Estimating Long-term Economic Statistics for Agricultural Products

Toshihiko Kawagoe

The purpose of this note is to discuss a guideline for estimating long-term economic statistics for Asian agricultural products. We will also consider the problems one can anticipate when compiling this kind of data, along with a few related issues. This note is based on Discussion Paper No. 96-9, "A Note on the Estimation Methods of Long Term Economic Statistics on Agriculture in Asian Countries" (in Japanese, with a detailed English appendix), to which the reader may refer for a more detailed discussion.

1. Framework for Analysis

For the Asian Historical Statistics Project (hereafter referred to as Asian-LTES), the Long-term Economic Statistics of Japan series (hereafter LTES) is an essential model. Kazushi Okawa, Miyohei Shinohara, and Mataji Umemura were the directing editors of Long-term Economic Statistics, published in fourteen volumes (in Japanese) by Toyo Keizai Shimposha from 1965 to 1988. Volumes dealing with agriculture include Volume 9, Agriculture and Forestry (hereafter LTES-9), which covers statistics related to production; Volume 8, Prices; and Volume 3, Capital Stock, which covers agricultural capital stock.

What was the framework in which the first historical statistical project was conducted? In the preface for Long-term Economic Statistics, the editors describe their general purpose as follows: "This series...takes modern economics as its foundation for framing national income accounts structures, and for statistically tracing the shape of development of the Japanese economy since Meiji..." They did not, they added, "simply gather national income statistics and their structural aspects, but comprehensively compiled and organized them..." (see "A Preface from the Editors" on p. iii of each volume).

Within this general framework, then, what plan was undertaken for compiling agricultural statistics? The objective, according to the introductory remarks by the authors of LTES-9, Agriculture and Forestry, was "to prepare a continuous time series going back to the early years of Meiji covering data related to agricultural and forestry production and inputs as well as factor costs, and do so in conformity with national income accounts" (LTES-9, p. vii). However, there are two qualifications regarding this goal.

The first qualification is that because the system of national income accounts constitutes a basic framework, anything more general such as, for example, institutional or structural statistics would be excluded. This qualification provides an important starting point for the Asian-LTES project. Of course, if general historical statistics could be included in institutional or structural form, their value would be immeasurable. Therefore, our basic approach should be to use the statistical time series estimated for LTES as a basis for Asian-LTES, and to compile other institutional or structural statistics on a supplementary basis.

The second qualification regards the point that "emphasis should be placed on continuous long-term statistical time series from the early years of Meiji." Since the 1950s, Japan's statistical data has been greatly improved in terms of both quality and quantity, so emphasis was placed not on comprehensive inclusion of the abundant statistics of recent years but on the continuity of statistical time series from early Meiji. In short, improving existing postwar time series was not the goal.

This second qualification is also applicable to Asian-LTES. As in Japan, statistical compilation in other Asian countries has been substantially improved at all levels in recent years, not just in breadth of coverage but in accuracy as well. Given this situation, attempting to improve statistical time series for recent years should not be the task of our project. On the contrary, our aim should be to compile time series using the reliable statistics from the earliest point in time for which they exist. However, we also want to compile compatible time series while maintaining reliability.

The second qualification is illustrated in Figure 1, which depicts a conceptual outline. The vertical axis, moving downward, represents the passage of time, while the horizontal axis represents the volume of existing raw data. As time progresses downward, the volume of the raw data grows and its precision improves apace. The lower part of the triangle includes this increasingly dense data. The equilateral rectangle represents the time series estimated by Asian-LTES. Considered from a more recent perspective, the scope appears to be limited, but if we are going to emphasize long-term continuity, we will need to conduct extensive estimations for earlier periods. Thus there is no point in uselessly extending the scope of the project.

2. What Did LTES Estimate?

What kind of agricultural statistical time series were estimated in the LTES project? While the LTES project as a whole aimed to estimate statistical data within a framework of national income accounts, the LTES project on agriculture also compiled time series of agricultural value added as final products. Not only the value added and related time series but calculations of basic time series data for items such as agricultural land and farm households were estimated as well. We will review what sort of time series were compiled for LTES on agriculture.

In Part 3 (Statistical Tables) of LTES-9, there are 39 statistical tables related to agriculture, and they include a total of 476 data time series. If we group the tables into major categories, we have (1) time series for agricultural output and prices (Tables 1-13); (2) times series for current inputs in agriculture (Tables 14-27); (3) time series related to agricultural capital stock (Tables 28-31); (4) basic statistical time series for arable land and other items not covered in the first three categories (Tables 32-34); and (5) other time series (Tables 35-39).

(1) Time series for agricultural output and prices

This category indicates the value of agricultural production, agricultural price indices, volume of agricultural production, and the amount of value added for agriculture. Only data for "production of agricultural products in volume" (96 items in ten categories) are compiled here, and remaining time series are estimated from this production data and from separately estimated price data and other information. Several issues or technical difficulties were raised in this process. For instance, the distinction between agricultural and forestry products is often unclear in the case of items like shiitake mushrooms. Another problem is determining whether products processed by farm households should be classified as processed agricultural goods (agricultural products) or as manufactures. LTES-9 classifies processed agricultural goods as manufactures, while straw goods are included with agricultural products. Since there are many regions throughout Asia where farm households extensively engage in processing agricultural products, the income generated from these activities is important in regional economies. The issue raised in LTES-9 is important and thus should be kept in mind for the Asian-LTES project.

(2) Time series for current agricultural input

Current agricultural input comprises various intermediate inputs which are converted into agricultural products. Or it is defined as deductions of depreciation for fixed capital from non-wage inputs to production. The current input series were calculated because (1) agricultural gross value added is computed by subtracting current agricultural inputs from the value of agricultural product; and (2) current inputs are necessary for estimating agricultural production functions (LTES-9, p. 53).

Current inputs are categorized as inputs supplied within the agricultural sector and as inputs provided by the non-agricultural sector. The former include seed, silkworms, green manure, domestic fodder crops (grains, pulse, and tubercrops) and others. The latter include pesticides, fertilizer, imported fodder crops such as bran, and oil cake. However, in LTES-9, internally supplied intermediate products like compost were excluded from the list. There is a large quantity of statistical data for the most important agricultural factor of production, fertilizer, and it is analyzed in detail.

(3) Time series related to agricultural capital stock

Statistical time series for agricultural capital stock was estimated on the basis of LTES Vol. 3, Capital Stock. This data is necessary to estimate net value added, but due to its limitations only provisional. results are presented.

(4) Basic statistical time series for arable land and other items

Time series were compiled for area cultivated (paddy and upland fields listed separately), agricultural laborers (men and women listed separately), farm households, agricultural wages, forestry wages, and land and rental prices. These time series are not used directly to estimate value added, but the arable land serves as basic data for estimating items such as production. Further, information on agricultural labor is necessary for the analysis of agricultural productivity in Part 1 of LTES-9.

3. Estimates for the Asian Historical Statistics Project

Having examined the process of estimating agricultural statistics for LTES, let us deal with the problems involved in estimating statistics for agricultural products for the regions in Asia.

The objective of Asian-LTES estimates for "agricultural products," as learned from LTES, should be: to compile estimates which conform to the national income account structure, while extending continuous long-term time series data for agricultural output, input, and price factors as far back in time as possible.

Given this objective, we can present two qualifications.

  1. To put the national income account system into a basic framework, statistical time series are estimated for at least the following items -- agricultural output, input, and factor prices -- while preliminary estimates can be prepared for other institutional and structural statistics. Of course, we do not exclude the estimation of more general historical statistics, such as institutional and structural statistics, since these statistics possess immeasurable value.
  2. Emphasize estimating continuous long-term time series from the beginning of the century (or earlier, according to the region). In other words, the objective of Asian-LTES should not be to simply put together statistical time series which are incomplete or limited to particular periods, nor should it be to recalculate recently published statistics which are known to be abundant and accurate.
To these two points, an additional qualification is necessary in the case of Asian-LTES. Since numerous countries are included in the Asian-LTES project, statistical time series for all countries need to be compiled in a manner which will enable us to aggregate the data. If it is not possible to compare or aggregate the statistics on estimated agricultural production across countries or regions, the data is much less useful. In order to compare and aggregate data, it is therefore important that we employ common codes and common systems of measurement in compiling our estimates.

(1) Classification codes for agricultural products

The trade classification system is the most internationally standardized commodity classification system. Trade being an economic activity which crosses national boundaries, it became necessary to compile internationally coordinated trade statistics, and from the 1930s classification codes were developed for use in collecting customs duties. This has meant that the trade classification system has been the most complete system available to provide an international standard for commercial products. One problem is that the nature of trade classifications means that they do not correspond to the stages of agricultural production.

Agricultural statistics are based in widely varied geographic and cultural differences of many countries reflecting a wide diversity of historical origins. Hence different classification systems tend to take on their own particularities. At the same time, those systems often convey important information regarding those countries' agricultural systems. Thus, we should not distort the distinctive agricultural information systems of different countries, but seek to develop a means of compiling statistics which will render this information mutually comparable.

Fortunately, the Food and Agriculture Organization (FAO) of the United Nations has compiled a detailed classification system for agricultural products. Each agricultural product is classified by a four-digit code (FAOSTAT CODE) in one of twenty groups. Further, there is a corresponding code for each trade classification code (Standard International Trade Classification, Revision 2, Revision 3, HS) for each FAOSTAT CODE. With these tools, we can estimate each country's agricultural products time series according to its customary classifications, and then, by annotating the appropriate FAOSTAT CODE to the estimated time series, we can seek to maintain compatibility between the trade statistics system and the time series data for different countries.

In brief, when more detailed or different classifications of a country's statistics are available, we need not process or aggregate the time series for that country; the best approach is to present the times series as they are with notes which identify the products' standardized classifications according to the FAOSTAT CODE.

(2) Measurement units

It is necessary that units of raw data for each region be gathered using a unified method. The metric system has been widely used to record production volumes and land areas in recent years, but prior to World War II countries used different measurement units. Further, we can expect to find that different units have been used in various regions and eras within a country. It will be impossible to aggregate data across regions using the different units of measurements, yet uniform conversion to the metric system is inconvenient for conducting regional analysis.

With regard to units of measurement, statistical time series have adopted units of measurement for area, weight, and volume used in the original time series while also using metric conversions to create parallel time series. Moreover, when data is provided primarily in terms of volume, it is necessary to employ a weight-volume conversion factor. Other conversion factors should be reported as necessary. For example, rice production statistics are often shown in terms of unhusked rice, so a conversion ratio to milled rice is necessary. Conversion ratios will vary by region, era, production season, and varieties. If a conversion rate is available, trade statistics, which are often shown in terms of milled rice, can be consolidated with production statistics and it becomes possible to conduct such tasks as making estimates of domestic consumption volumes and other matters.

4. The process of estimating agricultural product statistics

With reference to the process of estimating LTES agricultural product statistics, Figure 2 illustrates how the process of compiling agricultural statistics for Asian-LTES can be prepared on flow charts. The sections marked by thick grey lines indicate the most fundamental time series which must be calculated for Asian-LTES. Those time series are needed to compute gross value added time series. On the other hand, other time series which are important but for which it is difficult to compile reliable estimates are presented as "supplemental time series." One type is time series for capital stock, and another is institution- or structure-related time series.

Because time series for capital stock are needed to adduce "net value added," even LTES estimated them as provisional and partial estimates due to difficulties in estimating reliable data on depreciation. For Asian-LTES, it may be possible in some countries to estimate these series on the basis of certain assumptions.

Seikei University, Department of Economics