Standard Life Investments

Weekly Economic Briefing

US

Uneven transmission

Back

Although headline Consumer Price Index (CPI) inflation is currently running at 2.7% year-on-year (y/y), that rate currently embeds the substantial increase in oil and other energy prices from their February 2016 lows that will drop out of the figures over the next few months. As a consequence, headline inflation developments over the coming year are likely to be driven more by the trend in core inflation measures, which currently sit at 2.2% y/y in the case of the CPI and 1.7% for Personal Consumption Expenditures (PCE). Economists’ (including those at the Federal Reserve) forecasts for core inflation, are for the most part guided (either explicitly or implicitly) by Phillips Curve or labour market slack based models of the inflation-generating process, along with some augmentation for the impact of inflation expectations and persistence. However, such models face serious problems. Not only is the apparent responsiveness of core inflation to slack very weak, but its forecasting performance is poor, and their models are often beaten by much simpler autoregressive models in which the forecaster’s best guess for core inflation in the next period is whatever it was in the previous period.

Lack of consistency in lags

Given the importance of core inflation to the outlook for monetary policy and asset prices, we have begun looking into this issue more deeply. As a starting point we decided to look at how the correlation structure of core inflation with the change in the unemployment gap (the actual unemployment rate minus estimates of the NAIRU) had varied over time and across different measures of underlying inflation. Our analysis suggests that the apparent failures of slack-based models may be partly due to changes in how slack has been transmitted through its components. To illustrate, we began by measuring the correlation between the unemployment gap and 10 different measures of underlying inflation over eight different lag lengths between January 1987 and January 2017 (see Table 1). What we found was that the correlations varied significantly – from near zero for core goods CPI inflation to 0.45 for the Cleveland Fed’s weighted median CPI – as did the lag structure, though the indicators with the highest correlations generally peaked after 15 months. However, when we decompose the correlation structure into three sub-samples – 1987 to 1997; 1997 to 2007; and 2007 to the present – we see that the full sample masks a lot of change. For example, average correlations for each measure were generally higher in each sub-sample, though the transmission length varied a lot. In the 1987 to 1997 period, changing slack was more highly correlated with core goods inflation than in the latter periods – perhaps due to the effects of globalisation. The correlation with core services inflation was generally higher and more stable, though the correlation was especially high during the pre-crisis credit boom when OER inflation and the labour market were moving in lock-step. That also helps explain why the market-based core PCE inflation measure – which has a much lower weight for housing – had such a low correlation with the change in slack during that period and arguably why the Fed kept policy too loose for too long. The upshot is that when we take account of the changes in the transmission of slack to core inflation, and incorporate a wider range of inflation indicators into our analysis, a relationship can still be identified, underpinning our expectation for the domestic components of core inflation to continue to slowly grind up over the next 12 months.

Jeremy Lawson, Chief Economist