Inflation beliefs

One of the biggest apparent puzzles in macroeconomic policy today is why inflation remains so low when the unemployment rate is at multi-decade lows. The evidence clearly suggests that the trade-off between inflation and unemployment is far weaker today than it used to be or, as the economics profession would have it, the Phillips curve is flatter than it once was (here). But as the academic economist Roger Farmer has pointed out, the puzzle arises “from the fact that [central banks] are looking at data through the lens of the New Keynesian (NK) model in which the connection between the unemployment rate and the inflation rate is driven by the Phillips curve.” But what if there were better ways to characterise the inflation generation process?

Originally, the simple interpretation of the Phillips curve suggested that policymakers could use this trade-off as a tool of demand management – targeting lower (higher) unemployment meant tolerating higher (lower) inflation. However, much of the literature that emerged in the late-1960s/early-1970s suggested that demand management policies were unable to impact on unemployment in the long-run and that it was thus not possible to control the economy in this way. The reason is straightforward – efforts by governments (or central banks) to stimulate the economy might lead in the short-run to higher inflation, but repeated attempts to pull the same trick would result in a response by workers to push for higher wages which in turn would choke off labour demand and raise the unemployment rate. In summary, government attempts to drive the unemployment rate lower would fail as workers’ inflation expectations adjusted. One consequence of this is that the absence of any such trade-off implies the Phillips curve is vertical in the longer-term (see chart).

Another standard assumption of NK models, which are heavily used by central banks, is that inflation expectations are formed by a rational expectations process. This implies some very strict assumptions about the information available to individuals and their ability to process it. For example, they are assumed to know in detail how the economy works, which in modelling terms means they know the correct structural form of the model and the value of all the parameters. Furthermore they are assumed to know the distribution of shocks impacting on the economic environment. Whilst this makes the models intellectually tractable, it does not accord with the way in which people think about the real world.

But some subtle differences to the standard model can result in significant changes to the outcomes, which we can illustrate with regard to some recent interesting work by Roger Farmer. In a standard NK model the crucial relationship is that inflation is a function of expectations and the output gap, and produces the expected result that the long-run Phillips curve is indeed vertical. But Farmer postulates a model in which the standard Phillips curve is replaced by a ‘belief’ function in which nominal output in the current period depends only on what happened in the previous period (known as a Martingale process). Without going through the full details (interested readers are referred to the paper), the structure of this model implies that policies which affect aggregate demand do indeed have permanent long-run effects on the output gap and the unemployment rate, which is in contrast to the standard NK model. Moreover, Farmer’s empirical analysis suggests that the results from models using belief functions fit the data better than the results derived from the standard model.

The more we think about it, the more this structure makes sense. Indeed, as an expectations formation process, it is reasonable to assume that what happened in the recent past is a good indicator of what will happen in the near future (hence a ‘belief’ function). Moreover, since in this model the target of interest (nominal GDP) is comprised of real GDP and prices, consumers are initially unable to distinguish between real and nominal effects, even though any shocks which affect  them may have very different causes. In an extreme case where inflation slows (accelerates) but is exactly offset by a pickup (slowdown) in real growth, consumers do not adjust their expectations at all. In the real world, where people are often unable to distinguish between real and price effects in the short-term (money illusion), this appears intuitively reasonable.

All this might seem rather arcane but the object of the exercise is to demonstrate that there is only a “puzzle” regarding unemployment and inflation if we accept the idea of a Phillips curve. One of the characteristics of the NK model is that it will converge to a steady state, no matter from where it starts. Thus lower unemployment will lead to a short-term pickup in wage inflation. Farmer’s model does not converge in this way – the final solution depends very much on the starting conditions. As Farmer put it, “beliefs select the equilibrium that prevails in the long-run” – it is not a predetermined economic condition. What this implies is that central bankers may be wasting their time waiting for the economy to generate a pickup in inflation. It will only happen if consumers believe it will – and for the moment at least, they show no signs of wanting to drive inflation higher.

Mr Phillips is resting

Markets are increasingly concerned that central bankers may be about to take away the punch bowl rather earlier than they had previously anticipated. The Bank of Canada was the latest central bank to tighten policy, raising rates by 25 bps this week for the first time since 2010. There is also increased nervousness regarding the policy intentions of the ECB and BoE. But whilst there are good reasons for taking away some of the emergency easing put in place in the wake of the financial crash of 2008-09, it is proving much harder to justify tightening on the basis of inflation than most had expected.

This is a particular problem for the Fed which has nudged up the funds rate in four steps of 25 bps over the past 18 months, but is reliant on signs of higher inflation to justify ongoing policy normalisation. US core CPI inflation, which was running above the Fed’s 2% target rate last year, slipped back to 1.7% in May and June and is thus at the bottom end of the range in place since 2011. Wage inflation has also picked up, but here too the acceleration has been modest, with hourly earnings running at an annual rate of 2.8% in June which is only 0.5 percentage points higher than the average of the last three years.

For an economy which is running close to what appears to be full employment, this might appear rather surprising. But the headline unemployment rate, currently 4.4%, understates the degree of slack in the US labour market. The so-called U6 rate which adds in “marginally attached” workers – defined as “those who currently are neither working nor looking for work but indicate that they want and are available for a job and have looked for work sometime in the past 12 months” – is still at 8.6%. This is slightly higher than the previous cyclical trough in 2007 when it reached 8.0%, and significantly above the low of 6.8% recorded in October 2000. Arguably, therefore, the jobless rate can still fall a little further before wage and price inflation starts to become more of an issue.

The UK shows a similar – indeed, perhaps more extreme – picture with the unemployment rate in the three months to May at its lowest since 1975 whereas the rate of wage inflation, at 1.8%, is a full percentage point below that recorded last November. As in the US, there is a significant amount of spare capacity in the labour market. Currently, 12% of those working in part-time employment are doing so because they cannot find full-time employment. Whilst this is down from a peak of 18.5% in 2013, it is still higher than the 8-9% range recorded before the recession of 2008-09 and points to a certain degree of involuntary underemployment. This in turn suggests that there have been structural changes in the labour market which have impacted on the traditional relationship between headline unemployment and wage inflation.

For many decades, economists have focused on the negative relationship between wage inflation and unemployment first postulated by Bill Phillips in the 1950s. In its simplest form, this suggests that policymakers face a trade-off between unemployment and inflation. In practice, the relationship holds only in the short-term, if at all. What is notable, however, is that in the UK and US there has been a flattening of the curve in recent years, suggesting that any negative relationship between wages and unemployment is even weaker today than in the past. 

This is illustrated for the UK in the chart below, based on an idea presented in Andy Haldane’s recent speech entitled “Work, Wages and Monetary Policy.” The chart shows the trend derived from a linear regression of wage inflation on the unemployment rate over various periods. Two features are evident: Most obviously, the line has moved down reflecting the fact that over time inflation in the UK has fallen. But it is also notable that the slope of the line has become shallower. In other words, UK wage inflation has become less sensitive to changes in the unemployment rate. To illustrate the implications of this, we assess the wage inflation rate consistent with an unemployment rate of 5.5% and how this would change if unemployment fell to 4.5% (current levels).

The results are shown in the table (below). Simply put, an unemployment rate of 5.5% would be associated with wage inflation of 14% on the basis of the relationship over the period 1971-1997, falling to 4.1% between 1998-2012 and just 2.1% on the basis of the data for 2013-2016. But what is also interesting is a one percentage point fall in the jobless rate to 4.5% has a much smaller impact based on recent years’ data than in the pre-recession period. For example, this might have been expected to produce a 0.9 percentage point rise in wage inflation over the 1997-2012 period compared to a 0.5pp rise based on recent data.

Space considerations preclude a look at the reasons for the weaker sensitivity of wage inflation to labour market conditions. It may be the result of factors such as a lower degree of unionisation; the more widespread use of zero hours contracts and the rise of the gig economy, all of which have raised the degree of slack which the headline unemployment rate does not capture. But what the analysis does suggest is that policymakers can afford to spend less time worrying about the impact of low unemployment on wage inflation. There may be a case for higher interest rates but it is not to be found in the labour market.

As a final thought, I am struck by certain parallels with Japan. Following the bursting of the bubble economy, the Japanese authorities failed to spot the structural factors which led the economy to the brink of deflation, notably an ageing demographic profile which prompted a switch towards saving rather than consumption. The one factor we might be missing today is the impact of automation, which threatens a significant substitution of capital for labour and which could put downward pressure on the relative price of labour. I would thus not be in a hurry to raise interest rates to counter a wage inflation threat which has so far failed to materialise.