In this note, we review two recent academic papers. The first one (“Anatomy of the Phillips Curve: Micro Evidence and Macro Implications” by Gagliardone and coauthors) makes an important point on the New Keynesian Phillips Curve (NKPC): the slope of the primitive NKPC (which has the marginal cost as measure of real activity) is 4-5 higher than in the conventional form. We assess the policy implications.
The second paper (“Underlying Inflation and Asymmetric Risks” by Bihan et al.) is a statistical exercise that proposes a new measure of “underlying inflation” for the Euro area. We are skeptical of the suggested approach and we offer a critique of the paper.
Paper #1: Anatomy of the Phillips Curve: Micro Evidence and Macro Implications
A new paper (“Anatomy of the Phillips Curve: Micro Evidence and Macro Implications”) by Luca Gagliardone (NYU), Mark Gertler (NYU), Simone Lenzu (NYU), and Joris Tielens (Central Bank of Belgium) use a high-frequency micro-dataset to estimate the slope of the primitive form of the New Keynesian Phillips curve, which features marginal cost as the relevant real activity variable. The paper is relevant for 2 reasons: (i) it clarifies (opposite of what many tend argue) what a New Keynesian Phillips Curve (NKPC) is, and (ii) provides a quantitative estimate of NKPC slope.
The issue: what the NKPC is NOT
The NKPC, in its primitive form, is NOT a relationship between price inflation and the output or unemployment gap. Gagliardone et al. (2023) is an important reminder that most people make confusion when talking about the NKPC. Indeed, some conclude that “the (NK)PC does not exist” based on a low correlation between a form of slack (say, the unemployment rate) and consumers’ price inflation. Let us be direct: this is wrong. As Gagliardone et al. (2023) explain, in the theory, firms optimize their pricing policies in response to current and anticipated movements in marginal cost (around its trend). Put it differently, when deriving the micro-founded PC, the economist gets an equation that links inflation to the real marginal cost, taken as a measure of real activity. This equation is known as “the primitive form” of NKPC. On the other hand, the so-called “conventional form” of the NKPC (which is what most people have in mind) links inflation to a measure of excess demand, typically estimated using the output or the unemployment gap. The key point, generally lost in the discussion, is that in the model, the “conventional form” only holds under specific conditions (including flexible wages) that establish a proportional relationship between marginal cost and the output gap. It follows that if one observes a weak correlation in macro figures between consumers price inflation and (say) the output gap, one cannot conclude that “the PC does not exist”. Indeed, as Gagliardone et al. (2023) show convincingly, there is a clear strong micro primitive NKPC in the data; but the link between the marginal cost and the output gap is weak because some assumptions of the model do not hold in the real world (i.e. wages are not fully flexible).
(For the record: this is why it is almost impossible to abandon the NKPC. Indeed, if one asks any price setter “why did you raise that price?”, most likely the answer is “because my costs went up”, which is precisely what the NKPC is about!)
The micro evidence
Gagliardone et al. (2023) show that the (low) slope of the conventional NKPC does not stem from a limited transmission of fluctuations in marginal cost to inflation, but rather from the weak connection between output gap and marginal cost. The main result of the paper is shown in Figure 1 below. The estimates in Gagliardone et al. (2023) indicate that the slope of the primitive NKPC (the “lambda, k” line in Figure 1) is about 4-5 times higher than the slope in the conventional form (columns (2) and (4)). This result helps to reconcile the Phillips curve estimates available in the literature. The authors’ results also suggest that there should be a more theoretical and empirical focus on understanding the structural relationship between output and marginal cost, particularly so given that the elasticity connecting the two could be nonlinear and time-varying (unfortunately, on this point our understanding is still limited).
Figure 1. Main result of Gagliardone et al. (2023)
The policy implication
Disinflation takes time, policymakers will be careful. Gagliardone et al. (2023) has been recently mentioned by Isabel Schnabel (here, footnote 13). We agree with Schnabel’s take. Based on Gagliardone et al. (2023), the question is if the sharp decline in marginal costs on the back of the fall in energy prices (especially in the EA) would drive inflation down just as quickly as higher costs had pushed it up. According to Schnabel, there are reasons to believe that this may not be the case, as there is often an asymmetry between positive and negative cost-push shocks. While firms are quick to pass large cost increases on to consumers, they may be more reluctant to pass on declines in marginal costs. Research (see Karadi and Reiff (2019)) suggests that the presence of trend inflation, which is largely determined by the central bank’s inflation target, can help explain this stylized fact. If firms expect the general price level to rise over time, they will hesitate to reduce prices in the presence of negative cost shocks, or at least they will do so very cautiously. This is because trend inflation will over time re-align their prices with those of their competitors even in the absence of price cuts. One takeaway, therefore, is that a micro founded Phillips curve based on marginal costs rather than the output gap (as in Gagliardone et al. (2023)) can help explain the surge in inflation. But such a curve would likely point to price pressures fading more gradually. For this reason, we expect policy makers, including at the ECB, to be extra careful before claiming victory against inflation.
Paper #2: Underlying Inflation and Asymmetric Risks
A new paper (“Underlying Inflation and Asymmetric Risks”) by Hervé Le Bihan, Danilo Leiva-Leon, and Matías Pacce (ECB) proposes a new measure of “underlying inflation” for the EA. We are critical for three reasons: (i) the new indicator is a measure of “core” more than “underlying” inflation, (ii) it does not have superior properties than the alternatives, and (iii) it is unclear the contribution of the paper.
What the paper does
The authors propose a measure of “underlying inflation” based on estimating a regime-switching model at the sectoral level (11 sectors in total) over the sample 1999 to 2023. The intuition is simple: each sector is regressed on a trend and an idiosyncratic shock. The novelty is that the trend is modelled as a Markov-switching regime-dependent component estimated using Bayesian methods: some periods are identified as “high inflation” and other as “low inflation”. In this way, the persistent component of inflation can “jump” when the model estimates a regime switch. The authors claim that the new indicator (which they call “ICARIS” = Indicator of Core by Aggregating Regimes of Inflation Sub-components) has superior properties. The main result of the paper is reported in Figure 2 below, which shows the YoY of headline HICP (dashed blue line) and the authors’ “ICARIS” indicator.
Figure 2. YoY of headline HICP and authors’ “ICARIS”
A critique of the paper
While the authors’ exercise is (in theory) interesting, a number of limitations prevent us from recommending it. Here are, in our view, the main limitations in no specific order:
- The model is run on YoY figures. The issue is that if one is interested in getting a timely indicator of “regime changes”, the use of YoY figures limits the timeliness. The suspect is that by the time the YoY has increased/decreased enough, it is already evident to everyone the regime change.
- Core vs underlying inflation. The paper makes no distinction between the notion of “core” and the one of “underlying” inflation. To us, what the authors estimate is an alternative measure of “core”, not of “underlying” inflation.
- What is the contribution? The spirit behind the authors’ model is not new, as the authors themselves acknowledge. The main novelty is the Markov-switching regime-dependent component. But even considering that, a model as in Stock and Watson (2016) with a flexible or less persistent common component would produce similar results. Indeed, when trying to explain how the paper departs from Stock and Watson (2016), the authors write: “[…] our modelling approach is similar in spirit. However, instead of capturing common movements in inflation, we let the common factors represent sources of information that, according to economic theory, influence the current and future inflation dynamics”. Unfortunately, that sentence (which is crucial to understand the contribution of the paper) is meaningless to us.
- High inflation regime in 2017? In Figure 2, ICARIS is estimated at 2% in 2017-2019. We find this result (among others not mentioned in the same Figure) to be very puzzling, considering that HICPX averaged around 1% in the same period. We suspect that this happens because ICARIS includes food and energy items and the regime-changes end up being sensitive to those items.
- What is the advantage of using ICARIS? The main critique is that once ICARIS is compared to existing measures of core/underlying inflation, it is unclear what is its advantage. Figure 3 (from the paper) show a comparison. Table 2 and 3 in the paper report the statistics. ICARIS is way more volatile than HICPX (in fact of most measures of “core”), and it has worse forecasting properties than the PCCI. In the end, we are not sure why one should look at ICARIS at all.
Figure 3. Alternative measures of core inflation.