Japan Must Improve the Accuracy of Price Statistics Immediately

Faculty Fellow, RIETI

Deflation is continuing in Japan with consumer prices on a decline since the mid-1990s. Although cases of deflation in advanced economies are extremely limited, probably the most prominent episode is that of the United States during the Great Depression in the 1930s. At that time, U.S. consumer prices declined at the pace of 7% per year and fell 20% in three years. In contrast, the pace of price decline in Japan has been around 1% per year or 2% at the maximum. As such, Japan's deflation can be described as relatively moderate.

The difference between the current Japanese deflation and the U.S. deflation during the Great Depression reflects the difference in the magnitude of the demand-supply gap. This, however, does not fully explain the moderate pace of Japan's deflation. The responsiveness of the growth rate of consumer prices to the widening of the demand-supply gap has declined sharply since the mid-1990s. In other words, the deflation rate is too small relative to the increase in the demand-supply gap. If consumer prices were as responsive as they were in the 1980s, Japan's deflation rate should have reached 3%.

A moderate rate of deflation is not a bad thing per se. However, we can also say that the deflation rate is so small that it takes a long time for prices to hit the bottom, hence, the continuation of deflation for more than 15 years.

♦  ♦  ♦

Why is the deflation rate in Japan low? Various attempts have been made to explain this, and some people say that it is underestimated simply due to the use of inaccurate measurement of consumer prices. This skepticism over measurement accuracy is particularly strong among overseas researchers and practitioners.

To what extent do figures differ when different price measurements methods are used? Using an analogy, if we describe prices as the "body temperature" of the economy, consumer price statistics are the "thermometers." Then, if one thermometer is found to be unreliable, we should try out others to see whether and to what extent measurements vary.

A research team including myself conducted an experiment to re-measure the body temperature of the Japanese economy using different types of thermometers. With the help of Nikkei Digital Media, Inc., we collected daily price data (point-of-sale (POS) data) on all items—approximately 200,000 including foods and sundries—sold in about 200 outlets of Japanese supermarket chains for the period 2000-2010. Taking the prices of those items on our computers as the prices of all items sold in Japan, we then performed a simulation in which virtual price checkers were dispatched to collect price data, based upon which we compiled consumer prices statistics.

In doing so, we paid special attention to sampling methods. The best way to get the accurate picture of price trends is to check the prices of all items sold in all stores. However, conducting such an extensive survey is impossible in practical terms, and it is inevitable to compile statistics based on a survey of sampled commodities sold at selected stores. While such limitations are not unique to Japan but common to any country, the choice of the sampling method differs significantly across advanced economies.

In our experiment, we simulated using three different sampling methods: 1) one similar to the method used by the Ministry of Internal Affairs and Communications (MIC) in compiling the Japanese consumer price index (CPI), 2) one based on the MIC method with some differences including the treatment of bargain prices, and 3) the method used by the U.S. Bureau of Labor Statistics, an agency in charge of compiling the U.S. CPI.

The MIC method is characteristic in that the scope of items targeted for price data collection is narrowed down beforehand, a scheme applied in the United Kingdom and some other major economies. For instance, when we look at POS data, there are more than 300 different types of butter. By applying "200-gram block of unsalted butter packaged in paper boxes" as a criteria, we can narrow them down to about 30 types. In contrast, the U.S. method is to draw a sample from all commodities sold without narrowing down the coverage of items beforehand. In the case of butter, the probability of sampling is assigned to each of more than 300 types proportional to the number of units sold, and, according to that probability, a random sample is drawn from each type.

If we know which types of butter are popular beforehand, the MIC method is desirable. However, if it is not clear which types are popular or where the lineup of products changes quickly as does the popularity ranking, the U.S. method is advantageous.

♦  ♦  ♦

In our experiment, we obtained results close to published consumer price figures when the sample was taken in a way similar to the MIC method, and changing the treatment of bargain prices (excluding short-term bargain prices applicable only for one week or less) did not make much difference. However, using the U.S. sampling method produced significantly different results.

In the figure below, the horizontal axis represents the percentage year-on-year change in the monthly CPI based on the MIC method, and the vertical axis represents that based on the U.S. method. For instance, 1.0% on the horizontal axis means that a year-on-year inflation rate of 1% was posted for the month based on the MIC method, and the corresponding percentage figure on the vertical axis indicates that based on the U.S. method. The solid line represents the median value. The vertical axis value corresponding to 1% on the horizontal axis is 0.1%, which means that when the inflation rate based on the MIC is 1%, the median of the corresponding inflation rate based on the U.S. method is 0.1%. Likewise, the horizontal axis value of 2% corresponds to 0.4% on the vertical axis. As such, figures based on the U.S. method tend to fall below those based on the MIC method.

Comparison of inflation rates based on the MIC and U.S. methodsComparison of inflation rates based on the MIC and U.S. methods

In other words, there is a possibility that percentage changes in Japan's CPI would have been lower than those officially announced had the MIC used the U.S. method of sampling. However, the difference between percentage changes based on the MIC method and those based on the U.S. method is about one percentage point. Thus, even if the U.S. method had been adopted, the resulting figures would not have pointed to severe deflation such as the one observed in the United States during the Great Depression.

Under the U.S. method, a sample is selected stochastically, and therefore sampled items can differ each time. In order to illustrate the variability resulting from such variations in selected items, the 80% confidence interval is shown by the shaded area in the figure. For instance, the shaded area ranges from -0.5% to +0.5% on the vertical axis when the value on the horizontal axis is 1%. This means that when a year-on-year percentage change of 1% is obtained based on the MIC method, the corresponding figure based on the U.S. method falls within the range between -0.5% and +0.5% with an 80% probability.

What are the implications of those findings in terms of policy management? The Bank of Japan (BOJ) has set a 1% increase in consumer prices as a "goal" for monetary policy management. This scheme is consistent with the BOJ's insistence that a leeway of one percentage point would provide a sufficient safety buffer because the median inflation rate based on the U.S. method corresponding to an inflation rate of 1% based on the MIC method is positive. However, such is the case only in average terms. The corresponding confidence interval includes negative figures, indicating the possibility that the Japanese economy may not have exited deflation at that stage based on the U.S. method.

The lower limit of the confidence interval exceeds zero only when the inflation rate based on the MIC method reaches 2.1%. That is, we must wait until that point in order to declare with confidence that Japan is out of deflation whichever measure is used.

Meanwhile, from a fiscal management point of view, it has been pointed out that since fiscal 2000, pensioners have been paid excess benefits totaling 7.5 trillion yen because the amount of pension benefits has not been adjusted fully for deflation as it should have been under the price indexation scheme. Based on the U.S. method, Japan's deflation rate would be larger and so would be the amount of excess payments.

♦  ♦  ♦

As such, while developing a high-precision thermometer is crucial to monetary and fiscal policy management, measurement results differ significantly depending on the type of thermometer. Against this backdrop, an attempt to utilize digital data in constructing price indexes has been drawing substantial attention. Some European countries including Switzerland and Sweden are collecting POS data with the cooperation of distributors and have begun to use them in creating consumer price statistics. The use of POS data, which cover a vast array of products and include information not only on prices but also on sales volume, will likely improve statistical precision by providing an accurate picture of which products are selling well.

Meanwhile, some private-sector companies in Japan and abroad, including Google Inc. of the United States, have been attempting to create high-precision price indexes by collecting price data via the Internet. In the United States, it has been reported that such indexes, although still in the test stage, often move ahead of government statistics. If those attempts prove to be successful, we will be able to obtain price information on a daily basis, instead of monthly as is the case with government statistics. It is expected that such private-sector initiatives to build price indexes will contribute to the improvement of statistical precision.

>> Original text in Japanese

* Translated by RIETI.

September 13, 2012 Nihon Keizai Shimbun

October 10, 2012