No Sign Yet of Underlying Inflation Picking Up
And why you should pay attention to core PCE inflation
Markets are jittery because of the April reading of headline year-on-year consumer price inflation of 4.2 percent. That looks like a pretty large number. But it is mostly a base effect. Because consumer prices fell sharply last April, this month’s reading largely reflects growth from that low level. In any case, core CPI inflation is still lower than it was in the Trump reflation period.
Matt Klein argues that the headline number is ‘mostly a reflection of the economy’s reopening and the idiosyncrasies of the used-vehicle market.’ And investors would be well-advised to ignore it and instead pay attention to ‘what’s going on under the hood by examining the specific categories driving the changes in the price level.’ Most of the increase, he shows, is due to specific dislocations related to the pandemic — hotels, restaurants, car insurance and airline fares.
I like Matt’s attention to detail. But I prefer to think of inflation as a systemic phenomenon. The question for me is what is going on with the underlying, persistent component of inflation. What is the best index that captures this underlying component? Which index has the highest signal-to-noise ratio?
Some series are noisier than others. For instance, core inflation is more stable that headline inflation. Specific stochastic detrending methods, such as looking at year-on-year percentage change, are subject to base effects. In order to get to the bottom of what is going on with underlying inflation, we must not allow such noise to confound our judgement.
Here we will examine four price indices: CPI, CPI ex. food and energy, PCE, PCE ex. food and energy. We shall see why the Fed prefers the last metric. Instead of looking at year-on-year percentage change, we’ll consider two measures. First, we’ll look at month-on-month percentage change at an annualized rate. Second, we’ll look at the difference between log level of the index and the median log level of the index over the previous 12 months. The second is a robust method of stochastically detrending the price indices. This will be our main metric.
Comparing these two measures for CPI and core CPI shows that the April reading was unusually high. Something unusual happened this month. Matt’s probably right about the localized nature of the phenomena.
We don’t yet have the April numbers for PCE and core PCE. The numbers until March 2021 show no sign at all of inflation picking up. The bottom-right graph is our preferred measure. We’ll have to wait to see what the April reading is. But until March, there was no evidence of underlying inflation picking up.
Why does the Fed prefer to look at the core PCE index? Why should we use this instead of the headline PCE index or CPI? The answer is that core PCE has the highest signal-to-noise ratio. Specifically, it contains the strongest predictive information about future changes in all four price indices. In order to document this pattern, we compute predictive correlation coefficients with and without controlling for the lagged values of the response.
Table 1 shows the unconditional correlation between lagged values of the row variables and contemporaneous values of column variables. All variables are monthly changes in log index levels. The data is for Jan 2012 to March 2021.
We can see that core PCE is not only the most persistent — with an AR(1) parameter of 0.53, see the values on the diagonal — it also has the highest predictive content. The last column shows the mean predictive correlation for each predictor. Core PCE has the highest mean predictive correlation (r = 0.48). It even predicts CPI (r = 0.55) better than CPI predicts itself (r = 0.44). So even if you are interested in headline consumer price inflation, you’re better off looking at core PCE.
The results in Table 1 are subject to the critique that inflation has simply become an AR(1) process, as opposed to being governed by the Phillips curve. This is not exactly right. The Phillips curve is indeed dead. But global slack now governs inflation. Specifically, I have shown that a single global factor — the unweighted mean inflation rate in the advanced economies — accounts for almost all the variation in inflation rates, including in the United States. Still, we must control for the persistence term to isolate the signal in each of these series. Table 2 shows partial correlation coefficients controlling for lagged values of the column variables.
The mean predictive correlation controlling for persistence is more than twice as large for core PCE (r = 0.36) as CPI (r = 0.15). Core PCE inflation predicts CPI (r = 0.36), core CPI (r = 0.37), PCE (r = 0.33) and PPI (r = 0.39). No other metric even comes close. Thus, we can see that core PCE has the highest signal-to-noise ratio. No matter which deflator you are interested in, you should pay attention to core PCE above all.
So we should wait for the core PCE inflation reading for April before jumping to the conclusion that underlying inflation is picking up. As Fed officials have been saying and as Matt Klein has shown, the April reading is most likely a temporary idiosyncratic shock that should dissipate in the coming months.
Odd to me how everyone at the same time believes: 1) we have no good theory of inflation & 2) we can confidently forecast inflation over the next year.
My guess is inflation will retain it's upward pressure as housing costs catch up to market-based asking rents (very wide divergence there has been putting downward pressure on inflation figures)
I was under the impression that one reason for excluding certain items from CPI was their volatility. Do the various CPI formulations have similar long-term trends, but different volatilites (in layman's terms).