Is Labor Market Mismatch a Big Deal in Japan?

A1 Data

A1.1 Data definitions

Unemployed, Vacancies, and Hires: Data on the number of unemployed, vacancies (job openings), and hires by contract (regular and non-regular), employment (full-time and part-time), and occupational types, are obtained from monthly Employment Referrals for General Workers (Ippan Shokugyo Shokai Jokyo) conducted by the Ministry of Health, Labour, and Welfare (MHLW). The survey collects information on vacancies, unemployed person, and the number of people who found employment through the Public Employment Security Offices. Note that data on unemployed, vacancies, and hires exclude workers newly graduated from college. Owing to the exclusion of this category, the total number of unemployed in this survey is different from that in the survey by the Ministry of Internal Affairs and Communications (MIAC). The number of unemployed persons in this survey represents approximately 80 percent of the number of unemployed reported by the Cabinet Office for the period between April 2000 and March 2013. The latter is used to calculate the unemployment rate for Japan. All the data are seasonally adjusted using X-12-ARIMA.

Part-time Workers: Part-time workers in the Employment Referrals are those whose hours worked are less than those of regular workers at the same establishment. The MHLW category of part-time workers consists of regular part-time workers who have an indefinite period or longer-than-four-month period and temporary part-time workers whose contract is one month or longer but shorter than four months or whose employment period is fixed and normally responds to seasonal demand. Please go to the link, http://www.mhlw.go.jp/toukei/list/114-1_yougo.html, to read the detailed definition (in Japanese). While a standard measure of part-time vs full-time jobs is based on the number of hours worked in other countries, the part-time and full-time jobs in Japan would suggest some underlying differences in responsibilities, benefits, flexibilities in hours. Part-time workers receive less responsibilities, less benefits, but more flexible working hours. See more detailed discussions in Houseman and Osawa (1995).

Full-Time Workers: Those who are not part-time workers.

Job Separation Rate: Data on the job separation rate is obtained from the MHLW's Monthly Labour Survey (Maitsuki Kinro Toukei Chosa). This survey is conducted on about 33,000 establishments with five or more regular workers including both the private and public sector. The job separation rate is given by the total decrease of employees divided by the number of workers at the end of the previous month at establishment level and averaged across the sample. The total decrease in employees includes those who have retired as well as those who have been transferred to another establishment within the same firm. Details of the definition and the coverage are given in the following link: http://www.mhlw.go.jp/english/database/db-slms/dl/slms-01.pdf

Unemployment Rate: Data on the unemployment rate is provided by the Ministry of Internal Affairs and Communications.

Data Comparison: Survey of Employment Dynamics data

While the data from MHLW has been one of the most widely accepted data sources for understanding job market tightness in Japan—vacancy to unemployment ratio, there are two sources for potential biases: (i) the trend in the usages of private vs public agencies in hiring workers over the years and (ii) potential cyclical bias coming from different costs in posting vacancies between private vs public job agencies. We have checked to see if the vacancy and hires data from the Survey of Employment Dynamics (Koyo-Doko-Kihon-Chosa) are generally in line with the vacancy and hires data in our sample.

Survey of Employment Dynamics (SED) collect data from randomly selected 15,000 establishments with five or more employees, covering both private and public job agencies. The response rate of the survey is around 60 percent. Figure 4 plots the vacancy (in thousands) in our sample against the vacancy data from SED between 2000 and 2017 (the latest data available for SED). We find that while vacancy measures from the two different data sources vary in level, they both share the same cyclical and trend patterns. The vacancy data in our sample generally tracks the data from SED, particularly a sharp decline during the Global Financial Crisis, which is of great interest to our study. Moreover, there could be a slight difference in how vacancy is measured between the two surveys. While the vacancy (mi-jusoku-kyujin-su, unfilled vacancy) in the SED is based on self-reported number of vacancies at the establishment level, the vacancy (yuko-kyujin) in the Employment Referral for General Workers from the MHLW in this paper report the number of vacancies that are actually posted and reported to Public Employment Security Office. Although vacancy based on SED could capture all types of vacancy, the number of vacancy data from MHLW is higher than that in SED.

Figure 4:

Vacancy data: Survey of Employment Dynamics vs the Monthly Labour Survey (MHLW).

Figure 5:

Hires data: Survey of Employment Dynamics vs the Monthly Labour Survey (MHLW).

Figure 5 plots hires data in SED against those in the sample (the data from Public Employment Security offices) between 2000 and 2017 (the latest data available for SED). We find that while hires measures from the two different data sources vary in level, they both share the same cyclical and trend patterns, particularly until 2012, which is the primary focus of our analysis. Hires data in SED are based on the survey at the establishment level and distinguishes the channels through which hires happened, namely (i) Public Employment Security Office, (ii) Public Employment Security Office (Internet), (iii) Private Employment Introduction Office, (iv) school, (v) advertisement, and (vi) others. Note that SED began collecting hires data from “Public Employment Security Office-Internet” in 2007. Comparing the total numbers of hires from SED and the sample in this study (MHLW), we find that the hires data in our sample (Public Employment Security Office) seems to track the hires data from SED relatively well but less so since 2012.

Unfortunately, it is not possible to estimate the labor market mismatch via the methodology developed by Şahin et al. (2014) (SSTV) using Survey of Employment Dynamics (Koyo-Doko-Kihon-Chosa) (SED) because of the following three reasons: (i) SSTV methodology requires the data on vacancies, unemployed persons, and hires at occupation level; SED only contains information on hires at the industry or occupation level; (ii) industry/occupation codes in SED cannot be mapped 1-to-1 into occupation codes in MHLW data; and (iii) frequencies of the two datasets differ (SED is annual whereas MHLW is monthly).

A2 Descriptive statistics

This section provides background information on the Japanese labor market. We show matching function-based assessment of labor market conditions. In particular, we show how likely an unemployed worker in a submarket finds a job (job-finding rate) and how likely a firm fills a vacant position (vacancy yields) by contract-type, employment-type, and occupational group.

• –Vacancy-Unemployment Ratio

The vacancy-to-unemployment ratio (viui) or market tightness indicates how many vacancies are available per unemployed person (Figure 6). When it is greater than one, there is, on average, more than one vacant position for an unemployed person in that market, implying a higher chance for the unemployed workers to find a job and a lower chance for the firms to fill the position.^[20] Dispersion of the vacancy-unemployment ratios across different labor markets indicates misallocations of vacant positions and unemployed workers in the economy. When there is no misallocation, the vacancy-unemployment ratios should be equalized across submarkets if matching efficiencies are homogenous. A higher value of the ratio, however, does not necessarily translate into a higher number of matches formed. The approach in the main text takes into account such differences in matching efficiencies and thus provides an different perspective on labor market conditions.

• –Matching Function

To represent the relationship among vacancies (v), hires (h), and the number of unemployed (u), labor economists have used a constant returns-to-scale matching function as follows:

where ht represents the number of matches (hires), Θt represents a matching efficiency, V represents vacancies, ut is the number of unemployed, and α∈(0, 1) is the vacancy share. The matching function indicates that the number of hires (matches formed) is increasing in both the number of vacancies (vt) and the number of the unemployed (ut).

• –Job-Finding Rates

Figure 6:

Job Finding Rate and vacancy yields for aggregate series: 2000–2019.

Using the matching function, we can express the fraction of the unemployed who find a job in a given month by dividing the matching function by the number of unemployed. We call this the job-finding rate:

Figure 6 plots the aggregate job-finding rate in Japan between 2000 and 2019. Conditional upon matching efficiency, Φ and the vacancy share, α, the job-finding rate is increasing in the number of vacancies and decreasing in the number of unemployed. From a worker's perspective, the job-finding rate shows how easily an unemployed person can land on the job. In recession, the job-finding rate tends to decrease as there tend to be more unemployed workers, and firms tend to post fewer vacancies. Therefore, the series should be pro-cyclical. The job-finding rate in the data confirms this pro-cyclical pattern within the range of 5–10 percent since 2000 (Figure 6).

• –Vacancy Yields

From the firm's perspective, we can obtain an average probability of finding a worker by dividing the number of matches by the vacancy. By dividing both sides of the matching function by the vacancy, we obtain the vacancy yield.^[21]

Given the number of vacancies, more unemployed persons (a greater labor supply) would make it easier for an average firm to fill a vacancy (find a worker). Figure 6 plots vacancy yields for the period between 2000 and 2019. It increases toward the end of recessions because there are more unemployed persons searching for a job while less vacancies are posted.

• –Job-Finding Rates and Vacancy Yields for Disaggregated Data

In this section, we show job-finding rates and vacancy yields for different contract types, employment types, and occupational groups to understand how the sub-labor markets in Japan have been changing over time. We limit our descriptive statistics to the period between 2000 and 2019 to compare the data across different groups.

• 1.Contract Type: Regular vs Non-Regular Workers

Figure 7 plots job-finding rates and vacancy yields for regular and non-regular workers and positions. The probability of finding a job has been higher for non-regular workers than regular workers for the entire sample. While the job-finding rate for non-regular continued to be stable even during the 2008 recession, job finding rate for the regular positions declined and remained low for about two years. In terms of vacancy yields, hires per vacancy, before the GFC, it was easier for firms to fill regular worker's position than non-regular worker positions per vacancy. However, after the GFC, it became easier for firms to fill the non-regular contract vacancies and have converged to similar values in recent years.

• 2. Employment Type: Full-Time vs. Part-Time Workers

Figure 8 plots job-finding rates and vacancy yields for full-time and part-time workers. The probability of finding a job has been higher for part-time workers than for full-time workers. The job-finding rate for part-time workers declined after 2004 and continued to be stable, unaffected by the 2008 recession. From the firms' point of view, it was also easier to fill a full-time position than a part-time position. It has been easier to fill a position toward the end of the recession when there were many unemployed in the market. The probabilities of job seekers finding part-time and full-time positions became very similar after the GFC.

• 3. Occupations

Figure 9 plots the job-finding rates by occupational types. We see that the job-finding rates have been high for (i) production processing occupations and (ii) agricultural, forestry, and fishery workers. From the firms' perspective, the probabilities of filling a vacant position were higher in (i) agricultural, forestry and fishery occupations and in (ii) clerical positions.

This section provided a brief overview of the Japanese labor market. In particular, we showed job-finding rates, and vacancy yields by different contract types, employment types, and occupation.^[22] Based on these measures, labor markets were tighter (smaller vacancy-unemployment ratios) during recessions for full-time workers, regular workers, and workers aiming for certain occupations than others.

Figure 7:

Job Finding Rate for contract-type and vacancy yields for contract type: 2004–2019.

Figure 8:

Job-Finding Rate and vacancy yields by employment type: 2000–2019.

Figure 9:

Job-Finding Rate and vacancy yields by occupation: 2000–2013.

A3 Estimation of matching function and vacancy share

The vacancy share α in the aggregate matching function ht=ϕtvtαut1−αis estimated by two methods following Şahin et al. (2014).

The first method is to estimate the following equation:

where QTTt is a vector of four elements for the quartic time trend that is meant to capture shifts in aggregate matching efficiency. The second method is by following the procedure in Borowczyk-Martins et al. (2013) to account for the endogeneity. Both methods show that the estimate of the vacancy share for the Japanese data to be around 0.4 (0.34 based on GMM others being higher), which we use in our analysis. In the U.S. and U.K. studies, Şahin et al. (2014) and Patterson et al. (2016) set vacancy share, α, to be 0.5.

A4 Three measures of mismatch (M, Ms, Mϕ)

For comparison, the following conventional mismatch index is calculated as in Jackman and Roper (1987) and by the statistical bureau.

The mismatch index (12) only requires data on vacancies and unemployment. However, its measure is less accurate as it does not take into account vacancy share (α) and matching efficiencies (ϕit).

For comparison, Figure 10 plots three measures of mismatch: (i) Simple Mismatch Index based on Eq. (A12)(Ms)(i. e., the measure of mismatch used by statistical agencies in Japan), (ii) Baseline Mismatch Index in Eq. (6) (M) (i. e., mismatch in the absence of heterogeneity in matching efficiencies), and (iii) Mismatch Index in Eq. (5)(Mϕ). The three measures of mismatch (Mϕ, M, Ms) generally capture similar trends in the mismatch in Japan.

There are three main advantages in the proposed mismatch indices Mϕand M over the simple mismatch index, Ms. First, higher vacancy-unemployment ratios do not necessarily translate into a greater number of observed matches and thus may provide misleading implications (Appendix A2). Second, our preferred mismatch indices, Mϕ and M, have a clearer interpretation (fraction of hires lost) than the simple mismatch, Ms. Lastly, the mismatch index in the theory-based approach accounts for its contribution to the rise in the unemployment rates owing to mismatch.

Figure 10:

Three measures of mismatch in Japan: 2004–2019 (Contract-Type), 2000–2019 (Employment-Type), and 2000–2013 (Occupation).

In Figure 10, Simple Mismatch (orange dotted-dashed line) shows Ms , Baseline Mismatch (red solid line) shows M, and Mismatch (green dashed line) shows Mϕ series.