In this note, we first explain how to use FRB-US to simulate scenarios. In part II, we discuss two important results. First, we show that under a reasonable Taylor rule and estimated Phillips curve, inflation remains above the 2% target even with a terminal rate around 6%. The reason is that the process is now so persistent that disinflating faster entails a large output cost, which is not optimal. Second, under all simulations, it is virtually impossible to get 100bps cut of the FF rate in 2023, even in case of a recession triggered by a financial shock (at most, we get a 50-60bps cut with the FF rate around 4.8% in Q4:2023). The bottom line is that the model suggests that the FOMC remains constrained in its actions by the inflation dynamics, possibly more than perceived by the markets.
Please, meet Tilda
Before beginning, we are pleased to introduce you to Tilda Horvath. Tilda is a former Fed (Board) staff member. Tilda has 20years+ experience working with FRB-US in the modeling group at the Fed Board. She was integral part of the group who created FRB-US from scratch. In other words, Tilda is the expert when it comes to FRB-US. On top of this, Tilda is an incredibly nice person and very friendly to work with. Tilda is available for consultations. #AskTilda
The “consistent” FRB-US and the “add factors”
The publicly available FRB-US is “made up” to interpolate the SEP. After every SEP, and with a significant lag, the Fed staff publishes a FRB-US dataset that is “consistent” with the SEP. The latest dataset is available here. By “consistent” the Fed staff means that the forecast for real GDP, PCE inflation, and the unemployment rate published in the FRB-US dataset fits the published SEP dots. The “consistent” dataset is interesting for two reasons. First, by looking at the non-SEP variables one can get a sense of what needs to happen to get the SEP forecast. Second, and most importantly, we can infer how distant would be the “unconstrained” FRB-US forecast from the “consistent” dataset. What we mean by this, is that in order to fit the latest SEP dots, the Fed staff embeds “add factors” to a series of variables, so that the final result after running the model is the desired one. Put it differently: the Fed staff adds manually little “helps” (i.e. it manually changes the values of variable “x” in quarter “q”) to a certain number of variables according to the needs. Therefore, by removing the add-factors and re-running the unconstrained model, we get to see how distant the model prediction is from the latest SEP.
Why the “unconstrained” FRB-US is interesting
The unconstrained FRB-US delivers a better forecast. Before showing the model forecast right now, we show why this exercise is interesting by going back in time to the September SEP. Figure 1 shows the dataset “consistent” with the September SEP projections (the blue line), and the unconstrained FRB-US forecast conditional on the same information set. By construction, the blue line in Figure 1 interpolates the published September 2022 SEP dots and, as explained above, the orange line is obtained by re-running FRB-US with the September 2022 dataset but removing the add-factors. The main takeaway is that the unconstrained simulation delivered a much better forecast, as it correctly predicted that real GDP would have been (much) higher than expected by the FOMC, that the unemployment rate would have fallen, and that the terminal rate would have been higher than flagged by the FOMC (real GDP growth in Figure 1 is plotted as QoQ ar). The second important result is that the forecast for core PCE price inflation is nearly identical in the two runs. We will return to this point, which is crucial, in part II of this mini-series.
(For the record, the “consistent” and “unconstrained” forecast of other relevant variables, including 5y and 10y yield are available upon request)
Figure 1. FRB-US September 2022 SEP “consistent” (blue) and unconstrained simulation (orange)
Note: the figure shows the forecast of the publicly available FRB-US database consistent with the September 2022 SEP (blue line) and our simulation (the orange line). Real GDP growth is shown as QoQ ar. “Core inflation rate” refers to core PCE price inflation (YoY) at quarterly frequency. The Federal Fund rate and the unemployment rate are expressed in percent.
The December 2022 “unconstrained” forecast
Pre-SVB the model was calling for a terminal rate close to 6%. Figure 2 shows the same exercise reported in Figure 1 but using the December 2022 dataset updated through Q4, at least for the main variables (real GDP growth in Figure 2 is plotted as YoY to facilitate the comparison with the SEP). For the record, we have found 21 “add factors” in the dataset (the complete list of which is available upon request), which in our view is a good indication of the fact that interpolating the December SEP was not an easy task. Indeed, the unconstrained simulation (orange line) is again far from the SEP consistent. Specifically, conditional on Q4 in sample, the model forecasts a mild but long recession (which is triggered by the recent contraction in residential investment). Nevertheless, the FF rate reaches 5¾ percent in Q3 because core inflation is more persistent than in the September dataset, and the unemployment rate rises to 5 percent by 2025. Put it differently, the pre-SVB repricing of the terminal rate was unsurprising through the lens of the “unconstrained” FRB-US, even conditional on a significant slowdown of the economy.
(For the record: about two/three weeks ago Bloomberg circulated a note by Anna Wong claiming that the December 2022 “consistent” dataset contained a recession because the output gap becomes negative at some point in the forecast. Unfortunately, the claim is not correct. Indeed, the output gap in the SEP consistent dataset becomes negative but that happens because output grows less than potential (but still does not contract) for a protracted period of time).
Figure 2. FRB-US December 2022 SEP “consistent” (blue) and unconstrained simulation (orange)
Note: the figure shows the forecast of the publicly available FRB-US database consistent with the December 2022 SEP (blue line) and our simulation (the orange line). Real GDP growth is shown as YoY. “Core inflation rate” refers to core PCE price inflation (YoY) at quarterly frequency. The Federal Fund rate and the unemployment rate are expressed in percent.
Having said all of this, the reader should notice that the path of core inflation in Figure 2 is again nearly identical in the “unconstrained” model compared to the “consistent” version, despite the different trajectories of the FF rate, output and unemployment. How is it possible? After working with the model for several weeks, it turned out that the assumptions of the Fed staff are, to us, a bit too restrictive and end up forcing inflation towards target by construction (without entering into the details that we will discuss in Part II, the reader can see the main price equations of FRB-US and appreciate the complexity here).
Conclusion
In part II of this series, we embed a different Phillips curve in FRB-US and show the unconstrained simulation. The result is that the Fed staff and consequently the FOMC needs to be extra careful because they are, once again, at risk of underestimating the persistency of the process. In any case, even using the public specification, it is hard to see an aggressive cut of the FF rate because inflation returns only gradually at target. Finally, we show what happens in the model when we simulate a large financial shock. Again, the answer is that it is almost impossible to find a scenario in which the Fed cut the FF rate by 100bps in 2023; at most, we end up with a 50-60bps cut and the FF rate at 4¾% in the last quarter of this year.
For any question about FRB-US or requests of consultancy, you can reach Tilda at tilda.horvath@underlyinginflation.com (or tibaref@gmail.com). #AskTilda