The Reliability of Price Indexes: Japan, the U.S., the Philippines, and Indonesia

Kaoru Hosono


1. Introduction

Can we trust price indexes? This question has arisen from time to time. In Japan, especially in the 1990s as the number and kinds of discount stores began to expand, suspicions were voiced to the effect that there was an upward bias in the consumer price index (CPI). Similarly, the CPI became a center of controversy in America when the Advisory Group to Study the Consumer Price Index, a panel headed by Michael Boskin, a former economic advisor to President Bush, issued a report to Congress claiming that the US CPI has an upward bias of 1.1 percent annually. There has so far been almost no research on the reliability of price indexes in Asia. To judge from the recent experiences of the US and Japan, however, some careful study is urgently called for.

The price index is, needless to say, one of the most important economic indicators. Maintaining the stability of price indicators is always one of the principal objects of economic policymaking. They also directly influence public expenses by indexing social benefits and other disbursements. Estimates of price movements play a major role in wage negotiations. Moreover, price indexes form part of the base upon which real growth rates and other statistical materials are compiled. As pointed out above, if there is an upward bias in the CPI, a number of problems will arise. For example, monetary policy may tend to be too tight (on the other hand, if an overestimated CPI leads to an underestimate of the real growth rate, the result may be a constant lax bias in monetary policy). In addition, if wages are determined on the basis of publicly announced CPI, then rising wage costs may generate inflationary pressures in the real economy.

In this paper, I first outline the basic issues regarding the reliability of consumer price indexes in Japan and the US. Then, using data from the Philippines and Indonesia, I make a number of comments on the reliability of price indexes in Asia.


2. CPI biases in Japan and the US

The report prepared by Boskin's committee suggested four main causes of upward bias in CPI formation. These causes (at least in their qualitative senses) can also be applied to Japan.

    (1) A market basket of goods is fixed on a base year, so it does not reflect the substitution effect which occurs when large volumes of cheap goods are purchased (substitution effect bias).

    (2) Even though purchases are made relatively frequently at discount stores, the CPI does not adequately reflect the interstore substitution effect (discount store bias).1

    (3) When new products create changes in market patterns, the price surveys shift their survey subjects, but they cannot make adequate adjustments for changes in quality,2 resulting in overestimates of price rises (quality change bias).

    (4) Because the CPI survey items change only once every several years, when the base revisions are made, new products are not included in the CPI, or else they are included only after time lags. As a result, for several years from when new products are first marketed, the price-cutting effects they often bring are not reflected in the CPI (new product bias)3.

Still, opinions are divided over the quantitative extent of the biases created by these causes. To begin with, Boskin's group puts its estimate of the annual upward bias in CPI at the 1.1 percent range (from 0.8 percent to 1.6 percent). However, this finding is challenged by BLS (Dean Baker 1995), which emphasizes the following points:

    (1) The substitution effect bias is relatively small (from 0.1 percent to 0.2 percent).

    (2) The share of items sold at discount stores is relatively small (in the 15 percent range).

    (3) Changes in quality entail not only an upward bias, but also cases of downward bias (in particular, factors such as declining quality of service are not reflected in the price index).

    (4) New products diffuse widely after their price falls, and there is no need for them to be reflected in the CPI before then.

    Furthermore, BLS argues that there is in fact probably a downward bias because the costs of health insurance, expenses for individual businesses (legal and intermediary fees), and the costs of protecting the quality of life (for example, the added costs of living in safe areas) are not properly reflected in the index. BLS concludes that there is almost no basis for claiming that there is an upward bias in the CPI. To be sure, it is difficult to quantitatively grasp any biases other than the substitution bias, so the Boskin group's report is not necessarily persuasive.

Next, what about Japan's CPI? The early 1990s was a period when discount stores spread rapidly, possibly creating a large bias. Further, cellular phones, for example, are not included in the current CPI (1995 base year), suggesting that the CPI after 1995 includes an upward bias4. However, such examples are definitely only one part of the whole. Proving the total extent of bias is no easy matter. Shiratsuka (1995) estimated the degree of formula bias (derived from the above-mentioned substitution effect among different product items) and found that the fixed based year Laspeyres index had an upward bias relative to the chain base Tonquist index and the chain base Fisher index of 0.2 percent and 0.3 percent, respectively. Indications of other biases can be found as well, but the quantitative effect has not been clarified.


3. The Philippine and Indonesian CPIs

The Boskin group's report points out how difficult it is to accurately calculate CPI in a dynamically changing economy like America's, where new products and sales outlets constantly appear. So what about indexes in Asian countries? Given the lack of price data for most Asian countries, or else the lack of information about how their statistics are compiled, it is difficult to give a direct answer. This paper will, first, use Philippine data to consider what degree of differentiation in rates of CPI increases might result from different price index formulae. Second, it examines regional price differentials using data from the Philippines and Indonesia. This is a point which receives little attention in advanced countries, but it could become an important factor in judging price conditions in developing countries.

(1) Differences in Index Formulae

In order to eliminate the bias resulting from the product substitution effect, it is desirable to estimate price rises using an index which weights expenditures for every year for each item. However, due to mind the limitations of data availability, I use data for 1955 and 1965, the base years for postwar Manila's CPI, and try to determine what degree of bias are produced by changes in the weight of the five major categories of expenditure (foodstuffs, clothing, water and fuel, housing, others) (Table 1).

First, according to the Laspeyres index employed by the CPI, the annual rate of price increases in the ten-year period was 3.45 percent, but when recalculated with the Paasche index, the figure is 3.24 percent, about 0.2 percentage points lower. Further, using the Tornqvist index for the calculation yields 3.22 percent, lower, if only slightly so, than the Paasche index. As the use of five major categories of expenditure yields very rough calculations, these are difficult to compare to preceding research conducted in Japan and the US, but as far as can be realized from these calculations, the bias produced by using a fixed weight (i.e., the Laspeyres index) cannot be ignored5.

(2) Regional Differentials

As developing countries do not have fully formed communication and distribution networks, their regional price differentials are undoubtedly greater than those in advanced countries. It is therefore necessary to gather price information from as many regions as possible in order to estimate price levels for the whole of a single country, but in reality there are more than a few cases where price information from only a few regions has been collected. In the postwar Philippines, the Manila CPI has been in existence since 1949 but the index for the entire country has been compiled only since 1957. In such cases, where we suppose that changes in the index for some regions represent index changes for the entire country, there is concern that the regional changes actually deviate from those of the national changes. Given this concern, let us use the Indonesian and Philippine data made available so far and have a look at the degree of regional differentiation.

First, Table 2 compares price levels for the cost of living for four Indonesian regions using cost of living indexes for January 1964 through January 1967. The simple average value for the four regions for each year is put at 100. These figures indicate that, with different outcomes by year, the degree of differentiation between the lowest and highest cost regions ranges from 20 to 40, meaning a substantial level of difference. For the sake of comparison, Table 3 displays a consumer price regional differential index for the ten regions of Japan for 1992 to 1995 (exclusive of imputed rental income for homeowners, national average equals 100). The degree of differentiation is nine at most. Thus it appears that the degree of differentiation in Indonesia is quite large.

However, large differentials in regional price levels do not necessarily mean that rates of price increase differentials are also large. In fact, if we consider average annual increases in regional CPI for the Philippines (1957 to 1963), Indonesia (January 1964 to 1967), and Japan (from 1991 to 1996) (in Table 2, Table 3, and Table 4), we see that it is Japan which has the largest differential. Therefore, we can say that the comparative level of differentials in price rises is relatively small if we average them over several years. Assuming that the differential in the rate of price increase is small, then CPI compiled on the basis of price information from only some regions may be, to a certain extent, reliable as an indicator for price movements for the entire country.


4. Conclusion

This paper has discussed the reliability of consumer price indexes, using as its starting point debates in Japan and the US to investigate several points regarding the Philippines and Indonesia. Although data is limited, I found that the bias produced when a fixed weight (Laspeyres index) is adopted for the CPI cannot be ignored, while large levels of regional differentiation do not necessarily mean that long-term differentials in price increases will be large. Of course, since I could only utilize the limited data currently available, the analytical results presented are very provisional. Thus, my next goal is to gather more data in order to improve our understanding of CPI reliability.


    Sources

      Philippines: Central Bank of the Philippines, Statistical Year Book. Annual.

      Indonesia: Bank Indonesia, Report for the Financial Year. Annual.

      Japan: Statistics Bureau of the Management and Coordination Agency, Shohisha bukka shisu nenpo [Consumer Price Index Yearbook]. Annual.




    References

      Advisory Group to Study the Consumer Price Index (1996), "Toward a More Accurate Measure of the Cost of the Living," Final Report to the Senate Finance Committee.

      Dean Baker (1995), "The Inflated Case Against the CPI," The American Prospect, no. 24, pp. 86-89.

      Price Bureau of the Economic Planning Agency (1997), Bukka Repooto '97 [Price Report '97].

      Shiratsuka Shigenori (1995), "Shohisha bukka shisu to keisoku gosa -- sono mondai-ten to kaizen ni mukete no hosaku --" [Consumer price indexes and measuring errors: problems and a plan for making improvements]. Kinyu Kenkyu, vol. 14, no. 2, pp. 1-45.

      Statistics Bureau of the Management and Coordination Agency (1996), Heisei 7-nen kijun shohisha bukka shisu no kaisetsu [Commentary on the base year 1995 consumer price index].

Hitotsubashi University, Institute of Economic Research


Notes

(1) For the statistical survey of retail prices in Japan, a fixed number of retail stores are selected for each item in each survey district according to the order of sales volume and size of establishment, with selection starting from the largest establishments (Statistics Bureau of the Management and Coordination Agency (1996)). Thus, even if the sales volume of a discount store grows, it is not included in the survey if its sales share does not reach a certain level.

(2) When the brands surveyed for Japan's CPI change, but there is no difference in quality and volume between the old and new brands, the price indexes reflect the price differentials of the old and new brands. If there is an obvious difference in quality, the price indexes remain unchanged and do not reflect the price differentials of the old and new brands at all (Statistics Bureau of the Management and Coordination Agency (1996)). Thus, when quality and pure (quality adjusted) prices change at the same time, appropriate measures are not necessarily taken.

(3) The statistical survey of retail prices for Japan covers prices of around 520 items, including 810 different brands, for which an amount equal to at least 1/10,000th of total consumer expenditures was recorded in the annual household survey (Statistics Bureau of the Management and Coordination Agency (1996)).

(4) Included in "telephones" for the current CPI (1995 base year) are phones for household use (cordless answering machines), but cellular phones are not included. According to the Price Bureau of the Economic Planning Agency (1997), between 1994 and 1996 the store price of cellular phones fell from around 109,000 yen to around 28,000 yen, or 74.4 percent. Employing the Hedonic approach and considering quality improvements, it is computed that the price fell by at least an additional ten percent.

(5) Using data on the ten major items of expenditure for 1990 and 1995, both base years for Japan's CPI (exclusive of imputed rental income for homeowners) and applying the same formulae (Table 1), the Paasche and ToNrnqvist indexes are somewhat higher than the Laspeyres index. This is mainly because of an increase in the weight of expenditures of residences whose prices rose relatively quickly during the five-year period. The Statistics Bureau of the Management and Coordination Agency (1996) reports that compiling a Paasche index, which uses weights for every year, for individual items gives an average yearly rise of 1.210 percent for 1990 through 1995, lower than the Laspeyres index (1.248 percent). This suggests that the substitution effect works when individual items are considered.