealers using trade-date accounting under International Financial Reporting Standards (IFRS) could only net the payables and receivables arising from their purchases and sales with common counterparties. Hence, our counterfactual analysis of replacing trading counterparties with a single counterparty, which we apply to cash trades from the period reported under MiFID II regulations, illustrates the benefits that the Basel III modification could bring. These are shown in Chart 3.

Chart 3: Contribution of cash gilt trades to total exposures of UK dealers using trade-date accounting under IFRS during the DFC period

Netting across all counterparties would have eliminated four fifths of UK dealers’ exposures arising from cash gilt trades in the approach to the DFC and two thirds at the peak of it for dealers that were using accounting practices that did not allow for such netting. Although larger in proportionate terms, these exposure reductions due to comprehensive central clearing are smaller in absolute terms than for gilt repo trades. Hence, as previously, this suggests that the balance sheet effects of a central clearing mandate may not have been large enough to have alleviated leverage-ratio constraints at firm level, though they may still have been material to government bond trading desks.

Policy implications

Our analysis suggests that in the gilt-repo market broader central clearing could materially enhance dealer intermediation capacity. In the cash-gilt market, in contrast, recent changes to Basel III have already brought about a boost to the intermediation capacity of certain dealers that broader central clearing might otherwise have delivered. Besides dealer intermediation capacity, however, there are additional potential benefits and costs to broader central clearing to consider.

In terms of potential benefits, central clearing could facilitate more all-to-all trading – where market participants can trade directly with each other, rather than through a dealer – since the buyers and sellers would then hardly need to concern themselves with the credit risk of their counterparty. Of course, such trading does not require any dealer intermediation capacity at all. In addition, reducing the volume of exposures to be settled would reduce the risk of settlement fails, which could have a positive impact on market functioning and market efficiency.

In terms of potential costs, the main item is the additional counterparty risk that CCPs would have to manage. This would arise if market participants were to access CCPs directly or indirectly through a sponsor but while maintaining ultimate responsibility for settlement of their trades. Moreover, these market participants could include different types of non-bank institutions, which CCPs often have not dealt with to date. In addition, more-widespread central clearing could increase liquidity strains on market participants at times of stress to the extent that CCP initial margin requirements react more to volatility than non-CCP requirements and because CCPs require variation margin calls to be settled in cash.

These other benefits and costs of broader central clearing need to be balanced against the potential quantitative effects on dealer intermediation capacity that we have documented in this post.

Yuliya Baranova, Nicholas Vause and Georgia Waddington work in the Bank’s Capital Markets Division, Eleanor Holbrook and William Rawstorne work in the Bank’s Market Intelligence and Analysis Division and David MacDonald works in the Bank’s Post-Trade Policy Division.

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.

Álvaro Fernández-Gallardo, Simon Lloyd and Ed Manuel

Since the 2007–09 Global Financial Crisis, central banks have developed a range of macroprudential policies (‘macropru’) to address fault lines in the financial system. A key aim of macropru is to reduce ‘left-tail risks‘ – ie, minimise the probability and severity of future economic crises. However, building this resilience could influence other parts of the GDP-growth distribution and so may not always be costless. In our Working Paper, we gauge these potential costs and benefits by estimating the effects of macropru on the entire GDP-growth distribution, and explore its transmission channels. We find that macropru is effective at reducing the variance of GDP growth, and that it does so by reducing the probability and severity of excessive credit booms.

Measuring macroprudential policy changes

To estimate the effects of macropru, we first obtain a summary measure of policy actions. Unlike for monetary policy, there is no single macropru policy tool, or simple measure of the overall change in policy stance. So we construct a macropru policy index using the MacroPrudential Policies Evaluation Database (MaPPED). The database covers 480 policy actions taken between 1990 Q1 and 2017 Q4 for 12 advanced European economies, including the UK. The actions captured include bank-capital requirements, housing tools and risk weights.

Relative to other databases, such as the IMF’s Integrated Macroprudential Policy (iMaPP) database and the International Banking Research Network’s prudential policy database, MaPPED has several advantages for our purposes. In particular, the survey designed for MaPPED ensures that policy tools and actions are reported in the same manner across countries, allowing for cross-country comparability. Furthermore, MaPPED includes a wealth of information on each policy action, including announcement and enforcement dates, stance (loosening, tightening, or ambiguous), and whether it has a countercyclical design – which is crucial for our identification.

To construct our index, we follow the approach prevalent in the existing literature. Using the announcement date of each policy, we assign a value to each action, giving a positive value to tightening actions and a negative value to loosening actions. We assign different weights to different policy actions based on importance. Under this widely used weighting scheme, the first activation of each policy are given the highest weights. Changes to pre-existing polices are given lower weight.

The resulting index can be interpreted as a composite measure of the overall macropru policy in each of the selected advanced economies. We plot our macroprudential policy index at quarterly frequency over time for each country in the sample in Chart 1. The index displays significant heterogeneity across countries, reflecting the fact that different countries have chosen to tighten or loosen macropru to different extents over time.

Chart 1: Macroprudential policy indices by country

Identification: from correlation to causation

Armed with this macropru index in each country, we then address a second key challenge: identifying the causal effect of macropru on macroeconomic variables. In any statistical exercise, it is well-known that correlations between variables in the data do not necessarily capture causal relations: correlation is not causation. This issue is particularly pertinent in our setting, since macropru policy makers may respond to conditions in the macroeconomy.

Consider the following example. Suppose that a ‘tightening’ in macropru is effective at reducing financial-stability risks. But then suppose that policymakers only tighten macropru when they see financial stability risks rising. This could in turn mean that macropru is uncorrelated with measures of financial stability, since tighter macropru simply serves to offset any potential rise in financial stability risks. But this lack of correlation does not imply macropru has no causal effect – rather it would be evidence that macropru is an effective stabilisation tool.

To sidestep this issue, we use a ‘narrative identification’ approach. In particular, we use the fact that our data set includes a rich set of information on each macropru action – including whether policies were implemented specifically in response to changes in macroeconomic conditions. We strip out any policy that is implemented in response to the economic cycle, as this would run into the issue described above – labelling the remaining subset of macropru changes as macropru ‘shocks’.

To ensure our approach is ‘doubly robust’ we also control for a variety of variables that capture the state of the macroeconomy at the time macroprudential policies were implemented. This allows us to compare outcomes for different time periods and countries where macropru was set at different levels, despite underlying macroeconomic conditions being identical. Finally, we show that our results are robust to controlling for anticipation effects.

Three conclusions about the effects and transmission of macropru in the tails

Having dealt with identification issues, we then estimate the relationship between our macropru shocks and the entire distribution of the GDP distribution for all 12 countries in Chart 1 from 1990 to 2017. Like other studies, we rely on ‘quantile regression’, a statistical tool, to estimate this relationship. We regress GDP growth on our narrative macropru shocks as well as a range of macroeconomic control variables.

Our first finding is that tighter macropru significantly boosts the left tail of future GDP growth (reducing the probability and severity of low-GDP outturns, ie 1-in-10 ‘bad’ outcomes), while simultaneously reducing the right tail of GDP growth (reduces the probability of high-GDP outturns, ie 1-in-10 ‘good’ outcomes). Together, these effects serve to reduce the variance of future growth – making future GDP outcomes less extreme. Chart 2 demonstrates this visually, showing the distribution of future GDP growth in ‘normal’ times (blue), compared to a situation where policymakers tighten macropru (red). The effects on median growth (near the centre of the distribution) are muted, and generally insignificant. This suggests that tightenings in macropru to-date have not come at significant costs via restricting (mediN) GDP-growth.

Chart 2: Effect of macropru on GDP-growth distribution

Notes: Blue line shows distribution of 4-year-ahead GDP growth when all controls set to cross-country and cross-time average values, and macropru index is 0. Red line shows the same distribution when macropru index is +2.

We then repeat this exercise to look at the effect of macropru on intermediate outcomes such as credit growth and asset prices, instead of GDP, to unpick the transmission mechanisms. We find limited evidence for some of these channels. According to our results, macropru does not appear to significantly influence the composition of credit: we find macropru is effective at reducing excessive credit growth for both households and businesses. Moreover, we find limited evidence of transmission through asset prices (eg, financial conditions and house prices).

However, we do find an important role for the overall quantity of credit. This leads us to our second finding: that macropru is particularly effective at reducing the right tail of credit growth (reducing the probability of excessive credit ‘booms’, ie 1-in-10 high-credit-growth episodes), as Chart 3 illustrates.

Chart 3: Effect of macropru on credit-growth distribution

Notes: See Chart 2 notes.

We explore this result further, by assessing the extent to which high realisations of credit growth (formally, outturns above the 90th percentile of the credit-growth distribution) weigh on the left tail of GDP growth (formally, the 10th percentile of the GDP-growth distribution). To do so, we extend our quantile-regression framework to assess the extent to which the link between credit growth and the left tail of GDP growth changes when there is a credit boom (defined here as a realisation of credit growth in the top decile) or not.

The results from this exercise are shown in Chart 4, and highlight our third finding: faster credit growth (90th percentile or above) is associated with a significant reduction in the left tail (10th percentile) of annual average GDP growth and this effect is particularly strong when the economy is already experiencing a credit boom. This suggests that credit growth is strongly associated with a deterioration in the growth-at-risk over the medium term particularly in financial booms. Our empirical finding therefore suggests that the prevention and mitigation of credit booms plays a major role in explaining why macroprudential policy can be effective in defusing downside economic risks.

Chart 4: Effect of credit growth on left tail of GDP growth with and without credit booms

Notes: Estimated change in 10th percentile of annual average real GDP growth following a 1 standard deviation increase in credit growth when there is a ‘credit boom’ (two-year credit growth above its historical 90th percentile) and ‘no credit boom’ (two-year credit growth below its 90th percentile).

Conclusions

In this post, we have estimated the effects of macropru on the entire distribution of GDP growth by incorporating a narrative identification strategy within a quantile-regression framework. While macropru has near-zero effects on the centre of the GDP-growth distribution and therefore appears to have limited overall costs, we find that tighter macropru brings benefits. It does so by significantly and robustly boosting the left tail of future GDP growth, while simultaneously reducing the right. Assessing a range of potential channels through which these effects could materialise, we find tighter macropru reduces the probability of excessive credit booms, which, in turn, is important for reducing the probability and severity of future GDP downturns.

Álvaro Fernández-Gallardo is a PhD student at the University of Alicante. Simon Lloyd works in the Bank’s Monetary Policy Outlook Division. This post was written while Ed Manuel was working in the Bank’s Structural Economics Division.

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.

Ivan Yotzov, Philip Bunn, Nicholas Bloom, Paul Mizen and Gregory Thwaites

Inflation in 2023 remains elevated across many advanced economies. Existing studies have considered the contribution of profits to persistently high inflation in the US, euro area and UK. To add to this debate, we recently asked firms in the Decision Maker Panel about their profit margins over the past year and their expectations for the year ahead. This post summarises the key findings from these new questions, and links them to recent trends in prices. Firms reported a squeeze in profit margins over the past year, on average, but they expect to rebuild margins over the next year. Firms expecting to increase margins also expect slightly higher price growth, suggesting that margin rebuilding could make some contribution to inflation persistence.

The Decision Maker Panel (DMP) is a monthly survey of CFOs/financial directors in firms across the UK. It receives around 2,500 responses each month. The survey is frequently used to study business trends across the country and advise policymakers. Recent work using the DMP has analysed inflation during the Covid pandemic, firms’ responsiveness to CPI outturns, and firm price-setting behaviour. Since May 2023, firms were also asked about the evolution of their profit margins (defined as operating profits as a share of sales) over the past year, and their expectations for margins over the year ahead.

Recent trends in firm price growth

Firm annual own-price growth appears to have peaked around the end of 2022 and has begun to decline in recent months. In the three months to July, annual price growth across firms in the DMP was 7.4%. Looking ahead to the next 12 months, firms expect their own-price inflation to fall by around two percentage points (as seen by the gap between the navy and maroon lines in Chart 1). Average expected price growth was 5.2% in the three months to July 2023.

Chart 1: Firm annual price growth, expected price growth, and CPI inflation

When comparing the DMP and CPI inflation trends, it is important to note that the DMP covers prices set by firms across the whole economy, whereas CPI focuses on prices of consumer goods and services only. Certain sectors (eg energy, food) have a greater weight in the CPI basket than in the DMP sample, which can help explain the deviations between these two measures over the past year. This can help to explain why annual CPI inflation increased sharply during 2022 (Chart 1), and also why it may fall by more than the DMP measure over the coming months.

Firm profit margins

Firm-level data on profit margins are typically only available with a lag of at least a year due to the delay in publication of firms’ accounts. Two new questions on firm profit margins added to the DMP survey in May 2023 provide timely data to fill this gap, and they also allow us to study firms’ expectations of profit margins for the year ahead. Moreover, these data allow us to assess which types of firms are seeing their margins rise or fall and to look at whether the firms seeing an increase in margins are also firms who are raising prices the most.

Recent work has highlighted the importance of the precise measurement of profits, as well as the differences between national and firm accounting (see, for example, a recent speech by Jonathan Haskel). This is discussed in detail in a recent Bank Underground post, which also compares measures from firm-level accounts data with aggregate national accounts. It shows that the two comove, at least over the last 20 years, although the firm series are more volatile, on average.

Chart 2: Changes in profit margins over the past year and historical comparisons

Panel A: Changes in profit margins

Panel B: Distribution of profit margins using firm accounts and DMP data

Drawing on data from the DMP survey, Chart 2, Panel A shows that 41% of firms reported that their profit margins had declined over the year between 2022 Q1 and 2023 Q1, and 26% reported large declines (navy bars in Chart 2, Panel A). In contrast, around one third of firms reported their profit margins increased.

Looking to the year ahead, firms expect some increase in margins. Forty-five per cent of firms expect their profit margins to increase in the next 12 months, compared with 32% that expect no material change and 23% who expect profit margins to decline (maroon bars in Chart 2, Panel A).

To try to put these recent developments in margins into a historical context, we can compare them with firm accounts data up to 2021 (Chart 2, Panel B). Specifically, this Chart tracks the historical trends in profit margins for the same firms for which we also have data for in the DMP survey. It furthermore splits these trends across different percentiles of the distribution (calculated at the yearly frequency). The solid lines indicate data from firm accounts, and the dashed lines indicate the implied changes in margins in financial year 2022 (which runs to 2023 Q1) and the expected change in 2023 (which runs to 2024 Q1), based on the latest data from the DMP. The level of profit margins reported in the DMP was higher than accounts data where comparable data exist; we therefore splice changes in margins from the DMP onto the last accounts data rather than showing the actual levels from the DMP.

Looking at firms’ accounts, profit margins declined in the first year (FY) of the pandemic (FY2020) and then recovered in FY2021. This drop was present on average (black line), but most pronounced at the bottom of the distribution (navy line). The latest data from the DMP suggest profit margins declined slightly in FY2022, and that firms expect a rebuild in FY2023. However, these patterns vary across the distribution: firms at the 10th percentile reported the largest drop, and they only expect a partial recovery. In contrast, there has been less of a squeeze in the upper part of the distribution. For example, margins in the 90th percentile are flat over FY2022 and are expected to increase in the year ahead.

Going beyond aggregate trends, we next consider how changes in profit margins relate to other firm-level variables. We highlight three important relationships here. First, the relationship between annual price growth and changes in profit margins in the cross section is only weakly positive (Chart 3, Panel A). However, we find a strong positive relationship between annual real sales growth and changes in profit margins (Chart 3, Panel B). Note that these Charts are binned scatterplots, where each dot represents 3% of observations, or around 48 firms. Finally, there is also a negative correlation between unit cost growth and changes in profit margins.

Chart 3: Changes in profit margins over the past year

Panel A: Change in profit margins and annual price growth

Panel B: Change in profit margins and real sales growth

It is important to emphasise these relationships are correlations, and do not necessarily reflect a causal underlying mechanism. But, they do highlight that factors beyond price growth, such as demand and costs, can also have important impacts on margins at the firm level. Increases in demand, could for example, be associated with a reduction in average costs if firms face increasing returns to scale. Past work using the DMP finds that energy prices, supply disruptions, and labour shortages have been the largest recent drivers of inflation since 2021.

Looking ahead, firms expect some increase in their profit margins (Chart 2, Panel A). The expected improvement is broad-based across sectors of the economy. At the firm level, we find that firms who have experienced bigger falls in margins over the last year, expect more of an increase next year (Chart 4, Panel A). Like Chart 3, Chart 4, Panel A is a binned scatterplot, where each dot represents around 57 firm-level observations.

Chart 4: Realised and expected changes in profit margins

Panel A: Realised and expected change in profit margins

Panel B: Expected year-ahead price growth by expected change in profit margins

Finally, we find some evidence that firms that expect to increase margins also expect higher price growth over the year ahead (Chart 4, Panel B). This is particularly evident for firms that expect a large increase in margins. These differences suggest that margin rebuilding could make some contribution to inflation persistence going forward, although the implied magnitudes are relatively small given that only 13% of firms expect a large increase in margins. This relationship is also not necessarily causal, and the changes in margins and price growth could both be driven by a third factor. These impacts are already factored in to the average expected price growth presented in Chart 1 and suggest that expected inflation may have been slightly lower in a counterfactual world without margin rebuilding. We also find that expected real sales growth is robustly positively correlated with expected changes in profit margins and the relationship with expected unit cost growth is weakly negative. Therefore, monitoring all of these variables will be important to understand how margins evolve.

Summary

Firms in the DMP were recently asked about developments in their profit margins over the past year and about what they expect for the year ahead. On average, firms experienced a squeeze in margins between 2022 Q1 and 2023 Q1, and they expect to rebuild profit margins over the next year. Real sales growth has the strongest positive correlation with the evolution of margins. However, firms that expect to increase margins also have slightly higher expected price growth for the year ahead, suggesting margin rebuilding could make some contribution to inflation persistence.

Ivan Yotzov and Philip Bunn work in the Bank’s Structural Economics Division, Nicholas Bloom works at Stanford University, Paul Mizen works at King’s College London and Gregory Thwaites works at the University of Nottingham.

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.

Saleem Bahaj, Robert Czech, Sitong Ding and Ricardo Reis

Few topics captivate our attention like the enigma of inflation. Understanding where the market thinks inflation is headed is crucial for policymakers, investors, and anyone who wants to keep their financial ducks in a row. And that’s where inflation swaps come into play. They are like the crystal ball of inflation expectations, allowing traders to hedge against inflation risk and giving us a peek into the minds of market participants. In a recent paper, we delve into this thriving market to uncover the who, what, and why behind the prices of these swaps to shed light on the dynamics of inflation expectations.

Why, you might ask, are these swaps gathering such attention? Well, their significance lies in their ability to offer a complementary perspective on inflation expectations. While traditional measures such as breakeven inflation from index-linked government bonds have their merits, inflation swaps provide a complementary canvas for market participants to express their views on future price dynamics. By analysing these swaps, economists can refine their understanding of market sentiment and calibrate their decisions accordingly. Despite the market’s importance, not much is known about the relevant players and quantities behind the prices.

Inflation swap basics

But first, let’s go back to basics. Inflation swaps are derivative contracts that allow two parties to exchange a stream of payments at a fixed rate for another at a floating rate pegged to an inflation index (for example the UK Retail Price Index (RPI)). Most inflation swap contracts take the form of zero coupon swaps, where cash changes hands at the end of the contract. The majority of these contracts tend to mature at relatively short horizons (eg one to three years) or at long horizons of ten years or more.

The buyer of an inflation swap pays the fixed rate, which reflects the expected inflation at the contract’s end date. If inflation matches the fixed rate, no money changes hands, and both parties break even. The seller of an inflation swap pays the floating inflation rate. Since the actual inflation rate is uncertain, the seller’s liability is determined only at the end of the contract. When the realised inflation rate deviates from the fixed rate, one party has to compensate the other. For example, if realised inflation is higher than the fixed rate, the seller will owe the buyer at the end of the contract, and vice versa if realised inflation is lower than the fixed rate.

Unsurprisingly, the inflation swap market is not without its complexities. Market participants have different bargaining powers and risk-bearing capacities, and often demand extra compensation for trading such a risky derivative. This means that the swap breakeven rate not only reflects the markets’ pure inflation expectations, but is also likely contaminated by a liquidity premium – a catch-all term for market imperfections that can be large and vary over time. Given so, how useful are these swap breakeven rates as a measure of expected inflation?

Stylised facts on the market for inflation swaps

To lift the lid on this market, we harness the regulatory DTCC EMIR Trade Repository Data to obtain detailed trade-level reports on over-the-counter inflation swap contracts. The inflation measure for the UK inflation swap market is the RPI, which dominates nearly all (around 99%) swap contracts traded on UK inflation – consistent with the RPI’s role as the index used for inflation-linked gilts. Taking a closer look at the UK RPI swaps, one of the key facts that we document is the segmentation within the market. Pension funds and LDI funds emerge as the primary buyers of inflation protection, holding substantial positive net positions (predominantly at longer horizons of more than 10 years). Dealer banks, who are on the other side of the trade, actively sell inflation protection beyond their holdings of inflation-linked gilts. Hedge funds actively engage in short-horizon trading (≤ three years), leading to daily fluctuations in their net positions. This market segmentation sheds light on the strategies and trading motives of market participants across different trading horizons: hedge funds seem to engage in informed arbitrage trading in the short-horizon market, while pension funds seek to hedge their long-dated liabilities by buying inflation protection in the long-horizon market. As a result, dealer banks end up as substantial net sellers of inflation protection (as shown in Chart 1).

Chart 1: Net notional positions in the UK RPI inflation swap market

Source: DTCC Trade Repository OTC interest rate trade state files, from January 2019 to February 2023.

Our identification strategies

Backed by a theoretical model, we decompose swap prices into two components: (i) a liquidity premium; and (ii) a ‘fundamental’ component of inflation expectations. To estimate these components, we employ three different identification strategies, each leveraging different features of the data.

First, we take advantage of the high-frequency nature of our data and assume that hedge funds respond more to fundamental shocks than dealer banks within a trading day. Dealer banks, in turn, react more than pension funds to these shocks. This means that a fundamental shock resembles a demand shock in the short-horizon market (both prices and quantities rise) and a supply shock in the long-horizon market (prices rise but quantities fall). We also assume that demand shocks to the short-horizon market do not spill over to the long-horizon market within a trading day, and vice versa. Importantly, we are able to verify these assumptions in the data. Using a sign restriction approach, we are then able to identify supply, demand and fundamental shocks by observing fluctuations of prices and quantities in the swap market.

Second, we leverage the variation across market participants to construct granular instrumental variables, using investor-level shocks as instruments for the aggregate demand of each sector.

Finally, we exploit the fact that inflation swap rates exhibit larger fluctuations on inflation release dates. Assuming that these dates are dominated by fundamental shocks, we can use price movements over time to identify these shocks. For all three identification strategies, we then estimate vector autoregressions.

Results and policy implications

Our empirical analysis yields several robust findings across the three identification strategies. First, we find that all swap rates stabilize within two to three trading days after a fundamental shock. In other words, the inflation swap market seems to incorporate new fundamental information rather quickly. Second, the supply of inflation protection by dealer banks to pension funds at long horizons is very elastic, unlike their supply to hedge funds at short horizons. This means that shocks to the demand of pension funds determine the traded quantities in the long-horizon market, but are unlikely to have a significant impact on prices. Third, most of the movements in short-horizon swap prices are driven by liquidity frictions, particularly by shocks to the demand of hedge funds. Our results therefore suggest that short-horizon swap rates are relatively unreliable measures of expected inflation. In contrast, fundamental factors dominate price movements in the long-horizon market. This suggests that changes in the 10-year inflation swap rate are a better measure of fundamental expected inflation.

Moreover, we provide a novel time series for long-horizon fundamental expected inflation that is cleaned of liquidity frictions (see Chart 2). Our counterfactual estimates of expected long-term inflation suggest that an unfiltered reading of actual inflation swap prices will lead to an overstatement of movements in expected inflation. For example, actual measures overstated the risk of deflation during the pandemic, and they equally overstated the risk of sustained inflation during the energy crisis. In fact, our counterfactual measure of expected fundamental inflation has been lower and declining more rapidly than actual swap rates since Autumn 2022. Our counterfactual measure therefore suggests that long-run expectations of inflation are more stable than implied by actual swap rates alone.

Chart 2: Fundamental expected inflation

Finally, we find a significant dispersion in the beliefs about inflation among dealer banks and hedge funds. In the short-horizon market, we show that a handful of institutions respond to fundamental changes in inflation expectations by taking far larger positions in the market. This implies that their trading behaviour is likely to determine swap prices. In the long-horizon market, in contrast, pension funds tend to have more uniform beliefs and their price impact is more evenly distributed. Intriguingly, the inflation beliefs of individual dealer banks inferred from trading activity also line up remarkably well with their one-year inflation forecasts: the dealer banks that submit higher inflation forecasts tend to sell less inflation protection to hedge funds in the short-horizon market.

Conclusion

Understanding the dynamics of the inflation swap market is useful for policymakers and market participants alike. By shedding light on the key players, market dynamics, and expectations, we reveal the swift reactions of market participants to new information, the role of different institutions in trading inflation protection, and the impact of liquidity frictions and fundamental factors on price movements. These insights provide valuable guidance for understanding inflation expectations and making informed decisions in a rapidly changing economic landscape.

Saleem Bahaj works in the Bank’s Research Hub division and is an Associate Professor of Finance and Economics at University College London. Robert Czech works in the Bank’s Research Hub. Sitong Ding is a PhD student at the London School of Economics and Ricardo Reis is the A. W. Phillips Professor of Economics at the London School of Economics.

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.

Julian Oakland

Exchange-traded funds (ETFs) are supposed to be simple and straightforward, and for the most part they are, but one group punches well above its weight when it comes to market impact. In this post, I show that leveraged and inverse (L&I) ETFs generate rebalancing flows that: (1) are always in the same direction of the underlying market move; (2) grow significantly with both increasing and inverse leverage; and (3) must be transacted towards the end of the trading day. These features give rebalancing flows the potential to amplify market moves when markets are at their most vulnerable. L&I ETFs do not currently pose a risk to UK financial stability, but this could change if they grow in popularity.

What are ETFs?

ETFs are baskets of securities that can be traded continually on public exchanges at market prices, just like shares, and typically track an index, a sector, a commodity, or other assets. The first ETF launched in 1993 tracking the S&P 500 index, and since then, ETFs have grown in number, size, and popularity, enabling investors to gain passive exposure to all sorts of markets without significant management fees. However, as time passed, the ETF universe grew to include actively managed products, including rule-based trading strategies like L&I ETFs.

Turning to the topic of this post, Leveraged ETFs are designed to deliver a multiple of the return on a basket over a one-day period, typically the underlying basket 2* or 3*, while Inverse ETFs are designed to deliver the opposite of the return on a basket (basket return -1*). Leveraged inverse ETFs combine the two models and are designed to deliver a multiple of the opposite return on a basket (typically -2* or -3*). These numbers are the ETF leverage factor and can be found in the ETF name or within its prospectus.

What do L&I ETFs hold?

L&I ETFs use derivatives to achieve their aims, and if positively leveraged invest in the underlying assets in addition to entering into long swaps and futures positions to deliver the desired return. These funds also hold cash and money-market instruments to cover losses and margin calls on swaps and futures, and to provide a return to help fund the swaps and futures. The inverse and inverse leveraged funds similarly enter into short swap and futures positions, hold cash and money-market instruments, but hold none of the underlying assets.

A worked example:

Consider a -3* ETF on an index priced at 100 with starting net asset value (NAV) of 100.

At the start of the day, through a portfolio of cash-like instruments and short futures and swaps, the ETF has an Index exposure of -300 (ETF leverage factor * NAV).

To illustrate exposure, calculate the profit/loss to a 1% move up in the index: if you are long 100 units worth £1 each, you make: 1%*100*£1 = £1. If you are short 300 units worth £1 each, you make 1%*(-300)*£1 = -£3. And so for every £1 that the index earns, a -300 exposure to that index earns -£3.

Depending on the moves in the index during the day, the NAV and exposure of this ETF will change.

If the index decreases by 5% to 95, the new NAV = starting NAV + ETF return = 100 + 3*(100 – 95) = 115.

And the exposure is now: -3*(100 – 5) = -285.

If the Index increases by 5% to 105, the new NAV = 100 – 3*(105 – 100) = 85.

And the exposure is now: -3*(100 + 5) = -315.

At the end of each day the ETF return is added to starting NAV to get the NAV for the next day and the portfolio is rebalanced to ensure the leverage is -3* new NAV.

With NAV = 115 and exposure -285, we need -60 more exposure to maintain -3* leverage, and so need to sell futures/swaps to get to an exposure of -345.

With NAV = 85 and exposure -315, we need +60 exposure to maintain -3* leverage, and so need to buy futures/swaps to get to an exposure of -255.

Table A illustrates rebalancing for common L&I ETF leverage factors, including for a vanilla ETF (leverage = 1), and shows that rebalancing is formulaic, given starting NAV, the leverage factor, and the market move.

Table A: Calculations for different ETF Leverage, L, with starting NAV N = 100 and index move on the Day D = -5%

I’ll pause here to look at non-intuitive features of L&I ETFs:

(1) L&I ETFs generate procyclical rebalancing flows: they are always in the same direction as the underlying market move.

(L² – L) > 0 for all L > 1 and L < 0; therefore, rebalancing flows for these leverage factors always have the same sign as D, and so are always in the same direction as the market move, for both positive or inverse leverage factor: if the underlying index has sold off, all L&I ETFs must sell, and if the index has risen, all L&I ETFs must buy.

(2) The higher the leverage, and the more negative the leverage, the higher the percentage of starting NAV that needs to be traded to rebalance.

Using the rebalancing formula, Chart 1 illustrates rebalancing flows to show the power of increasing the leverage and of making it inverse.

Chart 1: Rebalancing flows for differently leveraged ETFs

(3) Rebalancing flows occur towards the end of the trading day, with leverage reset to the closing price of the underlying index.

Leverage is dynamic, so rebalancing too early in the day risks having to unwind trades at a loss if underlying moves are reversed – the closing price is key here.

What does this mean in the real world?

On 13 September 2022, equity markets had their largest sell-off since 2020, with the Nasdaq 100 down 5.2% and the S&P 500 down 4.3%. Using ETF fund data from etfdb.com and equity market data from www.cboe.com, I calculated rebalancing flows for all US equity L&I ETFs on this day to be just over US$20 billion of equities sold, representing around 3.8% of the total value of all S&P 500 shares traded (or the equivalent of just over 100,000 E-mini S&P 500 futures – as an old futures and options trader, these numbers made me gulp!). 

Nomura Holdings Inc. estimate L&I ETF rebalancing flows to be US$15.5 billion on the day, and although our precise estimates of the flows differ (it would be great to mark each other’s homework!), we agree that these flows are likely to have amplified the sell-off.

Selling flows hitting a market towards the end of a down day risk a sudden and severe drop in prices and liquidity, and significant losses can occur – a flash crash. If such conditions persist into the market close and beyond, effects can transmit to other jurisdictions (eg in the 1987 crash, portfolio insurance amplified equity market sell-offs).

In the past 50 years there have been two days of equity market declines greater than 10%: the most recent is a down 12% day on 16 March 2020 during the ‘dash for cash’, and the other is the 1987 crash. Based on fund data and own calculations, I estimate this move led to rebalancing flows equal to around 9.4% of the total value of all S&P 500 shares traded, or around 246,000 S&P E-mini futures contracts (8.5% of daily volume). US equity markets closed on their lows that day, with Covid news dominating, but it is possible that L&I ETF rebalancing flows amplified these moves.

Conclusions

Rebalancing must happen towards the end of the day to set L&I ETFs up with the correct leverage for the next day, and this can add pressure to weak markets when liquidity is already impaired, amplifying market moves.

These products are currently a tiny part of the UK financial system, with products focusing on UK markets representing <0.2% of the total. As such, financial stability risks to the UK appear limited but could arise through contagion from other markets; so it may be wise to monitor these products.

The L&I ETF segment continues to grow and innovate along with the broader ETF market and any acceleration in growth, or concentration on specific themes, means risks could materialise in the future.

Guarding against complacency, I quote Warren Buffet: ‘…derivatives are financial weapons of mass destruction, carrying dangers that, while now latent, are potentially lethal’.

This post was written while Julian Oakland was working in the Bank’s Capital Markets Division.

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.

Sophie Piton, Ivan Yotzov and Ed Manuel

How have profits behaved in this context of sustained level of inflation? In part, the answer depends on how ‘profits’ are defined. Some broad measures suggest increasing profits, but conflate market and non-market sector dynamics and omit important corporate costs. We construct an alternative measure of corporate profits to capture UK firm earnings in excess of all production costs. This measure has been declining since the start of 2022, consistent with evidence from historical energy shocks. This decline has not been uniform across firms, however: firms with higher market power have been better able to increase their margins; others have experienced large declines.

Profits versus excess profits: measurement and trends

Measuring profits is a challenging task: national accounts data are affected by the ‘mixed income’ of self-employed, taxes and subsidies, and conflate two different dynamics: those of the market and non-market (real estate and public) sectors.

We focus on the corporate sector and exclude self-employed and non-market sectors. This is similar to the measure used in Haskel (2023), but in contrast to IMF and ECB work that focuses on the total economy. We additionally abstract from the role of taxes and subsidies; as Haskel (2023) points out, they play a large role in the recent period. We thus focus on corporate gross value added (GVA, at factor costs) and initially split this into two components: employee compensations and ‘profits’, ie everything that is left after paying employees, ie corporate ‘gross operating surplus’ (GOS).

We then go further, following Barkai (2019), and decompose ‘profits’ into two components:

• Capital compensations, which capture firms’ cost of capital (eg the costs of maintaining the equipment and repaying the debt used to purchase it). Following the Hall and Jorgensen (1967) formula, we measure these costs as the sum of capital depreciation, changes in its replacement cost, and the opportunity cost of holding physical rather than financial capital captured through 10-year government bond yields.
• ‘Excess’ or, as Barkai (2020) calls it, ‘pure’ profits, ie what a firm earns in excess of all production costs (including not just labour costs, but also the cost of holding and maintaining the capital input). The share of excess profits in corporate GVA (which captures the price a firm sets relative to its average costs) is the closest concept in national accounts to the measure of mark-ups estimated in firm-level data (which captures the price a firm sets relative to its marginal costs).

We start in Chart 1 by looking at trends in the shares of labour and profits in corporate GVA since 2015, and decompose the profit share further into capital/excess profit shares. The profit share is broadly flat over the period, suggesting the IMF and ECB results reflect in part non-corporate sector dynamics, consistent with the findings in Haskel (2023). When decomposing this share into capital and excess profit parts, we can see excess profits increased in 2021 during the rapid post-lockdown demand recovery, consistent with mark-ups increasing during the high demand Covid recovery period. They started to decline, however, in 2022, when the Ukraine war started. This fall in excess profits partly reflects higher capital costs for firms who are now experiencing higher interest payments to service their debt (due to rising interest rates since start-2022).

Chart 1: Profit, capital and excess profit shares, 2015 Q1–2023 Q1

Percentage of corporate GVA at factor costs

Source: Authors’ calculations using ONS data.

Similarities and differences between current and past energy shocks

In Chart 2.a, we compare the evolution of these shares in the 2022 episode to the oil shocks in the 1970s. In Chart 2.b, we formalise the comparison using a regression framework to estimate the average response of firms’ profits and mark-ups following energy supply shocks over the period 1984–2022 (see Technical appendix for details). We use the oil-supply news shock series from Kanzig (2021) as our measure of an energy shock, identified via the response of oil-price futures in narrow windows around OPEC+ announcements.

Chart 2.b (top panel) describes the responses of labour, capital, and excess profit shares to the identified energy shock. Chart 2.b (bottom panel) estimates the impact on firm-level mark-ups, which is closest in concept to our measure of excess profits in the national accounts. We lack granular data on firms’ mark-ups since the onset of the Russia–Ukraine war, but Haldane et al (2018) estimate mark-ups for all UK-listed firms using data over 1987–2018 from Worldscope. The chart shows the estimated response of average (detrended) mark-ups, weighted by firms’ UK sales.

Chart 2: Profit, capital and labour shares around energy price shock episodes

Chart 2.a: Cumulative changes (percentage points (pp)) of the shares from the first quarter of the shock

Notes: Authors’ calculations using ONS data. The black vertical line denotes quarter five after the shock.

Chart 2.b: Estimated impulse responses of excess profit, capital and labour shares (top panel) and mark-ups (bottom panel) following a 10% increase in oil prices

Notes: Estimated impulse responses to energy shocks, using local projections (see Technical appendix). Results for excess profit, labour and capital shares estimated on quarterly data 1984 Q4–2022 Q4. Results for mark-ups estimated on annual data 1987–2018. Dashed lines denote 90% confidence intervals.

There are two main takeaways from this comparison:

• The labour/profit share response in the 2022 episode is different from the 1970s but similar to more recent energy shocks.

In the 1970s, the labour share increased in the first couple of years followed by a decline in the following years. The rise and fall in the labour share was, at the time, thought to reflect the failure of wages to adjust to the adverse supply shocks in the short run (Blanchard et al (1997)). Consistent with this story, this pattern was stronger in countries with more rigid labour markets, such as continental Europe and Japan. Blanchard et al (1997) also suggest that while the initial effect of the shock was to decrease the profit share, over time firms reacted by moving away from labour, leading to a steady increase in unemployment as well as a recovery, and even an increase, in profit shares.

By contrast, the labour share is broadly flat in both the 2022 episode and in the local projection results covering the period 1984–2022. This might suggest a different labour market now than in the 1970s, after labour market reforms took place in the 1980s.

• The excess profit share and mark-ups decrease across all energy shocks, including that in 2022.

All shocks exhibit a significant decline in the excess profit share in the first three years. The excess profit share declines by about 0.7 pp at the peak following a 10% oil price increase in the local projections; it suggests a 14 pp decline following the 200% price increase observed in 1973. By comparison, we observe a peak decline of 20 pp in excess profit after the 1973 oil shock, so the responses are of similar magnitude. Over the five quarters 2022 Q1–2023 Q1, we observe a 3.5 pp decline in the share. While the magnitude might seem smaller this time, the shock is also smaller (40% oil price increase), and the government introduced support schemes to soften the impact. Note that the falling excess profit share reflects at least in part the rise in interest rates – an important component of the cost of capital that rises across all episodes.

This fall also occurs for the average mark-up. The mark-up falls significantly in response to the energy shock with the peak response in year two (by around 0.8 pp), where both the magnitude and time-profile of the response is remarkably similar to our findings for the excess profit share. Ultimately it is mark-ups that matter for inflation, as inflation is proximately driven by changes in marginal costs and changes in mark-ups (over those marginal costs) desired by firms.

Different responses across sectors and firms

The aggregate fall in mark-ups masks significant heterogeneity across sectors: mark-ups rise significantly in the mining and quarrying sector (driven by oil and gas extraction firms), as well as in some other sectors (eg wholesale and retail) – Chart 3.

Chart 3: Estimated response of mark-ups by section-level industry at the one-year horizon

Notes: Dots show estimated response of mark-ups to an energy shock for each industry at the one-year horizon, bands are 90% conf intervals using Driscoll-Kraay standard errors. See Technical appendix for more details.

We extend our local projection framework and interact the mark-up response with firms’ characteristics. We find that, in the year after the energy shock, mark-ups rise by more (fall by less) for firms that are: (i) in more concentrated industries; (ii) less energy-intensive; and (iii) have less sticky prices.

We combine real-time firm-level data from the Decision Maker Panel (DMP) Survey with firms’ balance-sheet data to investigate these firm heterogeneities in the recent period. We lack the required information to estimate mark-ups. Instead, we focus on net operating profit share in sales. Despite some conceptual differences, this measure is closest to the concept for GOS share in gross value added (profit share, including both capital and excess profit), and the two measures commove strongly since the mid-1990s.

We investigate how profits evolve for the average DMP firm, and how they depend on firm characteristics. We find that firms in the DMP experienced a small decline in margins since the start of the Ukraine war (Chart 4). This contrasts with ONS gross profit measure (grey line in Chart 1) that exhibits a flat profit share over 2022–23. This could be related to the broad sample of firms in the DMP; eg there are usually few respondents from the oil sector, where profits tend to increase significantly following an oil shock.

Consistent with the local projections, we see that profits were more negatively affected for firms in high-energy intensive industries, and less negatively affected for firms in more concentrated industries. The DMP also asks about the importance of competitor prices for pricing decisions. Firms that answer that competitors’ pricing is among the most important factor for their price decision usually exhibit a lower decline in margins, but the significance is sensitive to the specification.

Chart 4: Changes in firm profit margins in the DMP conditional on sector/firm characteristics

Notes: The results on profit margins are based on the question: ‘In the first quarter of 2023 (January to March), what was your approximate ‘operating profit margin’ (in percentage terms)? And what was it one year ago, in the first quarter of 2022?’. Energy intensity is estimated using industry data on energy costs from ONS Supply and Use Tables. Industry concentration is measured using a Herfidahl-Hirschman Index at the SIC2 level. The results are weighted by employment and industry shares.

Technical appendix: details on the local projection exercise

In the main post, we use local projection regressions to estimate the response of various outcome variables to energy shocks. First, we estimate the response of a range of aggregate time-series data through the following local projection:

(1)

where is the h-period ahead cumulative change in the outcome variable of interest (eg the profit share), ε_t is a measure of an energy supply shock (which we take from Kanzig (2021)), and X_t are (lagged) control variables including lags of the dependent variable and other variables capturing the state of the macroeconomy (eg GDP and inflation). Estimates of β^h from equation (1) for the labour, capital and excess profit shares, and aggregate (detrended) mark-ups are shown in Chart 2.b in the main post.

We then estimate panel local projections to study the response of firms’ mark-ups at a more granular level. First we estimate the following regressions for subsets of firms in each section-level industry S:

(2)

We then collect and plot the one-year sector-specific mark-up responses to highlight the heterogeneity across sectors (Chart 3 in the main post).

Next, we extend equation (2) to investigate the drivers of this heterogeneity:

(3)

where 𝑍𝑖,𝑡 is a vector capturing a range of potential drivers of heterogeneity in firms’ mark-up response to energy shocks. Table A plots the coefficients for the estimated interaction terms δ^h at the one-year horizon, where we test for all potential sources of heterogeneity simultaneously. We find significant evidence that mark-ups rise by more (fall by less) for firms that are: (i) in more concentrated industries (as measured by the Herfindahl–Hirschman index from Savagar et al (2021)) ; (ii) less energy-intensive (based on firms’ intermediate consumption of energy goods); and (iii) have less sticky prices (based on firms’ reported frequency of price changes).

Table A: Drivers of heterogeneity in mark-up response

Notes: Standard errors calculated using Driscoll-Kraay. Asterisks indicate significance at 99% (***), 95% (**) and 90% (*) level.

Sophie Piton and Ivan Yotzov work in the Bank’s Structural Economics Division.

This post was written while Ed Manuel was working in the Bank’s Structural Economics division.

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.

Sophie Piton and Ivan Yotzov

How have profits behaved in this context of sustained level of inflation? In part, the answer depends on how ‘profits’ are defined. Some broad measures suggest increasing profits, but conflate market and non-market sector dynamics and omit important corporate costs. We construct an alternative measure of corporate profits to capture UK firm earnings in excess of all production costs. This measure has been declining since the start of 2022, consistent with evidence from historical energy shocks. This decline has not been uniform across firms, however: firms with higher market power have been better able to increase their margins; others have experienced large declines.

Profits versus excess profits: measurement and trends

Measuring profits is a challenging task: national accounts data are affected by the ‘mixed income’ of self-employed, taxes and subsidies, and conflate two different dynamics: those of the market and non-market (real estate and public) sectors.

We focus on the corporate sector and exclude self-employed and non-market sectors. This is similar to the measure used in Haskel (2023), but in contrast to IMF and ECB work that focuses on the total economy. We additionally abstract from the role of taxes and subsidies; as Haskel (2023) points out, they play a large role in the recent period. We thus focus on corporate gross value added (GVA, at factor costs) and initially split this into two components: employee compensations and ‘profits’, ie everything that is left after paying employees, ie corporate ‘gross operating surplus’ (GOS).

We then go further, following Barkai (2020), and decompose ‘profits’ into two components:

• Capital compensations, which capture firms’ cost of capital (eg the costs of maintaining the equipment and repaying the debt used to purchase it). Following the Hall and Jorgensen (1967) formula, we measure these costs as the sum of capital depreciation, changes in its replacement cost, and the opportunity cost of holding physical rather than financial capital captured through ten-year government bond yields.
• ‘Excess’ or, as Barkai (2020) calls it, ‘pure’ profits, ie what a firm earns in excess of all production costs (including not just labour costs, but also the cost of holding and maintaining the capital input). The share of excess profits in corporate GVA (which captures the price a firm sets relative to its average costs) is the closest concept in national accounts to the measure of mark-ups estimated in firm-level data (which captures the price a firm sets relative to its marginal costs).

We start in Chart 1 by looking at trends in the shares of labour and profits in corporate GVA since 2015, and decompose the profit share further into capital/excess profit shares. The profit share is broadly flat over the period, suggesting the IMF and ECB results reflect in part non-corporate sector dynamics, consistent with the findings in Haskel (2023). When decomposing this share into capital and excess profit parts, we can see excess profits increased in 2021 during the rapid post-lockdown demand recovery, consistent with mark-ups increasing during the high demand Covid recovery period. They started to decline, however, in 2022, when the Ukraine war started. This fall in excess profits partly reflects higher capital costs for firms who are now experiencing higher interest payments to service their debt (due to rising interest rates since start-2022).

Chart 1: Profit, capital and excess profit shares, 2015 Q1–2023 Q1

Percentage of corporate GVA at factor costs

Source: Authors’ calculations using ONS data.

Similarities and differences between current and past energy shocks

In Chart 2.1, we compare the evolution of these shares in the 2022 episode to the oil shocks in the 1970s. In Chart 2.2, we formalise the comparison using a regression framework to estimate the average response of firms’ profits and mark-ups following energy supply shocks over the period 1984–2022 (see Technical appendix for details). We use the oil-supply news shock series from Kanzig (2021) as our measure of an energy shock, identified via the response of oil-price futures in narrow windows around OPEC+ announcements. Chart 2.2 (top panel) describes the responses of labour, capital, and excess profit shares to the identified energy shock. Chart 2.2 (bottom panel) estimates the impact on firm-level mark-ups, which is closest in concept to our measure of excess profits in the national accounts. We lack granular data on firms’ mark-ups since the onset of the Russia–Ukraine war, but Haldane et al (2018) estimate mark-ups for all UK-listed firms using data over 1987–2018 from Worldscope. The chart shows the estimated response of average (detrended) mark-ups, weighted by firms’ UK sales.

Chart 2: Profit, capital and labour shares around energy price shock episodes

Chart 2.1: Cumulative changes (percentage points (pp)) of the shares from the first quarter of the shock

Notes: Authors’ calculations using ONS data. The black vertical line denotes quarter five after the shock.

Chart 2.2: Estimated impulse responses of excess profit, capital and labour shares (LHS) and mark-ups (RHS) following a 10% increase in oil prices

Notes: Estimated impulse responses to energy shocks, using local projections (see Technical appendix). Results for excess profit, labour and capital shares estimated on quarterly data 1984 Q4–2022 Q4. Results for mark-ups estimated on annual data 1987–2018. Dashed lines denote 90% confidence intervals.

There are two main takeaways from this comparison:

• The labour/profit share response in the 2022 episode is different from the 1970s but similar to more recent energy shocks.

In the 1970s, the labour share increased in the first couple of years followed by a decline in the following years. The rise and fall in the labour share was, at the time, thought to reflect the failure of wages to adjust to the adverse supply shocks in the short run (Blanchard (1997)). Consistent with this story, this pattern was stronger in countries with more rigid labour markets, such as continental Europe and Japan. Blanchard (1997) also suggests that while the initial effect of the shock was to decrease the profit share, over time firms reacted by moving away from labour, leading to a steady increase in unemployment as well as a recovery, and even an increase, in profit shares.

By contrast, the labour share is broadly flat in both the 2022 episode and in the local projection results covering the period 1984–2022. This might suggest a different labour market now than in the 1970s, after labour market reforms took place in the 1980s.

• • The excess profit share and mark-ups decrease across all energy shocks, including that in 2022.

All shocks exhibit a significant decline in the excess profit share in the first three years. The excess profit share declines by about 0.7 pp. at the peak following a 10% oil price increase in the local projections; it suggests a 14 pp decline following the 200% price increase observed in 1973. By comparison, we observe a peak decline of 20 pp in excess profit after the 1973 oil shock, so the responses are of similar magnitude. Over the five quarters 2022 Q1–2023 Q1, we observe a 3.5 pp decline in the share. While the magnitude might seem smaller this time, the shock is also smaller (40% oil price increase), and the government introduced support schemes to soften the impact. Note that the falling excess profit share reflects at least in part the rise in interest rates – an important component of the cost of capital that rises across all episodes.

This fall also occurs for the average mark-up. The mark-up falls significantly in response to the energy shock with the peak response in year two (by around 0.8 pp), where both the magnitude and time-profile of the response is remarkably similar to our findings for the excess profit share. Ultimately it is mark-ups that matter for inflation, as inflation is proximately driven by changes in marginal costs and changes in mark-ups (over those marginal costs) desired by firms.

Different responses across sectors and firms

The aggregate fall in mark-ups masks significant heterogeneity across sectors: mark-ups rise significantly in the mining and quarrying sector (driven by oil and gas extraction firms), as well as in some other sectors (eg wholesale and retail) – Chart 3.

Chart 3: Estimated response of mark-ups by section-level industry at the one-year horizon

Notes: Dots show estimated response of mark-ups to an energy shock for each industry at the one-year horizon, bands are 90% conf intervals using Driscoll-Kraay standard errors. See Technical appendix for more details.

We extend our local projection framework and interact the mark-up response with firms’ characteristics. We find that, in the year after the energy shock, mark-ups rise by more (fall by less) for firms that are: (i) in more concentrated industries; (ii) less energy-intensive; and (iii) have less sticky prices.

We combine real-time firm-level data from the Decision Maker Panel (DMP) Survey with firms’ balance-sheet data to investigate these firm heterogeneities in the recent period. We lack the required information to estimate mark-ups. Instead, we focus on net operating profit share in sales. Despite some conceptual differences, this measure is closest to the concept for GOS share in gross value added (profit share, including both capital and excess profit), and the two measures commove strongly since the mid-1990s.

We investigate how profits evolve for the average DMP firm, and how they depend on firm characteristics. We find that firms in the DMP experienced a small decline in margins since the start of the Ukraine war (Chart 4). This contrasts with ONS gross profit measure (grey line in Chart 1) that exhibits a flat profit share over 2022–23. This could be related to the broad sample of firms in the DMP; eg there are usually few respondents from the oil sector, where profits tend to increase significantly following an oil shock.

Consistent with the local projections, we see that profits were more negatively affected for firms in high-energy intensive industries, and less negatively affected for firms in more concentrated industries. The DMP also asks about the importance of competitor prices for pricing decisions. Firms that answer that competitors’ pricing is among the most important factor for their price decision usually exhibit a lower decline in margins, but the significance is sensitive to the specification.

Chart 4: Changes in firm profit margins in the DMP conditional on sector/firm characteristics

Notes: The results on profit margins are based on the question: ‘In the first quarter of 2023 (January to March), what was your approximate ‘operating profit margin’ (in percentage terms)? And what was it one year ago, in the first quarter of 2022?’. Energy intensity is estimated using industry data on energy costs from ONS Supply and Use Tables. Industry concentration is measured using a Herfidahl-Hirschman Index at the SIC2 level. The results are weighted by employment and industry shares.

Technical appendix: details on the local projection exercise

In the main post, we use local projection regressions to estimate the response of various outcome variables to energy shocks. First, we estimate the response of a range of aggregate time-series data through the following local projection:

(1)

where is the h-period ahead cumulative change in the outcome variable of interest (e.g. the profit share), ε_t is a measure of an energy supply shock (which we take from Kanzig (2021)), and X_t are (lagged) control variables including lags of the dependent variable and other variables capturing the state of the macroeconomy (eg GDP and inflation). Estimates of β^h from equation (1) for the labour, capital and excess profit shares, and aggregate (detrended) mark-ups are shown in Chart 2.2 in the main post.

We then estimate panel local projections to study the response of firms’ mark-ups at a more granular level. First we estimate the following regressions for subsets of firms in each section-level industry S:

(2)

We then collect and plot the one-year sector-specific mark-up responses β¹_s to highlight the heterogeneity across sectors (Chart 3 in the main post).

Next, we extend equation (2) to investigate the drivers of this heterogeneity:

(3)

where 𝑍𝑖,𝑡 is a vector capturing a range of potential drivers of heterogeneity in firms’ mark-up response to energy shocks. Table A plots the coefficients for the estimated interaction terms XXX  at the one-year horizon, where we test for all potential sources of heterogeneity simultaneously. We find significant evidence that mark-ups rise by more (fall by less) for firms that are: (i) in more concentrated industries (as measured by the Herfindahl–Hirschman index from Savagar et al. (2021)) ; (ii) less energy-intensive (based on firms’ intermediate consumption of energy goods); and (iii) have less sticky prices (based on firms’ reported frequency of price changes).

Table A: Drivers of heterogeneity in mark-up response

Notes: Standard errors calculated using Driscoll-Kraay. Asterisks indicate significance at 99% (***), 95% (**) and 90% (*) level.

Sophie Piton and Ivan Yotzov work in the Bank’s Structural Economics Division.

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.

Hela Mrabet and Jack Page

The rise in commodity prices after Russia’s invasion of Ukraine had a direct and noticeable impact on consumers’ bills for energy and food. But firms also felt the brunt of higher costs. How did firms pass on these cost shocks through the supply chain and all the way onto consumer prices? How much and how quickly can firms pass through such large cost shocks? In this blog post, we combine information from Supply-Use tables with a rich industry-level data set on input and output price indices to shed light on these questions.

How do cost shocks pass through the supply chain?

Imagine an economy with three sectors (and firms): an energy producer, a food manufacturer and a restaurant. Energy is a primary input into production, and the economy is hit by a large energy price shock. The restaurant will see its energy bills rise as a result; and will seek to pass it through to its customers – this is the ‘first-order’ supply-chain effect on inflation (solid arrow in Chart 1). But the restaurant will also see food prices go up as a result of the energy price shock, and will also attempt to pass this increase through to its customers – this is the ‘second- order’ supply-chain effect on inflation (dashed arrows in Chart 1).

So to generalise this idea for an economy with multiple sectors, an input price shock will generate interactions through the supply chain as the shock is passed to upstream sectors, and these interactions will all affect inflation.

Chart 1: The idea

Source: Authors’ calculations.

A representation through Supply-Use tables

One way to formalise this idea is to use Supply-Use tables. These describe how products are used as intermediate inputs to produce further products (either intermediates or final goods and services), and so allow us to estimate a given input cost pass-through from the full supply-chain interaction.

Let’s use energy (E) as a primary input again in an economy with n different products, and let’s assume a shock  to the price of energy. For each of the remaining n-1 products in the economy, the first-order supply-chain effect of the cost shock on the price of product j is the share of energy in the output of product j multiplied by the energy price shock. And the second-order effects and beyond are the price changes of all the other inputs used to produce product j multiplied by their share in output. So overall, the full effect captures how the energy shock ripples through to final products, both directly through first-order supply chain effects, and indirectly, through second-order effects and beyond.

The Supply-Use tables give us the increase in the price of 105 non-energy products following an energy price shock – goods and services in the economy are classified into 105 categories according to the Classification of Products by Activity (CPA). These 105 CPA categories do not perfectly match to CPI components (which are classified by purpose instead), so we use the ONS CPA-COICOP convertor.

We apply a similar methodology to obtain indirect food effects through the supply chain. Chart 5 below shows the contribution of indirect food and energy effects to CPI inflation.

How much and how quickly do cost shocks get passed through the supply chain?

At face value, the representation through Supply-Use tables described above assumes full and immediate pass-through of the energy price (or any other input) shock at each stage of supply chain interaction. We think this is a strong assumption, and might not properly reflect firms’ pricing decisions. For example, the Bank of England’s Agents Intelligence pointed to firms facing a margins’ squeeze over the past year immediately after the commodity price shock, and a gradual rebuild this year and next. This suggests the pass-through of the energy price surge is rather lagged, and possibly incomplete.

To address this, we add information on the scale and speed of pass-through from rich data sets on producer price inflation (PPI) and services producer price (SPPI) to capture firms’ pricing decisions. These provide input and output price indices for manufacturing and services sectors going back to 1997. For manufacturing sectors, we estimate industry-specific error-correction models (ECMs) of output prices on input prices. For services, there are sector-specific output prices, but not sector-specific input costs, so we use the aggregate manufacturing input price PPI on the right hand-side of the regressions instead. Equations 1a and 1b below describe the ECMs long-run relationship and short-run dynamics:               

Equation 1a – Long-run (LR) regression:

Equation 1b – Short-run (SR) Dynamics:

We estimate these regressions for around 70 sectors with quarterly data going back to 1997 (when available). We use the coefficients in equation 1a to underpin the long-run pass-through of an input cost shock into the output price of each sector i.

And we use the impulse response functions from the short-run dynamics in equation 1b to underpin the timing of this pass-through for each sector i.

Overall, our sector-level regressions suggest the pass-through of an input cost shock is incomplete (Chart 2), with long-run coefficients ranging from 0.4 (for services industries) to 0.8 (for most manufacturing industries).

Chart 2: Long-run pass-through coefficients by sector

Source: Authors’ calculations.

The dynamics also vary significantly across sectors. For each sector, we use the ECM regressions to plot the impulse response functions of the output price to an input price shock. Chart 3 shows the time (in quarters) needed to pass through 80% of the input price shock for each industry. Pass-through is found to be faster for manufacturing sectors, with eight quarters on average until 80% of the shock is passed through versus 15 quarters on average for services industries.

Chart 3: Time to pass through 80% of the input price shock by sector

Source: Authors’ calculations.

Does what go up go down?

Do firms change prices in the same way irrespective of whether input costs go up or down? This question is interesting from a policy perspective: if firms decide to pass through an input cost increase faster than an input cost fall, then there could be more persistence in inflation from the current commodity shock even as commodity prices start to fall.

We use the industry-level ECM regressions to check for asymmetry on the way down. To do so, we introduce dummy variables into the dynamic part of the equation to separate out periods when CPI inflation was above or below the mean, or alternatively rising or falling. We limit the estimation sample to 2019 Q4, such that it is not biased by the current episode of input cost shock.

We find evidence of asymmetry in the cost-push shock for most manufacturing industries, as well as some services industries (eg food and accommodation services in line with the Bank of England’s Agents Intelligence). Overall, input price shocks get passed into output prices with an additional two quarters’ lag when costs are going down versus going up (Chart 4).

Chart 4: Impulse response functions (IRFs) on the way up versus on the way down

Source: Authors’ calculations.

So how much of CPI inflation is driven by energy and food costs passing through the supply chain?

Taking all this together (Chart 5), we estimate that the pass-through of energy and food price shocks through the supply chain boosted CPI inflation by around 1 percentage point at peak (2022 Q4). And could be a source of persistence in inflation going forward, as firms continue to pass through past input shocks to rebuild their margins. Chart 5 also shows what a ‘full and immediate’ pass-through assumption would suggest, with a larger effect on inflation at peak, but also a lot more short-lived.

Chart 5: Contribution of indirect effects through the supply chain to CPI inflation

Source: Authors’ calculations.

Looking at differences across CPI components (Chart 6), the energy contribution to inflation has been largest for food and non-alcoholic beverages (FNAB); it is estimated to have peaked at approximately 3 percentage points in 2022 Q3 and to have moderated relatively quickly afterwards. Our forecast is consistent with significant further moderation in 2023 Q4. Energy has provided a significant boost to some services sector inflation, for example transport and restaurants & hotels (approximately 1 percentage point at peak). For these sectors, the contribution of energy is relatively persistent, reflecting the longer lags through the supply chain suggested by the PPI regressions.

Chart 6: Contribution of indirect energy effects to inflation across COICOP categories, 2022 Q3–2024 Q2

Source: Authors’ calculations.

Conclusion

In this blog post, we present a way of estimating the inflation effects of energy and food cost shocks through the supply chain, which combines information from Supply-Use tables as well as relationships between input and output prices from the PPI data set. Our key assumption is that the pass-through is gradual, incomplete and asymmetric; and our methodology captures the full set of interactions along the supply chain. The results show that energy and food effects through the supply chain have had a sizeable contribution to inflation over the past year, and – given the asymmetric time lag in passing on cost shocks coming down (slower) versus going up (faster) – might be a source of persistence over the next 12 months as firms try to rebuild their margins.

Hela Mrabet works in the Bank’s Monetary Policy Outlook Division and Jack Page works in the Bank’s External MPC Unit.

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.

Gerry Gunner and James Waddell

Buy-Now-Pay-Later (BNPL) is a relatively new form of consumer credit that you might have noticed as a payment option when shopping online or in person. However, there is little analysis in the public domain about who is using BNPL credit in the UK and its contribution to total household debt. We have used the Bank’s NMG Consulting survey to reveal that BNPL borrowers are typically younger adults and renters, and are more likely to report signs of financial distress.

What is BNPL?

Defining exactly what BNPL is can be tricky. The only thing all BNPL products have in common is they allow users to defer payment across multiple instalments when buying goods or services.

Many BNPL products are exempt from regulation because they do not charge interest on repayments. Revenue is generated by charging retailers a fee for facilitating the transaction. Some lenders supplement this by charging borrowers fees for late payments. Typically these products are provided by newer fintech firms, for whom BNPL makes up most or all of their lending.

Some other BNPL products are regulated and charge interest on repayments. These products are typically provided by more established lenders who have banking licences. For these lenders, BNPL products are relatively new and make up a very small part of their lending business.

Who uses BNPL?

Given BNPL is relatively new and much of the market is unregulated, there is little publicly available data on its use. In 2021, the FCA released the Woolard Review which covered unregulated BNPL products. It found that the use of BNPL products nearly quadrupled in 2020 to transactions worth £2.7 billion. Data collected from some BNPL providers revealed 25% of users are 18–24 and 50% are aged 25–36.

To shed further light on who uses BNPL we analysed the Bank’s NMG household survey, which uses weights to be nationally representative. In the March 2023 wave of the survey we asked respondents if anyone in their household owed any money, and if so how much, on BNPL products defined as:  

Buy now pay later agreements; where payment is deferred or split into instalments, often with no interest (eg Klarna, Clearpay, Monzo Flex, Instalments by Barclays etc).

The NMG Survey is run at the household level, whereas the FCA’s data was collected at the user level. This means our results will not be directly comparable to the FCA’s.

For the purposes of this article we define ‘users’ as households reporting BNPL use. We have filtered respondents to only include those who are sole or joint financial decision makers in their household.

On aggregate, the NMG data suggests that 11% of households, or 3.1 million households across the UK reported owing money on BNPL. Among users the mean balance was £866, implying an outstanding aggregate BNPL balance of around £2.7 billion. The distribution of balances is skewed. Lots of BNPL users report balances of a few hundred pounds or less, with a small number reporting much larger balances. The median balance of users is £300 and the 90th percentile is £2,000.

In addition to aggregate statistics, the NMG data allows us to gain insight into how BNPL use varies across different households. There are four characteristics we analyse: age, income, housing tenure and self-reported financial difficulty.

Chart 1a: Share of households using consumer credit products by age group

Chart 1b: Mean household BNPL balance by age group

Age

In Chart 1a, we see BNPL use is most common among 25–34 year olds. We have included credit and store cards as a reference point. BNPL has overtaken store cards for all age groups, but remains less popular than credit cards. There are a number of reasons why BNPL use could be higher for younger people; BNPL is frequently integrated into online shopping, which is more popular with younger people, and older people already have greater access to credit through credit cards. 

In Chart 1b, we see that 35–44 year olds report the highest BNPL balances. Around 37% of households who owe £2,000 or more on BNPL are aged 25–34, the most of any age group.  

Chart 2a: Share of households using BNPL by income decile

Chart 2b: Mean household BNPL balance by income decile

Income

We expect BNPL to be more popular with lower income households. Unregulated BNPL providers do not have to carry out affordability checks on borrowers (although they may choose to do so), so poorer households might turn to BNPL if they are excluded from other forms of regulated credit.

In Chart 2a there is no clear relationship between the BNPL usage rate and household income. Households with incomes between 45,000–54,000 are the most likely to report using BNPL. BNPL balances increase with household income (Chart 2b). The top two income deciles account for about 38% of households who owe £2,000 or more. There are a small number of low income households with high BNPL balances, although the average BNPL balance for this group is about the same as for all households.

Chart 3a: Share of households using BNPL by tenure

Chart 3b: Mean household BNPL balance by tenure

Housing tenure

It is important to understand the relationship between tenure and consumer credit products as mortgage debt and rent payments might be prioritised over payments for consumer credit. These housing obligations could make mortgagors and renters more likely to miss repayments or default on consumer credit.

In Chart 3a we see renters are much more likely to be BNPL users than other groups, and outright owners are the least likely. This is partially explained by the relationship between housing tenure and age, as younger people are more likely to be renters. However, we can control for this relationship by looking at use by tenure across younger households only. After doing so we see that renters are still more likely than other groups to be BNPL users. In Chart 3b we see that renters report the lowest balances, while outright owners report the highest balances. This is partially explained by the relationship between housing tenure and income, as renters are much less likely to have very high incomes than mortgagors or outright owners. Mortgagors account for around 38% of households owing £2,000 or more and renters 34%.

Financial difficulty

Younger people and renters are more likely to be BNPL users, and these groups tend to have less resilient finances. The data confirms that BNPL users are more likely to display signs of financial vulnerability. 68% of BNPL borrowers are concerned about their level of borrowing, compared to 45% of other borrowers. BNPL users are more likely to report falling behind on any of their unsecured debt by two months or more over the past year. 21% of BNPL users report having been in arrears, compared to 6% of other borrowers.

Conclusions

Data on BNPL is limited, partly due to its exemption from regulation. Household survey data reveals financially vulnerable groups such as renters and those aged 18–34 are more likely to be BNPL users. They are also significantly more likely to report signs of financial distress. 

On the other hand, the poorest households are no more likely than average to be BNPL users. Currently BNPL use is not as widespread as other forms of consumer credit and average balances are small. Regulated lenders’ exposures to losses from BNPL lending are limited.

Gerry Gunner and James Waddell work in the Bank’s Macrofinancial Risk Division.

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.

Harvey Daniell and Andre Moreira

The latest developments in the labour market are often central to monetary policy decisions. We outline a framework for mapping labour market indicators to near-term employment and pay growth, drawing on established insights from the ‘nowcasting’ literature. The key benefits of our approach are: the ability to map a range of ‘soft’ and ‘hard’ indicators of different frequencies to quarterly official data; the empirical determination of how much weight to place on each indicator; and the ability to shift those weights flexibly as more data become available. This framework beats simple benchmark models in our labour market application.

Understanding the latest developments in the labour market is often key for monetary policy decisions. In May, for example, the Monetary Policy Committee linked further tightening in monetary policy to, among other things, the tightness of labour market conditions and the behaviour of wage growth.

However, official data on the labour market are published with a lag. To improve understanding of current conditions when setting policy, it’s necessary to extract the signal from more contemporaneous indicators to ‘nowcast’ – ie to predict current and near-term developments in – the labour market.  

What are the key insights of our approach?

1: Hard and soft indicators, of different frequencies, can be mapped individually and directly to the target labour market variable.

The growth of high-frequency data over the past decade or so means we have indicators of the labour market at quarterly, monthly, and even weekly frequency. These indicators can be mapped individually to a target variable, such as quarterly employment growth, using ‘mixed-data sampling’ (or ‘MIDAS’) techniques. MIDAS techniques are ‘essentially tightly parameterised, reduced form regressions that involve processes sampled at different frequencies‘.

These techniques have the advantage of being able to handle data of different frequency, avoiding the need to convert variables to the same frequency, such as reducing weekly data to quarterly. This means we avoid any loss of information (or use of additional assumptions) that transforming variables to the same frequency entails.

A further benefit of our particular approach is the ability to map each indicator individually to the variable of interest, before combining those into an overall model nowcast. In other words, we start by obtaining a different nowcast from each indicator, which is often of interest to policymakers, as it allows us to discuss features such as the dispersion across individual indicators.

As an example, Chart 1 illustrates the nowcasts for quarterly employment growth from MIDAS regressions using a range of labour market indicators. The grey lines illustrate the individual nowcasts, which track the target variable, in the blue line, fairly well, despite the volatility in the latter.

Chart 1: Indicator-based nowcasts for quarter on quarter employment growth (per cent) ^(a)

(a) Indicator-based nowcasts are based on Bank of England Agents’ scores, the Lloyds Business Barometer, ONS/HMRC PAYE payrolls growth, S&P Global/CIPS Purchasing Managers Index and KPMG/REC Report on Jobs.

2: The steer from different indicators can be combined into one overall view, where the weights are informed by the information content of the relevant indicator.

Different indicators often point to different nowcasts for the same variable, so it’s vital to know how much weight to place on each indicator. In our approach, we weight the steer from each indicator according to its relative forecast performance in the recent past – a common approach in the forecasting literature, which we find also works well in this application to the UK labour market. In other words, more accurate indicators receive a higher weight.

Chart 2 and Chart 3 plot a measure of forecast performance for a range of indicators of quarter-ahead employment and pay growth. They show that ‘soft’ indicators like the monthly employment balance of the S&P Global/CIPS Purchasing Managers Index and the monthly KPMG/REC Report on Jobs perform the best at predicting quarterly employment and pay growth, two quarters ahead. Crucially, though, the combined steer from all indicators outperforms relying on any individual indicator alone.

Chart 2: Forecast performance of indicators of employment growth, two-quarters ahead ^(a)

Chart 3: Forecast performance of indicators of pay growth, two-quarters ahead ^(a)

(a) ‘RMSE’ is root mean squared error, a standard measure of historical forecast performance. ‘BCC’ is British Chambers of Commerce quarterly economic survey. ‘Vacancies’ are ONS three-month vacancies growth. ‘Lloyds’ is Lloyds Business Barometer. ‘GDP’ is ONS three-month GDP growth. ‘Payrolls’ are ONS/HMRC PAYE payrolls and median pay growth. ‘REC’ is KPMG/REC Report on Jobs. ‘PMI’ is S&P Global/CIPS Purchasing Managers Index. ‘Agents’ are Bank of England Agents’ scores for employment and pay growth. ‘Job-to-job moves’ are ONS quarterly flows data. ‘Cycle’ indicator is sum of CPI inflation and the (inverse) unemployment gap from Bank of England Monetary Policy Reports.

It’s interesting to note that, within the range of individual nowcasts, soft indicators tend to perform much better than past observations of the ‘hard’ data – ie available monthly observations of the target ONS variable. Fundamentally, this is because the hard data are volatile, and are therefore not necessarily very informative even about their own trajectory. 

3: The weights placed on each indicator change over the data cycle, which is particularly important for the relative weights placed on ‘hard’ versus ‘soft’ indicators of the labour market.

When new data become available, the mapping between a given indicator and the target variable can also change, sometimes in quite an important way. So it’s crucial to re-specify and re-estimate the underlying relationships each time the information set changes. Acknowledging this, the model is also set up in a way that allows it to optimally shift the weight put on ‘hard’ information (such as ONS data) versus ‘soft’ information (such as business surveys) over the data cycle. This optimal weighing takes place in a simple second-stage OLS regression.

Taking a step back, updating the near-term outlook for an expanding information set is an important source of reducing forecast error, not least because monthly observations of the hard data, which the ONS also publishes, are a direct input to the calculation of the quarterly outturn.

Early in the data cycle, when no intra-quarter official data are available, the model tells us to place more weight on the timelier soft indicators such as the monthly business surveys. Once intra-quarter monthly hard data become available, however, these begin to receive a much higher weight.

Under this approach, the full model’s average forecast error progressively declines each time new data become available but, unsurprisingly, the improvement is particularly striking at the juncture where the hard indicators come in to supplement the steer from soft indicators. Chart 4 illustrates this point for a nowcast of employment growth.

Chart 4: Average forecast error for quarterly employment growth declines as more data become available

Viewed in this way, the framework can tell us both the marginal impact of new data on a point estimate of the nowcast, as well as the extent to which the new data also reduce uncertainty around that nowcast.

4: The approach significantly outperforms simple benchmark models for forecasting the labour market.

The key insights outlined above are important sources of reducing forecast error. For example, in a forecast for employment and pay growth two quarters ahead, the combined forecast in which a collective steer is taken from a range of indicators generally outperforms relying on any single indicator alone.

Moreover, the combined nowcast from the full model outperforms a simple autoregressive benchmark model – ie a simple model of quarterly employment/pay growth based on lagged employment/pay growth. This is generally true at all states of the data cycle, with the reduction in root-mean-squared errors increasing to around 65% prior to the release of the outturn.

Indeed, the model’s forecast performance at the quarter-ahead horizon is similar to the one-quarter ahead forecasts from successive Monetary Policy Reports. This reflects the fact that the approach outlined here formalises many of the existing heuristics that forecasters at the Bank employ, such as placing more weight on softer indicators early in the data cycle, and relying on a wide range of data.

Interestingly, performance at the one-year-ahead horizon of the combined model actually slightly outperforms successive Monetary Policy Reports. This is likely to reflect two factors: data-driven approaches to forecasting may outperform more structural models like those supporting the Monetary Policy Report at the one-year horizon; and Monetary Policy Report forecasts are conditional forecasts, whose forecast accuracy is only one requirement of the model among many other requirements.

Harvey Daniell and Andre Moriera work in the Bank’s Current Economic Conditions Division.

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

Kim Nyamushonongora and Oscar Spencer

99.9% of UK businesses are small and medium-sized enterprises (SMEs), employing 61% of the UK population. Yet, we know so much more about large businesses, how they function and particularly how they finance themselves. SMEs have been referred to as the backbone of economies around the world. Therefore, SME’s access to finance is systemically important. Using the SME Finance Monitor, a cross-sectional survey by BVA BDRC on 4,500 SMEs each quarter, we dive into how many SMEs use finance, what finance types they used prior to Covid and during Covid, what characteristics make them more likely to use finance and other relevant questions around SME financing. SMEs are defined as having 249 or less employees.

What types of finance do SMEs use and what are they used for?

Our pre-Covid analysis is conducted over the economically stable period of mid-2018 to end-2019 and the Covid analysis looks at 2020 Q4. We found that 46% of SMEs used external finance. A higher proportion of medium sized companies – employing 50–249 individuals – used finance compared to small companies – which employ fewer than 50 employees. This is in line with expectations, given that the British Business Bank found in the three years leading up to 2019, 10 percentage points more medium companies were likely to seek finance than small companies. The European Central Bank also found in 2021 that the use of all types of finance increased with company size.

Chart 1 investigates and highlights the most popular forms of finances used by SMEs prior to Covid. Bank overdrafts and credit cards are most used. Chart 2 shows the most popular reasons for needing finance, if a company stated they had a need for external finance. Charts 1 and 2 link together as different types of finance tend to be used for different purposes. In terms of relating the charts to one another, economists believe that bank overdrafts and credit cards tend to be used for cash flow related purposes and short-term funding gaps, though some SMEs may use them for investment too. Business investment, like capital expenditure (capex), business expansion and research and development (R&D) is better financed by larger facilities such as a loan, making a bank loan more suitable. Leasing and hire purchase facilities are also used for capex and are a more popular form of finance than bank loans. Bank overdrafts and credit cards are likely the most popular forms of finance used by SMEs because of the ease of obtaining them – current accounts tend to automatically come with an overdraft facility and credit cards can be applied for easily. We find under 5% of companies who need finance state R&D as a reason for needing finance. R&D is typically seen as a very important driver for productivity, a lack of finance for R&D could be contributing to low productivity in the UK. This could be food for thought for policymakers. Chart 2 shows the most popular reasons stated for needing finance; we exclude other reasons which were stated by fewer SMEs from the chart. These include: to take on staff, to fund expansion overseas, to fund new premises, to take over another business, approached by a bank offering funding, a safety net/for safety, IT/online/technology update, stock, marketing/advertising, vehicles, refurbishment/renovation and other.

Chart 1: Companies that had a need for external finance: finance types used by SMEs prior to Covid

Source: BVA BDRC – SME Finance Monitor.

Chart 2: Companies that had a need for external finance: reasons for needing finance prior to Covid

Source: BVA BDRC – SME Finance Monitor.

Nonetheless, things changed during the pandemic. Chart 3 shows the most popular forms of finance used by SMEs in 2020 Q4. 41% of SMEs who had a need for finance used bank loans during Covid compared to 25% before the pandemic. With over 50% of these SMEs stating working capital for cash flow as the main reason for needing finance. Other surveys find similar, the British Business Bank found in 2021 that 25% of SMEs used loans compared to 10% in past years. This was all in large part due to the Coronavirus Business Loan schemes. Will this start a new trend of SMEs using bank loans more? The latest data so far does not suggest so. The latest SME Finance Monitor reports shows that use of bank loans has decreased from 16% of all SMEs in 2020 Q4 to 11% in 2022 Q4. During Covid, lending was guaranteed by the government and there was a desire for banks to lend as much as possible, making it much easier for SMEs to access finance as banks were likely less rigorous in their risk assessment of borrowers. The decrease in use of finance by SMEs now may just show a return to the normal financing conditions. Charts 1 and 3 both show the most popular types of finance used by SMEs, we exclude other types of finance which are used by fewer SMEs from the chart. These include: commercial mortgage, grant, export/import finance, crowd funding, selective/single invoice finance, asset-based lending and ‘other’.

Chart 3: Finance types used by SMEs in 2020 Q4

Source: BVA BDRC – SME Finance Monitor.

What increases a company’s likelihood of using finance?

After looking into the types of finance used by SMEs and the reasons for needing external finance, we ran a logit regression (a regression showing how different variables impact the likelihood of something happening) to delve into who the SMEs using external finance are. Our outcome variable was whether a company used external finance or not. Our results show how a range of different variables impact the probability of a company using external finance. Chart 4 displays the results from the regression. We report the marginal effects – these tell us if the change in probability of a company using finance if the independent variable increases by 1. All our regressors are binary variables, so the marginal effect tells us the increase in the probability of using finance if say, a company is making profit as opposed to if they are not.

Chart 4: Logit regression results ^(a)

Source: BVA BDRC – SME Finance Monitor.

(a) Dummy variables excluded to prevent perfect multicollinearity – 1 employee and 2–5 years.

We find being a larger, older, ‘ambitious’ (the company agreed to the statement they had ‘long-term ambition to be a significantly bigger business’) company increases your likelihood of using external finance. Larger companies are more likely to use finance, though the marginal increase in the likelihood of using finance starts to decrease after reaching 51–100 employees. Older companies are also more likely to use finance, though, companies younger than two years old have been excluded from this regression as they are unable to provide a growth rate due to not being old enough to generate a growth rate estimate. This fits with our expectations that more larger, older companies use finance. From the supply side of finance, larger, older companies are likely to have lower credit risk compared to smaller, younger companies. The five-year survival rate of companies born in 2016 was 38%, therefore finance providers may not be as willing to lend to younger companies as over 60% of them fail within the first five years.

Having a positive turnover growth rate and having ambition to grow increase the likelihood of using finance, too. Companies with a positive growth rate were 9 percentage points more likely to use finance compared to companies who do not. A thought-provoking result is that having used personal funds in the past year to fund the business also increases the likelihood of using finance, increasing the probability by 11 percentage points. Immediate reactions would have been that using personal funds to fund the business is an alternative to using external finance, however, the regression suggests they are complementary. Additional control variables included in the regression but not shown are sector, region – whether they are London based or not, a dummy variable equal to one if company is a sole trader in professional services sector, current account holdings and the regional savings ratio. We find some sectoral heterogeneity – companies in Agriculture, Health and Social work, Transport, Storage and Communication are more likely to use finance. After controlling for regional savings ratios, we do not find any significance of being based in London.

Conclusions

We have much less knowledge about SMEs and their use and access of external finance. From the resources available to us, we find that larger, older, growing SMEs are more likely to use finance. We also see that bank overdrafts and credit cards were popular among SMEs before Covid with many using finance for cash-flow purposes, though during Covid we see bank loans being popular due to their increased availability through Covid loan schemes. Bank loans are now less popular and the British Business Bank show that SMEs are using less of all forms of core finance – bank overdrafts, bank loans/mortgages, credit cards, leasing and hire purchase and grants – as of 2022 Q3.  The terms on which finance was available to SMEs was very different in Covid and we are now seeing increased costs of finance due to rising interest rates. The recent decrease in SMEs using loans fits with the higher cost of borrowing making loans more expensive and reducing activity, but how SME access to and use of finance will change as SMEs navigate the cost of living crisis is an important issue we plan to follow.

Kim Nyamushonongora and Oscar Spencer work in the Bank’s Financial Stability Strategy and Projects Division.

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.

Marco Garofalo, Simon Lloyd and Edward Manuel

The economic consequences of the Russia-Ukraine war have brought the importance of sharp changes in commodity prices, such as oil, to centre stage. While many have focused on understanding the impact of these developments on the central projection for the macroeconomic outlook, this post investigates the balance of risks arising from oil-supply shocks, asking: could these lead to more severe or persistent changes in output growth and inflation, in rare events? Through the lens of a simple statistical model of Inflation- and GDP-at-Risk, we quantify the macroeconomic risks to inflation and GDP growth associated with (exogenous) changes in oil supply, showing that these shocks have more pronounced effects on the upper tail of the inflation distribution than at the centre.

A simple model of Inflation- and GDP-at-Risk

To do this, we capitalise on developments in the academic literature, as well as the growing use of the GDP– and Inflation-at-Risk frameworks by international institutions and at the Bank of England.

Inflation– and GDP-at-Risk offer summary statistics for the overall level of tail risk, capturing the severity of inflationary outturns and potential downturns, respectively. Oil price shocks tend to push output and inflation in opposite directions. So, for our analysis, we define the former as the 95th percentile of the predictive conditional distribution of CPI inflation, and the latter as the 5th percentile of the predictive conditional distribution of GDP growth. In other words, these are extremely high-inflation and low-growth realisations, respectively, that can occur with a ‘1-in-20’ probability, and so capture severe, and potentially costly, tail events.

We use a statistical tool called ‘quantile regression‘ to estimate the relationship between changes in oil prices and the tails of the distributions of inflation and GDP growth. This can ‘weigh up’ the impact of various indicators to provide an overall assessment of the level and drivers of tail risks to inflation and GDP. It comes with some limitations though. For example, it relies on historical data to predict future tail risks, so may struggle in the face of unprecedented events (eg, the Covid pandemic).

To investigate the evolution of the tails of inflation and GDP-growth distributions conditional on oil-supply developments, we need an exogenous ‘shock’ measure. The literature commonly characterises oil-supply shocks as sudden disruptions in the current or future availability of oil, triggering an increase in oil prices. Researchers have developed several strategies to identify such shocks, ranging from the construction of narrative-shock series (Caldara et al (2019); Hamilton (2003); and Kilian (2008)) to SVAR models of the oil market (Baumeister and Hamilton (2019); Kilian (2009); and Kilian and Murphy (2012)). Key to a successful identification strategy is that one can confidently assume that the measures used correlate with oil-supply disturbances, and no other macroeconomic factor drives them.

With that in mind, we capitalise on state-of-the-art work by Känzig (2021), and use his oil-supply news shocks series, obtained through a novel identification design. This exploits high-frequency data on oil-supply surprises based on oil futures prices changes in a narrow window around OPEC announcements. We then estimate a local-projection quantile regression estimating the responses of the tails of the distribution of UK inflation and GDP growth over the three-year horizon to oil-supply shocks. We show results detailing the response of inflation and GDP to oil-supply shocks both at the mean and at the tails.

Effects on inflation

Chart 1 shows the results for inflation. In line with a range of previous work, we find that, on average, UK inflation (blue line) rises significantly in response to oil-supply shocks. Interestingly, we find that Inflation-at-Risk (red line) rises much more in the near term – with the coefficient at the right tail about 50% larger than at the mean. This implies that the oil-supply shock not only shifts the inflation distribution to the right, but also makes the distribution more right-skewed, with much greater probability now in the right tail.

Chart 1: Response of expected inflation (blue) and Inflation-at-Risk (red) to oil-supply shock

Notes: Shaded red area (blue dashed lines) denotes 68% confidence interval for Inflation-at-Risk (mean) estimates.

Chart 2 demonstrates this visually, showing the response of the entire inflation distribution to an oil-supply shock at the one-year horizon. Relative to ‘normal times’ when the oil shock is set to zero and all covariates set to their historical mean (green distribution), a positive shock to oil supply (red distribution) significantly widens the right tail, while leaving the mode (ie, the most likely outcome for inflation) broadly unchanged.

Chart 2: Response of inflation distribution to oil-supply shock at one-year horizon

Notes: Probability density function of UK four-quarter ahead CPI inflation (%) for scenario with all covariates at their historical mean (green line) and with all covariates set to historical mean plus three standard deviation shock to oil supply (red line). The mode of each distribution is the highest point.

Importantly, the change in shape of the inflation distribution and the larger response of the right tail could point to important non-linearities in the macroeconomic relationships underpinning prices. For example, these findings could be consistent with non-linearities in the Phillips curve, ie, the theoretical structure that has inflation mainly determined by economic slack and cost-push shocks, such as oil shocks. Specifically, the latter could be associated with larger increases in inflation when either the shock is large or inflation is high to start with.

Effects on GDP

We now turn to the response of UK GDP. Chart 3 estimates the response of cumulative GDP growth, both at the mean and at the left tail. Again, we find similar results to previous work when focusing on the mean: on average, GDP growth (blue line) falls significantly in response to an oil-supply shock. And importantly, we find the response is more negative in the left tail (red line) than at the mean, although not always significant.

Chart 3: Response of expected GDP (blue) and GDP-at-Risk (red) to oil-supply shock

Notes: Shaded red area (blue dashed lines) is 68% confidence interval for GDP-at-Risk (mean) estimates.

Chart 4 shows how the GDP-growth distribution at the one-year horizon changes in response to an oil-supply shock. The modal path is broadly unchanged in response to the shock, but the left tail becomes significantly longer, pointing to greater downside risks to activity.

Chart 4: Response of GDP distribution to oil-supply shock at one-year horizon

Notes: Probability density function of UK four-quarter ahead GDP growth (%) for scenario with all covariates at their historical mean (green line) and with all covariates set to historical mean plus three standard deviation shock to oil supply (red line). The mode of each distribution is the highest point.

Implications and potential trade-offs

Our results highlight the importance of developments in oil prices for policymakers. It is well-established that these types of shocks may lead to a difficult trade-off for monetary policy between quelling inflation and supporting economic activity. Building on this, our findings stress the specific effect of oil-supply shocks of worsening the trade-offs at the tails. For a policymaker concerned with risk management – and specifically the avoidance of (high) inflation and (low) GDP disasters – the trade-off becomes even starker than when focusing solely on the mean response.

Marco Garofalo works in the Bank’s Global Analysis Division, Simon Lloyd works in the Bank’s Monetary Policy Outlook Division and Edward Manuel works in the Bank’s Structural Economics Division.

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.

Nicholas Vause and Carolin Pflueger

Recently, Pflueger, Siriwardane and Sunderam (2020) proposed a new measure of investor risk perceptions based on the cross-section of stock prices. Using that measure, they found that when risk perceptions are high, the cost of capital of risky firms is high and subsequently real investment and employment decline in the United States. In this post, we show that similar relationships exist in the United Kingdom. In 2023 Q1, the UK measure fell to its lowest level since the outbreak of the Covid pandemic, indicating higher risk perceptions and potentially foreshadowing weaker economic activity. This indicator may be helpful for policymakers, as it could serve as a useful measure of risk perceptions relevant for future economic developments and monetary policy.

Introducing the price of volatile stocks as a measure of risk perceptions

Economists such as Keynes, Minsky and Kindleberger have highlighted the importance of investor risk perceptions in driving economic fluctuations. In such accounts: (i) a negative economic shock causes perceptions of risk to rise; (ii) investors then value the safety of government bonds and charge risky firms a higher cost of capital; and (iii) firms invest less and employment and output decline.

Pflueger, Siriwardane and Sunderam (2020) introduce a new measure of risk perceptions, the price of volatile stocks (PVS), motivated by a stylised model of (i), (ii) and (iii). Most simply, PVS is defined as the difference between the average book to market ratio (ie the accounting value of a company relative to its market value) of low minus high-volatility stocks. Intuitively, when investors perceive more risk: risk appetite is low, the price of volatile stocks falls, the book to market ratio of these stocks rises relative to less-volatile stocks, and PVS is low. This matters for the real economy: as investors perceive more risk, they require higher expected returns to supply capital to risky firms, their investment falls and employment and output decline.

The PVS in the UK

We expand the analysis to the UK, up to and including 2023 Q1. To calculate the PVS in the UK, we follow four steps. First, at each quarter-end, we collect book to market ratios for all stocks in the FTSE All-Share index. Second, using daily data on equity prices for the previous two months, we compute the return volatility for these stocks. Third, we group stocks into quintiles based on their return volatilities. Fourth, we compute the PVS as the difference between the average book to market ratio of stocks in the lowest and highest-volatility quintiles.

Chart 1 shows the time series of the UK PVS since the start of 2000, from when the coverage of our data has been reasonably comprehensive. For comparison the US PVS is superimposed. The correlation between the two series is quite high at 53%, suggesting that investor risk perceptions are driven by global factors to a significant degree, potentially consistent with a global financial cycle (Miranda-Agrippino and Rey (2020)). It also shows that the UK PVS fell sharply in 2023 Q1 amidst the banking turmoil in the US and Europe. It declined by 2.2 standard deviations (which means that sharper falls occur in only 1.4% of quarters), reaching levels not seen since the outbreak of Covid in 2020 Q1.

Chart 1: The price of volatile stocks

The relationship between PVS and economic activity in the UK

To investigate how the PVS relates to economic activity, we estimate local projection regressions of the form:

where y_t+h denotes a variable related to economic activity at h quarters ahead of the current quarter (t), which is either (i) the investment ratio, defined as the ratio of gross fixed capital formation to gross capital stock net of depreciation; (ii) the output gap as estimated by the Office for Budgetary Responsibility or (iii) the change in the unemployment rate. In addition, RR denotes the real risk-free interest rate, which is an alternative driver of economic activity that we control for, and which is measured as the yield on two-year inflation-indexed gilts. All the variables on the right-hand side of the equation are standardised, so we can interpret their coefficients as the response of economic activity in h quarters’ time to a one-standard-deviation shock to the right-hand-side variable.

Chart 2 shows the estimated responses of the investment ratio, output gap and unemployment rate to a one standard deviation positive shock to the PVS in the UK. The solid lines show the central estimates and the shading shows 95% confidence bands. A positive shock means that investor risk perceptions have decreased. This boosts the investment ratio and the output gap and leads to a fall in the unemployment rate, with peak effects 3–8 quarters after the shock. The magnitudes and timings of the estimated responses are similar to those found in the US by Pflueger, Siriwardane and Sunderam (2020), suggesting that risk perceptions are similarly relevant for economic activity in the UK as in the US.

Chart 2: Estimated responses of economic activity to a one standard deviation positive PVS shock in the UK

In light of these relationships, the 2.2 standard deviation decrease in the UK PVS in 2023 Q1 may foreshadow a tangible decline in economic activity. While these predictions clearly come with significant uncertainty attached, point estimates based on the results above suggest a peak decline in the investment ratio of 0.4 percentage points from its 2023 Q1 level of 3%, a peak decline in the output gap of 1.2 percentage points and a peak increase in the unemployment rate of 0.5 percentage points.

Nicholas Vause works in the Bank’s Capital Markets Division and Carolin Pflueger works at the University of Chicago.

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.

Mahmoud Fatouh and Ioana Neamțu

Similar to the Deutsche Bank’s episode in 2016 and the Covid stress in 2020, AT1 spreads over subordinated debt rose rapidly and sharply following the Credit Swiss rescue deal. Beyond these three cases, AT1 spreads have been stable. In this post, we focus on conversion risk of AT1 bonds (also known as contingent convertible, CoCo, bonds) to explain the sharp rise in AT1 spreads in these three cases. Conversion risk is the main additional risk of AT1 bonds, compared to subordinated debt. It arises from the potential wealth transfer from AT1 bondholders to existing shareholders when AT1 conversion is triggered, conditional on the solvency of the issuer. We show that, in normal times, investors believe conversion risk is very low, but major events can change this significantly, largely due to higher uncertainty.

Understanding AT1 prices and yield movements

The recent rescue of Credit Suisse included a write-off of its entire US$17 billion AT1 capital (CoCo bonds). This led to rapid and sharp drops in prices (and increases in yields) of other issuers’ AT1, to levels not seen since the Covid stress in March 2020, as Chart 1 shows. They recovered quickly afterwards following the prompt statements from the EU’s financial authorities and the Bank of England. Similar scenes occurred in early 2016, when investors thought Deutsche Bank was about to cancel its AT1 coupons, after being hit by a US$15 billion fine by authorities in the US.

Chart 1: AT1 price movement, January 2015–March 2023

Source: Refinitiv Eikon.

The price movements mirror yields movements, which are usually assessed relative to yields of bank’s subordinated bonds of similar duration. As Chart 2 demonstrates, spreads of AT1 bonds over subordinated bonds of similar duration (measured by differences in yield to worst) have been generally stable, except during and around these three episodes in 2016, 2020 and 2023.

Chart 2: Yield to worst of AT1 versus subordinated bonds

Source: Refinitiv Eikon.

To explain these patterns, we should look at the unique characteristics of AT1 bonds, compared to subordinated debt. AT1 bonds have a lower level of seniority, and hence rank below subordinated debt in terms of pay-out ranking, if the issuer was liquidated. Beyond seniority, holding AT1 bonds involves three additional risks. First, to count as AT1 capital, AT1 bonds must be perpetual, unlike subordinated debt, which can have fixed maturity of five years or more. However, all AT1 bonds are issued with recall covenants, allowing issuers to recall them five years after their issuance. Issuers are generally expected to exercise the call options, when they are activated, but may choose not to if interest rates are relatively high, causing a loss of potential higher return to AT1 bondholders. This is referred to as extension risk. Extension risk should have no effects on AT1 spreads over subordinated debt (Chart 2). This is because yields to worst for AT1 bonds are always equal to yields to call, whose calculation assumes bonds are called at the first call opportunity.

The second risk – coupon cancelability risk – arises from the possible (partial/full) cancellation of coupon payments. The cancellation can happen automatically, when the issuer does not fully meet its capital buffer requirements.

The third risk reflects the possible wealth transfer from AT1 bondholders to existing shareholders when the loss-absorbing mechanism (LAM) of AT1 bonds is triggered, which we refer to as conversion risk. LAM is triggered at a certain capitalisation level (7% CET1 ratio in the UK). The effects on AT1 bondholders and existing shareholders depend on the type of LAM the bonds involve. LAM can be either conversion to equity (CE), where AT1 bondholders get equity shares in exchange of their bonds (at a pre-specified conversion rate), or principal write-down (PWD), where the principal of the bonds is written down. The triggering of LAM can transfer wealth between AT1 bondholders and existing shareholders. PWD bonds (like Credit Swiss AT1) always transfer wealth to shareholders. CE bonds may be dilutive to existing shareholders, if the price at which the bonds convert to equity was lower than the market price of shares. However, given that equity prices are likely to be significantly low during times of stress, we posit that CE bonds are non-dilutive. In our staff working paper, which empirically assesses the link between the AT1 bonds issuance on risk-taking of issuers, we estimate the wealth transfer between shareholders and AT1 holders at the point of conversion for AT1 bonds issued by UK banks, predominantly CE. Our estimates show that, on aggregate, the conversion of these bonds would imply that existing shareholders would gain at the expense of AT1 holders at conversion (ie, AT1 bonds are non-dilutive to existing shareholders). In other words, the central expectation of investors should be that either coupon cancellation or LAM triggering (‘conversion risk’) would generate a loss to AT1 holders, which would be significantly larger for PWD bonds.

It is key to note that the three risks (extension, coupon cancellation and conversion) would matter only if they are expected to materialise while the bank is solvent. In insolvency, the difference in the losses suffered by subordinated debt and AT1 holders is only driven by seniority, and not any of these three risks. This has two implications. First, changes in the creditworthiness (probability of default) of the issuer reflect on the yields of subordinated debt in the same manner, and hence would not have strong effects on the spread differential between AT1 bonds and subordinated debt. Second, the three additional risks would affect AT1 yields and (hence) spreads only if investors believed they would take losses due to these risks, while the issuer is solvent. We argue that this explains the patterns AT1 spreads over subordinated debt show. That is, in normal conditions in AT1 market, investors believe the additional AT1 risks are very low. Market developments, like those seen in 2016, 2020 and 2023, can change investors’ beliefs significantly, leading to spikes in spreads, largely due to higher uncertainty.

While the three risks can affect AT1 spreads, we think such effect would be mainly determined by conversion risk. This risk is linked to the principal of AT1 bonds, rather than their returns, making potential losses due to this risk much larger than those anticipated from coupon cancellation and extension risk. Moreover, given that AT1 spread over subordinated debt (Chart 2) is measured by difference in yield to worst, it should not be affected by extension risk. Hence, we focus our assessment on conversion risk.

In the rest of this post, we estimate the probability of conversion risk conditional on the issuer being solvent, which we use as a measure of the ‘mechanical level’ of conversion risk in normal market conditions.

How do we estimate conversion risk

We use data of eight AT1 issuing UK banks between 2013 H2 and 2021 H1. The data is collected from multiple sources, including share market data, published financial statements and regulatory returns.

Since our analysis approaches the issue from investors’ perspective, we focus on solvency from the market’s perspective and assume that a bank would be solvent if the market-implied value of its assets is greater or equal to the value of its debt. Our aim is to estimate the probability of conversion while the issuer is solvent. To do so, following the approach we used in our paper, for each bank in each period, we calculate the probability of its capital (CET1) ratio falling from its concurrent level to 7% (probability of conversion) and 0% (probability of default).

Both conversion and default probabilities rely on the value of a bank’s asset falling below certain thresholds. Investors would rely on the market value rather than the book value of assets when assessing possible conversion and default in the future. However, the market value of many bank assets is unobservable (eg mortgages). We handle this by estimating the market value of assets and their implied market volatility using the Merton model. The model states that under limited liability, equity can be seen as a European call option on the firm’s assets, with a strike price equal to total debt of the firm and maturity equal to the average maturity of that debt. For a one-year horizon, the convention is to estimate the debt by half of the long-term liabilities together with the full short-term debt amount from a bank’s balance sheet. Even though we can calculate the asset variables daily, the debt information is only available quarterly. Hence, we compute the distance to conversion/default at a quarterly frequency; that is, how far are a bank’s assets from being below the AT1 conversion threshold, and respectively from insolvency. Finally, we extract the probability of conversion/default from the respective distance, assuming the values to be normally distributed.

Having estimated both sets of probabilities, we regress the probability of conversion on the probability of solvency. We use the regression coefficients as estimate of the target probability (probability of conversion conditional on solvency).

Results

Table A present the estimation results. As the table shows, the probability of conversion conditional on solvency is extremely low at about 0.22% on average for all bank-time combinations in the sample. We sort the banks by their relative CET1 ratio compared to their peers. We find that the conditional probability is higher for banks with a lower CET1 ratio but stays below 2% for bank-time combinations with the 25% lowest CET1 ratios in the sample (column (d) in Table A).

Table A: Estimating the probability of conversion while the issuer is solvent

Note: Coefficient estimates of probability of conversion on probability of solvency. Standard errors reported between parentheses, * p<0.10 ** p<0.05 *** p<0.01.

With this estimation in mind, we argue that the perceived conversion risk remains very close to its ‘mechanical level’ in normal times. However, when major shocks with implication for AT1 conversion hit, such as the conversion/write-down of a major AT1 issuer, the perceived conversion risk can become significantly higher than its mechanical level, increasing AT1 spreads over subordinated debt. We think that these sudden changes in the perceived conversion risk can plausibly explain the patterns in AT1 spreads in Chart 2.

Summing up

In summary, major events affecting AT1 bonds market can increase uncertainty or create panic. This can cause an unfounded rise in investors’ perception of conversion risk (and coupon cancellation risk) relative to its mechanical level, and drive AT1 spreads over subordinated debt upward sharply.

Mahmoud Fatouh works in the Bank’s Prudential Framework Division and Ioana Neamțu works in the Bank’s Banking Capital Policy Division.

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.

May Rostom

On average, parental contributions help children buy homes four years earlier than those without them. Out of every 100 new homeowners below the age of 30, 16 will have had help from ‘the Bank of Mum and Dad’, or Bomad for short. That rises to one in four new homeowners under the age of 25. Those who have had help from their parents put down a deposit twice as large, bought bigger first homes, and had smaller mortgage payments than those who did not. Anecdotes about cash assistance from Mum and Dad have recently been backed up by evidence from Legal & General, which implies Bomad plays a non-trivial role in the housing market. I attempt to investigate its prevalence.

I perform a simple calculation using administrative data that records borrower demographics and loan details on all new mortgages issued in the UK between 2015 and 2017. First, I estimate the cumulative savings each borrower could have amassed from earnings alone since entering the workforce. Then I compare that number to the size of their down-payment. If it is greater than their estimated savings, I assume they got help. Otherwise, I assume they did not.

I observe demographic information only at the point of mortgage issuance, not before. So to estimate cumulative savings, I make three rough assumptions. One, I assume people are in full-time education until they are 20, after which they start working. Two, I assume no one is ever unemployed. Three, I assume that past incomes and expenditures were at least as high as what was reported at the time of mortgage issuance. These will overestimate income and expenditure, but together should give a reasonable read on savings. Of course, the reality is far more complex. However, the assumptions are sufficiently conservative in that they are likely to understate the importance of parental support.

These are the takeaways.

First, getting help is fairly common. Chart 1 shows over 10% of first-time buyers (FTBs) younger than 45 are getting financial help from someone else. This number rises to 28% for the under 25s.

Chart 1: Proportion of FTBs getting help in 2015–17

Source: Product Sales Database. Data are pooled from 2015 Q1–2017 Q1 inclusive for FTB borrowers. Getting help is an indicator variable equal to one if after-tax income minus expenditures is greater than the deposit.

Second, the support is substantial. Chart 2 shows that, on average, deposits are two and a half times larger, loans are 30% smaller, and houses cost £15,000 more for those getting help, compared with those who are not. This means ‘Bomad borrowers’ are typically less-leveraged and have lower mortgage payments, leaving more leeway for them to save or spend their incomes on other things.

Chart 2: Purchased house price by help

Source: Product Sales Database. Data are pooled from 2015 Q1–2017 Q1 inclusive for FTB borrowers. Getting help is an indicator variable equal to one if after-tax income minus expenditure is greater than the deposit.

Third, recipients of financial support buy their first homes earlier – on average four years earlier, at the age of 26 instead of 30. And, as above, they tend to buy more expensive homes.

We can work out just how much earlier those with assistance buy a house for a given price by looking at the horizontal distance between the pink and blue lines in Chart 3. The results are extraordinary. The average 26 year old with help paid about £254,000 for their first home. Those with no help waited a decade – until they were 37 – to buy a property for an equivalent sum.

Chart 3: House prices by age

Source: Product Sales Database. Data are pooled from 2015 Q1–2017 Q1 inclusive for FTB borrowers. Getting help is an indicator variable equal to one if after-tax income minus expenditure is greater than the deposit. This chart plots the mean property price paid (y-axis) for FTBs by the age (x-axis) when they purchased the property.

Here is a true story. I have three British friends, all the same age, all unrelated. They all earn roughly the same amount of money in similar jobs, which they worked hard to get and work harder to keep. They are all equally smart and have achieved comparable schooling. My friends are obviously not identical in many ways, but there is, however, one big difference between them.

The first one got significant help at a young age to buy a flat. With this investment, they were able to upsize into a large home in a leafy London neighbourhood about 10 years ago. The second got some – but less help to buy a flat and it occurred several years later. A couple of years before the pandemic, they moved into a small house an hour’s commute from London (this may matter less now, but it did at the time). The third has received no financial help at all. They have been trying to buy a flat for as long as I can remember.

What are the consequences of these differences in timing and loan size? There are many, and economists should think about what they mean.

Many organisations are speaking about the struggles of UK housing affordability. I have written about it too, for example in the context of generational imbalances. This piece adds to this work, demonstrating that whether and when you receive a gift can affect your entire homeownership trajectory – exacerbating the differences not just across generations, but within them.

May Rostrom works in the Bank’s Monetary Policy Outlook Division.

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.

Tomas Key

During the recovery from the Covid pandemic, the demand for workers rose to unprecedented levels in the UK. The number of jobs that firms were looking to fill increased to 1.3 million in the middle of 2022, 60% higher than the level in the past three months of 2019. The amount of job vacancies has fallen substantially over the past year, but remains at a high level. This post discusses how those changes to the demand for workers have affected the unemployment rate. In particular, it outlines how an equilibrium model of the labour market can help to explain why there appears to have been a change to the relationship between job vacancies and unemployment in recent years.

The Beveridge curve

Before turning to the model, let us first take a look at the data. In Figure 1, I have plotted the vacancy and unemployment rates that have been observed over the past 20 years or so. This shows the striking recent increase in the vacancy rate that I mentioned. It also shows that before the pandemic, there was a reasonably stable negative relationship between the vacancy and unemployment rates. When firms are looking to fill more positions, it is easier for unemployed workers to find a job, and so there tends to be fewer of them. This relationship is known as the Beveridge curve.

Figure 1: Vacancy and unemployment rates

Notes: Data is from the three months to June 2001 to the three months to April 2023: latest observation highlighted in red. Vacancy and unemployment rates are as a proportion of the labour force. I use unemployment and labour force data for those aged 16–64 to be consistent with the inputs to the modelling exercise.

Source: ONS.

Based on that pre-pandemic relationship, it would have been reasonable for a casual observer to expect that the very high vacancy rate in 2022 would have been accompanied by a much lower unemployment rate than was the case. Below, I will outline how a fairly standard model of the labour market can help to explain: (i) why the post-pandemic increase in the vacancy rate did not produce a lower unemployment rate; (ii) why the substantial fall in the vacancy rate over the past year has only been accompanied by a relatively modest increase in the unemployment rate; and (iii) the impact that a further decline in the vacancy rate is likely to have on the unemployment rate.

A model of the labour market

The framework that can be used to interpret labour market developments is based on the transitions – or flows – between employment, unemployment and ‘inactivity’ – a catch-all term for anyone that is not currently working or actively searching for work. Lots of people experience these transitions every quarter in the UK. For example, around a quarter of a million people moved from employment into unemployment in every quarter of 2022. Changes to the rate at which people are making these transitions are what generate movements in the employment, unemployment and inactivity rates.

At the heart of the model is an aggregate matching function. This is a device that is useful for summarising how the time that it takes to find a job – or match – is determined by the number of vacancies relative to the number of job seekers as well as the level of ‘matching efficiency’ – the productivity of the matching function. It captures the fact that it takes considerable time and effort for job seekers to find a suitable vacancy, and that this is affected by both the number of opportunities that are available and how many other people are competing to fill them.

The measure of job seekers that I use when estimating the matching function includes unemployed workers as well as some employed and inactive individuals. In the case of inactive people, that might seem odd as I mentioned above that these are individuals who report that they are not actively searching for work. However, many of them do move into employment over a three-month period, perhaps because their circumstances change or they are lucky enough to find a job without having to search for one. Accounting for these ‘passive’ job seekers among the inactive, as well as an estimate of the number of employed individuals searching for work, has been shown to be important in recent research.

After estimating the parameters of the matching function, I can use it to describe how the level of the vacancy rate affects the rate at which people transition into employment. When combined with values for the other flow rates – such as the rates at which individuals are entering unemployment from employment and inactivity – this gives a framework that can be used to trace out the impact of changes to the vacancy rate on the steady-state, or equilibrium, unemployment rate. That is the rate that is obtained once the system has fully adjusted to the changes in the flow rates.

Figure 2: Simulated relationships between the vacancy and unemployment rates

Source: Author’s calculations.

Two illustrations of this are shown in Figure 2. The model produces the negative relationship between the vacancy and unemployment rates seen in the data. That is due to the impact of the vacancy rate on the speed with which unemployed workers find jobs – their ‘job-finding rate’. Holding the other transition rates constant, a higher vacancy rate will raise the job-finding rate of unemployed workers, and so reduce unemployment. This figure also demonstrates that, in this framework, changes to the other flow rates or to matching efficiency will lead to a shift in the position of the simulated Beveridge curve. They will change the level of the unemployment rate that is produced by any level of the vacancy rate.

Another important feature of the simulated relationship between the vacancy and unemployment rates produced by the model is that it is non-linear, or convex. This reflects the fact that as the number of vacancies increases relative to the number of unemployed, it becomes increasingly difficult for firms to fill them. That is something that many companies in the UK have become familiar with in recent years.

Explaining recent labour market dynamics

It is now time to bring together the simulated relationship between the vacancy and unemployment rates produced by the model and the data. I have done that in Figure 3. The simulated Beveridge curve on this plot is produced by the framework I described when calibrated with flow rate estimates from the past year – it is not an attempt to fit a curve using all of the data shown on the chart. The fact that the simulated Beveridge curve does not fit through all of the data makes clear that the changes in the unemployment rate that have been seen over time have not only been due to the impact of changes in the vacancy rate. They have also been due to changes to other flow rates, such as the rate at which people are moving from employment to unemployment, and to matching efficiency – factors that act to shift the position of the curve produced by the framework that I have described.

Figure 3: Simulated Beveridge curve and vacancy and unemployment rates

Notes: Data is from the three months to June 2001 to the three months to April 2023: latest observation highlighted in red. Vacancy and unemployment rates are as a proportion of the labour force. Simulated Beveridge curve is produced using data from 2022 Q1 to 2023 Q1. Data on labour market stocks and flows is for those aged 16–64.

Sources: Author’s calculations and ONS.

So how can this help to explain recent developments? Well, over the past year or so, changes in the vacancy rate have been the main factor generating changes in the unemployment rate. That means that the data have moved down the simulated Beveridge curve. As the vacancy rate is currently very high relative to the unemployment rate, the portion of the curve along which the data have moved is relatively steep. That is why the substantial fall in the vacancy rate over the past year has only been accompanied by a fairly modest increase in the unemployment rate.

The reason that the very high level of the vacancy rate in 2022 did not produce a lower unemployment rate reflects two factors. First, the steepness of the curve that I just mentioned. Second, the fact that the simulated Beveridge curve has ‘shifted out’ from its position before the pandemic. The reason for that shift is that there has been both an increase in flows from inactivity into unemployment, which act to increase unemployment for any level of the vacancy rate, and a reduction in matching efficiency.

The impact of further falls in the vacancy rate will depend on whether the data continue to move down a stable Beveridge curve, or the curve shifts position once more. The current position of the curve suggests that the unemployment rate might settle at a level higher than immediately before the pandemic, once the demand for workers has returned to a more normal level.

Conclusion

Although some recent movements in the UK vacancy and unemployment rates appear odd at first glance, they can be well-explained by a standard model of the labour market. That framework also provides some guidance about the future direction of the labour market – about the impact of further falls in the vacancy rate on the unemployment rate. That impact will depend on whether the data continue to move down a stable Beveridge curve, or whether changes to matching efficiency or to other features of the labour market lead to a deviation from that path.

Tomas Key works in the Bank’s Structural Economics Division.

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.

Lydia Henning, Simon Jurkatis, Manesh Powar and Gian Valentini

Autumn 2022 saw some of the largest intraday moves in gilt yields in history. It was then that jargon normally confined to financial stability papers entered into mainstream commentary – ‘LDI’, ‘doom loop’, ‘deleveraging’. And it was then that the Bank of England engaged in an unprecedented financial stability motivated government bond market intervention. What happened and why has been set out in detail in official Bank communications. This article instead hovers a magnifying glass over transaction-level regulatory data on derivative, repurchase agreements (repo) and bond markets to quantify liability-driven investment (LDI) and pension fund behaviour and enrich our understanding of these exceptional few weeks of stress.

By examining flows in core sterling markets we uncover more detail regarding the main sources of leverage for LDI and pension funds in the run-up to the stress, the origin of margin calls that they faced, and the scale and timing of the deleveraging that followed. We find that the majority of margin calls came from repo. In addition to gilts, large asset sales were also observed in corporate bond markets, especially from pension funds themselves.

Given the scope of the work, we consider only LDI and pension funds with an open gilt repo or interest rate derivatives position during the stress. Our results may differ from other research focused on this period due to differences in the data cleaning procedure or sectoral classification.

But let’s take a step back…

What exactly is LDI?

LDI strategies are intended to help ensure the value of investments moves roughly in line with the value of liabilities. This has made them popular with defined benefit (DB) pension schemes, helping them to meet promised payments to pensioners – aka their liabilities – no matter how interest rates and inflation turn out.

Why and how do liability-driven investors deploy leverage?

Leverage – created by borrowing or arising ‘synthetically’ through the use of derivatives – allows schemes to use less capital to hedge the same amount of liabilities, freeing up cash to be invested in higher returning assets. This extra return grows the value of their assets, helping to close the funding gap. However, leverage also comes with risks and needs to be dynamically managed.

Liability-driven investors deploy leverage through a combination of repo and derivatives. They borrow cash in repo markets, securing this borrowing using their long maturity and inflation-linked gilt holdings, and use this cash to fund additional gilt investments. And they use derivatives – in this case, interest rate and inflation swaps – to further increase their market exposure. While swaps have historically been more popular, in recent years funds have relied increasingly on repo as the cheaper option.

As shown in Chart 1, liability-driven investors in our sample reported a total of £205 billion of net gilt repo borrowing (sum of red bars) coming into the stress, or around 60% of the total net gilt repo borrowing by non-banks. On aggregate, they also reported a net notional position of £167 billion in interest rate swaps (sum of green bars) – receiving a fixed rate while paying floating – and £57 billion in inflation swaps (sum of blue bars).

We estimate that around 50% of net repo borrowing by the sector was secured with longer-dated gilts (20+ years) and more than 70% with inflation-linked gilts. The value of this collateral was therefore particularly sensitive to movements in long-term rates, as we’ll come on to. Derivative positions on the other hand were less sensitive – they had a shorter ‘duration’. Interest rate swaps held by liability-driven investors were generally shorter maturity and less sensitive overall to movements in interest rates than repo.

Chart 1: Net notional of outstanding swap positions (by contract maturity) and net repo borrowing (by collateral maturity) as of 22 September 2022 ^(a)

Sources: SMMD and EMIR TR data, and Bank calculations.

(a) Note that for repo the x axis refers to the maturity of the collateral securing the transaction, whilst for swaps the x axis refers to the life of the contract.

What happened when yields increased?

Higher yields are generally good for DB schemes – specifically when they aren’t fully hedged – as they reduce the value of liabilities. However, the Autumn 2022 stress has shown that rapid price moves can cause challenges for leveraged investors with poor liquidity management.

Following the announcement of the Government’s ‘Growth Plan’ on 23 September, gilt yields increased sharply. The speed and scale of the rise over the following days were unprecedented.

As yields shot up, the gilts securing repo borrowing and interest rate swaps held by liability-driven investors rapidly lost value, driving increased collateral and margin calls. We estimate that between the announcement of the ‘Growth Plan’ and the announcement of the Bank of England intervention, liability-driven investors faced approximately £66 billion in calls for variation margin, around 80% of which related to repo positions. Given data completeness challenges and assumptions underlying the estimation technique, the actual variation margin calls faced by liability-driven investors over this period were likely higher than this.

Chart 2: Cumulative variation margin on net repo borrowing and derivatives positions held by liability-driven investors

Sources: ICE Data Indices, SMMD and EMIR TR data, and Bank calculations.

How did liability driven investors react?

Rising yields also meant liability-driven investors were faced with a rapid increase in their leverage. In response to this, they could either reduce their leverage or recapitalise, by calling capital from pension fund investors. Where additional funds could not be raised quickly enough, some liability-driven investors were forced to start selling gilts into a market that had quickly become very one-sided. As the market struggled to absorb further sales, gilt market functioning became severely challenged.

On 28 September, the Bank of England announced a temporary and targeted programme of purchases of long-dated gilts to restore orderly market conditions. This programme was subsequently extended to cover the inflation linked market. Following the announcement, yields initially fell back sharply, reducing the scale of margin calls for a period (see Chart 2), providing LDIs and pension funds with time to make the necessary adjustments.  

We estimate that in the short period between the announcement of the ‘Growth Plan’ and the start of the Bank of England intervention, liability-driven investors sold approximately £6 billion of gilts on a net basis, as shown in Chart 3. Gilt sales accelerated towards the later stages of the intervention. By this time, LDI managers had greater clarity from their pension fund investors on the amount of capital that could be raised. Over the entire intervention period, liability-driven investors net sold approximately £37 billion of gilts, around 70% of which were inflation linked.

Chart 3: Cumulative net gilt sales by LDI funds and pension schemes, some of which (£19.3 billion in total) were sold to the Bank of England via dealers ^(a)

Sources: Bloomberg Finance L.P, MiFID data and Bank calculations.

(a) The white dashed line shows cumulative gilt purchases by the Bank of England as part of its temporary gilt market operation.

These gilt sales enabled LDI and pension funds to reduce the leverage of their portfolios. This can be seen in the vast reduction in net repo borrowing shown in Chart 4. Overall, we estimate that liability-driven investors reduced their repo positions by around £25 billion (or 12%) between the 22 September and the end of the intervention on the 14 October. While gilt sales plateaued following the end of the intervention, liability-driven investors continued to reduce their repo borrowing in the following days, making the necessary adjustments to rebuild their resilience and rebalance their portfolios.

Chart 4: Cumulative reduction in net repo borrowing by LDI funds and pension schemes

Sources: SMMD data and Bank calculations.

What role did corporate bond sales play in raising liquidity?

While a lot of analysis of the Autumn 2022 stress has focused on gilts, credit markets also played a key role in the episode. Liability-driven investors, mostly pension funds, sold non-gilt assets held outside of LDI portfolios to generate liquidity. While a broad set of asset classes was involved, we focus here on corporate bond data. Chart 5 shows significant net sales of corporate bonds by pension and LDI funds, totalling around £10 billion, or around 30% of the size of gilt sales over the period.

Chart 5: Cumulative net sales of corporate credit by LDI funds and pension schemes

Sources: MiFID data and Bank calculations.

Wrapping up

This article provides quantitative evidence on the origin of margin calls during the Autumn 2022 stress and how these were managed via sales of assets across bond markets and the unwinding of repo positions. It also illustrates how a combination of granular regulatory data sets can help to deepen understanding of stress events.

Lydia Henning works in the Bank’s Market Intelligence and Analysis Division and Simon Jurkatis, Manesh Powar and Gian Valentini work in the Bank’s Capital Markets Division.

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.

Nikoleta Anesti, Marco Garofalo, Simon Lloyd, Edward Manuel and Julian Reynolds

Understanding and quantifying risks to the economic outlook is essential for effective monetary policymaking. In this post, we describe an ‘Inflation-at-Risk’ model, which helps us assess the uncertainty and balance of risks around the outlook for UK inflation, and understand how this uncertainty relates to underlying economic conditions. Using this data-driven approach, we find that higher inflation expectations are particularly important for driving upside risks to inflation, while a widening in economic slack is important for downside risks. Our model highlights that rising tail-risks can become visible before a turning point, making the approach a useful addition to economists’ forecasting toolkit.

To the mean and beyond: a fan chart story

The Bank of England pioneered the approach of including information on uncertainty and risks around their forecast with their inflation ‘fan chart’ – first published in February 1996 (Chart 1). It remains a staple of the quarterly Monetary Policy Report (MPR) to this day. The ‘fan’ sets out the MPC’s assessment of the outlook for inflation and the risks around it over the forecast horizon. The inner dark red band reflects the ‘central projection’ – the MPC’s view of the most likely outcome for inflation. The lighter bands reflect less likely – but still possible – outcomes. The chart is constructed such that inflation is expected to lie somewhere within the entire width of the fan on 90 out of 100 occasions.

Chart 1: The first inflation ‘fan chart’ (February 1996)

Changes in the size and shape of the fan reflect changes in the MPC’s views on the level of uncertainty and balance of risks. A symmetric widening of the fan to the upside and downside implies a greater degree of overall uncertainty around the outlook. Alternatively, a one-sided widening in the fan above or below the dark red central scenario implies changes in the balance of risks. For example, a widening in the fan above the dark red band implies an increase in the level of risk specifically that inflation might turn out higher than expected.           

The MPC uses a range of statistical tools and judgement to construct its fan chart. There are a number of challenges involved in any forecasting exercise, and such challenges become even starker when trying to construct estimates for risks around the central projection. The issue is that standard statistical tools (eg linear regression) are designed to produce forecasts for the expected, ie mean, path of macroeconomic variables. They typically do not provide a direct estimate of the uncertainty around these paths. While a measure of uncertainty can be constructed by examining historical forecast errors from these types of model, this does not help in understanding which variables drive the uncertainty, nor can it capture changes in uncertainty over time driven by changing economic conditions.

We want to go beyond this approach and explicitly estimate the level and drivers of risk around inflation over time.

A new approach to quantify risks: Inflation-at-Risk

In order to do so, we borrow an approach from recent work in academic and policy circles aimed at monitoring risks to financial stability: ‘GDP-at-Risk’. Like other central banks that have adopted similar approaches, we rely on quantile regression, a statistical tool that allows us to estimate the relationship between a range of indicators and the whole distribution of possible inflation outcomes. Through this, we determine which variables are particularly important, not just for explaining changes in the expected path for inflation, but also in shaping the overall level of risk around that path. We also employ a local-projection framework, which allows us to estimate the level of risk across different forecasting horizons.

We include various macroeconomic indicators that are typically considered important for driving inflation dynamics, specifically: lagged inflation, inflation expectations (for a combination of households and corporates), the estimated output gap, and world export prices. Our choice of variables mirrors those that feature in an Open-Economy Philips Curve. The quantile regression model allows us to investigate how changes in each of these variables affect the whole distribution of possible inflation outcomes across a range of forecast horizons. To estimate our model we rely on data from a number of advanced economies (US, UK, euro area and Japan) with a variety of historical inflation experiences.

Results: tales of tails

Among our main results, we find that inflation expectations and the output gap are particularly important for shaping risks around the central projection in the near term.

Chart 2 shows the estimated coefficients from those two variables across five different quantiles (ie different parts of the inflation distribution) reported on the x-axis. They show how the outlook for future inflation one quarter ahead – and the risks around it – respond to changes in each of the variables. If the line for a coefficient is broadly flat and non-zero, it means that changes in the corresponding variable are associated with a shift in the whole distribution. In contrast, if the line is not flat, then changes in the variable contribute to a change in the balance of risks. For example, the variable may have a larger effect on the left or right tail of the distribution than at the mean. These results refer to the predicted conditional inflation distribution one quarter ahead, but the picture over other short-run horizons is very similar.

We find that higher inflation expectations today contribute to an increase in the central forecast for inflation next quarter, but they also shift the balance of risks to the upside, increasing the likelihood of inflation coming out above the central projection. On the other hand, a more negative output gap (ie a greater degree of economic ‘slack’) contributes to a reduction in the central projection for inflation while simultaneously shifting the balance of risks to the downside.

In contrast to these two variables, we find lagged inflation and world export prices have significant effects over the entire the predicted inflation distribution. Higher past inflation or inflationary pressures from the rest of the world contribute to an increase in the central projection for inflation without affecting the overall balance of risks the forecast.

Chart 2: Inflation expectations, the output gap and the balance of risk

Notes: Coefficient estimates across quantiles at the one quarter ahead horizon. Blue line shows point estimates and shaded area is 68% confidence interval. Model is estimated using data from UK, US, euro area and Japan from 1995–2022.

We can also use the model to produce forecasts for possible UK inflation outcomes. Chart 3 shows the estimated distribution of possible inflation outcomes one quarter ahead for each period over 2019–22 from our model. Notably the model estimates a rise in upside inflation risk over the later period of 2020 – the model thus detects upside risks early on that then materialised over 2021.

Chart 3: Model forecasts for UK inflation over Covid

Notes: One quarter ahead probability distributions for year-on-year inflation (%); distributions fitted from quantile-regression output using non-parametric approach.

Conclusion

Our analysis highlights how quantile regression can be used to assess the level and drivers of risks around the inflation outlook. We show that higher inflation expectations matter more for upside risks to inflation, while slack is more relevant for downside risks in the near term. Our model picks up upside inflation risks rising steadily over the course of 2020 before eventually materialising in 2021. Thus, this framework is particularly well suited for calibration of fan charts produced by central banks and policy institutions.

Nikoleta Anesti works in the Bank’s Current Economic Conditions Division, Marco Garofalo and Julian Reynolds work in the Bank’s Global Analysis Division, Simon Lloyd works in the Bank’s Monetary Policy Outlook Division and Edward Manuel works in the Bank’s Structural Economics Division.

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.

Marcus Buckmann, Galina Potjagailo and Philip Schnattinger

Understanding the origins of currently high inflation is a challenge, since the effects from a range of large shocks are layered on top of each other. The rise of UK service price inflation to up to 6.9% in April might potentially reflect external shocks propagating to a wider range of prices and into domestic price pressures. In this blog post we disentangle what might have contributed to the rise in service inflation in the UK using a neural network enhanced with some economic intuition. Our analysis suggests that much of the increase stems from spillovers from goods prices and input costs, a build-up of service inflation inertia and wage effects, and a pick-up in inflation expectations.

Linear models can face limitations in explaining large, unprecedented fluctuations in inflation. At the same time, machine learning tools have become increasingly popular in forecasting and monitoring inflation and GDP growth. Such models can accommodate complex non-linearities and a larger number of variables, which makes them particularly appealing during periods of swift turning points and large shocks, and when exploring granular data. We employ a neural network Phillips curve model proposed in a recent working paper by Philippe Goulet Coulombe. The method allows us to extract signals for inflation from many variables, fed into the model according to the familiar building blocks of the Phillips Curve. Whereas the working paper also proposes a strategy to separate out a Phillips curve slope from an output gap estimate, our focus lies in the contributions from different Phillips curve components to the forecast and the signals that the model exploits over time rather than a structural identification.

An economically intuitive neural network

We use a neural network to approximate the Phillips Curve. Neural networks learn to recognise patterns in large data sets and make predictions. They pass data through several layers of interconnected nodes, where each connection between nodes is a parameter that is learned from data. For example, when predicting inflation, the parameters are calibrated to minimise the difference between the model’s final prediction of inflation and observed inflation. Most of the recent advances in artificial intelligence, such as ChatGPT, are based on huge neural networks with billions of parameters.

The model we use is of relatively small size, but still exploits a large set of variables and is able to learn a rich array of non-linearities. Rather than letting the model be purely driven by data, we impose economically motivated reduced-form assumptions by feeding the network with data split into sub-categories. Each sub-network learns to distil the information contained in the input variables and their non-linear interlinkages into a final neurons, or ‘latent’ components relevant for forecasting service inflation. The components represent the building blocks of a Phillips curve (Chart 1),  parcelling up the drivers of inflation into past inflation dynamics, inflation expectations, a domestic output gap, and international prices to account for global price pressures affecting inflation in an open economy. For past domestic inflation, we further distinguish between a) domestic input costs and goods price inflation to reflect indirect effects from goods prices into services via for instance input-output linkages, and b) nominal inertia that captures past service price inflation, output costs and pay growth. Each component is derived from a set of aggregate (eg: unemployment rate, total service inflation) and disaggregate series (eg: two-digit industry output), also including lags and moving average transformations of each series. For example, rather than using one particular output gap measure, we use a range of real activity indicators that are likely to contain information about the output gap. The sum of the latent components forms the model’s overall inflation forecast.

Chart 1: A neural network with Phillips Curve structure

We estimate the model on a quarterly basis over the period 1988 until the first quarter of 2023, and after training over the first 12 years of the sample using the out-of-bag approach, we then run predictions in an out-of-sample exercise. We evaluate the model out-of-sample and re-calibrate the model every quarter to incorporate incoming data.

Service inflation forecast decomposition into Phillips curve contributions

The model forecasts service inflation relatively closely, with a good out-of-sample performance. Chart 2 decomposes the forecast into contributions from Phillips curve components. The forecast (black line) reproduces the main fluctuations in service inflation (grey dashed line in the left panel) over the sample period such as the surge during the late 1980s, the decline during the Covid-19 pandemic and the recent rise. According to the model, the surge of inflation during the end-1980s was associated with elevated inflation expectations (light and dark purple bars) and domestically generated inflation (light and dark green bars). After the subsequent policy interventions and establishment of inflation targeting, all contributions come down and the role of inflation expectations is diminished for the rest of the sample period.

Chart 2: Decomposition of service inflation via the neural Phillips Curve

Notes: 1-q ahead out-of-sample forecasts from 2000, out-of-bag cross validation up to 1999. Quarterly growth rates are annualised, with quarterly fluctuations smoothed out. Contributions to forecast (solid line) from PC components, relative to the mean of service inflation of 3.3% during 1997–2019. Dashed grey lines shows service inflation outturn. Left: zoomed in decomposition since 2020 Q1. 2023 Q2: current forecast period.

Throughout, we observe spill-overs from tradable goods into service inflation via input-output linkages and cost-push effects (light green bars), in particular during the 2000s and in 2014–15, yet these have typically not translated to inertia in service prices and pay growth (dark green bars). The effects from good prices and input costs were weak during the pandemic, reflecting that service inflation was falling but goods price inflation was being pushed up by supply shortages. International prices (blue bars) contribute only little overall, which is in accordance with service inflation being a measure of domestic price pressures not directly affected by terms of trade shocks (we find stronger effects for headline inflation).

The intrinsic dynamics of service prices and pay growth, or nominal inertia, have shifted infrequently in the past. During the decade following the financial crisis, nominal inertia (dark green bars) dragged on service inflation, likely due to timid wage growth and weak dynamics of service inflation during that period. But this has now reversed with the most recent rise in services prices. Since the beginning of 2022 the model detects jointly increased contributions from spillovers from input cost and from nominal inertia. These two contributions have been the largest at play. Since mid-2022, the inflation expectations contributions (light and dark purple bars) have also picked up, but the size of the effect remains relatively small compared to the early episode in the early 1990s, suggesting that expectations remain anchored. The output gap effect has contributed slightly positively, but the size of this effect has not grown since the initial post-pandemic recovery.

Recently strong synchronised signals from input costs and wage growth

We further dissect the signals that the model exploits from each input variable over time via Shapley value heat maps, derived from the out-of-sample exercise since 2000 (Chart 3). Apart from the year 2008, signals from variables related to goods prices and input costs have rarely been both as strong and synchronised as in the recent period (Panel a). And during earlier periods, positive signals from input and energy costs did not translate into rises of nominal inertia (Panel b). Apart from some variation in output and accommodation prices, all signals in the nominal inertia component were jointly pointing to below-mean service inflation, with the strongest negative signals stemming from average weekly earnings and lagged service dynamics. Since early 2021 this started to shift, initially via positive signals from earnings, followed by accommodation and catering, two industries where prices recovered post-pandemic. Since the second half of 2022, all nominal inertia signals have been synchronised and clearly positive.

Chart 3: Signals to the service inflation forecasts over time

Notes: Signals are derived as Shapley values for each quarter and variable. Darker red indicates stronger positive signal, darker blue indicates stronger negative signal. Indicators with stronger average signals are ordered on top, although relevance can shift over time. Panel c: Shapley values from additional 11 sectoral output series not shown for readability.

According to the model, the inflation-relevant output gap has contributed only slightly to inflation across most of the sample period (Panel c). During the global financial crisis in 2009, the model reads mostly positive signals and misses the fall in service inflation, albeit the Bank of England’s output gap series itself provided a negative persistent signal. Challenges in empirically detecting a strong Phillips curve relationship are not new and might be related to identification issues around the role of monetary policy. During the pandemic, our model detects a strengthened role of the output gap, with drag from the output gap explaining much of the fall of service inflation. Signals also became temporarily more dispersed, in line with the heterogeneous nature of the Covid-19 shock. Recently, signals have been mostly positive, but not very pronounced.

Summary and implications

The neural network Phillips curve model suggests that the recent rise in UK service inflation has been associated with a rise in nominal inertia related to lagged service inflation dynamics and pay growth. This component has fluctuated little and represented a drag on service inflation in the past. Its rise could reflect second-round effects affecting domestic inflation beyond the direct effects from external shocks, although the degree to which we can distinguish between the two in a non-structural model is limited. As in any empirical model, the number of observations following high inflation is limited, and so uncertainty around model predictions for these periods is higher.

Marcus Buckmann works in the Bank’s Advanced Analytics Division, Galina Potjagailo and Philip Schnattinger work in the Bank’s Structural Economics Division.

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.