Consumption smoothing in the working-class households of interwar Japan

07/16/2018
by   Kota Ogasawara, et al.
0

I analyze Osaka factory worker households in the early 1920s, whether idiosyncratic income shocks were shared efficiently, and which consumption categories were robust to shocks. While the null hypothesis of full risk-sharing of total expenditures was rejected, factory workers maintained their households, in that they paid for essential expenditures (rent, utilities, and commutation) during economic hardship. Additionally, children's education expenditures were possibly robust to idiosyncratic income shocks. The results suggest that temporary income is statistically significantly increased if disposable income drops due to idiosyncratic shocks. Historical documents suggest microfinancial lending and saving institutions helped mitigate risk-based vulnerabilities.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

08/23/2019

Eco-Mobility-on-Demand Fleet Control with Ride-Sharing

Shared Mobility-on-Demand using automated vehicles can reduce energy con...
06/05/2017

A Possibilistic and Probabilistic Approach to Precautionary Saving

This paper proposes two mixed models to study a consumer's optimal savin...
04/10/2019

Association of night-waking and inattention/hyperactivity symptoms trajectories in preschool-aged children

Objective: To study the longitudinal associations between inattention/hy...
06/28/2021

Measuring close proximity interactions in summer camps during the COVID-19 pandemic

Policy makers have implemented multiple non-pharmaceutical strategies to...
11/17/2021

Case study of SARS-CoV-2 transmission risk assessment in indoor environments using cloud computing resources

Complex flow simulations are conventionally performed on HPC clusters. H...
12/05/2019

Modeling and Prediction of Iran's Steel Consumption Based on Economic Activity Using Support Vector Machines

The steel industry has great impacts on the economy and the environment ...
02/19/2019

A primer on statistically validated networks

In this contribution we discuss some approaches of network analysis prov...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

Consumption smoothing is one of the most important means at the household level of dealing with unforeseen shocks; this is especially the case in industrializing countries. In developing economies, a collapse of smoothing impedes human capital accumulation and the demographic structure itself (Foster 1995; Rose 1999; Gertler and Gruber 2002; Dercon and Krishnan 2002). Therefore, households’ risk-coping mechanisms have been widely studied, ever since a set of pioneering studies provided a systematic empirical design by which to test consumption smoothing (Rosenzweig 1988; Mace 1991; Cochrane 1991; Townsend 1994).111 See Dercon (2004) for a review of earlier studies. See also Attanasio and Pistaferri (2016) and Meyer and Sullivan (2017) for reviews and the latest discussions on consumption and income inequality.

In terms of providing a historical context, despite there being a number of studies on intrahousehold resource allocation, only a few studies have investigated consumption smoothing among households in Western countries.222Examples of studies on intrahousehold resource allocation include Horrell and Oxley (1999; 2000; 2012) and Horrell et al. (2009). See Horrell and Oxley (2013) for a comprehensive review of these studies. For instance, Kiesling (1996) determined the importance of informal sources of income assistance (e.g., savings, transfers, and charity) in Victorian Lancashire.333See also Boyer (1997) for an alternative view of the importance of informal assistance. Horrel and Oxley (2000) found evidence that British industrial households made use of self-help organizations such as sickness and health benefit clubs, which complemented family employment in addressing economic hardship in the late 19th century. Scott and Walker (2012) also found that interwar British working-class households extensively used risk-sharing institutions such as clubs and hire purchases to smooth expenditures.444This finding is consistent with the reduction of absolute poverty among urban British working-class households in the early 20th century (Gazeley and Newell 2012). See also O’Connell and Reid (2005) for details of working-class consumer credit in the United Kingdom at that time.

Saaritsa (2008; 2011) found that informal assistance, credit, and saving accounts had supported income smoothing among worker households in interwar Helsinki. As for Asian countries, James and Suto (2011) found that in Japan in the 1910s and 1920s, savings correlated positively with transitory income. According to their estimates, the savings level was higher than that of working-class households in the United States in the early 20th century. They suggest that the widespread use in prewar Japan of postal saving accounts have encouraged or facilitated saving, relative to institutions available to US workers.

These findings within the literature confirm that risk-sharing institutions and self-insurance behavior played an important role for working-class households in mitigating idiosyncratic shocks in periods of historical economic development. The current study builds on this historical view within the literature. However, I seek to contribute by addressing in the following ways the potential issues that have not yet been adequately assessed.

First, the current study adopts a systematic empirical design to test risk-sharing. Since the degree of consumption smoothing can reflect the levels of maturity of formal and informal financial markets in a given economy, a systematic way of testing risk-sharing would allow for comparable estimates for each objective economy, and thus offer plentiful information by which to know and understand the diversity of historical economic developments (Dercon 2004). In the field of economic history, however, there is a dearth of research into the extent to which risk has been shared in specific countries and eras. In the current study, therefore, I employ the stylized empirical strategy for testing a complete market (Mace 1991; Cochrane 1991) and compare the elasticity (i.e., the responsiveness of consumption changes to idiosyncratic income shocks) to that in Japan’s economy of the 1990s, which was a near-complete market. Another advantage of this strategy is that the degree of risk-sharing can be estimated for each expenditure item. In this way, the current study examines each item’s sensitivities to idiosyncratic income shocks.

Second, the current study uses a unique household-level monthly expenditure panel dataset captured through a survey of Osaka city completed around 1920. Since consumption smoothing is a concept pertaining to the dynamics of household behavior over time, monthly panel data on the household budget would be essentially needed to test a household’s short-term smoothing behavior (e.g., Mace 1991). Unfortunately, however, it is difficult to compile such household-level budget data in the historical context, due to data unavailability (James and Suto 2011; Scott and Walker 2012). While Saaritsa (2011) used a quarterly panel dataset of 142 households in 1928 Helsinki, but lower-frequency aggregation over time could lead to difficulties in capturing the short-term responses of the households. To overcome this issue, I use monthly variations in the household budget to investigate risk-coping behaviors among working-class households.

Third, the current study assesses the risk-coping strategy choices of the households. There are usually several types of risk-coping instruments, such as microfinancial lending institutions and self-insurance. These are important means by which to deal with the uncertainties inherent in developing economies (Carroll 1997; Fafchamps and Lund 2003). However, the systematic test of full risk-sharing, as discussed above, provides little information as to how risks are shared. Therefore, the current study looks to investigate whether temporary income from borrowings, as well as other sources (e.g., gifts and assets use) serve in sharing risk efficiently. While economic history studies have analyzed what types of informal insurances have correlated with income or consumption, the current study is the first to test the short-term responses of household-level risk-coping behavior while using a historical household-level monthly panel dataset.555Kiesling (1996) found a negative correlation between regional-level cotton consumption and the use of public relief. Saaritsa (2011) found a negative correlation between quarterly household income and the use of informal assistance. Scott and Walker (2012) found a positive cross-sectional correlation between household income and both clothing club and hire purchase expenditures.

I found that while the null hypothesis of full risk-sharing for total expenditures was rejected, the responsiveness of the consumption changes to the idiosyncratic income shocks was six times greater than that of Japan in the 1990s. Despite this sensitivity, factory worker households could maintain themselves, in the sense that they could cope with making payments for essential expenditures such as rent, utilities, and commutation. Children’s education expenditures were also less likely to be affected by shocks. Temporary income was increased if disposable income was reduced by the idiosyncratic shocks. Historical documents support the evidence that microfinancial lending institutions and savings institutions had been used by households at that time as risk-coping strategies.

This paper is organized as follows. Section 2 provides a brief overview of the historical background and explains the road map of the current study. Section 3 describes the theoretical framework and empirical strategies used herein. Section 4 introduces the data used, and Section 5 provides the results. Section 6 discusses the findings.

2 Background

In early 20th-century Japan, the typical employment contract was fragile and stipulated no fixed term of employment; thus, the labor mobility of factory workers—comprising not only unskilled workers, but also skilled workers—was extremely high (Moriguchi 2000). After World War I, large companies began to introduce comprehensive corporate welfare programs for factory workers, in order to accumulate firm-specific human capital; additionally, the Retirement Allowance Fund Law of 1936, which obligated employers to set up a retirement allowance fund for their employees, complemented these enterprise-based welfare programs (Moriguchi 2003). As a result, the turnover rates of large companies began to decline after the war, and fell to below 10% in the late 1920s (Hyodo 1971).

However, the factory workers employed by such large companies comprised only about 20% of all production workers in the late 1920s (Moriguchi 2003, p. 644), and the unemployment insurance bill was not passed during the interwar period (Kase 2006). The turnover rate of small and medium enterprises was approximately 30%, and the average annual turnover rate of factory workers between 1918 and 1939 was 28.3% (Hyodo 1971; Taira 1970). Therefore, labor mobility remained at high levels across the country (Odaka 1999). With the exception of a small number of favored workers among the large companies, worker uncertainty in the labor market was high in interwar Japan, at which time a comprehensive social security system had not yet been established.666In general, the introduction of government social insurance systems led to reduction in the use of private insurance, such as private insurance purchase and precautionary savings (Kantor and Fishback 1996; Emery 2010). In terms of official social welfare assistance, however, comprehensive public assistance did not exist in prewar Japan (Ogasawara and Kobayashi 2015). In interwar London, pension payments constituted an income below the poverty line (Baines and Johnson 1999). This underscores the importance of risk-sharing institutions, as studied by Scott and Walker (2012) and as described in the introduction.

In spite of this uncertainty, it is known that the standard of living of the factory workers was established around 1920 and was maintained through the interwar period, while poor households were forced to experience fluctuations in their standard of living (Nakagawa 1985). Even inequality among members of the working class—measured in terms of the Gini coefficient—had decreased between the early 1920s and the early 1930s, while the disparity in wealth among poor household had increased (Yazawa 2004; Bassino 2006). As a result, throughout the interwar period, the household saving rate and the average years of education increased, whereas the fertility rate declined (Mosk 1979; Godo 2011; Minami 2002). This “odd” trend implies that working-class households had coped with multiple risks relatively well, and had therefore contributed to the accumulation of human or physical capital; that capital became a driving force of economic growth in postwar Japan (World Bank 1993; Mason 1997; Bloom and Williamson 1998).

By leveraging the detailed information available on working-class households—as captured through surveys conducted after World War I—the current study investigates the risk-coping behaviors of factory worker households in Osaka city. It does so by undertaking a two-step process. In the first step, I test the “full insurance hypothesis.” If idiosyncratic shocks are well insured by not only financial markets but also communities or governmental assistance, households can smooth their consumption. An important strategy in coping with idiosyncratic shocks is, therefore, the insurance contracts offered in market and nonmarket mechanisms (Fafchamps 2003; Dercon 2004). Since full risk-sharing among consumers suggests that a representative agent model can describe the real economy, testing consumption insurance has been explored in the consumption literature. Although many previous studies reject the hypothesis of full risk-sharing and found that individual consumption is affected by idiosyncratic shocks, they reveal that the risk is pooled to a considerable degree, or that the risk insurance hypothesis was not rejected for certain expenditure categories.777Examples of studies investigating full risk-sharing include those of Cochrane (1991), Mace (1991), Deaton (1992), Townsend (1994), Hayashi et al. (1996), and Kohara et al. (2002). While considering this overall background, the current study investigates whether the full risk-sharing hypothesis is accepted, the extent to which it holds, and what categories of consumption are robust to shocks.

In the second step, I investigate the efficiency of risk-sharing. Although in a complete market the risk-coping strategy typology is no longer important, in an imperfect market, the choice of insurance instruments depends on the market imperfections at hand. However, the typical test of full risk-sharing, as explained above, provides little information as to how risks are shared. Regarding the choice of risk-coping instruments, self-insurance is the primary way in which households deal with risk (Carroll 1997). Microfinancial lending institutions and informal gifts are also important means of dealing with uncertainty (Islam and Maitra 2012). In the current study, I separately analyze associations between idiosyncratic income shocks and either net borrowing or other income sources, including gifts and changes in assets.

3 Theoretical framework

If risk is shared efficiently in an economy, individual consumption should be unaffected by idiosyncratic variations in income. As a first step, I test the hypothesis of full risk-sharing to analyze whether risk had been shared efficiently, and what categories of consumption were robust to shocks at that time. To derive the empirical specifications needed to test the hypothesis, I reviewed the necessity of full risk-sharing for any interior Pareto-efficient allocation in a simple closed-exchange economy lacking storage, in the spirit of Mace (1991) and Cochrane (1991). Mace (1991) found that if in the case of a constant absolute risk aversion preference, individual i’s consumption can be written as:

(1)

where is the coefficient of constant absolute risk aversion, is individual i’s welfare weight, is preference shock, and , , and are aggregate variables of consumption, the planner’s weight, and preference shock, respectively (see Appendix A for details of the derivation). This equation implies that individual-level consumption should vary positively with aggregate consumption, while individual-level income should have no effect on individual consumption in a competitive equilibrium with complete markets.

The first difference of equation (1) will eliminate the individual fixed effect to yield:

(2)

This equation implies that changes in individual-level consumption should depend only on the aggregate consumption variable, and not on any idiosyncratic shock variables. Thus, using the change in individual income as a proxy for idiosyncratic shock, I characterize the empirical specification as:

(3)

where is a disturbance term that includes both the time-varying preference shock, which affects individual-level consumption, and measurement errors in the data. Similarly, I can also express equation (3) under the case of a constant relative risk aversion preference:

(4)

For both equations, if risk is efficiently shared among individuals, the coefficient on the change in individual-level income or the growth rate of individual-level income becomes 0, while the coefficient on the change in aggregate consumption or the growth rate of aggregate consumption becomes 1. Hence, the null hypothesis of full risk-sharing is and . Furthermore, one can surmise that estimates of possibly range from 0 (for full insurance, where idiosyncratic shocks are perfectly insured) to 1 (for the total absence of insurance, in the case of the growth specification). To investigate the degree of insurance, therefore, I use the growth specification of equation (4) to test the full risk-sharing hypothesis.

With the second specification, I intend to investigate risk-coping mechanisms. Following Fafchamps and Lund (2003), I consider that household consumption can be defined as follows:

(5)

where and indicate net borrowing from lending institutions, and net income from other temporary sources (including profits from gifts and the sale of assets).888Although Fafchamps and Lund (2003) consider net gifts and changes in household assets separately, I aggregate these sources, given the limited availability of historical data (Section 4). and are permanent income and transitory income, respectively. Equation (1) can then be rewritten by substituting equation (5):

(6)

The time-constant components (, ) and individual-constant components (, ) can be replaced by the individual fixed effect () and the time fixed effect (), respectively. The transitory income and preference shock (, ) can be replaced by the idiosyncratic income shock () and the observed family characteristics (), respectively. Under these assumptions, the empirical specification is given as follows:

(7)

where is a random error term. If the idiosyncratic shocks are well compensated by borrowings, gifts, and/or the sale of household assets, the estimated coefficient on the shock variable should be negative and statistically significant. To assess which channels were effectively used at that time, therefore, I regress the borrowing () and other temporary sources () separately on the shock variable, in the spirit of Fafchamps and Lund (2003).

4 Data

After World War I, many household surveys were conducted by local governments in the Japanese cities, to better understand the rapid formation among households of living styles (Nakagawa 1985). Analyzing the data obtained from these surveys helps reveal household management strategies in urban areas. To investigate household expenses in terms of risk-coping strategies, I compiled a unique survey report of the Report of Labor Research (RLR), which documents the monthly family budget of working-class households between July 1919 and July 1920.999In the current study, I use a reprinted edition of the original archives, as found in Tada (1991). Although James and Suto (2011) used 99 households in the RLR as an annual-level cross-sectional dataset, the current study is the first to use monthly data from this survey. In this survey, the Municipal Bureau of Labor Research of Osaka investigated the monthly income and expenditures of the households of 411 wage-earning and salaried workers within the Osaka city area. Although details in the sampling method are, unfortunately, not recorded—as is the case with other historical household survey datasets—the households had been selected through a labor union or directly at the factories. Accordingly, approximately % of the household heads were artisans. In Appendix B.1, I confirm that relative to the occupation structure in Osaka city at that time, all the occupations of the RLR households were biased towards manufacturing industries. Therefore, the current study targets these factory worker households.

Level Log-difference
Observations Mean Std. dev. Observations Mean Std. dev.
Consumption 1880 88.72 38.54 1574 0.02 0.37
     Food 1880 41.72 15.10 1574 0.01 0.33
     Housing 1880 7.86 4.92 1574 0.02 0.50
     Utilities 1880 3.76 2.85 1368 0.04 1.25
     Furniture 1880 2.27 4.96 1136 0.02 1.91
     Clothes 1880 10.74 15.56 1519 0.08 1.69
     Education 1880 0.53 1.51 528 -0.05 1.17
     Medical expenses 1880 3.79 4.46 1557 0.01 0.89
     Entertainment expenses 1880 5.58 6.84 1473 0.03 1.31
     Transportation 1880 1.14 1.90 1130 0.05 1.43
     Other 1880 5.68 11.42 1538 0.02 1.32
Disposable income 1880 91.26 41.51 1574 0.02 0.44
Disposable income (except for and ) ()
     Households receiving borrowings () 599 87.46 36.82
     Households receiving other temporary income () 1711 92.48 41.84
Income from borrowing () 599 5.09 13.07
     In borrowed months 154 19.78 16.88
Income from other income sources () 1711 10.62 22.92
     In received months 1023 17.76 15.22
Family structure
     Household size 237 4.00 1.61
     Children aged 0–5 (%) 237 14.83 16.02
     Children aged 6–9 (%) 237 7.40 11.26
     Children aged 10–12 (%) 237 4.13 9.09
     Children aged 13–16 (%) 237 5.43 10.38
     Men aged 17+ (%) 237 33.52 14.02
     Women aged 17+ (%) 237 34.69 15.72

Notes: The consumption and income figures listed in Column 1 are in Japanese yen. Disposable income is income excluding tax payments. The group of children aged 0–5 (%) is used as the reference group in the regression.
Table 1: Summary statistics

I extracted those households whose heads worked in the factories during the initial survey period.111111Although information on occupation had been subsequently revised if the household head changed occupations, 101010red

I focused on the households that were classified as being in the manufacturing industry throughout the sampled period. I did so because the current study looks to analyze dynamic behavior in the households of factory workers.

Correspondingly, 335 households remained of all 411 households. I then excluded 18 households for which there was no information on monthly income or family structure. An interesting characteristic of the RLR is that the investigatory periods reached 13 months. Thus, the investigator visited households monthly to check for data omissions and to collect survey books. While such a long-term panel survey allows me to test the full insurance hypothesis, a set of households dropped out—and were subsequently compensated for—during the study period. In fact, 78 households were dropped because they had been observed for only one month during the survey period. Finally, two households were dropped on account of having reported unrealistic income and expenditure values (i.e., exceeding 600 yen per month). Accordingly, data pertaining to 237 factory worker households were used in my empirical analyses.121212I conducted a two-sample

-test with unequal variances between the full sample and subsample of the RLR households. Differences in all household characteristics reported in Table 

1—such as income, consumption expenditures, and family structure among 411 households and the 237 factory worker households—were not statistically significant at the 1% level (not reported). Although I focus on the factory worker households, this finding implies that the RLR households have similar characteristics and thus could be classified into a similar social class.

To provide an overview of the sample’s characteristics, I first describe the family structure features. The current study focuses on nonsingle households. While items pertaining to income and expenditures had been recorded every month, information on family structure had been investigated in the initial survey month.131313To maintain the quality of the survey, the investigators visited all households and instructed them all once or twice per month (Tada 1991, pp. 11–12). This means that my measure of consumption was less likely to contain measurement error, and this in turn ensured the quality of the RLR survey would be high. In addition, my aggregate measure of income and expenditures captured macroeconomic trends, as shown in Figure 1 below. Table 1 shows the summary statistics of the data used. As for family structure, the average number of family members was four, which approximates the figures reported in the 1920 population census in Osaka city and those of a similar survey for factory worker households in Kyoto in the 1920s (Appendix B.2

). Almost all of the households include more than two persons, excepting three single-person households. However, this does not mean that the distribution of household size was highly skewed; rather, it had good variation (Appendix 

B.2). Second, the living standards in my sampled households could be classified as being in the standard range of factory worker households at that time. While the average monthly income for the RLR households was yen, that of factory worker households in Kyoto city between September 1926 and August 1927 was yen (Bureau of Social Welfare 1930, p. 47).141414I used consumer price indices for 1919–1920 and 1926–1927 (Bank of Japan 1986, p. 436). The household size in this survey is , suggesting household characteristics similar to those among RLR households. Overall, nonsingle urban working-class households employed in the manufacturing industry are the primary interest of the current study.

(a) Income and expenditures (in yen)
(b) Excess and deficiency (in yen)
Figure 1: Monthly income and expenditures
between June 1919 and June 1920

I next discuss trends in household income and expenditures. Figure 1 illustrates monthly income and expenditure information between June 1919 and June 1920. Excess values and deficiencies (i.e., total monthly income minus expenditures) are also shown in the figure, by month. First, overall, there were increasing trends in terms of both income and expenditures. This is consistent with the increasing trend, worldwide, in terms of living standards after World War I (Appendix B.3). However, one can see that both income and expenditures seemed to decrease after April 1920. This trend reflects the well-known recession that followed the war after March 1920 in Japan (Takeda 2002, pp. 9–11). These macroeconomic trends will be effectively controlled for in my empirical analyses.

Second, seasonality is clearly observed in December 1919 and January 1920. Both income and expenditures steeply increased in December. Although differences between income and expenditures were largely positive but fluctuated, this net income became statistically significantly negative in January (Figure (b)b). This was because, in order to prepare for the New Year events and customs, both labor income and consumption expenditures increased in December. Accordingly, workers took New Year’s holidays, and this could have significantly reduced their earnings. Third, the difference between income and expenditures also became large in March. This might relate to the fact that the fiscal year ends in March and starts in April.

Finally, Table 1

reports descriptive statistics for the variables used in my quantitative analysis. Ten categories of expenditures were used to test the full risk-sharing hypothesis—namely, food, housing, utilities, furniture, clothes, education, medical expenses, entertainment expenses, transportation, and other. As discussed in Section 

3, I use the first difference of the log of disposable income as a measure of idiosyncratic income shocks. Both the mean of the first difference of the log of disposable income and total consumption were positive in my RLR sample; this is consistent with the increasing trends discussed above.151515Note that the mean of the growth rate of aggregate consumption was equal to that of the growth rate of individual consumption. For the second specification in testing risk-coping mechanisms, I focused on the households that had income from borrowings and other sources. Income from other sources included both informal gifts and asset sales, such as money drawn from savings. This variable might include temporary earnings from family members other than the spouse and children (e.g., grandmothers or grandfathers). Such aggregation of income categories could disturb interpretations of the results. However, the average share of earnings by nonworking-age family members was indeed reported to be substantially small in the RLR sample. Unfortunately, the details of these additional sources of income (i.e., who earned this income and when) had not been documented, and the average earnings by these family members was reportedly 0.13 yen per diem, which was approximately only 3.7% of the total daily per-capita income (Municipal Bureau of Labor Research of Osaka 1921, p. 37). Therefore, one can consider the income from these sources as being mainly from gifts and asset sales. This means that I could still separately test the use of loans and/or other types of temporary income by using both variables.

5 Results

5.1 Descriptive analysis

(a) Disposable income
(b) Expenditure
(c) Correlation
Figure 2: Log-difference of disposable income and expenditure

Before formally testing the full risk-sharing in the regression framework, I first looked into the relationship between changes in expenditures and income, in a descriptive manner. Figure (a)a and (b)b show the log-difference of monthly disposable income and expenditures, respectively. The mean values of the log-difference of disposable income and expenditures were approximately and , respectively. While these values imply that both disposable income and expenditures increased on average during the sample period, Figure (a)a provides evidence that the households had still experienced negative income shocks. Accordingly, the households frequently experienced the loss of expenditures, as reported in Figure (b)b. Changes in the log of disposable income indeed ranged from roughly -2.5 to 2.0 (Figure (a)a), whereas changes in the log of expenditures ranged from roughly -1.5 to 1.5 (Figure (b)b).

Figure (c)c describes the relationship between the log-differences of monthly expenditures and disposable income, which shows a positive linear relationship. To further delve into this relationship, Figure 3 decomposes the total expenditures into 10 subcategories, as described in Table 1. A quick review of these figures suggests that there were similar positive relationships between changes in income and subcategories, except for housing and education expenditures.161616Appendix B.2 describes in finer detail the distribution of the subcategory expenditures. This suggests that there might be a few subcategories that had been less prone to being affected by income shocks. Based on these findings, in the next subsection, I start to test for full risk-sharing.

(a) Food
(b) Housing
(c) Utilities
(d) Furniture
(e) Clothing
(f) Education
(g) Medical
(h) Entertainment
(i) Transportation
(j) Other
Figure 3: Relationship between changes in disposable income and expenditure

5.2 Full risk-sharing

: , Number of
Coef. Std. error Coef. Std. error -Stat. -value observations
Panel A: June 1919 and June 1920
Total consumption 0.559 [0.067]*** 0.336 [0.041]*** 33.605 0.000 1574
Food 0.769 [0.084]*** 0.127 [0.041]*** 5.437 0.005 1574
Housing 0.142 [0.305] 0.000 [0.034] 4.151 0.017 1573
Utilities 0.743 [0.279]*** 0.117 [0.104] 0.746 0.475 1136
Furniture 0.751 [0.150]*** 0.470 [0.169]*** 3.903 0.022 1136
Clothes 1.106 [0.075]*** 0.283 [0.133]** 7.353 0.001 1519
Education 0.717 [0.087]*** 0.101 [0.129] 5.336 0.006 528
Medical expenses 0.703 [0.177]*** 0.199 [0.061]*** 7.111 0.001 1557
Entertainment expenses 0.954 [0.085]*** 0.362 [0.095]*** 7.366 0.001 1473
Transportation 0.994 [0.138]*** 0.268 [0.130]** 2.217 0.112 1130
Other 0.706 [0.182]*** 0.503 [0.131]*** 7.553 0.001 1538
Panel B: June 1919 and March 1920
Education 0.681 [0.096]*** 0.040 [0.142] 5.484 0.006 368

Notes: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Cluster-robust standard errors are in brackets. Estimates for the intercepts are not reported, but are included.

Table 2: Results of testing for full risk-sharing

Panel A of Table 2 presents the results for equation (4). As described in Section 3, if risk is efficiently shared, the coefficient on the growth rate of individual income becomes 0, while the coefficient on the growth rate of aggregate consumption becomes 1. This means that the null hypothesis of full risk-sharing is and . The -statistic -values reported in Column 7 of Table 2 show that the full risk-sharing hypothesis was rejected in most of the categories. For most of the rejected categories, idiosyncratic income shocks had statistically significantly positive impacts on household consumption growth. For instance, the estimated coefficient on the change in disposable income for total consumption was . Note again that the estimate of can range from 0 (for perfect smoothing) to 1 (for the total absence of insurance). Thus, the estimate obtained herein related to conditions far from perfect consumption smoothing. As discussed in the introduction, an important advantage of my testing method is that it allows me to compare my estimates to those obtained in previous studies. If I compare the estimates with those in the 1990s Japanese economy (i.e., a near-complete market), as produced by Kohara (2002), the estimate obtained here () was roughly six times larger. For food expenditures, the estimated coefficient was , which is roughly times greater than that in 1990s Japan. This result implies that factory worker households in Osaka city at that time could not effectively deal with idiosyncratic shocks. This finding is implicative, because in the 1920s, Osaka was as urbanized an economy as Tokyo.171717In fact, Osaka city was the second-largest city in Japan at that time. The population figures of Osaka and Tokyo cities in 1920 were and , respectively (Statistics Bureau of the Cabinet 1925, 1929b). These two cities therefore accounted for approximately % () of Japan’s total urban population at the time (Statistics Bureau of the Cabinet 1929a).

However, for two categories (i.e., utilities and transportation), the null hypothesis was not rejected at the 5% level. This result suggests that idiosyncratic shock was more likely to take place among these consumption categories, given the fixed nature of those goods. In addition, the estimated coefficient on the change in disposable income for housing was very close to 0 and statistically insignificant, suggesting that the rent expenditure was less likely to respond to idiosyncratic income shocks. In fact, expenditures for rent, utilities, and transportation fees, especially for commutation, needed to be paid, regardless of fluctuations. The distributions of these items, as reported in Appendix B.2, indeed indicated smaller variances in these subcategories.

It is also noteworthy that education expenditures were less likely to respond to shocks. The estimated coefficient on the change in disposable income for education was , and therefore not statistically significantly different from 0. This implies that the educational investment was relatively robust against idiosyncratic income shocks; thus, children in working-class households might have been able to attend school, regardless of idiosyncratic shocks. However, since Japan’s academic year starts in April and ends in March, some of the children might have graduated from schools at the end of March 1920, which would have disturbed education expenditures. To address this issue, I also ran a regression while using an alternative cut-off period between June 1919 and March 1920, in Panel B of Table 2. One can confirm that the result remains unchanged, and this suggests that graduations did not disturb the results. This finding is also consistent with the raw descriptive relationship reported in Figure (f)f.

In using the testing specification used in this subsection, however, I assumed that the aggregate measure of consumption ( in Eq 4) can effectively capture macroeconomic shocks. Moreover, Martin Ravallion and Shubham Chaudhuri (1997) examined how parameter estimates will be biased downwards if the first difference of the log of individual income contains both idiosyncratic and aggregate income effects.181818See also Ligon (2008) for details on the problems surrounding private information and limited commitment. The omitted variable bias—which is caused by ignoring the consumption of nontradable goods or self-produced goods, in the case of international consumption risk-sharing and village economies—has been signaled (Lewis 1996). However, the current study focused on purchased consumption, since self-production was rare among working-class households in large cities at that time. Although it was difficult to remove idiosyncratic seasonalities, I further considered the two-way fixed-effect model to separate aggregate risk from the idiosyncratic income effect (Appendix C.2). I confirmed that the results would remain largely unchanged if I were to control for macroeconomic shocks by using month–year fixed effects rather than the aggregate measure of consumption.

5.3 Risk-coping strategies

The foregoing results suggest that housing, utilities, transportation, and education expenditures were less likely to be affected by idiosyncratic shocks. One plausible explanation for this result is intrahousehold resource allocation: households could allocate their resources for these consumption categories to food, clothes, entertainment, and so on.191919In addition to the resource allocation among categories, households could also alter their diet in order to adjust food expenditures (Öberg 2016). However, another possibility was that this result implies that certain risk-coping mechanisms could mitigate fluctuation in subcategories of household consumption, even though the full risk-sharing hypothesis was rejected. Since the results of the above test of full risk-sharing provided little information as to how risk was shared, I investigated whether borrowings and/or other temporary sources of income serve to efficiently share risk among households.

(2) Borrowings (3) Other sources
Household characteristics (1) Full sample Yes No Difference Yes No Difference
Average monthly disposable income (yen) 79.54 70.51 82.95 12.44 79.71 78.56 -1.15
Size 4.00 4.22 3.92 -0.30 4.05 3.69 -0.37
Children aged 0–5 (%) 14.83 15.82 14.46 -1.37 15.00 13.88 -1.12
Children aged 6–9 (%) 7.40 10.16 6.35 -3.81 7.41 7.29 -0.13
Children aged 10–12 (%) 4.13 4.56 3.97 -0.59 4.20 3.74 -0.45
Children aged 13–16 (%) 5.43 6.54 5.01 -1.53 5.42 5.48 0.06
Men aged 17+ (%) 33.52 30.10 34.81 4.71 33.64 32.81 -0.83
Women aged 17+ (%) 34.69 32.81 35.40 2.59 34.33 36.79 2.46
Number of households 237 65 172 202 35
Notes: and represent statistical significance at the 1% and 5% levels, respectively. The average monthly disposable income is the average of disposable income (excluding temporary income from borrowings and other sources). Data on family structure were collected in the initial month of the survey.
Table 3: Comparison of the means of variables among subsamples

I first investigated the observable household characteristics among various groups with respect to risk-coping strategies (Table 3). Column 1 indicates the means for all households, while Column 2 compares the means of households with borrowings to those of other households. Column 3 compares the means between the households that received temporary income (other than borrowings) and otherwise.

Column 2 implies that there was a statistically significant difference in average monthly disposable income between the two groups. The monthly disposable income of households with borrowings was approximately 12.4 yen lower than that of the other households. Differences in family characteristics were relatively unclear, but among households with borrowings, the share of those with children aged 6–9 was higher than that among other households, by roughly 3.8 percentage points. Additionally, among households with borrowings, the share of those with men aged 17+ was lower than that among other households, by 4.7 percentage points. This implies that the households with more school-aged children and fewer adult men were more likely to rely on loans; this in turn suggests that households with younger heads had to borrow money to take care of their children. Although I could determine from the binary choice models no such clear relationship between borrowing and family structure (Appendix C.2), I tried to include the family structure variables in the following statistical analysis.202020In fact, the additional labor supply would also be an alternative means of coping with income shocks (Horrell and Oxley 2000; Moehling 2001). The increased dependence on the household head’s earnings due to urbanization could be associated with the increased demand for market purchases of insurance (di Matteo and Emery 2002). In contrast, Column 3 shows that there were no statistically significant differences between households that received any other temporary sources of income and the other households. This may indicate that the withdrawal of savings and receiving gifts were less likely to be influenced by credit constraints.

(a) Borrowings
(b) Other sources
Figure 4: Average monthly net income and temporary income,
between June 1919 and June 1920

Figure 4 then illustrates average monthly net income and temporary income between June 1919 and June 1920. Figure (a)a presents net income and borrowings, while Figure (b)b presents net income and temporary income from other sources. Figure (a)a shows a relatively clear negative correlation between net income and borrowings. The average monthly income from borrowings was approximately yen, as reported in Table 1. If I were to exclude the household–month cells without borrowings, however, this value would become yen, accounting for approximately % of the average disposable income (; Table 3). This implies that borrowings might have constituted an important risk-coping strategy in Japan’s working-class households at that time. In contrast, the correlation illustrated in Figure (b)b is unclear.212121Despite this ambiguity, I confirmed a weak negative correlation upon excluding the observation of December 1919 in Figure (b)b, which reflected clear seasonality. See Appendix C.3 for details of the result. The average monthly income from these income sources was approximately yen, thus accounting for approximately % of the average disposable income (; Table 3). Although the share was smaller than that of the borrowings, it comprised one-fifth of total disposable income.

Borrowings Other sources
(1) (2) (3) (4) (5) (6) (7) (8)
Disposable income () -0.115** -0.122** -0.671** -0.516*** -0.192*** -0.198*** -0.418*** -0.457***
(0.049) (0.047) (0.262) (0.143) (0.040) (0.039) (0.100) (0.097)
Model Linear Linear Nonlinear Nonlinear Linear Linear Nonlinear Nonlinear
Household and month–year FE Yes Yes Yes Yes Yes Yes Yes Yes
Family structure Quarter FE No Yes No Yes No Yes No Yes
Observations 599 599 599 599 1711 1711 1711 1711
Censored 445 445 445 445 688 688 688 688

Notes: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Results from the fixed-effects Tobit model, as proposed by Honoré (1992), are reported in Columns 3–4 and 7–8. A quadratic loss function was applied for the estimation, to ensure computational tractability. Robust standard errors are in parentheses. For the linear models, standard errors are clustered at the household level.

Table 4: Results of testing risk-coping mechanisms

Finally, I looked at the results for the second specification of equation (7).222222 Note again that I limited the sample to the households that had received any income from borrowings and/or other sources during the study period as reported in Table 3. As mentioned, although I found the tendency for households with higher disposable income to be less likely to borrow money, correlations between family structure and the choice of these additional income sources were statistically insignificant (Appendix C.2). Columns 1–4 present the results for borrowings, while Columns 5–8 present the results for the other sources. Columns 2, 4, 6, and 8 also include interaction terms between the family structure variables listed in Table 1 and quarter fixed effects.232323 I used the quarter fixed effects rather than month–year fixed effects, because optimization in the nonlinear fixed-effect model is computationally demanding, especially when the models are complex. Nonlinear specifications with month-year fixed effects are indeed no longer computationally practical. In the linear models, however, I confirmed that the results would remain unchanged if I used the interaction of the month–year fixed effects (not reported). Columns 3–4 and 7–8 employ the fixed-effects Tobit model proposed by Honoré (1992), to address the potential attenuation effects induced by censoring.

In Column 1, the estimate is negative and statistically significant. This result remains unchanged if I include the family structure variables in Column 2. This finding implies that the composition of family members might be irrelevant in these matters. The absolute value of the estimate is increased approximately six-fold () if I take data-censoring into account in Column 3. This is indeed consistent with the fact that more than 70% of the observations were censored, as reported in the final row of Table 4. The estimate in Column 4 indicates that a 1-yen increase in disposable income reduced temporary income from borrowings by yen.

A similar relationship can be seen for the other income sources. The estimate is statistically significantly negative and is also robust against including the family structure variables (Column 6). The absolute value of the estimate is increased approximately two-fold () after dealing with the censoring issues in Column 7.242424A smaller increment compared to the case of borrowings was found to be consistent with the lower proportion of censored observations (i.e., 40% []). The estimate in Column 8 implies that a 1-yen increase in disposable income reduced the temporary income from the other sources by yen. A relatively smaller magnitude than that for borrowing is consistent with the raw descriptive relationship illustrated in Figure (b)b.

These results suggest that the magnitude of idiosyncratic shocks on borrowing was slightly greater than that on the other temporary income sources, such as gifts and savings withdrawals. Despite the smaller amount, therefore, the borrowing might have functioned well in response to income shocks at that time.252525I confirmed that the results would remain largely unchanged if I were to include the household-specific time trend. See Appendix C.4 for the results. This may be because my sample period was relatively short (i.e., 13 months), and also because the macroeconomic trend observed in Figure 1 was effectively captured by the month–year fixed effects.

6 Discussion

To analyze whether the risk had been shared efficiently and which categories of consumption were robust to shocks at that time, I tested the hypothesis of full risk-sharing. The results suggest that the housing, utilities, and transportation expenditures failed to reject the null hypothesis. While these categories have a durable nature, I also found that children’s education expenditures were robust to idiosyncratic shocks. This is consistent with the fact that the average years of schooling increased throughout the early 20th century, and accelerated Japan’s rapid economic growth in the postwar period (World Bank 1993). To analyze the mechanism of risk-sharing, I estimated the income elasticity of the temporary income sources. I found that temporary income from borrowings and from other sources (e.g., savings and gifts) were statistically significantly increased if household income were reduced on account of idiosyncratic shocks. The estimated magnitude was larger on the borrowings than on the other temporary income sources. Although I could not break down the latter income sources, this means that microfinancial lending institutions, saving institutions, and informal gifts might have played important roles in mitigating income shocks.

Considering these results, I next examined the kinds of lending institutions that had been widely used among Japan’s working-class households at that time. Regarding loan institutions, one can claim the importance of pawn shops (shichiya). Although pawnbroking had been in decline in the early 20th century in Europe, pawn shops remained an essential microloan institution throughout 20th-century Japan; see Shibuya et al. (1982), Murhem (2015), and Kenttä (2016). Official report of pawn shops document the fact that there were 983 pawn shops in Osaka city in October 1919; this represents a pawn shop “density” of 1,660 citizens per shop. According to the report, artisans were the most frequent users of the shops, accounting for % of all users.262626In all, % are workers in commerce. In terms of average borrowing, % of the total amount loaned by the shops had been borrowed by artisans (Osaka City Office 1920, p. 102, 104). The average amount of money borrowed by artisans per event was yen (Osaka City Office 1920, pp. 122–23). This figure accounts for approximately % () of the mean value of borrowings among my RLR households (Table 1). If I use the median value of borrowings, this figure reaches % (). Thus, pawn shops clearly constituted a key microfinancial lending institution in Japan at that time, and they were widely used by factory workers. Indeed, the report of the Osaka City Office states that the pawn shop was a “very major financial institution among the artisans” (Osaka City Office 1920, p. 153). Since inexpensive clothes were the dominant pawns at that time, the workers were able to borrow money without risk of fall into heavy debt (Shibuya et al. 1982; Saito 1989). The pawn shops had thus been an access-friendly loan institution for the working-class households.

This historical fact suggests, furthermore, that it had been difficult to borrow money from other lending institutions. One example of a lending institution other than a pawn shop is a money-lending business (kinsen kashitsuke gyō). In 1926, there were 196 money-lending institutions in Osaka city, and the per-event loan amount generally ranged from 20 to 200 yen—an amount considerably larger than that offered at pawn shops (Osaka City 1934, pp. 240–41). These lending institutions including banks were less likely to be used by the working-class households due to the costly collateral (Shibuya 2000; 2001). Although the other potential example is the informal gifts, the gifts from their personal networks had also been less likely to be used among working-class. A official survey in Tokyo in 1933 indeed shows that approximately 70% of total borrowing in the working households was from pawnshop, whereas that ratio from their relatives and friends was only 21% (Inoue 2019).

In contrast with these formal and informal institutions, precautionary saving had been an important alternative risk-coping strategy in urban working-class households in Japan (James and Suto 2011). Postal saving (yūbin kyoku) and saving banks (chochiku ginkō) were saving institutions widely used around 1920.272727See Okazaki (2002) and Tanaka (2014) for historical overviews of these saving institutions. The number of postal saving accounts in Osaka at the end of December 1920 was , accounting for % of Osaka’s total population (Osaka Chamber of Commerce 1922, p. 6).282828The number of people in Osaka prefecture as per the 1920 population census was (Statistics Bureau of the Cabinet 1928, p. 2). The average savings per account was yen—an amount roughly equivalent to % () of the average monthly disposable income of RLR households (Table 1). As for saving banks, there were 36 banks in Osaka city in 1921.292929The description of saving banks is obtained from the Tokyo Institute for Municipal Research (1925, pp. 82–95). In 1923, there were approximately 62 accounts per 100 Osaka citizens; this implies that more than one-half of the city’s citizens had an account in a saving bank. The average savings set aside by households whose head was employed in manufacturing was reportedly yen; again, this amount is roughly % () of the average monthly disposable income of RLR households. These facts suggest that despite having only small amounts of savings, factory workers were able to draw on their savings in the event of economic hardship.

In summary, urban factory worker households in Osaka in the early 1920s might have been able to maintain themselves in the face of financial hardship, in the sense that they could cope with making payments for essential expenditures such as rent, utilities, and commuting. I suggest that they were able to cope with idiosyncratic income shocks by using microfinancial lending institutions, their precautionary savings, and presumably, informal gifts, although gifts are fundamentally unobservable. My estimates suggest that borrowing responds more sensitively than the other temporary income sources. Despite these findings, the full risk-sharing hypothesis is rejected for overall expenditure. This suggests that the households prioritized fixed payments—such as rent, utilities, and commutation fees—when facing income shocks. Although they could not fully smooth their consumption, they might have managed to live daily life by leveraging microfinancial and saving institutions.

An important finding of the current study is that expenditures relating to children’s education were also robust to idiosyncratic income shocks. Since education expenditures include both school fees and the cost of textbooks and stationery, this result implies that parents might have considered investments in their children’s education an important expenditure. This speculation is consistent with Japan’s documented sustainable improvements in terms of average years of education, which helped accelerate Japan’s postwar rapid economic growth (see the introduction). Therefore, the results herein suggest that microfinancial lending institutions and saving institutions might have contributed to Japan’s accumulation of human capital during its period of industrialization.

These findings help fill a gap in the current body of knowledge within the literature, as described in Section 2; in so doing, they help explain why the living standards of Japanese factory worker households remained relatively stable across the interwar period. Future research should further investigate the details of intrahousehold resource allocation among working-class households in times of economic hardship.

References

  • [1] Attanasio, Orazio P., and Luigi Pistaferri. “Consumption Inequality.” Journal of Economic Perspective 30, no. 2 (2016): 3–28.
  • [2] Baines, Dudley., and Paul Johnson. “Did They Jump or Were They Pushed? The Exit of Older Men from the London Labor Market, 1929–1931.” Journal of Economic History 59, no. 4 (1999): 949–71.
  • [3] Bassino, Jean-Pascal. “Inequality in Japan (1892–1941): Physical Stature, Income, and Health.” Economics and Human Biology 4, no. 1 (2006): 62–88.
  • [4] Bloom, David E., and Jeffery G. Williamson. “Demographic Transitions and Economic Miracles in Emerging Asia.” World Bank Economic Review 12, no. 3 (1998): 419–55.
  • [5] Boyer, George R. “Poor Relief, Informal Assistance, and Short Time during the Lancashire Cotton Famine.” Explorations in Economic History 34, no. 1 (1997): 56–76.
  • [6] Carroll, Christopher D. “Buffer-stock Saving and the Life Cycle/Permanent Income Hypothesis.” Quarterly Journal of Economics 112, no. 1 (1997): 1–55.
  • [7] Cochrane, John H. “A Simple Test of Consumption Insurance.” Journal of Political Economy 99, no. 5 (1991): 957–76.
  • [8] Deaton, Angus. Understanding Consumption. New York: Oxford University Press, 1992.
  • [9] Dercon, Stefan. Insurance Against Poverty. New York: Oxford University Press, 2004.
  • [10] Dercon, Stefan., and Pramila Krishnan. “In Sickness and in Health: Risk Sharing within Households in Rural Ethiopia.” Journal of Political Economy 108, no. 4 (2002): 688–727.
  • [11] di Matteo, Livio., and J.C. Herbert Emery. “Wealth and the Demand for Life Insurance: Evidence from Ontario, 1892.” Explorations in Economic History 39, no. 4 (2001): 446–69.
  • [12] Emery, J.C. Herbert. “’Un-American’ or Unnecessary? America’s Rejection of Compulsory Government Health Insurance in the Progressive Era.” Explorations in Economic History 47, no. 1 (2010): 68–81.
  • [13] Fafchamps, Marcel. Rural Poverty, Risk and Development. Cheltenham: Edward Elgar Publishing, 2003.
  • [14] Fafchamps, Marcel., and Susan Lund. “Risk-sharing Networks in Rural Philippines.” Journal of Development Economics 71, no. 2 (2003): 261–87.
  • [15] Foster, Andrew D. “Credit Markets and Child Growth in Low-income Rural Areas.” Economic Journal 105, no. 430 (1995): 551–70.
  • [16] Gazeley, Ian., and Andrew Newell. “The End of Destitution: Evidence from Urban British Working Households 1904–37.” Oxford Economic Papers 64 (2012): 80–102.
  • [17] Gertler, Paul., and Jonathan Gruber. “Insuring Consumption against Illness.” American Economic Review 92, no. 1 (2002): 51–70.
  • [18] Godo, Yoshihisa. “Estimation of Average Years of Schooling for Japan, Korea and the United States.” PRIMCED Discussion Paper Series 9 (2011): 1–53.
  • [19] Hayashi, Fumio, Joseph Altonji, and Laurence Kotlikoff. “Risk-sharing between and within Families.” Econometrica 64, no. 2 (1996): 261–94.
  • [20] Honoré, Bo E. “Trimmed LAD and Least Squares Estimation of Truncated and Censored Regression Models with Fixed Effects.” Econometrica 60, no. 3 (1992): 533–65.
  • [21] Horrell, Sara, David Meredith, and Deborah Oxley. “Measuring Misery: Body Mass, Again and Gender Inequality in Victorian London.” Explorations in Economic History 46, no. 1 (2009): 93–119.
  • [22] Horrell, Sara., and Deborah Oxley. “Crust or Crumb: Intra-household Resource Allocation and Male Breadwinning in Late Victorian Britain.” Economic History Review 52, no. 3 (1999): 494–522.
  • [23] Horrell, Sara., and Deborah Oxley. “Work and Prudence: Household Responses to Income Variation in Nineteenth-century Britain.” European Review of Economic History 4, no. 1 (2000): 27–57.
  • [24] Horrell, Sara., and Deborah Oxley. “Bringing Home the Bacon? Regional Nutrition, Stature, and Gender in the Industrial Revolution.” Economic History Review 65, no. 4 (2012): 1354–79.
  • [25] Horrell, Sara., and Deborah Oxley. “Bargaining for Basics? Inferring Decision Making in Nineteenth-century British Households from Expenditure, Diet, Stature, and Death.” European Review of Economic History 17, no. 2 (2013): 147–70.
  • [26] Hyodo, Tsutomu. The Evolution of Labor–Capital Relations in Japan. [in Japanese] Tokyo: University of Tokyo Press, 1971.
  • [27] Inoue, Tatsuki. “The role of pawnshops in risk coping in early twentieth-century Japan.” arXiv: 1905.04419v1 (2019).
  • [28] Islam, Asadul., and Pushkar Maitra. “Health Shocks and Consumption Smoothing in Rural Households: Does Microcredit Have a Role to Play?” Journal of Development Economics 97, no. 2 (2012): 232–43.
  • [29] James, John A., and Isao Suto. “Early Twentieth-century Japanese Worker Saving: Precautionary Behaviour before a Social Safety Net.” Cliometrica 5, no. 1 (2011): 1–25.
  • [30] Kantor, Shawn E., and Price V. Fishback. “Precautionary Saving, Insurance, and the Origins of Workers’ compensation.” Journal of Political Economy 104, no. 2 (1996): 419–42.
  • [31] Kase, Kazutoshi. “A Study of Proposals of Unemployment Insurance Schemes in Pre-war Japan.” [in Japanese] Shakaikagaku Kenkyu 58 (2006): 125–55.
  • [32] Kenttä, Tony. “When Belongings Secure Credit…: Pawning and Pawners in Interwar Borås.” Ph.D. diss., Uppsala University, 2016. http://www.diva-portal.org/smash/record.jsf?pid=diva2%3A972458&dswid=8381
  • [33] Kiesling, Lynne L. “Institutional Choice Matters: The Poor Law and Implicit Labor Contracts in Victorian Lancashire.” Explorations in Economic History 33, no. 1 (1996): 65–85.
  • [34] Kohara, Miki, Fumio Ohtake, and Makoto Saito. “A Test of the Full Insurance Hypothesis: The Case of Japan.” Journal of the Japanese and International Economies 16, no. 3 (2002): 335–52.
  • [35] Lewis, Karen K. “What Can Explain the Apparent Lack of International Consumption Risk Sharing?” Journal of Political Economy 104, no. 2 (1996): 267–97.
  • [36] Ligon, Ethan. “Risk Sharing.” In The New Palgrave Dictionary of Economics, Second Edition, edited by Steven N. Durlauf and Lawrence E. Blume. New York: Palgrave Macmillan, 2008.
  • [37] Mace, Barbara J. “Full Insurance in the Presence of Aggregate Uncertainty.” Journal of Political Economy 99, no. 5 (1991): 928–56.
  • [38] Mason, Andrew. “Population and the Asian Economic Miracle.” Asia–Pacific Population and Policy 43 (1997): 1–4.
  • [39] Meyer, Bruce D., and James X. Sullivan. “Consumption and Income Inequality in the U.S. since the 1960s.” NBER Working Paper Series, Vol. 23655 (2016): 1–48.
  • [40] Minami, Ryoshin. The Economic Development of Japan. [in Japanese] Tokyo: Toyō Keizai Shinpō Sha, 2002.
  • [41] Moehling, Carolyn M. “Women’s Work and Men’s Unemployment.” Journal of Economic History 61, no. 4 (2001): 926–49.
  • [42] Moriguchi, Chiaki. “The Evolution of Employment Relations in U.S. and Japanese Manufacturing Firms, 1900–1960: A Comparative Historical and Institutional Analysis.” NBER Working Paper Series, Vol. 7939 (2000): 1–69.
  • [43] Moriguchi, Chiaki. “Implicit Contracts, the Great Depression, and Institutional Change: A Comparative Analysis of U.S. and Japanese Employment Relations, 1920–1940.” Journal of Economic History 63, no. 3 (2003): 625–65.
  • [44] Mosk, Carl. “The Decline of Marital Fertility in Japan.” Population Studies 33, no. 1 (1979): 19–38.
  • [45] Murhem, Sofia. “Credit for the Poor: The Decline of Pawnbroking 1880–1930.” European Review of Economic History 20, no. 2 (2015): 198–214.
  • [46] Nakagawa, Kiyoshi. A City Lower Layer of Japan. [in Japanese] Tokyo: Keisō Shobō, 1985.
  • [47] Öberg, Stefan. “Did the Poor Pay More? Income-related Variations in Diet and Food Quality among Urban Households in Sweden 1913–1914.” Rivista Di Storia Economica 32, no. 2 (2016): 211–48.
  • [48] O’Connell, Sean., and Chris Reid. “Working-class Consumer Credit in the UK, 1925–60: The Role of the Check Trader.” Economic History Review 58, no. 2 (2005): 378–405.
  • [49] Odaka, Konosuke. “Japanese-style Labour Relations.” In The Japanese Economic System and Its Historical Origins, edited by Tetsuji Okazaki and Masahiro Okuno-Fujiwara. Oxford: Oxford University Press, 1999.
  • [50]

    Ogasawara, Kota, and Genya Kobayashi. “The Impact of Social Workers on Infant Mortality in Inter-war Tokyo: Bayesian Dynamic Panel Quantile Regression with Endogenous Variables.”

    Cliometrica 9, no. 1 (2015): 97–130.
  • [51] Okazaki, Tetsuji. “Selection, Exits and Efficiency in the Japanese Banking Industry: An Historical Perspective.” [in Japanese] In Strategy for the financial revitalization in Japan, edited by Makoto Saito. Chuō Keizai Sha, 2002.
  • [52] Ravallion, Martin, and Shubham Chaudhuri. “Risk and Insurance in Village India: Comment.” Econometrica 65, no. 1 (1997): 171–84.
  • [53] Rose, Elaina. “Consumption Smoothing and Excess Female Mortality in Rural India.” Review of Economics and Statistics 81, no. 1 (1999): 41–49.
  • [54] Rosenzweig, Mark R. “Implicit Contracts and the Family in Rural Areas of Low-income Countries.” Economic Journal 98, no. 393 (1998): 1148–70.
  • [55] Saaritsa, Sakari. “Informal Transfers, Men, Women, and Children: Family Economy and Informal Social Security in Early 20th Century Finnish Households.” History of the Family 13, no. 3 (2008): 315–31.
  • [56] Saaritsa, Sakari. “The Poverty of Solidarity: The Size and Structure of Informal Income Smoothing among Worker Households in Helsinki, 1928.” Scandinavian Economic History Review 59, no. 2 (2011): 102–27.
  • [57] Scott, Peter M., and James Walker. “Working-class Household Consumption Smoothing in Interwar Britain.” Journal of Economic History 72, no. 3 (2012): 797–825.
  • [58] Shibuya, Ryuichi, Kameji Suzuki, and Shojiro Ishiyama. Pawn Shops in Japan. [in Japanese] Tokyo: Waseda University Press, 1982.
  • [59] Shibuya, Ryuichi. Kōrigashi kinyū no tenkaikōzō [in Japanese] Tokyo: Nihon Tosho Center, 2000.
  • [60] Shibuya, Ryuichi. Shomin kinyu no tenkai to seisakutaiō. [in Japanese] Tokyo: Nihon Tosho Center, 2001.
  • [61] Saito, Horoshi. Shichiyashi no kenkyū [in Japanese] Tokyo: Shinhyōron, 1989.
  • [62] Taira, Koji. Economic Development and the Labor Market in Japan. New York and London: Columbia University Press, 1970.
  • [63] Takeda, Harehito. “Business Cycle and Economic Policy.” [in Japanese] In Japanese Economic History, Vol. 3, edited by Ishii Kanji, Hara Akira, and Takeda Haruhito. Tokyo: University of Tokyo Press, 2002.
  • [64] Tanaka, Hikaru. “Postal Savings and Mass Saving Behavior in the Early 20th Century: A Case Study of Mishima City in Shizuoka Prefecture.” [in Japanese] Rekishi to keizai 54, no. 2 (2012): 16–31.
  • [65] Townsend, Robert M. 1994. “Risk and Insurance in Village India.” Econometrica 62(3): 539–91.
  • [66] World Bank. The East Asian Miracle: Economic Growth and Public Policy. New York: Oxford University Press, 1993.
  • [67] Yazawa, Hirotake. Income Distribution and Family Economy in Modern Japan. [in Japanese]. Tokyo: Nihontosho Center, 2004.

Documents

  • [1] Bank of Japan. The Hundred Year History of the Bank of Japan, Source Material, Vol. 1. [in Japanese] Tokyo: Bank of Japan, 1986.
  • [2] Bureau of Social Welfare. Report for the Living Standards Surveys. [in Japanese] Tokyo: Bureau of Social Welfare, 1930.
  • [3] Municipal Bureau of Labor Research of Osaka. “Report of Labor Research (Series 2, 4, 5, and 8).” [in Japanese] In Collection of Household Budget Surveys in the Taisho Era, edited by Yoshizo Tada. Tokyo: Seishi Sha, 1919–1920.
  • [4] Municipal Bureau of Labor Research of Osaka. “Report of Labor Research (Series 10), Cost of Living among Laborers in Osaka, Japan.” [in Japanese] In Collection of Household Budget Surveys in the Taisho Era, edited by Yoshizo Tada. Tokyo: Seishi Sha, 1921.
  • [5] Osaka Chamber of Commerce. Annual Statistical Report, 1920 Edition. [in Japanese] Osaka: Osaka Chamber of Commerce, 1922.
  • [6] Osaka City Office. “Survey on Pawn Shops.” [in Japanese] In Report of Labor Research, Series 4, edited by Municipal Bureau of Labor Research of Osaka. Osaka: Osaka City Office, 1920.
  • [7] Osaka City Office. History of Osaka City, Vol. 4. [in Japanese] Tokyo: Osaka City Office, 1934.
  • [8] Statistics Bureau of the Cabinet. Population Census of Japan, Osaka Prefecture, 1920 Edition. [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1925.
  • [9] Statistics Bureau of the Cabinet. Population Census of Japan, Vol. 1, 1920 Edition. [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1928.
  • [10] Statistics Bureau of the Cabinet. Population Census of Japan, Vol. 2, 1920 Edition. [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1929a.
  • [11] Statistics Bureau of the Cabinet. Population Census of Japan, Tokyo Prefecture, 1920 Edition. [in Japanese] Tokyo: Statistics Bureau of the Cabinet, 1929b.
  • [12] Tada, Yoshizo. Collection of Household Budget Surveys in the Taisho Era. [in Japanese] Tokyo: Seishi Sha, 1991.
  • [13] Tokyo Institute for Municipal Research. City Saving Bank. [in Japanese] Tokyo: Tokyo Institute for Municipal Research, 1925.

Appendix Appendix A Theory appendix

Let us assume that there are N individuals named in the economy. Individual i receives an uncertain income , where stands for the state of the world at time t and derives instantaneous utility from consumption . The expected lifetime utility of individual i is expressed as

(8)

where is the social planner’s weight, which is the reciprocal of the marginal utility of each agent, and satisfies (Negishi, 1960); is the discount factor;

is the probability that state

takes place at time t; and is a preference shock.

The social planner maximizes the objective function (8) by choosing an allocation of consumption across individuals, subject to the aggregate resource constraint of the form:

(9)

By postulating a constant absolute risk aversion preference, , where is the coefficient of constant absolute risk aversion, I can obtain the first-order condition for individual i:

(10)

where is the Lagrange multiplier on the resource constraint (9) at time t. Taking the log of equation (10) and aggregating over agents, I obtain individual i’s consumption as follows:

(11)

For simplicity, I use the conventional notation for a random variable

and . Finally, equation (11) with this notation becomes:

(12)

where

(13)

Appendix Appendix B Data appendix

b.1 Household head’s occupation

To discuss the sampling feature of the RLR, I compare the share of occupation among household heads within the RLR households to relevant statistics obtained from the national population census. Table B.1 shows the share of men’s occupation among all households in Osaka prefecture or city in 1920, and that of RLR households in the first month of the survey.303030In the RLR dataset, information on the household head’s occupation can be obtained for 406 households (of all 411 households) in the first month of survey. Occupations were classified into nine social class categories, using the industrial classification of the first population census, which was conducted in 1920 (Statistics Bureau of the Cabinet 1929a, pp. I–IV).

Name of survey (1) 1920 Population census (2) 1920 Population census (3) The RLR
Survey area Osaka prefecture Osaka city Osaka city

Survey subject
Complete survey of the prefecture Complete survey of the city Sample from the city area
Survey month and year October 1920 October 1920 June 1919 to June 1920

Agriculture
14.1 0.8 0.0
Fisheries 0.5 0.1 0.0
Mining 0.3 0.3 0.0
Manufacturing 42.5 45.6 82.5
Commerce 25.8 34.0 2.0
Transport 8.7 10.6 4.2
Public service and professions 6.2 6.7 6.7
Housework 0.1 0.1 0.0
Other industry 1.8 1.8 4.4
Notes: Occupations were classified using the industrial classifications of the first population census, conducted in 1920. Individuals whose occupation were classified as “unknown” were dropped. Sources: Calculated by the author from the RLR dataset. Statistics Bureau of the Cabinet 1929a, pp. 8–11; Statistics Bureau of the Cabinet 1929b, pp. 84–85, 108–09.
Table B.1: Industrial structure in RLR households and population censuses

Column 1 presents the share of occupation of those men who participated in the work force across all of Osaka prefecture. Although 14% of men worked in the agricultural sector, most men worked in other industries. Since this figure for all of Japan was %, Osaka was a fairly well-industrialized prefecture at that time (Statistics Bureau of the Cabinet 1929, pp. 8–11). In fact, as shown in Column 2, approximately 90% of working men in Osaka city worked in the manufacturing, commerce, or transportation industries. Bear in mind, again, that the shares of men who worked in the manufacturing, commerce, and transportation industries across Japan were 22%, 12.8%, and 5.8%, respectively (Statistics Bureau of the Cabinet 1929a, pp. 8–11). Thus, these numbers reflect the fact that Osaka was at that time a representative industrialized city, on par with Tokyo city.

Monthly income (in yen)
Occupation Observations Median Std. dev.
Miscellaneous 185 72.18 39.20
Train crew 121 79.50 41.14
Artisan 2060 83.18 44.95
Commerce 29 127.60 203.86
Public official and teacher 55 141.45 50.07

Notes: “Miscellaneous” includes laborers, servants, and those engaged in domestic piecework.
Table B.2: Household head’s occupation and monthly income

Column 3 in Table B.1 contains data on the prevalence of occupation among RLR household heads, as a percentage share of all such households. Clearly, in terms of occupation, the RLR household heads are concentrated in the manufacturing industry; % of the men are artisans. To investigate the living standards of these households in the manufacturing sector, Table B.2 lists the median monthly income of each occupation, among the RLR households. The median monthly income of artisans was approximately yen. The occupations with lower monthly income than artisans were train crew members and workers in miscellaneous industries. While 17 heads were either train crew members or a supervisor with Osaka City Trams, 34 heads were laborers or servants, or engaged in domestic piecework. The median monthly income of these occupations was approximately and yen, respectively. In addition, eight household heads within the RLR worked in commerce. The median monthly income for commerce was higher than that of artisans, albeit from a small number of observations. In terms of median monthly income, public officials and teachers earned more than 140 yen per month. Therefore, artisans in the factories in Osaka city at that time, as observed in the RLR dataset, could be classified as being of the working class, but not as heading poor households.

b.2 Distribution of household size, income, and expenditure

Figure B.1: Numbers of family members in RLR households
Note: This figure describes the distribution of the average number of family members among RLR households.

As shown in Table 1, the average number of family members in RLR sample households was four. According to the first population census, executed in 1920, the average household size in Osaka prefecture and city was and , respectively (Statistics Bureau of the Cabinet 1925, pp. 2–3). The figure for Osaka city is close to that in the RLR sample. A similar survey of 231 factory worker households in Kyoto city, conducted in 1926, reports that the average household size was

; this too is in the vicinity of the value in my dataset (Bureau of Social Welfare 1930, pp. 46–47). These data findings assure me that my data sample does not contain a set of outliers or otherwise unusual household size values. In fact, Figure 

B.1 indicates that the distribution of the number of family members has a reasonable shape and is not highly skewed.

(a) Disposable income
(b) Expenditure
(c) Correlation
Figure B.2: Disposable income and expenditures (yen)

Figure (a)a and (b)b show the monthly disposable income and expenditures of the RLR households, respectively. The distribution of both figures is skewed rightwards, thus showing the typical distribution of income and expenditures. The mean values of disposable income and expenditures were and yen, respectively. Since a few outliers have been excluded (as noted in the main text), there are no specific observations here that take extremely large values. Figure B.3 presents the log-difference of monthly expenditures for the subcategories. There is plenty of variation in the differences of expenditures, except for housing (i.e., rent) and utilities expenditures. Figure B.4 describes the distribution of monthly income from borrowing and from other sources (including gifts and assets).

(a) Food
(b) Housing
(c) Utilities
(d) Furniture
(e) Clothing
(f) Education
(g) Medical
(h) Entertainment
(i) Transportation
(j) Other
Figure B.3: Distribution of the log-difference of the subcategories
(a) Borrowings and the others sources
(b) Borrowings
(c) Other sources
Figure B.4: Income from borrowings and other sources,
including gifts and changes in assets (yen)

b.3 Per-capita personal consumption expenditure

Figure B.5 illustrates the five-year moving average of per-capita personal consumption expenditures, in 1934–1936 prices and in yen, between 1912 and 1928. There was a clear, increasing trend after World War I and until the end of 1920s. As noted in Section 3, the recession after April 1920 and corresponding deflation took place in the early 1920s, and the Great Kantō Earthquake occurred in September 1923. Nonetheless, per-capita personal consumption expenditures increased during the early 1920s (Nakamura and Odaka 1989, pp. 36–37).

Figure B.5: Per-capita personal consumption expenditure,
in 1934–1936 prices (yen)
Note: The five-year moving average of per-capita personal consumption expenditures, in 1934–1936 prices, is illustrated in the figure. Source: M. Shinohara (1967, pp. 140–41).

Appendix Appendix C Empirical analysis appendix

c.1 Results for the two-way fixed-effect model

The baseline specification for testing the full risk-sharing hypothesis in Subsection 5.2 assumed that the aggregate measure of consumption captures macroeconomic shocks. To determine the sensitivity of this assumption, I use the two-way fixed-effect model instead of the baseline first difference model. The specification is as follows:

(14)

where is consumption, is disposable income, is a household fixed effect, is a month–year fixed effect, and is a random error term. Table C.1 presents the results. The results are similar to those reported in Table 2. The estimated coefficients on the changes in disposable income are statistically insignificant for housing, utilities, and education, as seen in Table 2.

Disposable income
Coef. Std. error -Squared Observations
Total consumption 0.358 [0.043]*** 0.4704 1574
Food 0.130 [0.043]*** 0.2921 1574
Housing 0.014 [0.037] 0.0883 1573
Utilities 0.195 [0.120] 0.0601 1136
Furniture 0.436 [0.201]** 0.1438 1136
Clothes 0.327 [0.148]** 0.2935 1519
Education -0.045 [0.161] 0.3082 528
Medical expenses 0.326 [0.088]*** 0.0790 1557
Entertainment expenses 0.501 [0.114]*** 0.1759 1473
Transportation 0.330 [0.175]* 0.1231 1130
Other 0.526 [0.155]*** 0.1214 1538
Notes: Household and month–year fixed effects are included in all specifications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Cluster-robust standard errors are in brackets.
Table C.1: Results of testing for full risk-sharing: two-way fixed-effect model

c.2 Differences in household characteristics

(1) Borrowing and the other sources (2) Borrowing (3) Other sources
Average monthly disposable income -0.001 -0.004*** -0.001
(0.001) (0.001) (0.001)
Size 0.023 0.038 0.033*
(0.018) (0.029) (0.019)
Share aged 6–9 (%) -0.001 0.004 -0.001
(0.002) (0.004) (0.002)
Share aged 10–12 (%) -0.000 -0.000 -0.000
(0.003) (0.003) (0.003)
Children aged 13–16 (%) 0.002 0.003 -0.001
(0.002) (0.003) (0.002)
Men aged 17+ (%) 0.002 -0.001 0.002
(0.002) (0.002) (0.002)
Women aged 17+ (%) -0.001 0.002 -0.000
(0.002) (0.002) (0.002)
Notes: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. The number of observations is 237. Robust standard errors are in parentheses.
Table C.2: Results of testing risk-coping mechanisms

Table C.2

provides the results from the linear probability model. Note that the results remained largely unchanged when I used nonlinear binary choice models, such as Probit and logistic regression models. Column 1 contains the results for a specification that uses a binary dependent variable that takes 1 for households that received any borrowings or other temporary income. Column 2 contains the results for a specification that uses a binary dependent variable that takes 1 for households that received any borrowings. Column 3 contains the results for a specification that uses a binary dependent variable that takes 1 for households that received any other temporary income (such as precautionary savings and gifts). The family structure variables show no clear relationships with the use of risk-coping strategies. An obvious result herein is that the estimated coefficient on the average monthly disposable income is statistically significantly negative in Column 2. The estimate implies that a 1-yen increase in the average monthly disposable income increases the probability of using any loan by

%. Since the standard deviation of the average monthly disposable income is

yen, this means that an increase of one standard deviation in this variable increases the probability by approximately %.

c.3 Correlation between net income and temporary income

Figure C.1 presents the relationship between average monthly net income (i.e., income minus expenditures) and the temporary incomes reported in Figure 4. Figure (a)a shows the correlation between net income and temporary income from borrowings. Figure (b)b shows the correlation between net income and temporary income excluding borrowings, while Figure (c)c replicates Figure (b)b but excludes December 1919.

(a) Borrowings
(b) Other sources
(c) Other sources: excluding December 1919
Figure C.1: Relationship between average monthly net income and temporary incomes,
between June 1919 and June 1920

c.4 Results from alternative specification

Borrowings Others
(1) (2) (3) (4)
Disposable income () -0.115** -0.120** -0.192*** -0.194***
(0.049) (0.060) (0.040) (0.049)
Household and month–year FEs Yes Yes Yes Yes
Household-specific time trend No Yes No Yes
Observations 599 599 1711 1711
Notes: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Cluster-robust standard errors are in brackets.
Table C.3: Risk-coping mechanisms: results from an alternative specification

Table C.3 presents the results from an alternative specification for testing risk-coping strategies. Columns 1 and 3 show the baseline results reported in Table 4. Columns 2 and 4 add a household-specific linear time trend that is not used in the baseline specification reported in Table 4. Clearly, my baseline results remain unchanged if I include the household-specific trend. As discussed in the main text, the optimization in the fixed-effect Tobit model is computationally demanding when the models are complex. Unfortunately, estimating the specification that includes the household-specific trend is indeed no longer computationally practical. The results should, however, remain largely unchanged if I were to include the trend term, as censoring tends to induce attenuation effects.

References used in the Appendices

  • [1] Nakamura, Takafusa., and Konosuke Odaka. Dual Structure. [in Japanese] Tokyo: Iwanami Shoten, 1989.
  • [2] Negishi, Takashi. “Welfare Economics and Existence of an Equilibrium for a Competitive Economy.” Metroeconomica 12, nos. 2–3 (1960): 92–97.
  • [3] Shinohara, Miyohei. Personal Consumption Expenditure, Long Term Economic Statistics, Vol. 6. [in Japanese] Tokyo: Tōyōkeizai Shinpōsha, 1967.

Statistical reports used in the Appendices

  • [1] Bureau of Social Welfare. Report for the Living Standards Surveys. [in Japanese] Tokyo, 1930.
  • [2] Statistics Bureau of the Cabinet. Population Census of Japan, Osaka Prefecture, 1920 Edition. [in Japanese] Tokyo, 1925.
  • [3] Statistics Bureau of the Cabinet. Population Census of Japan, Vol. 2, 1920 Edition. [in Japanese] Tokyo, 1929a.
  • [4] Statistics Bureau of the Cabinet. Population Census of Japan, Tokyo Prefecture, 1920 Edition. [in Japanese] Tokyo, 1929b.