Global Slack, US Inflation and the Fed's Policy Error
After signaling four rate rises in 2016 this time last year, the Fed finally managed to deliver a quarter-point hike in a unanimous decision. It also signaled that it would raise rates thrice in 2017. Markets promptly responded by selling off bonds and EM assets, and strengthening the US dollar. Meanwhile, core inflation remains well below target. The Fed has, in effect, made clear that it regards the 2 percent inflation target as a ceiling; not a symmetric target.
The Fed is eager to hike because it thinks inflation is just around the corner. The FOMC has consistently predicted higher inflation than has obtained in this recovery. How can the Fed be making a systematic error of this magnitude? The answer is surprisingly simple: The Fed, along with most other central banks and international macroeconomics institutions (IMF, OECD, World Bank), relies on the standard Phillips curve theory that relates the rate of change of inflation to national measures of slack such as the output gap or the unemployment rate. In particular, labor market tightness is expected to generate wage pressures, which in turn would cause inflation to accelerate.
The problem is that the standard, domestic, accelerationist Phillips curve no longer captures the inflation process. Figure 1 shows the US output gap vs. change in core PCE inflation over four periods: 1960-1973, 1973-1990, 1990s, 2000-2016. We have used quarterly data from the CBO and FRED. We see that the model worked from 1960 through 1990, weakened in the 1990s, and then disappeared after 2000.
Figure 1. The Phillips curve weakened in the 1990s and broke down after 2000.
There are two competing explanations for the demise of the standard, domestic, accelerationist Phillips curve. The first is that because inflation expectations have become firmly anchored at the target (no one doubts the Fed's willingness or ability to keep inflation in check), the inflation process has mutated so that the relationship that works now is between domestic slack and the level of inflation rather than changes in inflation. There is some evidence to support this hypothesis. See Figure 2.
Figure 2. The return of levels?
The competing explanation is the Global Slack Hypothesis which says that due to the integration of global markets, what now drives inflation is not domestic slack but rather global slack. Due to competition from global rivals, domestic producers in the tradable sector cannot raise prices when the domestic labor market tightens and wage pressures build. Instead, they either rebalance their global supply chains and off-shore production; or they lose business to their foreign rivals. In either case, domestic inflation is determined as much by global slack as by domestic slack.
The evidence is mounting that the second explanation is the right one. What is especially compelling is the evidence that global slack is statistically significant in ALL countries for which data is available while domestic slack is significant is NONE since 2000. See Figure 3. The last column corresponds to global slack ("foreign gap"); no stars means the variable is not significant; three means it is significant at the 1 percent level.
Figure 3. From Manopimoke (2015).
I wanted to check for myself so I did a straightforward exercise. I took output gap data for twenty countries (not including China) from the IMF and constructed a global slack measure for the United States defined by the average output gap of US trade partners weighted by trade intensity (exports+imports of state i to the US/ sum of exports+imports of all 20 states to the US). I use these weights not because inflation propagates through trade but because global integration tracks trade intensity. I use annual data from 1985-2016. Figure 4 displays the simple linear relationships between domestic and global slack vs. change in US inflation. We see that global slack is a stronger predictor of US inflation than domestic slack.
Figure 4. Domestic vs. global slack as predictors of changes in US inflation.
US output gaps are, of course, contemporaneously correlated with global slack. Figure 5 shows the relationship.
Figure 4. US output gap vs. global slack.
In order to examine the independent influence of the two variables we (1) project the US output gap onto global slack, and then, we (2) project changes in inflation onto the fitted values from (1) (which represent global slack) and the residuals (which capture variations in the US output gap orthogonal to global slack). The results are extraordinarily revealing. See Table 1 below.
Table 1. What explains changes in US inflation?
Estimate SE tStat pValue const -0.053 0.058 -0.915 0.367 global slack 0.129*** 0.038 3.363 0.002 domestic gap -0.006 0.046 -0.129 0.898 We see that global slack is significant at the 1 percent level, while variation in domestic output gap orthogonal to global slack is completely insignificant (and bears the wrong sign). Meanwhile, the inclusion of domestic slack in the model reduces the adjusted-R^2 from 0.256 to 0.231. In other words, the weak relationship between domestic slack and changes in US inflation depicted by the lower chart in Figure 4 is entirely due to the fact US domestic slack is contemporaneously correlated with global slack.
Returning to the return of the levels model, we can do the same exercise we did above. See Table 2.
Table 2. What about return of the levels model?
Estimate SE tStat pValue const 2.274*** 0.152 14.962 0.000 global slack 0.193* 0.100 1.933 0.063 domestic slack -0.246** 0.120 -2.046 0.050 The levels model is even more of a disaster for the standard model. The coefficient for domestic slack is significant but bears the wrong sign after we control for global slack! (Note that we are using annual data here while Figure 2 uses quarterly data. Figure 6 at the bottom shows the annual data. There again, we see that global slack is a significant predictor of the rate of inflation but domestic slack is not. Also, Figure 7 shows that the output gap is strongly correlated with unemployment. One can use either in a Phillips curve. Figure 8 shows the time-variation in the measures for domestic and global slack for the United States. Figure 9 and Table 3 revisit the levels model and document the absurdity that the Canadian output gap is a better predictor of the level of US inflation than the US output gap!)
We have not included any slack measures from China and other emerging markets. There is good reason to believe that their inclusion would make global slack an even stronger predictor of US inflation.
I hope to have at least sown some doubt in your mind about the standard model of the inflation process. The Fed's model error is a significant risk for the economy and markets. The Fed will likely fail to deliver the three promised hikes. But if it does, it would be running a significant risk of plunging the US economy into recession. Ironically, it might be Trump's Keynesian shock that kills the recovery by bringing out the hawks at the Fed.
***
Figure 6. Levels, annual.
Figure 7. US output gap and headline unemployment.
Figure 8. Domestic and global slack for the United States (1985-2016).
***
P.S. More on the return of the levels model.
Quarterly output gap data seem to be unavailable for other countries besides Canada, Japan and Sweden. Among these the US' most important trading partner is Canada. Turns out, the Canadian gap is strongly correlated with the US' global slack metric at the annual frequency. What this means is that we have the following absurdity. The Canadian gap is a slightly better predictor of the level of US inflation than the American gap! See Figure 9. The data is for Q2'2001 to Q2'2016.
Figure 9. Absurdity (levels model): The Canadian output gap is a better predictor of US inflation than the US output gap!
And just to really rub in the point, we repeat the decomposition exercise we tried above for the US and Canada. We project the US output gap onto the Canadian output gap and then project US inflation on the fitted values (representing Canadian/global) and residuals (American orthogonal to Canadian/global). Table 3 displays the estimates. Note that the constant is more or less equal to the Fed's target, which is a good sign because this is a period of anchored inflation expectations. The Canadian/global gap coefficient is more statistically significant and more than twice as large as the coefficient of US gap orthogonal to the Canadian/global gap. Fun stuff!
Table 3. Absurdity continued: Canadian gap is a superior predictor of US inflation than the US gap. Estimate SE tStat pValue const 2.066*** 0.048 43.455 0.000 Canadian gap 0.167*** 0.018 9.248 0.000 US gap 0.064** 0.027 2.318 0.024