Having dropped to 5.1%, the unemployment rate has reached the longer-run employment goal of the Federal Reserve’s Open Market Committee (FOMC). So, starting to raise interest rates would seem to be in the cards. And, many observers expect policymakers to act soon, possibly very soon.
The key sticking point, and it is a big one, is that inflation – as measured by the personal consumption expenditure price index (PCE) favored by the FOMC – has been consistently below their stated 2% medium-term objective since early 2012.
Tightening monetary policy for the first time since 2006 requires confidence that inflation will in fact head back up (see, for example, the July FOMC statement and Fed Vice Chair Fischer’s recent comments). The difficulty is that confidence requires reliable forecasts. And, as it turns out, precise forecasts of inflation are hard to come by. (See our earlier post for the mechanics of how the Fed will tighten policy when the time comes.)
First, what’s been happening to inflation? In the chart below, we plot annual headline, or all-items, PCE inflation (thick black line); along with two indicators of trend inflation, the PCE excluding food and energy (red line); and the trimmed mean PCE constructed by the Federal Reserve Bank of Dallas (blue line). (For a discussion of core inflation measures see here.) You can see the problem right away. It has been over three years since inflation hit 2%, while the core measures recently have remained stubbornly around 1½%. And this has occurred amid remarkably low inflation across most of the world and deflation risks in several advanced economies.
Price index of personal consumption expenditures (percent change from year ago), 2000-July 2015
Source: FRED.
The question facing the FOMC is how secure are they that inflation will rise back to 2% in the fairly near future? Their own forecasts, published in June in what is known as the “Summary of Economic Projections,” show inflation rising to 2% by the end of 2017, while the unemployment rate edges down to 5%. But these “central tendency” projections were made before China’s recent devaluation and the third-quarter drop in oil prices. They also conceal wide disagreement. FOMC participants differ considerably about the level of the policy interest rate they believe to be consistent with these inflation and unemployment projections. For example, one participant projects the mid-point of the federal funds rate target range at the end of 2017 at 2%, while two others expect it to reach 3⅞%. The gap in the projected policy rates at the end of 2016 (namely, 2½ percentage points) is even larger.
We suspect that the primary reason for this disagreement is the difficulty in forecasting inflation. And, if anything, that forecasting challenge has increased. We will try to explain what has happened.
Inflation models are typically based on the notion that the tighter the labor market, the more likely it is that inflation will rise. The idea is that, when the labor market is unusually stretched, employers are forced to bid up wages to attract qualified workers. Higher wages mean higher costs; and higher costs mean higher prices. Following this logic to its obvious conclusion, there should be a degree of labor market tightness – say, corresponding to a particular level of the unemployment rate – that is consistent with stable inflation. That’s what is sometimes called the “natural rate of unemployment.” You might also think of it as the equilibrium (or frictional) rate of unemployment that would prevail in a world of completely flexible prices and wages.
The natural rate of unemployment is unobserved, so no one knows what it is. What we do know is that it changes considerably over time. For example, since they tend to spend more time looking for jobs, the more young people there are in the labor force, the higher the natural rate will be. A lengthening of unemployment benefits, which tends to occur during recessions, will also drive it up. And anything that diminishes firms’ incentives to hire workers, like increases in licensing requirements (see here), also raises the natural rate. These changes are difficult to calculate, and even more difficult to predict.
The Congressional Budget Office (CBO) produces a measure of the natural rate of unemployment as a byproduct of their estimate of potential GDP (see our earlier post on the problems with that!). The CBO estimates that the natural rate was 5.0% before the crisis, rose to 6.0 % by the beginning of 2012, and is just below 5.2% today (see here).
It is important to understand, however, that most estimates of the natural rate are computed as the unemployment rate consistent with inflation neither rising nor falling. So, if you use them to predict inflation, you may simply be reverse engineering the method used to construct the data in the first place!
Returning to the difficulties of forecasting inflation, the biggest problem turns out to be changes in the inflation process itself. Over the years, researchers have found that two things have changed dramatically and unpredictably. The first is the degree of persistence. This is the answer to the question: if inflation unexpectedly rises by one percentage point today, how much of this increase is likely to remain tomorrow? If the answer is something close to one percentage point, then persistence is high. Put differently, when the inflation process is persistent, a surprise alters the trend of inflation. When the process it not persistent, a surprise leaves the trend unchanged.
A simple example may help explain the role of persistence. Imagine a world in which the central bank targets 2% inflation, and inflation is in fact 2%. Not only that, but for the past few years, when inflation has gone up a little and down a little, it has reverted back to 2%. Now, suddenly, inflation rises to 3%. Because of its past success, the central bank has gained credibility, so no one really thinks inflation is going to stay at 3%. In fact, so long as expectations are unchanged and most people believe inflation will quickly fall back to where it started, then there will be virtually no persistence whatsoever.
The second thing that has changed in a way that makes forecasting even more difficult is the relationship between inflation and specific measures of labor market tightness, such as the unemployment rate. How low does the unemployment rate have to go before the tighter labor market starts to exert upward pressure on wage and price inflation? Is this level sensitive to changes in various aspects of the labor market, such as the average duration of unemployment (see, for example, here)? Or, as some very recent research suggests, does it depend on complex interactions with financial conditions?
To show the role of changes in the persistence of inflation and in inflation’s relationship with unemployment, we examine how well one can predict inflation using an exceedingly simple model based only on past inflation and the current and past level of the civilian unemployment rate. We (somewhat arbitrarily) divide the data set into three 10-year time periods starting in 1985. And, because the goal is to predict the trend of inflation, we use the trimmed mean PCE price index.
The next chart shows the fit of these models. The fit is best in the first period (R-squared=0.80), very poor in the second period (R-squared=0.09), and reasonably good in the latest period (R-squared=0.75). From 1995 to 2004, the best estimate of inflation is simply a constant 2.1%: in this period, there was absolutely no persistence, and movements in the unemployment rate were irrelevant. Prior to 1995, inflation persistence was high, while after 2004, inflation has been negatively related to inflation the prior year (albeit with limited statistical significance).
Three inflation models: actual and fitted values over three time periods (1985-2015)
Source: Authors' calculations based on quarterly observations of annual inflation from FRED.
Before continuing, we should emphasize how stable inflation has become since the early 1990s. For most of the past quarter century, trimmed mean PCE inflation has remained between 1% and 3%. By contrast, from 1980 to 1995, the range was 2¼% to 8½%. This narrowing of the range of trend inflation close to the FOMC target constitutes a remarkable monetary policy success that should not be underestimated. Compared to that, today’s concern about what should be done to nudge inflation up from about 1½% to 2% is really just a sign of how much the Fed has achieved in the past 20 years.
Turning to the most recent period, 2005 to 2015, we can ask whether either of the two relationships (between inflation and its past, and between inflation and labor market tightness) from the earlier intervals has stood the test of time. To do this, we look at the level of inflation that each of the two earlier models would now predict. In the following chart, we plot these projections along with the actual data (black) and the fitted values from the model estimated over this period (dashed gray). Two things to notice: first, all of these models over-predict recent inflation; and second, none of them fits very well. Using normal measures of fit for forecasting models, the earliest model actually fits the recent data best.
Forecasting trimmed mean inflation: predicted and actual values, 2005-2015
Source: Authors' calculations based on quarterly observations of annual inflation from FRED.
Our bottom line is that the models we have simply aren’t very good at forecasting inflation – at least not to the precision we would need to distinguish a change of trend inflation of one-half of one percentage point over the next two years. And, what was challenging a decade ago has gotten more difficult since then. (For a very careful and technical analysis, see here.)
But, central bankers need inflation forecasts to do their job. And for want of better models, they will use the ones that they have. They just need to be humble about the models’ reliability.
We have written before about how uncertainty concerning both the natural rate of interest and the trend rate of economic growth can lead to a policy that focuses on changes rather than levels, resulting in a heightened degree of policy inertia (see here and here). Today, policymakers are looking at an environment in which the unemployment rate has fallen steadily since its October 2009 peak of 10.0%. FOMC members presumably expect that a continued tightening of labor markets eventually will start to exert inflationary pressure. So far, however, there is little indication of a pickup in the trends of wage or price inflation. Maybe this is a temporary consequence of the fall in energy prices, the appreciation of the dollar, and the slowdown in China. Or, maybe it is something else. We surely don’t know, and we assume they don’t either. This kind of uncertainty probably also favors inertia. (For the most recent word on this, both from academics and policymakers, see the papers and speeches at the 2015 Jackson Hole Economic Symposium.)
Ironically, the FOMC’s success at keeping inflation low and stable creates another important risk that naturally favors policy delay and inertia. When the next recession hits, the scope for conventional policy easing will depend on the level of the policy rate at the time. Unless inflation has risen above the 2% target, that policy scope will be more limited than it was in past downturns. We doubt that the FOMC looks forward to another extended bout of quantitative easing any time soon. Given that, the temptation is to err on the side of providing too much stimulus in the hopes of reaching or even temporarily exceeding their medium-term inflation objective.
In the end, given that they are unlikely to be terribly confident in their inflation forecast, FOMC members will have to decide which of two risks they prefer to take. Are they willing to allow inflation to rise briefly above their objective, in which case they will tighten policy more slowly? Or, are they more concerned that inflation might rise too far for too long, in which case they will go more quickly? In either case, the precise timing of the first move doesn’t matter much. What does matter is how fast and how far the policy rate rises. And regardless, the earlier they start, the slower they are likely to proceed.