Who faces higher prices? An empirical analysis based on Japanese homescan data 1. Kyosuke Shiotani (Bank of Japan 3 ) Abstract

Transcription

1 Who faes higher pries? An empirial analysis based on Japanese homesan data 1 Naohito Abe 2 (Hitotsubashi University) and Kyosuke Shiotani (Bank of Japan 3 ) Abstrat On the basis of Japanese household-level sanner data (alled homesan), we onstrut a household-level prie index and investigate the auses of prie differenes aross households. Similar to the results of Aguiar and Hurst (2007), we observe large prie differentials aross households. However, the differenes aross age and inome groups are small. In addition, we find that elderly people fae higher pries than younger ones do, whih ontradits Aguiar and Hurst (2007). The most important determinant of the prie level is the reliane on bargain sales; doubling the proportion of purhases at bargain sales dereases the prie level by about 2%, while shopping frequeny has limited effets on the prie level. JEL Classifiation Codes: D12, D30 Keywords: Prie Index, Household Consumption, Sanner data 1. Introdution Owing to reent tehnologial developments in data reation, numerous 1 We are grateful to omments from Andrew Leiester, Yukiko Abe, Sahiko Kuroda, Colin MKenzie, Tsutomu Watanabe, Yukinobu Kitamura, Mihio Suzuki, and seminar partiipants at Osaka University, annual meeting of the Japanese Eonomi Assoiation, and Asian Eonomi Poliy Review Conferene. This researh is an outome of the JSPS Grants-in-Aid for Young Sientist (S) Naohito Abe: The Institute of Eonomi Researh, Hitotsubashi University. Naka, Kunitahi Tokyo. nabe@ier.hit-u.a.jp. Phone: b 3 The views expressed in this paper are those of the authors and are not refletive of those of the Bank of Japan. 1

2 ommodity-prie researhers have begun to use not only traditional aggregates, suh as the onsumer prie index, but also information on miro-level ommodity pries. To date, ommodity-level prie data are used in various eonomi fields, suh as maroeonomis (Nakamura and Steinsson, 2008), international eonomis (Haskel and Wolf, 2001), and industrial eonomis (Baye et al., 2004). Reently, on the basis of ommodity-level homesan data, Aguiar and Hurst (2007) (hereafter, AH) found a violation of the law of one prie aross different age groups. More preisely, in the United States, elderly families fae lower pries for the same ommodities than younger families do. AH interpret their results as onsistent with the standard life-yle model of onsumption with endogenous deisions of shopping time. The mehanism is straightforward. Sine the opportunity osts of shopping for retired people are lower than they are for younger people, the elderly an spend more time searhing for lower pries. AH s findings have provided us with answers to several famous puzzles suh as the retirement-savings puzzle and the exess sensitivity of expenditure to predited inome shoks. Compared to standard onsumption panel data suh as the Panel Study of Inome and Dynamis, homesan data by AC Nielsen and Kantar provide us with detailed and frequent information on purhases at the household-ommodity level. However, most homesan data sets do not update household harateristis suh as inome and employment status regularly. In other words, for household data on inome and employment status, homesan data lak within variation. For this reason, AH did not ontrol for household-level fixed effets, rather, they used age and inome as instrumental variables (IVs) to deal with endogeneity in the determination of shopping frequeny. Beause IV estimates by AH are about 20 times larger than estimates without IVs, areful examinations of the effets of households' unobservable harateristis are required. This study onsiders the relationship between shopping behaviors and prie 2

3 levels on the basis of Japanese ommodity-level homesan data. The advantage of using Japanese homesan over those in the US is that in Japanese data, household harateristis are updated every year, enabling us to make more robust estimates. It is also worth noting that many previous researhers have onfirmed the existene of the retirement-savings puzzle or exess sensitivity in Japan 4. Therefore, by investigating the relationship between shopping behaviors and prie levels in Japan, we an hek whether the mehanisms proposed by AH play important roles in eonomies outside the US. As in the US, we find that ommodities are traded at various different pries in Japan. Figure 1 illustrates the distribution of the relative ommodity prie index, onstruted following AH. 5 The index takes a value of unity if the reorded prie is equal to the regional average prie. A value of 1.2 implies that the prie is 20% higher than the average. 6 The figure learly shows that the same produts are sold at very different pries. We also found that the prie level inreases with age, in sharp ontrast to AH s findings. Among several potentially important determinants of the prie index, the proportion of expenditures spent on bargain sales is the most important fator. By doubling the proportion of purhases at bargain sales, people an enjoy a redution in their prie level of about 2%, whih is onsistent with Griffith et al. (2009), who find signifiant savings from purhasing at bargain sales in the United Kingdom. Other shopping behaviors, suh as the frequeny of shopping, the degree of mass purhasing, and a preferene for high quality goods are all statistially signifiant. However, these 4 See Wakabayashi (2008) and Ogawa (1990) for studies on Japanese onsumption. 5 The figure shows the distribution of the household-level monthly prie index. The definition of the index will be given in the next setion. 6 In another strand of researh on household-level heterogeneity in the prie level, rather than the differenes in prie, but differenes in weights are onsidered. See Kitamura (2008) and Kuroda and Yamamoto (2010) for the Japanese ase. 3

4 behaviors are not quantitatively important. Our empirial results suggest that the prie-redution mehanism based on the opportunity osts of shopping provided by AH is not observed in Japan. In ontrast to the US, elderly people in Japan, who are supposed to have lower opportunity osts for shopping, tend not to use bargain sales and thus fae higher pries than do the young. This suggests that further investigation into shopping strategy, partiularly the determinants of purhasing at bargain sales, is neessary to understand the mehanism behind the prie-level differential aross families. 2. Data The data are from the Household Consumer Panel Researh (hereafter SCI) data set ompiled by Intage, a marketing ompany in Japan. SCI ontains the daily shopping information of approximately 12,000 households, randomly seleted from all prefetures (exept Okinawa) in Japan. 7 The sample households are restrited to married ouples. Using a barode reader, households are asked to san the barode of every ommodity they purhase. 8 In SCI, for every ommodity purhased, we an observe (1) the Japanese Artile Number (JAN), a unique ommodity identifier, 9 (2) the date of purhase, (3) the prie and quantity, and (4) the name of the store from whih the ommodity was purhased. Fresh foods (e.g., meat, fish, and vegetables) without 7 The sample households are seleted based on three-stage stratified area sampling. 8 Compared to the standard panel survey based on reolletion, SCI requires sample households to spend more time for data reation. Therefore, one might suspet that SCI is subjet to large sample seletion bias. To hek the potential bias, Intage uses the information of sales based on point of sales data of more than 4,000 retailers. Aording to Intage, the differenes between the two sanner data, one from households, and one from retailers, are not large. 9 JAN (Japanese Artile Number) ode is managed by The Distribution Systems Researh Institute. The ode is ompatible with the Universal Produt Code (UPC). Although the JAN ode is supposedly a unique identifier, some ompanies use the same JAN ode for different produts. Intage reates its own additional ode to deal with the repeated use of JAN odes. We use both JAN and Intage odes to identify ommodities. 4

5 barodes are exluded. This limitation is shared by AC Nielsen s US homesan data. The data we use in this paper over three years, from 2004 to Table 1 shows the distribution of the family omposition and omparisons with the Census and the KHPS (Keio Household Panel Survey) 10. Compared with the Census, the sample households of SCI ontain more family members. A similar bias an be found in the KHPS. Table 2 shows the age distribution of the sample wives in eah surveyed year, whih shows that the age distribution of SCI is lose to that in the Census. Table 3 reports the employment status of the wives, whih also shows that the proportion of housewives who have any paid job is lose to the figure in the Census. From Tables 1 3, it is safe to say that the distribution of households in SCI is not so different from the Census or other survey data Relative Prie Index Following AH, we onstrut the prie index as follows. Let us onsider a ommodity that belongs to a produt ategory C. Denote the prie of good i I purhased by household j J on date t T by j p, i t,, and the quantity by y j, i,t. The total expenditure by the household during time interval m an be written as, X j m j, i t C, i I, t m j, p, yi, t. (1) would be If the household purhases eah produt at the average prie, the expenditure 10 For details of KHPS, see 11 Abe and Niizeki (2010) ompared expenditures in SCI with those in the miro data of the Family Inome and Expenditure Survey by the Statistial Bureau. Consistent with the previous studies in the United Kingdom (Leiester and Oldfield, 2009), although the amount of expenditure from sanner data is smaller than that in the diary base survey, the orrelation between the expenditure and age or employment status are quite similar to eah other. 5

6 X j m C, i I, t m p, y, (2) i m j, i, t where p i, m j J, t m p j, i, t y j, i, t j, yi, t j J, t m is the weighted average prie paid for a good in ategory during time interval m. We define the prie index for the household as the ratio of atual expenditure divided by the expenditure at the average prie p i, m : the month: ~ j j X m pm. (3) j X m Finally, we normalize the index by dividing by the average prie index within ~ j j pm p m. (4) 1 ~ j p j m J This household-level prie index shows the prie level eah household faes relative to the average prie level. 12 Figure 2 shows the life-yle profile of this prie index. The horizontal axis shows the age of the wife, while the vertial axis indiates the prie index. As is lear from the figure, the prie index inreases with age; it does not derease, as found by AH. Moreover, the slope is very small, whih implies that the differenes in pries aross age groups are extremely limited; the absolute value of the slope is approximately one-third that estimated in the US. Figure 3 also shows the relationship between the prie index and household inome. Similar to Figure 2, we an observe a slightly upward line of prie over inome, whih implies that households with 12 When alulating the average prie for eah ommodity, we use the regional average that divides entire Japan into 10 different regions. 6

7 greater inomes fae moderately higher pries than low-inome families. Columns (1) (2) in Table 4 shows the regression oeffiients for inome and age dummies when the dependent variable is the natural logarithm of the prie index. 13 The effets of age and inome group dummies on the prie index are stable and highly signifiant. However, the values of the oeffiients are generally not large. Aording to olumn (1) in Table 4, households whose inome is over 9 million yen fae point higher pries than do the lowest inome group. It is worth noting that this prie index annot apture the movements of pries over time beause the average of the prie index is always unity. 4. Shopping Behaviors One of AH s main results is that elderly people an lower their pries by inreasing how frequently they shop. In this setion, in addition to the shopping frequeny, we introdue six other shopping behaviors that might affet the relative prie index introdued in the previous setion. As the measure of shopping frequeny, we use the number of stores households use, (ln_trip). More preisely, we first ount the number of different stores that a sample household visits eah day. Next, we alulate the sum of the number for eah month, 13 In their regression analyses, AH ontrolled for shopping needs beause a shopper who spreads shopping time over numerous goods have less time to find the best bargain prie. Following AH, in subsequent regression analyses we inlude the natural logarithms of the number of ommodities, the number of produt ategories, and the total expenditure per month as the shopping needs. 7

8 whih gives the index for shopping frequeny. Ln_store aptures the variety of shops eah household patronizes. Note that this variable does not inlude information regarding frequent shopping at the same store. This variable an be used as a proxy for searh intensity, whih might lead to a lower prie index, to find the people who use some stores in searh of better pries. Next, we onstrut the Herfindahl Hirshman Index (HHI) to apture the onentration of spending. The HHI is a measure of the amount of ompetition in an industry. We use it as an indiator of the degree of onentration of stores where the households purhase goods. For example, onsider two households. Both families go to three stores a month. One of the families relies on a large supermarket and spends 90 % of the monthly expenditure at the supermarket, while the other family spends evenly aross the three stores. Our HHI aptures the differene in suh shopping behaviors. The HHI is defined as K j 2 HHI j m k=1 S k,m, (5) j where S k,m is the share of store k K in monthly total purhases of household j. Next, we onsider the monthly total number of goods a household buys: Quantity j j, m = C,i I,t m y i,t. (6) It is reasonable to suppose that a family buying many goods an enjoy volume disounts more, thus dereasing the prie level. To observe the effet of buying at bargain sales, we onstrut a measure for bargains. As one might expet, a household an derease its prie index by purhasing more goods at bargain sales. Beause of the lak of store-level flags for bargain sales in our dataset, it is neessary to define the prie at bargain sales based on information 8

9 regarding the movements of store-level pries. In this paper, we adopt the store-level monthly minimum prie for eah good, min P i,t, as the prie at bargain sales. Then, the following index is used, where j, j, j, C,i I,t m )pi,t yi,t j, j, C,i I,t m p i,t yi,t bargain m j = I(P i,t I(P j, i,t ) = { 1, P j, i,t = mi n P i,t and min P i,t max P i,t 0, Otherwise, (7) shows the proportion of expenditures at bargain pries. A household with a large bargain index is purhasing produts at lower-than-normal pries, whih lowers the relative prie index. It is worth noting that this measure aptures the importane of temporal redution within a month. If pries are stable for several months, or if bargain sales last more than one month, this index fails to apture the importane of bargain sales 14. Generally, most produts an be purhased at both luxury stores and disount stores. The movement of pries differs aross stores to a great extent. Abe and Tonogi (2009) show that pries move very differently aross stores based on a large point-of-sale database of Japanese stores. Suppose a high-inome family has greater opportunity osts for shopping than do low-inome families. Also, suppose that a high-inome family tends to use luxury stores. Then it is probable that luxury stores sell ommodities at higher pries than standard supermarkets beause ustomers an redue their shopping osts by buying goods at one shop even if they know other stores have set lower pries for exatly the same goods. However, disount shops annot set higher pries for 14 When a household purhases very rare items that are sold only one in a given month and store, we annot identify whether this prie is a bargain prie or a regular prie. Our variable regards this as a regular prie. To hek the importane of the rarely traded goods, we try several different definitions of bargain sales and find that our main results in later setions do not depend on rarely traded goods. 9

10 ommon goods beause ommon goods are their main produts, so these shops expet ustomers to hange their favorite shops if one shop inreases its pries for ommonly used goods. Thus, it is worth examining the effets of the quality of stores on the prie index. We define the index for the quality of eah store, k K, by following the same basi proedure as we did for the relative prie index. The store quality index is the ratio of the hypothetial sales if the store sells the goods at their average prie P i,m to the sales if the store sells the goods at their ategorial average prie. More preisely, we first obtain the average prie for a given good in ategory C: k, k, y p i,t i,t i I,k K,t m y i,t P m = i I,k K,t m k,. (8) Next, assuming that the stores sell the average goods in eah ategory at the average prie, we obtain the total sales: Z m k = p m k, C,i I,t m y i,t. (9) Then, we alulate the total sales of store k if it sells the goods at their average pries = p i,t p i,m k, k, y i,t k K,t m k,, (10) k K,t m y i,t k, y i,t Z k m = C,i I,t m p i,m. (11) Now, the index for the quality of goods sold at store k is defined: q mk Z m k Z mk. (12) Finally, we normalize the index by dividing by the average monthly quality index, q m k k q m k K q mk, (13) whih gives us the quality index of a store k during the time interval m. 10

11 We employ the average of the store quality index weighted by the share of eah store in the monthly total purhases of a household j: j Store hoie j m k K S k,m T. (14) q m k The greater the store hoie index, the higher the likelihood of using luxury stores, whih leads to a higher prie index. By hanging stores to households in the previous index, we an reate a household-level monthly average quality index. The quality index for households is defined as the ratio of the hypothetial expenditure if the household purhases the goods at their average prie, p i,m, to the expenditure if the household purhases the goods at their ategorial average prie, p m. Formerly, define the total expenditure by household j when assuming the household purhase goods at the ategorial average prie: Z m j = p m j, C,i I,t m y i,t. (15) Next, define the hypothetial expenditure if the household purhases the goods at their ommodity level average prie: C,i I,t m (16) Z m j = p i,m y j, i,t, The index for quality of goods bought by household j is defined as follows: q m j Z j m j. (17) Z m We normalize the index by dividing by the average monthly quality index as follows: q m j q m j j J j q m. (18) It is expeted that the greater this quality index, the higher the prie index will be. As noted previously, this measure is not affeted by other shopping strategies of 11

12 eah household, suh as buying at sales, beause it uses the average prie of eah good. In this index, we assume all households enounter the same pries for speifi goods, so a higher index value does not imply that a household buys goods at higher pries than another household does. Table 5 reports the desriptive statistis of these shopping behavior variables and the relative prie index aross different age and inome groups. On average, Japanese families shop 14.4 times a month. The standard deviation of the number of trips is large, at 9.5, whih implies that families are highly heterogeneous in their shopping frequeny. Figure 4 onfirms the heterogeneity. Some families shop more than 100 times a month. It is important to note that this index ounts multiple trips to the same store within the same day as only one trip, so the number of shopping trips in this table is the lower bound of the atual number of trips. Aording to Table 5, households with a wife of years old shop more frequently than do younger households, whih is onsistent with the results found by AH. We an also observe that the shopping frequeny inreases with inome. Not surprisingly, the proportion of bargain purhases dereases with age and inome. The standard deviation is also large. Figure 5 shows the distribution of the proportion of expenditures spent on bargain sales. We an observe a mass point at zero, whih implies that many families always purhase goods at higher pries rather than at the monthly minimum prie. 15 The shopping onentration measure, HHI, dereases with age and inome, implying that elderly and high-inome families tend to disperse their expenditures aross different stores. 15 The monthly minimum prie is defined as the ommodity-store level minimum prie eah month. 12

13 5. The Relationship between the Relative Prie Index and Shopping Behavior Columns (3) (5) in Table 4 report the results of ordinary least squares. Beause of the endogeneity in shopping behaviors, we should be areful in our interpretations of the oeffiients of shopping behaviors, suh as the frequeny of trips. Beause of the large sample size, some of the t-values exeed fifty. Exept for the Hafindahl -Hirshman Index (ln_hhi) and the number of different stores (ln_store), the sign of the shopping behaviors are generally onsistent with the asual hypotheses raised in the previous setion. For example, the oeffiient of the frequeny of trips (ln_trip) is negative, whih implies that households that shop often fae lower pries. Moreover, the size of the oeffiient, in olumn (3), is similar to the same result in the ordinary least squares (OLS) regression performed by AH. AH used dummies for inome and age as instrumental variables to ontrol for the endogeneity of shopping behavior. Unfortunately, in our dataset, the two-stage least squares estimates with these instrumental variables are quite unstable and annot pass the over-identifiation tests. Thus, rather than relying on instrumental variables, we adopt a fixed-effets model, whih enables us to omit the biases due to unobservable family-level effets. Columns (6) (8) in Table 4 show the estimation results. Robust and stable relationships between shopping behaviors and the relative prie index an be found in ln_quantity, bargain, ln_store_hoie, and ln_quality. The effets of age and inome beome muh smaller than those reported in olumns (1) (5), probably beause the fixed effets absorb the age effet. The effet of shopping frequeny beomes positive signifiant. Although the positive effet is diffiult to interpret, the magnitude of the 13

14 effet is negligible. The only non-negligible effet omes from bargain sales. The average proportion of purhases at bargain sales is about 14%. Thus, by doubling this proportion, households an enjoy about a 2% redution in the prie level. It is worth noting that the R-squared in olumn (3) of Table 4 is approximately 20%, whih implies that approximately 80% of the differenes in the relative prie index annot be explained by the observed variables. As shown in Figure 1 and Table 4, there is a notable amount of heterogeneity in the relative prie index aross households. We need more information on the households shopping behavior and preferenes to study the ause of this heterogeneity in more detail. 6. Why do the Elderly Fae Higher Pries? Table 4 shows a very stable and robust effet of age on the prie level. The positive signifiant effets of the dummies on elderly households remain even after ontrolling for various shopping strategies. Thus, we have to seek different mehanisms behind the age effet. In this setion, we onsider an alternative shopping strategy whereby younger households may shop at disount stores more than older households, whih might explain the prie level differentials between age groups. Table 6 reports the share of total expenditures purhased at different types of store by household age. We observe that elderly households spend more at speialized stores and supermarkets, and less at drug stores and home-improvement stores than young households do. Table 7 ompares the OLS estimates with and without store hoie variables. Although the age dummies remain positive and signifiant in olumns (3) (4), the dummy for 60 years or older beomes smaller by more than 10%. Therefore, the store hoie an partially 14

15 explain prie-level differentials between households. However, Table 8 also shows that the elderly fae higher pries than the young do even within the same type of store, whih might appear in olumns (3) (4) in Table 7. Thus, the mehanism behind the higher pries paid by the elderly needs further investigation. This ould possibly be explained by the existene of unobserved variables that orrelate with age. Finanial assets are one suh example. Sine the homesan data we use do not ontain suh information, we need to seek different data sets, a task for future researh. 7. Poliy Impliations Consumer prie indies are some of the most important eonomi indiators for monetary poliy. When the inflation rate is lose to zero, a slight hange in the rate, perhaps as small as a tenth of a perent, attrat strong attention among poliy makers and investors. However, it is well known that prie indies are subjet to various types of measurement error (see Boskin et al., 1996, Shiratsuka 1999, Lebow and Rudd 2003). Our researh shows that the law of one prie, whih the standard onsumer prie index theory assumes, is violated. Prie differentials aross households are large and orrelated with household harateristis suh as inome and age, and with their shopping strategies. Abe and Tonogi (2010) observed that prie hange rates and bargain ratios are extremely heterogeneous aross stores, even for the same ommodity. These results suggest that inflation an have different effets on different types of household. It is likely that the welfare loss aused by inflation for households that an find better bargain pries or disount shops is smaller than the welfare loss for households that have limited hoies to seek lower pries. Aording to our results, it is 15

16 probable that inflationary monetary poliy reates a bigger loss for elderly and high-inome people. When artifiially hanging the inflation rate, poliy makers should onsider that suh polies have effets not only on the average of household welfare, but also on its distribution. 8. Conlusion This study used Japanese sanner data to investigate household-level prie differenes. The data reveal that the law of one prie is violated to a great extent; differenes in pries aross households exist for the same ommodity. These results are onsistent with previous studies based on US data. However, the prie level is negatively orrelated with shopping frequeny, while it is positively orrelated with inome and age, whih is inonsistent with US results. The proportion of purhases at bargain sales is a delining funtion of age. After ontrolling for purhases at bargain sales, the age effets on the prie level beome very small, suggesting that elderly households fae higher pries than young households do beause elderly people use less bargain sales. The fixed effets estimates show very small signifiant effets of the shopping frequeny on the prie level, whih is inonsistent with previous studies based on US data. Many tasks remain to be ompleted in this line of researh. In this study, the produt-level information is not fully utilized. The variation in household harateristis, suh as employment status and family omposition, may also be important in explaining the differenes in pries aross households. Finally, following Broda and Romalis (2009), the heterogeneity in the movements of the prie level, that is, 16

17 the heterogeneity in household-level inflation, needs to be investigated. Referenes Abe, N. and T. Niizeiki (2010) Consumption Data on Homesan: Comparisons with Other Consumption data, (in Japanese) The Eonomi Review, 61(3), pp Abe, N., and A. Tonogi (2010) Miro and Maro Prie Dynamis in Daily Data, Journal of Monetary Eonomis, 57, pp Aguiar, M. and E. Hurst (2007) Life-Cyle Pries and Prodution, Amerian Eonomi Review, 97(5), pp Baye, M., J. Morgan, and P. Sholten (2004) Prie Dispersion in the Small and in the Large: Evidene from an Internet Prie Comparison Site, Journal of Industrial Eonomis, 52, pp Boskin, Mihael J., E. Dulberger, R. Gordon, Z. Grilihes, and D. Jorgenson (1996) Toward a More Aurate Measure of the Cost of Living, Final Report to the Senate Finane Committee. Broda, C. and J. Romalis (2009) The Welfare Impliations of Rising Prie Dispersion, mimeo. Goldberg, P. K. and V. Frank (2005) Market Integration and Convergene to the Law of One Prie: Evidene from the European Car Market, Journal of International Eonomis, 65(1), pp Griffith, R., E. Leibtag, A. Leiester, and A. Nevo (2009) Consumer Shopping Behavior: How Muh Do Consumers Save? Journal of Eonomi Perspetives, 23(2), pp

18 Haskel, J. and H. Wolf (2001) The Law of One Prie - A Case Study, Sandinavian Journal of Eonomis, 103, pp Kitamura, Y. (2008) Kakeibetsu Bukkashisu no Kouhiku to Bunseki, (in Japanese) Kinyukennkyux, 27(3), pp Kuroda, S., and I. Yamanoto (2010) Kakeibetsu Infle Ritsu No Bonpu To Sono Jizokusei, (in Japanese) in Seko et al. ed., Nihon Kakei Kodo no Dainamizumu, Keio University Press, pp Lebow, D. E. and J. B. Rudd (2003) Measurement Error in the Consumer Prie Index: Where Do We Stand? Journal of Eonomi Literature, 41(1), pp Nakamura, E. and J. Steinsson (2008) Five Fats about Pries: A Reevaluation of Menu Cost Model, Quarterly Journal of Eonomis, 123(4), pp Ogawa, K. (1990) Cylial Variations in Liquidity Constrained Consumers: Evidene from Marodata in Japan, Journal of Japanese and International Eonomies, 4, pp Shiratsuka, Shigenori (1999) Measurement Errors in the Japanese Consumer Prie Index, Monetary and Eonomi Studies, 17(3), pp Unayama, T. and M. Keida, (2011) Koureisha-Setai no Shouhi Koudou To Bukka Shisuu, (in Japanese) RIETI Disussion Paper Series 11-J-047/. Wakabayashi, M. (2008) The Retirement Consumption Puzzle in Japan, Journal of Population Eonomis, 21, pp

19 Figure 1: Distribution of the relative prie index aross households Note: The definition of the prie index is given in Setion 3. 19

20 Figure 2: Life-Cyle Profile of the Prie Index Note: The horizontal axis is the age of the wife in the household. Figure 3: Household Inome and Prie Index ~ ~ Note: The horizontal axis is household inome in units of 1,000 yen. 20

21 Density Figure 4: Distribution of the Frequeny of Shopping per Month Figure 5: Distribution of the Proportion of Expenditures on Bargain Sales bargain1 21

22 Table 1: Family Composition of SCI and Other Data Family members year 2005 SCI Census KHPS Note: SCI is homesan data by Intage. KHPS stands for Keio Household Panel Survey. Table 2: Distribution of Wife's Age Wife Age ~29 30~34 35~39 40~ year 2005 SCI Census Wife Age 45~49 50~54 55~59 60~ year 2005 SCI Census

23 Table 3: The Relation between Wife's Age and Job Status SCI Census 2005 Age Full Part Self Time Time Employed Agriulture Sideline Non Non Working Working ~ ~ ~ ~ ~ ~ ~ ~ Note: The figures are average for the sample period:

24 Table 4: The Relation between Prie and Shopping Behavior (1) (2) (3) (4) (5) (6) (7) (8) Regression Type OLS OLS OLS OLS OLS FE FE FE ln_trip ( ) ( ) (9.590) (9.533) ln_store (-5.582) (1.647) ln_hhi (11.712) (10.571) ln_quantity (-3.254) (0.257) bargain ( ) ( ) ( ) ( ) ln_store_hoie (23.159) (10.046) ln_quality (54.630) (27.956) Dummy for Inome (1) 4,000-5, (6.503) (5.262) (2.472) (5.368) (5.421) (0.403) (0.149) (0.294) 5,500-6, (18.546) (18.676) (10.086) (16.128) (14.845) (0.521) (0.318) (0.378) 7,000-8, (22.234) (22.389) (12.938) (19.650) (18.561) (0.825) (0.602) (0.756) 9, (38.275) (40.347) (23.881) (34.824) (30.196) (0.423) (0.197) (0.354) Dummy for Age (2) (-3.509) (-4.859) (-3.820) (-3.492) (-2.980) (1.027) (1.151) (0.959) (3.931) (4.833) (3.356) (4.267) (4.082) (0.976) (1.353) (0.944) (2.721) (8.522) (2.590) (3.372) (3.211) (1.084) (1.370) (1.097) (6.571) (17.705) (6.100) (7.229) (6.952) (1.176) (1.554) (1.174) (7.051) (19.368) (9.366) (9.576) (8.004) (1.029) (1.337) (1.003) (10.121) (23.035) (14.360) (13.660) (12.157) (0.733) (1.043) (0.767) (19.154) (33.644) (23.030) (23.197) (20.894) (0.927) (1.206) (0.977) Constant (16.107) (18.527) (6.135) (-3.524) (24.422) (19.655) (26.536) (28.340) Household Charateristis Yes No Yes Yes Yes Yes Yes Yes Needs Yes Yes Yes Yes Yes Yes Yes Yes Loation Dummies Yes No Yes Yes Yes Yes Yes Yes Time Dummies Yes No Yes Yes Yes Yes Yes Yes Observations 371, , , , , , , ,367 R-squared Number of monitor_ode 14,442 14,442 14,442 14,442 14,442 14,442 14,442 14,442 Note: Ordinary least squares and fixed effets estimations based on Japanese homesan provided by Intage. The dependent variable is the Household-Level Prie Index Clustering t-statistis are in parentheses. Household harateristis inlude dummy variables for the number of family members and the number of Loation hild Dummies inlude dummy variables for ity size dummies and prefeture dummies Needs inlude the natural logarithms of the number of ommodities, the number of produt ategories, and the total expenditure per month. The data is onverted to household-level monthly data. (1) The unit is 1000yen. The base is the inome below 4,000. (2) The age of the wife in the household. The base is the dummy for below 30.

25 Table 5: Desriptive Statistis of Shopping Behaviors inome (2) age of wife (1) inome age of wife Note lnprie Prie Index (Level) ln_trip Number of Trips Number of Stores (ln) mean sd mean sd mean sd mean sd mean sd ~ ~ ~ ~ ~ ~ ~ ~ Total ~ ~ Total Bargain ln_quantity ln_hhi ln_store_hoie ln_quality mean sd mean sd mean sd mean sd mean sd ~ ~ ~ ~ ~ ~ ~ ~ Total ~ ~ Total (1) The age of the wife in the household. The base is the dummy for below 30. (2) The unit is 1000yen. The base is the inome below 4,000 thousand yen.

26 Table 6: The Expenditure Shares in Different Store Types and Age age of wife age of wife Conveniene store Speialized store Pharmay Home Improvement Store Home delivery & Door to door sales mean sd mean sd mean sd mean sd mean sd ~ ~ ~ ~ ~ ~ ~ ~ Total Supermarket Others mean sd mean sd ~ ~ ~ ~ ~ ~ ~ ~ Total

27 Table 7: The Relative Prie by Store Type and Age age of wife age of wife Conveniene store Speialized store Pharmay Home improvement store prie share prie share prie share prie share ~ ~ ~ ~ ~ ~ ~ ~ Total Home delivery & Door to door sales Supermarket Others prie share prie share prie share ~ ~ ~ ~ ~ ~ ~ ~ Total

28 Table 8: The Effets of Store Choie (1) (2) (3) (4) Dummy for Inome (1) 4,000-5, (6.503) (5.262) (6.123) (4.833) 5,500-6, (18.546) (18.676) (17.902) (17.813) 7,000-8, (22.234) (22.389) (21.141) (21.320) 9, (38.275) (40.347) (37.060) (39.190) Dummy for Age (2) (-3.509) (-4.859) (-4.041) (-5.820) (3.931) (4.833) (2.582) (2.655) (2.721) (8.522) (0.786) (5.536) (6.571) (17.705) (5.197) (15.412) (7.051) (19.368) (4.965) (16.078) (10.121) (23.035) (8.006) (19.570) (19.154) (33.644) (17.013) (29.676) Constant (16.107) (18.527) (18.129) (21.406) Household Charateristis Yes No Yes No Needs Yes Yes Yes Yes Loation Dummies Yes No Yes No Time Dummies Yes No Yes No Store Choie No No Yes Yes Observations 371, , , ,367 R-squared Number of monitor_ode 14,442 14,442 14,442 14,442 Note: Ordinary least squares estimates based on Japanese homesan provided by Intage. The dependent variable is the Household-Level Prie Ind. Clustering t-statistis are in parentheses. Household harateristis inlude dummy variables for the number of family members and the number of hildren. Needs inlude the natural logarithms of the number of ommodities, the number of produt ategories, and the total expenditure per month. Loation Dummies inlude dummy variables for ity size dummies and prefeture dummies. Store Choie is the expenditure share by store type. The data is onverted to household-level monthly data. (1) The unit is 1000 yen. The base is the inome below 4,000. (2) The age of the wife in the household. The base is the dummy for below 30.