1. Military expenditures

There are four sources that report military expenditure data--ACDA, GFSY, IISS, and SIPRI. These sources use different definitions of “military” or “defense” expenditure, with the differences frequently reflecting different purposes for which the statistics are collected or different primary data sources. The most commonly used definitions of defense expenditure are the GFSY definition, based on the UN’s Classification of the Functions of Government (COFOG), and the NATO definition. GFSY relies on COFOG; SIPRI uses the NATO definition as a guideline ^1/; ACDA and IISS use the NATO definition for NATO countries but for non-NATO countries they generally use data from each country’s defense budget, which may be based on yet a different set of definitions. ^2/ The more important specific differences between the GFSY and NATO definitions are as follows:

• The cost of pensions for military personnel is included in government expenditure on “social security and welfare” in GFSY, but in NATO it is included in defense expenditure.

• Operations within the government sector--for example social security contributions paid by one part of the Government to another--are excluded through consolidation in GFSY, but are included in NATO.

• GFSY includes reserves and auxiliary forces, but puts police, border and coast guards in a separate category, as expenditure on “public order and safety”. NATO includes police and paramilitary forces in defense expenditures if they are trained and equipped for military operations.

• Civil defense is included in defense spending in GFSY, although it is separately identified in the detailed breakdown of defense into “military defense affairs” and “civil defense affairs”. Civil defense spending is not included in the NATO definition.

• Military grants-in-kind are included in the defense expenditures of the donor country, but not the recipient country, in both definitions. However, the treatment of cash grants is not the same under the two definitions: GFSY includes military expenditures financed by cash grants in the expenditures of the recipient country, while the NATO definition does not.

The coverage of government also varies, both between sources and across countries within a particular source. Reported military expenditures may include only the budgetary expenditures of the central government, or they may include off-budget expenditures or expenditures of provincial or other governmental levels. Reported “budgetary” defense expenditures may also exclude some budget expenditures due to under-reporting or misreporting. ACDA notes, for example, that “there are indications or reasons to believe that the military expenditures reported by some countries consist mainly of recurring or operating expenditures, and omit all or most capital purchases” (p. 150).

Defense expenditure in some countries may include dual use equipment or infrastructure (e.g., roads, hospitals), dual purpose forces (e.g., military police), and dual purpose industries (e.g., cross-subsidization of military production by civilian output in the same plant). This is particularly the case in some developing countries where the military plays a major role in the economy, and where it is difficult to identify what truly are defense-related expenditures.

Finally, it should be noted that interest on military debt is excluded from defense expenditures under both the GFSY and NATO definitions.

2. Arms transfers

A major problem in identifying trade in military equipment is that the trade classifications used by most countries--the Standard International Trade Classification (SITC) or the Harmonized System (HS)--do not clearly identify military-related trade. For example, no distinction is made between military and civilian imports or exports in the case of dual use equipment such as aircraft. More importantly, for many countries a substantial amount of military trade is not reported by type of product but is lumped together under SITC group 931 “special transactions and commodities not classified according to kind”, and thus cannot be separately identified. In addition, some military trade may be hidden in other SITC groups, as the result of importers submitting false documents. It is also likely that some military trade does not go through customs at all. As a result, for most countries, customs data in their present form cannot be used as a source for information on military trade. Publishers of data on arms trade instead rely primarily on sources such as data compiled by defense ministries or specialized government agencies, confidential government information on third country trade, and trade publications.

The four statistical sources that report arms transfer data (ACDA, CRS, SIPRI and the UN) generally include arms transfers regardless of how financed. Arms transfers data may be more inclusive than the expenditures data because the transfer data includes receipts of grants-in-kind. At the same time, the arms transfers data do not include the military’s total use of imported goods, as they do not include foodstuffs, medical equipment, petroleum products, and other general supplies.

Each of the four statistical sources report on transfers of conventional arms, but the extent of the coverage of merchandise trade varies, and some sources include some services as well. The CRS data have the most inclusive product coverage, covering all categories of both large and small weapons, ammunition, military equipment and other ordnance, construction and other military related services, and military assistance and training programs. The coverage of ACDA is similar, except that ACDA has less extensive coverage of services. The SIPRI arms trade data cover only five categories of major weapons or systems, as “Publicly available information is inadequate to track [smaller] items satisfactorily.” (SIPRI, p. 352). The UN Register covers only major weapons.

ACDA, CRS, and SIPRI to varying degrees report data both in terms of arms transfers agreements (orders) and arms deliveries. Given the long lead time necessary for delivery of the larger, more sophisticated weapons, there can be substantial differences in the two sets of data, with the agreements data more closely reflecting current political developments and the deliveries data corresponding to actual trade flows. The agreements data give an indication of the likely future level of arms transfers on the assumption that most of the agreements are implemented. On the other hand, deliveries are more observable than agreements, particularly when there is an effort to hide the transfer. There can also be systematic differences between countries in the relationship between the agreements and deliveries data. For example, there seems to be a lower utilization rate of agreements for the United States than for other exporting countries, particularly in comparison with estimates for formerly centrally planned countries, where the agreements data tend to be based on delivery estimates. For CRS, the information on transfer agreements is a major focus of the analysis and is presented in the same detail as the data on arms deliveries. In the ACDA reports, the agreements data do not include as much country detail as the deliveries data. In the SIPRI report, the focus is also on deliveries data, with limited information on transfer agreements included in the section on the quantity of arms transfers. The UN reports deliveries data only.

The aim of the CRS report is to provide input to U.S. policy discussions, and the focus is therefore on timely information on trends in arms transfers to the “Third World”, broadly defined. As a result, the CRS data do not cover global arms trade, as they do not include imports by industrial countries, eastern Europe, and the former Soviet Union. CRS also provides only limited data on individual countries. Most CRS data is presented in regional subtotals or for cumulative time periods. ACDA reports information on global arms transfers, and also has the most country detail and the most annual data. As a result, it is the most useful source on trade data for someone undertaking most country-specific analysis. SIPRI and the UN Register fall somewhere between ACDA and CRS in terms of both timeliness and country coverage.

Some of the most politically sensitive arms transfers data are those showing the source of arms for individual importing countries. One of the advantages of the UN Register is that it includes the trading partner for each reported transaction, allowing the data to be cross-checked. ACDA, CRS, and SIPRI each provide some bilateral data on arms transfers, but only for cumulative periods of four or five years.

Statistics on arms transfers have some of the same definitional problems as do military expenditures, including, for example, how to classify dual use equipment which has application in both military and civilian activities, and the distinction between government and non-government buyers. In general, dual use equipment is included in military when its primary mission is identified as military, but this distinction can be difficult to make. In the SIPRI data the criterion is the identity of the buyer, which must be either armed forces, paramilitary forces, intelligence agencies or police force of another country. SIPRI notes that arms supplied to guerrilla forces can pose a definitional problem, and in the SIPRI data these are included in the importing county’s total but with the local recipient noted. In the ACDA data, arms transfers to subnational groups are not included. In the CRS data for United States exports, decisions about dual use exports do not arise because the export data are based strictly on the U.S. Foreign Military Sales (FMS) Program, and do not include data on U.S. commercial military sales, as it is felt that there are no commercial agreements data comparable to the agreements data available for the FMS program. Annual totals for U.S. commercial arms deliveries, based on shippers’ export documents and export licenses, are listed separately in the CRS report but without a country of destination breakdown. The ACDA data include commercial deliveries of items on the U.S. Munitions Control List, a small amount of which may be intended for civilian rather than military use.

For sources that report data in value terms, the approach to valuation can vary. CRS data are based on the estimated purchase price or delivery value of the items, except for data on the U.S. where actual prices are used. ACDA also uses actual or estimated purchase price, as available. The SIPRI values, in contrast, are based on the average production costs of the weapons; in cases where production costs are not available, estimates are made based on technical comparables. SIPRI stresses that its valuation system is not comparable to official economic statistics such as public expenditure and exports and imports. Rather, its arms transfers data were designed “as a trend-measuring device” for the total flow of major weapons and its geographic pattern (SIPRI, p. 354). Because of the valuation issue, and because “much of the international arms trade involves offset or barter arrangements, multiyear loans, discounted prices, third-party payments, and partial debt forgiveness” (ACDA, p. 154), ACDA warns that close comparisons of data on arms trade with other economic variables are not warranted.

The trade data do not present explicit conversion problems (from national currency to a common currency like U.S. dollars) because they are generally reported in U.S. dollars. However, particularly for trade among former Warsaw Pact countries, there are likely to be underlying valuation difficulties that were encountered by the compiler of the data. ACDA notes, for example, that its data on Soviet arms transfers are taken from sources that present the data directly in U.S. dollars and therefore particular caution should be used in comparing this data with other Soviet data.

Differences between sources can also reflect timing differences. Because of the lumpiness of military equipment sales, a difference in the time period covered (notably calendar year versus fiscal year) can result in significant differences in the data reported for the period. In most cases, data are reported by calendar year, the exception being ACDA’s coverage of the arms exports of the United States which are based on fiscal year data. Even when the time period is the same, a difference in the timing of recording of large year-end sales can result in significant differences in data for particular years and complicate the interpretation of trends, particularly for individual countries.

Timing can also be a factor because of differences in reporting lags or the lags between revisions. For example, ACDA (p. 154) notes that the arms transfers estimates for the most recent year, and to a lesser extent for several preceding years, tend to be understated, and can be expected to be revised upward somewhat in succeeding issues. There were in fact substantial upward revisions between the 1990 and the 1991-92 ACDA publications.

1. Difficulties in making comparisons

Annual military expenditure, as reported in each of the four sources, is shown in Appendix I, Table 1 for covered countries from 1988 through 1991. In addition to the definitional differences mentioned above, there are two other problems with comparing data across sources.

First, an obvious problem in comparing aggregates across sources is that the set of countries covered in each source is not the same. ^1/ Largely due to the differences in country coverage, comparisons of global levels of military expenditure across sources is uninformative. For example, total global military expenditure for 1988 varies by over 100 percent across the four sources, with data for countries reporting to GFSY totalling less than half the global figures reported by ACDA. If such global comparisons were attempted using only countries for which all four sources had data, the resulting comparisons would exclude so many countries as to no longer be “global” in any meaningful sense.

In addition to problems stemming from country coverage, comparison of data across sources is sometimes hampered by the form in which the data is presented. Military expenditure data is sometimes converted to U.S. dollars for publication. Local currency data can be converted to current or constant U.S. dollars in one of three ways--using current exchange rates, using historic exchange rates, or using purchasing power parity (PPP) exchange rates. SIPRI and GFSY data are presented in local currency in current prices. For the purposes of the comparisons below, these have been converted to U.S. dollars at the prevailing period average exchange rate, as reported in the IMF’s International Financial Statistics (IFS). IISS also uses IFS exchange rates for most conversions but uses PPP exchange rates where they feel official exchange rates are unrealistic. ^1/ However, IISS also publishes local currency figures, allowing the user to determine the conversion rates used and to apply different conversion rates if so desired. Thus, for these three sources, direct comparisons are possible.

ACDA, however, uses a rather unique methodology--based on “base year” exchange rates--for converting local currency expenditures to the current and constant dollar expenditures: “current-price national currency data for an entire series of years is converted to current U.S. dollars through the use of a single (base year) exchange rate and two price indexes, one national and one U.S.” (ACDA, p. 154). ^2/ This methodology, which is carefully explained in ACDA’s statistical notes, presents two problems to a researcher wishing to compare data sources, or to use ACDA data in combination with other data.

First, ACDA does not present either the conversion rates it uses or the local currency figures. Thus, in comparisons of ACDA with other data sources, differences in reported military expenditure data will reflect not only differences in underlying military expenditures, but also differences in exchange rates. Without knowing which conversion rates ACDA used for each country for each year, it is impossible to isolate the two sources of difference. Only ACDA’s base year data--1991 in the most recent publication--are directly comparable to data from other sources. ^3/

Second, each new ACDA publication uses a different updated base year exchange rate, with the result that ACDA’s historic data on current dollar military expenditures can change significantly across editions. For example, ACDA’s 1989 edition reported Argentina’s current dollar military expenditures in 1988 at US$1.9 billion, while ACDA’s latest edition reports Argentine current dollar military expenditure in 1988 at US$3.9 billion. Other examples of shifts in data for 1988 due to different conversion rates between the two recent ACDA editions include Egypt (US$5.6 billion to US$2.1 billion), France (US$33.4 billion to US$37.0 billion), and Sweden (US$4.9 billion to US$5.7 billion). These large swings make it difficult to interpret ACDA’s data, and they also make any comparisons with ACDA highly sensitive to the edition of ACDA which is used in the comparisons.

2. Global and regional trends in military expenditure

Against this background, we undertook comparisons of percentage changes in data reported by the various sources. Appendix I, Table 2 shows the year-to-year percentage changes in reported military expenditure based on global and regional aggregates. Aggregation involved summing the current dollar expenditures of the relevant countries shown in Appendix I, Table 1. Because of the problems of missing observations for individual countries, all discussions of year-to-year changes in military expenditure focus on countries for which data are available from a source in both years--e.g., SIPRI’s estimate of a 0.9 percent increase in military expenditures in 1989 refers to those countries for which SIPRI has military expenditure estimates for both 1988 and 1989. There continues to be differences in the list of countries included in the calculations for each source. Where these differences have a large effect on the comparisons, the differences in coverage are noted below.

The four sources present significantly different pictures of movements in global military expenditure from 1988 through 1991, both for individual years and for the entire period. The widest discrepancy--IISS’s very large global increase in 1989--is primarily (almost 90 percent of the total reported increase) accounted for by a one time adjustment in IISS’s reported level of Soviet Union military expenditure. Omitting this source of difference, the percentage change in 1989 is broadly similar across all sources. The same cannot be said for 1990 or 1991, however, and for the period as a whole estimates range from virtually no change (ACDA) to a 4.2 percent average annual increase (GFSY).

Regional subtotals ^1/ also show significant differences in trends in military expenditures as reported by the various sources in almost every year, as well as for the period as a whole. In most cases, such differences are largely accounted for by differences--either in coverage or reported military expenditures--of one or two countries.

With regard to Africa, the majority of the differences between SIPRI and ACDA reflect differences in estimates for South Africa in 1989 and Ethiopia in 1990. ^1/ For Asia, differences are largely due to IISS estimates for China, SIPRI estimates for Japan, and SIPRI’s lack of data for China and Taiwan, which heavily influence the trend in the other sources. For the Middle East and North Africa, all sources generally report large increases for the period as a whole, although the rate and trend of such increases differs significantly across the sources. ^2/ The differences are primarily due to inconsistencies across sources in reported military expenditures of Iran, Iraq, Kuwait, and Saudi Arabia. For the Western Hemisphere, the data are reasonably consistent in 1989 and 1990, but vary significantly in 1991 due to different estimates of United States military expenditure in 1991.

The explanation of differences is not quite as simple for western Europe and eastern Europe and the former Soviet Union. For eastern Europe and the former Soviet Union, there is a high level of uncertainty regarding military expenditures during this period (reflected in very different reported levels and changes). In addition, all sources have data for only a few countries in this region, so slight differences in coverage can make large differences in reported numbers. For western Europe, there are large differences across sources in the level and change in military expenditures for a number of countries--some of the largest are for Belgium, France, Italy, Spain, United Kingdom, and West Germany.

3. Trends in countries classified by level of economic development

In the above analysis, based on expenditures aggregated across countries, countries with the largest military expenditures tend to dominate the outcome and one or two countries can dominate the trend for a region. This section attempts to address that problem by analyzing movements in military expenditure relative to GDP ^{3/ 4/} for various economic groupings of countries. By using unweighted averages of ratios, each country is given equal weight. The comparisons in this section thus focus on the average of ratios of military expenditures to GDP. As with the previous section, the results reveal important similarities as well as differences across sources.

Military expenditures as a percentage of GDP are shown in Appendix I, Table 3 by country and data source. For some countries or years, the lack of a reported ratio is due to lack of GDP data rather than of military expenditure data. The same data, organized by economic groupings of countries, are shown in Appendix I, Table 4. Looking first at global data, ACDA shows a significant rise over the period in military expenditures relative to GDP, IISS shows a smaller increase, GFSY shows little change, and SIPRI shows some decline. By 1991, estimates of military expenditure as a percent of GDP range from just over three percent (GFSY) to just under 6 percent (ACDA).

Comparing country groupings, ACDA, GFSY and IISS all report lower military expenditures relative to GDP for industrial countries than for developing or east European countries; SIPRI reports east European countries as having the lowest military expenditure/GDP ratios. All sources report little change in industrial country military expenditure/GDP ratios throughout the period. Reflecting the limited availability of information on the level of east European military expenditure, there are wide differences in the reported ratios of this expenditure to GDP.

There is also a great deal of variation in reported movements in military expenditures relative to GDP for developing countries as a group, ranging from significant increases reported by ACDA and, to a lesser extent, IISS, to modest declines reported by GFSY and SIPRI. Within the developing country grouping, all sources agree that the Middle East developing countries have the highest military expenditure relative to GDP, while the Western Hemisphere developing countries have the lowest military expenditure relative to GDP.

Regarding changes, all sources show significant increases over the period in military expenditures relative to GDP for Middle East developing countries and small to moderate increases for developing countries in Asia. In contrast, all sources show northern Africa and Western Hemisphere developing countries decreasing military expenditures relative to GDP. The sources show different pictures regarding movements in sub-Saharan African developing countries’ military expenditures relative to GDP.

4. Differences for individual countries

Similarities across sources in reported global or regional totals or trends may be evidence of an underlying consistency in the data sources, or they may be the result of offsetting differences in the figures reported for individual countries. A comparison of reported expenditures by country shows that, at least for changes in military expenditure and as a percentage of GDP spent on military, similarities at the regional or global level are largely the result of offsetting differences at the country level.

Comparisons of reported military expenditures by country were done in three ways. First, we examined differences in the reported numbers, comparing the highest and lowest level of military expenditure for each country in each year in the four sources. For over 58 percent of the countries examined, the highest and lowest estimate of military expenditures varied by more than 30 percent in at least one year. For over 40 percent of the countries, the maximum difference in reported military expenditures was at least 50 percent. And for almost 23 percent of the countries, the highest estimated military expenditure was at least double the lowest estimated military expenditure in at least one year.

Nevertheless, it is possible that a single source is frequently responsible for the divergence. Therefore, for each country we eliminated the source with the most divergent estimate of military expenditures, and compared the three remaining sources. For 65 percent of the countries, the difference across the remaining sources was less than 20 percent in all years. There was no discernible pattern as to which source was eliminated.

Differences in levels could be do to definitional differences, and would perhaps not be too significant if the sources showed similar growth rates of military expenditure. However, a comparison of percentage changes in reported military expenditures shows even larger discrepancies. For 72 percent of the countries, the highest reported percentage change in military expenditures exceeded the lowest by at least 10 percentage points (e.g., 10 percent growth versus 0 percent growth) in at least one year. For over 57 percent of the countries, the maximum difference was over 20 percentage points, and for 29 percent of the countries the maximum difference was over 50 percentage points (e.g., 50 percent growth versus no change).

At the extreme, for 11 countries--Algeria, Argentina, Botswana, Burma, Colombia, Ethiopia, Hungary, Papua New Guinea, Rwanda, Soviet Union and Venezuela--the maximum difference exceeded 100 percentage points (e.g., 100 percent growth versus no change, or 50 percent growth versus 50 percent reduction) in at least one year. In seven of the eleven cases, it appears the large discrepancy may be the result of a one time jump in the data reported by a source, to a level more consistent with the other sources. In other words, it may reflect more complete information, and not an actual estimated increase. However users of the data have no way of confirming this. In the four remaining cases, analysis of the time trend of the data shows no obvious explanation for the huge discrepancy.

Our third method of comparing reported military expenditures by country was to compute annual pairwise correlation coefficients for the numbers reported for each country by two sources. Correlation coefficients of reported levels of military expenditure were generally quite high--.91 or greater in all but one of the 24 pairwise comparisons. This is not surprising; all sources recognize, for example, that the United States has far higher military expenditures than the United Kingdom, which has far higher military expenditures than Uganda, etc.

Correlation coefficients for reported military expenditure as a percentage of GDP--in the range of .65 to .89 in about half the cases, and .90 to .96 in the other half--are generally lower than the correlation coefficients for levels of military expenditure, but still high enough that the sources are conveying a broadly consistent story. Unfortunately, that is not true of correlation coefficients for percentage change in military expenditures. Of the 18 annual pairwise correlations coefficients of the reported percentage change in military expenditures, ten are below 0.5, with three negative correlations and two more approximately zero. Only five of the 18 are above 0.8, and only three are above 0.9.

For researchers whose primary concern is order-of-magnitude cross-country comparisons of the levels of military expenditures, all four data sources reviewed here will tell a broadly consistent story. This is certainly true for dollar values of military expenditures, and to a lesser extent for military expenditures as a percent of GDP. Also, for researchers concerned about one or more individual countries, for many countries in many years the sources provide similar information. For a number of countries, however, this is not true. Further, if the reason for looking at alternative data sources on military expenditures is, at least in part, to assess trends in military expenditures, the conclusions one comes to will depend greatly on which source is used.