Implications of current wage inflation − speech by Jonathan Haskel

Speech

Introduction

It’s my pleasure to be here at the Bank of England Watchers’ Conference, and my thanks to the Qatar Centre for Global Banking & Finance and the Macro Money and Finance society for organising the event and inviting me to participate.

Since the panel is on the labour market, I want today to set out briefly why I think wage growth is as high as it currently is.

But first, why should a monetary policy maker care about wage growth?

CPI inflation in the year to August 2023 was 6.7%. This is much lower than the 10-11% we had over the winter, but still well above the MPC’s 2% target. The MPC expects inflation to fall sharply over the rest of 2023 and the start of 2024, largely as past increases in energy and food prices drop out of the annual comparison.

But we cannot be complacent. Our remit is to use monetary policy to ensure that CPI inflation returns to the 2% target sustainably in the medium term. This we will do.

One risk to our inflation forecast comes from wage growth. To be clear, wage growth is a good thing, justly rewarding people for their efforts. But from an inflation point of view, wage growth that is unmatched by productivity gains risks inflation, hence the interest in wage growth.

Measuring wage growth

Chart 1 (solid line) shows average private sector regular wages, as measured by the Average Weekly Earnings (AWE) regular pay series, grew by 8.0% in the year to June-August 2023. This is down very slightly from the growth rates in the past couple of months, but roughly equal to the fastest rate of growth since the early 1990s. It also shows (dashed line) fitted values from a wage equation, which I shall return to shortly.

While Chart 1 shows official ONS data on wage growth, the November 2023 Monetary Policy Report makes clear that there are also other sources of data which the MPC monitors. While all measures point to unusually high levels of wage growth currently, there are some subtle differences between them. Key amongst these is the recent trajectory.

Chart 1: Annual growth in private sector regular wages, data and fitted values

Quarterly, 1990 Q1 to 2023 Q2 ^(a)

Footnotes

• Source: ONS, Bank of England, author’s calculations.
• (a) Notes: Data are quarter on same quarter a year ago growth rates in private sector regular pay AWE. From 2001 onwards this is based on data published by ONS, with the exception of 2020 Q1 to 2022 Q3 which is based on an estimate of “underlying” pay constructed by Bank of England staff to strip-out compositional and furlough effects. Before 2001 it is based on a modelled backseries produced by Bank of England staff. See Appendix for more details. Furlough period shaded (2020 Q1 to 2021 Q3). Latest data point 2023 Q2.

While other measures, such as those from the tax system (HMRC Real Time Information) and the Decision Maker Panel survey of businesses, show a high level of wage growth, they do not show an increase in growth rates in recent months. That is, they show annual wage growth relatively constant at about 7% since early 2023 (see Chart 2.13 in the November 2023 MPR).

By contrast, the AWE pay growth data in Chart 1 has increased in recent months. So what can we say about these data?

First, the source of the wage data, namely the AWE, does not depend on the Labour Force Survey (LFS), which has been suffering from declining household response rates. The AWE, by contrast, depends on a survey of businesses, known as the Monthly Wages and Salaries Survey (MWSS). Chart 2 shows the response rates for the MWSS and LFS since 2013. While response rates for the LFS have fallen steadily, and especially so during and since the pandemic, the response rates for the MWSS are high and stable.

Second, MWSS surveys only firms with 20 or more employees. ONS makes an adjustment for smaller firms using data from the Annual Survey of Hours and Earnings (ASHE) – another survey of businesses, which covers firms of all sizes. While it is well-known that smaller firms tend to pay a lower level of wages than larger firms, it is not clear that they would have different rates of wage growth. If smaller firms had different wage growth rates than larger firms, then the MWSS may give the wrong signal on overall wage growth. However, the divergence between small and large firms would have to be quite large to make a substantive difference to AWE growth. While plausible, this does not seem likely to be a substantive issue.^{footnote [1]}

Chart 2: MWSS and LFS response rates

Quarterly, 2013 Q1 to 2023 Q2 ^(a)

Footnotes

• Source: ONS – LFS performance and quality monitoring reports and AWE bulletins, author’s calculations.
• (a) Notes: MWSS response rates reported in monthly bulletins from May 2020 (survey for March 2020) onwards. Data show calendar quarters – for MWSS, the average of the associated months; for LFS, three-month periods. Before 2020 Q2, the dashed line shows the target and pre-covid norm response rate of MWSS of 83%, as reported by ONS. Latest data point April-June 2023.

Third, AWE doesn’t cover the income of the self-employed. Data on self-employment earnings are more difficult both conceptually and in practice than data on employee earnings. While this is a potential short-coming, it should be borne in mind that the other wage measures mentioned previously also do not include the self-employed, so this explanation for the apparent divergence of AWE relative to other measures. The
self-employed are also a relatively small share of the workforce, accounting for only around 13% of all workers in the latest data, and less than 10% of total labour income in the economy.^{footnote [2]}

Fourth, the AWE distinguishes between “total pay” and “regular pay”, where regular removes bonus pay and arrears. Bonuses can be volatile and distort the pay data, so regular pay measures can better reflect the underlying trend. In fact, private sector total pay growth is currently lower than private sector regular pay growth, since bonuses payments have fallen since last year. So bonuses are not the reason for high currently pay growth. I note that other measures of pay, such as those from the tax system, do not distinguish between bonuses and regular pay, and annual growth in the AWE total pay measure matches the signal from the tax data well.

Fifth, it seems unlikely to me that so-called “cost of living payments” are pushing up the current regular pay data. The exact treatment of these in the data is unclear, as it will depend on how the responding firms classify such payments – whether as bonuses, arrears, or regular pay. However, it seems unlikely that such payments are currently elevated – Agents’ intelligence^{footnote [3]} suggest these were most prevalent in late 2022 and early 2023, when energy prices and CPI inflation were highest. For such temporary payments to be inflating annual growth rates in summer 2023, they would have to have been paid during summer 2023, at a time when inflation and energy prices were falling, which seems unlikely.

Sixth, since AWE is essentially a mean average across employees (i.e. an estimate of total pay of all employees, divided by an estimate of the total number of employees), a few very high paying businesses (or highly paid individuals) could be skewing the results. Again, while plausible, this effect would have to be very large. For instance, the total pay of Premier League footballers (excluding bonus payments) is in the region of £30m per week, which accounts for around 0.2% of total pay in the UK. This isn’t enough to make a material impact on the aggregate growth rates.

Thus I see no reason based on current evidence to suggest measurement problems are greatly exaggerating current wage growth. What then explains high current wage growth?

Explaining wage growth

A simple wage growth model

To get a handle on the strength of wage growth in an economic framework, we estimated a wage equation in the spirit of the famous Phillips curve. Our dependent variable will be the growth of private sector regular pay, as described previously.

We will try to explain the growth in wages on a quarterly basis, compared with the same quarter of the previous year, such that these are rolling annual changes. Our explanatory variables will be as follows, with details of the measures used in the Appendix.

1. Trend productivity growth. In the long run, wages grow as productivity grows.
2. Wage growth from the previous year. It takes time for wages to adjust to new conditions in the labour market
3. Inflation expectations. Quite understandably, wage bargainers will strive for high nominal wages if they expect higher nominal prices.
4. The unemployment gap. Low unemployment signals a “tight” labour market making firms more willing to offer higher wages to avoid having to replace workers. But that unemployment has to be relative to a baseline or "underlying” rate, which is the unemployment rate to be expected as part of the natural process by which workers take time to find vacancies and that consistent with wage and price stability. This “underlying” rate goes under various names in the economics literature^{footnote [4]} and we refer to it as U* (U star) for convenience.

What is this baseline “underlying” unemployment rate, U*? Chart 3 sets out a simple approach to this, which is simply to fit a statistical filter through the actual unemployment rate. The difference between actual unemployment and this measure of U* is the “unemployment gap”, shown in the grey bars.

This is an intentionally simple approach, and many more sophisticated methods exist – indeed, the MPC’s view of U* is motivated by a range of approaches (see for instance Inflation Report February 2018, Box 4). For the purposes of these short remarks, a more practical empirical approach is sufficient for the points I wish to make.

Chart 3: Unemployment, U*, and the unemployment gap

Quarterly, 1990 Q1 to 2019 Q4 ^(a)

Footnotes

• Source: ONS, author’s calculations.
• (a) Notes: Unemployment gap (U-U*) = Unemployment rate (U) minus Underlying unemployment rate (U*). U* estimated using a Hodrick-Prescott (HP) filter with smoothing parameter 8000 on unemployment rate. See Appendix for more details. Latest data point 2019 Q4.

A sharper test of this model is to ask: how does this model predict wage growth “out of sample” from 2020 onwards? Chart 4 shows the prediction for wage growth with this model, along with the actual data. The model predictions are the orange bars (“Explained”), and actual wage growth is the white line, with the difference being the blue bars (“Unexplained”). The pandemic is the shaded area, where wage growth is very hard to predict given the unusual circumstances – sharp changes in the economy, the furlough scheme, etc.

After the end of the furlough scheme, from 2021 Q4 onwards, the Chart shows actual wage growth (the white line) has been far higher than predicted by the model (orange bars), leaving large positive residuals (blue bars). Since the start of 2022 the positive residuals have averaged 2.0pp. Indeed, in 2023 Q2, the latest data point, wage growth was 3.3pp higher than the model prediction.^{footnote [5]}

Chart 4: Annual wage growth and model decomposition

Quarterly, 2019 Q1 to 2023 Q2 ^(a)

Footnotes

• Source: ONS, Bank of England, author’s calculations.
• (a) Notes: Explained is the fitted values from the wage equation described in text and in Appendix; Unexplained is difference between fitted values and actual (i.e. residuals). Furlough period shaded (2020 Q1 to 2021 Q3). Latest data point 2023 Q2.

What might have changed in the economy?

To be clear then, the puzzle in this context is not high wage growth per se. Rather it is high wage growth relative to that which can be explained using past relationships. Thus the question is: does the breakdown of the model signal that the economy has changed?

Various factors may be relevant. Inflation expectations could have increased more than measured, perhaps related to the distribution of expectations rather than only the median of the distribution (Mann, 2023; Reis, 2021). Workers could be placing greater weight on past unexpected inflation than on future expected inflation, akin to the “catch-up” mechanism in Bernanke and Blanchard (2023). Relatedly, households may become more attentive to inflation when it is high.^{footnote [6]} This, in turn, could be interacting with a tight labour market, by which workers are better able to protect their real wage level in the face of high inflation when their have a high degree of bargaining power. There could also be non-linearities in the relationship between labour market tightness and wage growth, i.e. we are on a steep portion of the wage Phillips curve. All these explanations would point towards some change in the structure of the labour market, and more work on all of these topics is warranted. A further consideration is the role of the National Minimum Wage (which is usually increased by around the rate of inflation) and impacts on the wages of workers just above the threshold through maintaining pay relatives.^{footnote [7]} All of these factors could be relevant. For now, I want to consider just one explanation – an increase in U*.

It is worth noting that a model using a different measure of labour market tightness, namely the job vacancies-to-unemployment (V/U) ratio, fits the data better during the post-pandemic period. While the unemployment gap and the V/U ratio track each other closely in the past (see Appendix), V/U increased much more during 2022 than the unemployment gap fell. Thus, V/U signalled a greater degree of tightness than the unemployment gap, and might therefore better explain high current wage growth. Chart 2.14 in the November 2023 MPR shows that using the V/U ratio in this way can explain wage growth up to 2023 Q1. However, it cannot explain the persistently high level of wage growth in 2023 Q2, even as the V/U ratio declines. If the continued, and indeed increased, strength in pay growth in 2023 Q2 in the AWE cannot be explained by measurement issues (as discussed above and in the Appendix), then what can explain it?

Has U* risen?

A natural explanation for the under-estimate of wage growth in the model is that U* has risen since the end of 2019. Specifically, if U* has increased, then the unemployment gap is actually larger (more negative) than we have included, representing a tighter labour market, and thus should be exerting more upward pressure on wage growth. So we can ask: is it plausible that U* has risen?

One way to start thinking about this is to ask: what would be the U* required to eliminate the residuals from the model (on average) since the start of 2022? Chart 4 shows that the model under-predicted wage growth consistently since the start of 2022, by an average of 2.0pp. To eliminate these residuals, i.e. for the model to produce predictions of wage growth that match true wage growth on average since early 2022, U* would have to be about 1.7pp higher, in the absence of any other changes to the model. That would make U* around 6% rather than “just above 4%” – the MPC assumption before the pandemic (Inflation Report February 2018, Box 4).

Is this a plausible rise? A number of points are worth making. First, recent US work (Crump et al., 2022; Ball, Leigh and Mishra, 2022; Blanchard, Domash and Summers, 2022) estimates increases of U* during the pandemic of between 1.3pp and 2.0pp. Our estimate of 1.7pp sits neatly in this range.

Second, Chart 3 suggests that U* was above 6% for several years in the aftermath of the global financial crisis and higher still in the 1990s. While a rise to around 6% would be a sharp increase, it would take U* to a level not inconsistent with the past.

That said, this is likely to be an upper bound estimate for the increase in U*, given that other models have smaller residuals (but still have unexplained wage strength in recent quarters) – for instance, the wage model with the V/U ratio in the November 2023 MPR, Chart 2.14.

Explaining an increase in U*

In the latest Monetary Policy Report (November 2023), the MPC has raised its estimate of the medium-term equilibrium rate of unemployment, albeit by much less than the 1.7pp suggested above.^{footnote [8]} But if U* has risen, what has caused that? One possibility is that labour “mismatch” has worsened: for example, workers with available skills are not so well suited to the available vacancies. A summary measure of the frictions between the unemployed and the vacancies they seek is the position of the Beveridge Curve.

Chart 5 shows a scatter plot of the vacancy rate and unemployment rate. The points form a loose ‘curve’ from top-left to bottom-right – vacancies tend to be high when unemployment is low, and vice versa. I have discussed these data in a previous speech (Haskel, 2021).

Since the end of the furlough scheme, the points lie at the top-left of the chart (coloured purple) – that is, high levels of vacancies and low levels of unemployment. The latest five data points (covering April to August 2023) are in gold. These show a decline in vacancies and a small increase in unemployment so far. The current level of unemployment would historically have been associated with a vacancy rate of just under 2.5%, but instead we currently have a vacancy rate of around 3%.

This could be consistent with being on a very steep part of the Beveridge curve, or a shift out/right of the Beveridge curve, or both.^{footnote [9]} A rightward shift of the curve would indicate that fewer matches (hires) were made for a given constellation of the unemployed and open vacancies. This would generate higher wage growth for a given level of unemployment, consistent with an increase in U*.

Chart 5: Vacancy rate and unemployment rate

Monthly, June 2001 to August 2023 ^(a)

Footnotes

• Source: ONS, author’s calculations.
• (a) Notes: Vacancy rate defined as the number of vacancies divided by the active labour force (i.e. employed plus unemployed). The unemployment rate is defined analogously, and covers all people aged 16 and over. Data points reflect three months to given month for unemployment. Last data point covers June to August 2023, and uses the unemployment rate published by ONS on 26 October 2023. Points for March 2020 to September 2021 adjusted to include 10% of furloughed workers as unemployed – see Haskel (2021) for more.

A second possibility results from changes in economic openness. As the economist Dani Rodrik has pointed out, a more open economy with more capital mobility improves the bargaining position of firms, since they can credibly move production elsewhere in response to wage rises.^{footnote [10]} The UK data are noisy, but most suggest a decline in openness since the pandemic. This makes the demand for labour more “inelastic”, improving the bargaining position of workers.

Some preliminary work, set out in the Appendix, supports this idea. We looked at the responsiveness of employment to wages across 11 European countries over the last 20 years. Controlling for a number of factors, we found the elasticity of employment to wages to be around -0.7 at average openness, rising (in absolute value) to -0.85 with a one standard deviation decrease in openness. This might have contributed to a rise in U*.

Conclusion

Recent wage data is consistent with the idea that the “underlying” unemployment rate, U*, has risen, and that is reflected in recent MPC judgements. In real time it is hard to be precise on the causes of such a rise and future work on better and longer runs of data will be important. One possible cause is the decline in openness.

But the main message is that the capacity of the labour market to match workers with available vacancies appears to have deteriorated. With an impaired labour market, interest rates would have to remain higher for longer than would otherwise be the case.

My thanks to Josh Martin for help with the material for this speech. Thanks also to Tom Key for help with data and insights on Beveridge curves, and to Harvey Daniell on wage data. Thanks to Andrew Bailey, Sarah Breeden, Natalie Burr, Fabrizio Cadamagnani, Alan Castle, Harvey Daniell, Chris Duffy, Mike Goldby, Tom Key, Huw Pill, Doug Rendle, Martin Seneca, Brad Speigner, and Danny Walker and for helpful comments. Any errors are my own.