Revisiting Baumol’s growth disease in Japan

Abstract

This study analyzes the cause of secular stagnation in Japan after the economic bubble and its association with industrial structural change, specifically Baumol’s growth disease (BGD). Our analysis updates Nishi and focuses on the period between 1995 and 2018 using the latest version of the JIP (Japan Industrial Productivity) database. Then, we ask “Is Japan suffering from BGD?” Our approach includes decomposition of the aggregate labor productivity growth rate, fixed-share growth rate (FSGR) measurement, and a transition probability matrix. The decomposition reveals that the within-sector effect (WSE) is the main source of aggregate labor productivity. The reallocation growth effect (RGE), which measures BGD, was negative for both the short- and long-run averages of the post-1995 period. This period is different from the previous one, as the WSE declined significantly, and the considerable decline in the WSE can be seen as a major factor in Japan’s stagnation over the past 30 years. During this stagnation, BGD silently undermined the Japanese economy, albeit as a secondary effect. Furthermore, our FSGR and transition probability matrix show that persists and is infectious in Japan. Sectors that undergo BGD tend to persist in this state, and even when one sector is in a non-BGD state, it is more likely to develop BGD.

Keywords:

• Baumol’s model
• Japanese economy
• labor productivity growth
• structural change

Acknowledgement

I am grateful for the editor and anonymous referees for useful suggestions to revise the original version of this paper. The earlier version of this paper is presented as “Is Japan suffering from Baumol’s growth disease?” at Musashi Online International Symposium on Secular Stagnation as a New Great Depression: Japan and Beyond Part I, and I also appreciate Alan Freeman and Nobuharu Yokokawa for their valuable comments. Of course, all remaining errors are my own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 EU KLEMS stands for EU-level analysis of capital (K), labor (L), energy (E), materials (M) and service (S) inputs.

2 For more detail, visit RIETI’s website on JIP database.

3 Nishi (2019)’s classification depends on Franke and Kalmbach (2005) and Uemura and Tahara (2015). In classifying them, I appreciate Dr. Shinji Tahara of the Chiba University of Commerce to provide me with a useful converter for Excel files.

4 Besides, the generally exact and additive decomposition (GEAD) by Tang and Wang (2004) and Dumagan (2013) and Diewert (2015)’ GEAD are employed as an alternative way of detecting the symptoms of BGD at a point in time. The CSLS decomposition is a more appropriate way to measure the degree of BGD than these two formulas because of the reasons mentioned here. Regarding the advantages and disadvantages of each formula, see Nishi (2019).

5 For instance, the period of 1995–2000 includes 5 years, and hence the effects are divided by five to obtain the contribution and growth rate of this period. Likewise, the long-run period of 1995–2018 is of 23 years, and hence the effects are divided by 23 to obtain the long-run contribution and growth rate.

6 Rowthorn and Wells (1987) identified “positive de-industrialization” and “negative de-industrialization.” Both lead to employment shifts from the industry to the service, but the former may sustain the output and employment level in the industry, whereas the latter necessarily decreases the output and employment level in the industry with increasing unemployment. The mechanism of the BGD may be close to the negative de-industrialization, but not all. Additionally, Uemura and Tahara (2015) introduce the third and fourth types of de-industrialization, highlighting the industry-service linkages and long-term shifts in effective demand between different sectors, respectively. In the former, structural changes proceed with the manufacturing industry with increasing intermediate inputs from the “business-related services” industry. In the latter, the long-term increase in demand for services with changes in the productive system and lifestyle also causes de-industrialization. Although our scope is limited to a simple “negative de-industrialization," this perspective is useful to provide further insight into the linkage between BGD and de-industrialization.

7 Nordhaus (2008) measures the FSGR for labor and total factor productivity by using nominal output shares for a given year as the weight. Here, we continuously use the labor input share (i.e., man-hours share) along with the previous sections to consider if the long-run labor productivity growth accompanies a positive or negative structural change.

8 The FSGR analysis supposes a (non-) progressive sector based on the long-run average growth rate of labor productivity in each sector, which is a somewhat different approach from the RGE we have employed so far.

9 The CSLS decomposition shows that the business-related services sector realized negative RGE in each period, whereas the FSGR presented a positive structural change. The difference comes from the measurement base. The former measures the contributions by the differences between sectoral and aggregate productivity changes, whereas the latter measures them based on the average growth rate of labor productivity in each sector. It implies that the subsectors might be non-progressive compared to the aggregate average in the short period, but they realized a high growth rate on the long-run average and absorbed labor inputs. In fact, the man-hours share in the business-related services sector increased from 21.58% to 25.71% between 1995 and 2018.

10 Transition probability matrix describes the transition probabilities from one state (e.g., state $i$) to other states (e.g., state $j$) in the corresponding element (e.g., $(i, j)$ element). Consider a certain sector is initially in a BGD state. If BGD is a persistent phenomenon, this is manifested in a higher probability of BGD in the next period in this matrix. By contrast, if BGD is a curable phenomenon, this is manifested in a higher probability of non-BGD in the next period.

11 Table 7 is calculated based on the productivity growth rate differential between each sector and aggregate level to define the progressive and non-progressive sectors and types of BGD. We also calculated these probabilities based on the productivity changes between each sector and aggregate level, as we did to measure the RGE in the CSLS decomposition to check the robustness. These three results also hold for this alternative calculation, although we do not report them in detail. The Stata dataset for this is available upon request.

Additional information

Funding

Financial support from the Japan Society for the Promotion of Science KAKENHI [Grant Number 21K01495] is gratefully acknowledged.