“Hope is not a strategy”
In this note we update our medium-term models for Japan and run some sensitivity analyses. The bottom line is that the latest BoJ forecast (July) already appears too low. Not only, but the models are at risk of revising up again the forecast next quarter. The evolution of nominal wage growth and long-term expectations seem more important than the estimate of the output gap. Overall, the BoJ continues to appear behind the curve and does not seem to be well positioned in terms of risk management.
(Note: an Excel file containing all figures and underlying data presented in this note can be downloaded here)
Medium-term forecast
The inclusion of Q3 in-sample has triggered an upward revision of the forecast. We have updated our models to include (our nowcast of) Q3 in-sample. Our nowcast is based on the correlation between the Tokyo-CPI and the corresponding Japan index (the correlation of the MoM saar for the index ex fresh food is here, for the index excluding fresh food and energy is here, and for the index excluding food and energy is here). For the index excluding fresh food and energy we assume that it will expand 4.5% (QoQ ar) in the current quarter. Figure 1 shows the updated model-based forecasts for the 3 measures of core inflation. As a reminder, the model used for Japan follows BoJ Hogen, Kawamoto and Nakahama (BoJ review, 2015). Our “main” Phillips curve model adapted to Japan delivers similar findings. Table 1 shows the FY forecasts comparing the previous model-based forecasts to the current ones and to the latest BoJ figures. The bottom line is that the inclusion of the new quarter in-sample has triggered an upward revision of the forecast across indexes. The new forecast is already above the July BoJ projections at all horizons. Significantly, the forecast of the index excluding fresh food and energy now remains well above 2% at the end of the medium-term (more on this below).
(For the record: we have implemented a small change. We are now running the models on the indexes that correct for tax changes. This is done to avoid the big spikes observed in the left-hand-side variables. Results are not materially affected)
Figure 1. Medium-term model-based forecasts.
Index ex fresh food (BoJ)
Index ex fresh food and energy
Index ex food and energy (US-style core)
Table 1. Summary of model-based forecasts
Note: The figure shows the model-based forecast of three measures of core CPI. The model is based on Hogen, Kawamoto and Nakahama (BoJ review, 2015). All figures are YoY percent changes. The yellow shadows are intervals of confidence calculated as quasi-out-of sample exercises. The summary table shows the average of the YoY model-based in each fiscal year (Q2, Q3, Q4, and Q1 of the following calendar year). “BoJ” in Table 1 refers to the latest projection of the BoJ.
Sensitivity analysis
Sensitivity analysis: monitor expectations and wages; risks are to the upside. We have run a set of alternative scenarios to assess the sensitivity of the model to different assumptions. The idea of this section is to answer the question: what do we need to see for the models to deliver a higher forecast? Can the BoJ be surprised again to the upside going forward? We consider 4 scenarios: (i) the inclusion of Q4 in-sample, (ii) assuming a higher output gap in the forecast, (iii) assuming an acceleration of inflation expectations, and (iv) an acceleration of nominal wage growth. For this sensitivity analysis we use the index excluding fresh food and energy. Similar results apply to the other 2 indexes. Results are presented in Table 2. (Note: in each scenario we have changed only one assumption. In other words, each column in Table 2 should be compared to the baseline in Table 1) We comment each scenario separately.
Table 2. Model-based scenarios.
Note: the table presents the results of model-based scenarios for the index excluding fresh food and energy. In each column we have changed one assumption with respect to the baseline (that is, each column in Table 2 should be compared to the column “current” for the index ex fresh food and energy in Table 1).
The models are quite optimistic for Q4, risks are to the upside. In scenario “Q4 in-sample” we let the models see Q4, assuming that the left-hand-side variable (ex. fresh food and energy) will growth at 4.5% (QoQ ar) not only in the current quarter but also in the final quarter of the year. In this scenario, the model revises up the forecast because in the baseline the model expects Q4 at 3.3% (QoQ ar). Given that the 3m/3m (saar) of the index is estimated at 4.75% in June, in our estimates it appears possible to have additional upward revisions of the model-based forecast in the next few months.
The contribution of the output gap is small. In scenario “Higher Ygap” we assume a higher path of the output gap (the model measure of “slack” is the BoJ estimate of the output gap). Specifically, while in the baseline the output gap increases from -0.3% in 2023:Q1 to 1.85% at the end of the medium-term, in this scenario we assume that the gap is 1 percent higher in each quarter (i.e. in 2025:Q4 the gap is 2.85% vs 1.85%). Table 2 shows that in this scenario the forecast is only marginally higher than in the baseline. The reason is that the estimated elasticity of core inflation to the output gap is very low (about 1 tenth for each percentage point of output gap). This is not surprising and somehow in line with the estimates for the US and the EA (note: using the Urate, the Urate gap, or a non-linear specification does not help). For this reason, we think that the output gap cannot be particularly informative for policy makers (and/or the BoJ is at risk of underestimating the core dynamics if the judgment is based on the output gap).
Expectations are crucial. While the debate (especially for the US) on inflation dynamics often focuses on the role of the labor market, the low frequency component of the series is determined by the behavior of (long-term) inflation expectations. In these models, including in Hogen, Kawamoto and Nakahama (BoJ review, 2015), a movement of expectations eventually translates 1 to 1 in core inflation space. (In the model, expectations are proxied using TANKAN – outlook for general prices – 5 years ahead, all enterprises, all industries, average). In the baseline, we random walk the latest TANKAN value, while in the “Higher Tankan” scenario we assume it jumps to 3% in the current quarter and stays there throughout the medium-term. Unsurprisingly, the model delivers a forecast which is significantly higher than the baseline. The bottom line is simple: for the dynamics of the model (and possibly also for policy makers), expectations seem to matter much more than the output gap.
A 3% nominal wage growth can raise the forecast. While a correct analysis of the passthrough from nominal wage growth to consumers prices would require the estimation of orthogonal shocks, we have built a scenario to evaluate the correlation between the two. In scenario “higher wage growth”, we augment Hogen, Kawamoto and Nakahama (BoJ review, 20215) with a measure of nominal wage growth (cash earnings). Using other measures of wages and total compensation produces similar findings. Specifically, we assume that nominal wage growth not only enters into the model but that accelerates to 3% in the current quarter and stays there until the end of the forecast horizon. In this scenario, the model delivers a forecast which is comparable to “higher Tankan”. We take this result as an indication that the recent BoJ communication on nominal wage growth does have empirical support.
Conclusion
The ingredients are there. Will the BoJ trigger? Putting everything together, it seems we have a central bank which is behind the curve with its forecast and prone to additional upward surprises starting from this quarter. The models can revise up again the medium-term forecast this quarter and the next. Focus should be on incoming data, especially on expectations and nominal wage growth. The ingredients are there. Will the BoJ trigger?