Publication - The labour market - speech by Michael Saunders
This talk focusses on the labour market, and in particular the limited response of wage growth to falling unemployment. At first glance, the labour market now looks very tight. The jobless rate is down to 4.8%, slightly below both the 2000-07 average and the MPC’s estimate of the equilibrium rate, which are about 5%1 (see figure 1).
The jobless rate has only been below current levels for a few months in the last 40 years2 . The short-term jobless rate is the lowest since data began in 1992. The number of job vacancies is around a record high, and the ratio of unemployment to vacancies matches the 2005 low (see figure 2).
However, even with relatively low unemployment, average weekly earnings growth remains modest, at 2-3% YoY3 . Unit labour cost growth is perhaps still slightly below the pace consistent with the inflation target over time4 . There is little sign of significantly higher pay growth for 2017 (see figure 3).
The labour market models used by most forecasters, which presumably are based on past behaviour, have not done well in explaining and forecasting the modest trend in pay growth. In recent years, pay growth has repeatedly undershot consensus expectations, OECD forecasts and BoE forecasts5 despite falling unemployment (see figures 4 and 5). Indeed, the UK has seen the biggest average undershoot in pay growth of any country compared to the OECD’s forecasts in recent years (see figure 6).
Most forecasts have looked for the relation between unemployment and pay growth to return to something like that seen before the 2008-09 recession, whereas in practice this relation seems to have shifted downwards again (see figure 7)
Of course, all forecasts are fallible. But such repeated and widespread forecast errors are interesting because they hint that the underlying behavior of wage growth has changed. This is also evident in figure 8, showing the gap between average pay growth in 2014-16 and projections based on simple wage Phillips curves (i.e. regression of pay growth on unemployment) fitted over 2000-13 across OECD countries – with a marked undershoot in the UK.
I suspect the subdued trend in wage growth in part reflects structural changes which these models may not fully capture, including greater labour market flexibility and insecurity, extra labour supply, increased under-employment, broader educational attainment, and changes to the tax and benefit system. These changes probably imply greater downward pressure on pay growth for any particular jobless rate than previously.
The BoE’s 5% estimate of the equilibrium jobless rate is based on a mix of empirical observation from the pre-crisis period (the last time the economy had a near-zero output gap), time series analysis and a labour market search and matching framework. This search and matching framework consists of a Beveridge Curve (BC), which shows combinations of unemployment and vacancies for a particular level of search efficiency, and a job-creation curve (JCC), which shows V-U combinations of equal profit for firms. In this model, higher unemployment makes it more attractive to create vacancies and hire people as the match rate is higher, but there are diminishing returns to creating vacancies as the jobless rate falls and the labour market becomes ‘congested’. The intersection of these lines is the equilibrium jobless rate. The gap between the actual jobless rate and this equilibrium is used as an input to a standard wage Phillips curve model (with some role for inflation expectations).
There was an apparent rightward drift in the Beveridge curve in 2013-14 (see figure 10), which suggested reduced labour market matching efficiency7 . Recent data bring us roughly back to the 2002-05 trends. Hence, assuming the job creation curve has not shifted, one could infer that the natural jobless rate is about the same as the pre-crisis period.
The search and matching model has advantages of solid theoretical underpinnings. However, as noted, the forecasts for pay growth derived from this natural rate estimate have not been very accurate in recent years.
One issue is that the derived natural jobless rate has a sizeable margin of error, especially because parameters of the job creation curve are poorly observed. This also means that these models are quite slow to reflect structural changes. And, as Peter Diamond argues, the US Beveridge curve often shifts out as unemployment starts to fall, but these shifts have not been useful predictors of the jobless rate that the economy attained at the end of subsequent expansions8 . In addition, the search and matching model aims to measure a long-run equilibrium. At any particular time, this may not necessarily be the same thing as the NAIRU, defined loosely as the jobless rate consistent with on-target inflation once shocks from the exchange rate etc have worked through9 . For example, increased labour supply might shift the NAIRU for a period, especially if new entrants behave differently to the existing pool of workers.
page source http://www.bankofengland.co.uk/
