General issues

23. As a starting point, issues that need to be explored both within the compiling agency and with market participants are as follows:

• What objectives are being established for the survey at the national level, for instance, information sought, timeliness of results, frequency (if the survey is to be repeated)?

• Which residents invest in and/or own nonresident securities? In a broad sense, who are the largest investors?

• How do residents arrange for the custody of their nonresident securities? With domestic custodians? Or directly with nonresidents?

• How are nonresident securities data, both transactions and position, currently collected? Is it possible to adapt an existing survey, or is there a need to develop a new approach?

• What are the record-keeping and internal arrangements of potential survey respondents? What type of survey would most naturally fit these arrangements?

• Should the data be collected security-by-security or in aggregate?

• Would one type of collection system be inherently more efficient in terms of producing good quality data at lower cost than the other? What is the view of market participants?

• What budget and resource constraints does the compiler face?

24. The answers to the above questions will assist compilers in making an informed decision about the type of collection system to adopt.

The household sector

25. National compilers are advised to examine domestic sources of information to establish, in a broad sense, the saving pattern of the four major sectors of their respective economy: the household, financial, nonfinancial, and government⁶ sectors. This examination might provide some indication of which sectors are likely to be the largest owners of nonresident securities. In this regard, it is particularly important to gauge the extent to which the household sector acquires nonresident securities, and how they are acquired and held.

26. The household sector is invariably a major source of savings in many economies but is difficult to survey directly, given the size of the potential reporting population and the inherent problems in building up a representative sample. Thus, national compilers are encouraged to investigate the channels through which this sector invests in, and holds, nonresident securities. Does the household sector own nonresident securities through domestic pension funds, mutual trusts, etc.? Do custodians and fund managers manage nonresident securities directly for the household sector? Does the household sector transact and hold securities directly with nonresident financial entities, not least for tax minimization reasons?

27. In the experience of the Task Force, national compilers may take various approaches to covering private household sector investment in nonresident securities. One approach is surveying the own account holdings of collective investment funds, for instance, mutual funds and investment trusts, used by households to invest in nonresident securities. Because the household sector has a claim on the fund, and the fund holds nonresident security assets, the compiler is effectively covering household investments by surveying the fund. A second approach is surveying financial entities, like custodians who hold nonresident securities for households or like fund managers who manage investments on behalf of households. The next section places these various options into context—options that are then illustrated using the experience of various national compilers.

28. There remains a problem, however, of covering household investments in nonresident securities either kept in self custody or transacted through and held with nonresident financial entities. In some countries, these are insignificant issues; in others, very significant. For the latter group of countries, there are no simple solutions, and none are provided in this Survey Guide. Probably, the only way that the problem of investment through nonresident financial entities can be approached is through international cooperation—perhaps, in the first instance, through bilateral discussions.

An “end-investor” survey

30. An “end-investor” survey focuses primarily on the beneficial owner of nonresident securities. This approach should provide good coverage when investment in nonresident securities is concentrated in institutional investors, for instance, banks, security dealers, mutual funds, and pension and insurance funds. The quality of the data provided should also be good because end-investors are probably best informed about the size, composition, and value of their own portfolio.

31. In addition, if the current method of collecting portfolio investment data is through an end-investor survey, and it is possible to amend or supplement existing statistical collections, then this approach should be cost-effective. Data from the survey could be cross-checked against existing returns as a method of quality control (see chapter 4, section VI).

32. If, however, collecting portfolio investment data through an end-investor survey is a new approach for the national compiler, then additional preparation will be required. Most importantly, the compiler will need to prepare a comprehensive register of potential survey respondents; otherwise, there is the possibility of undercoverage. Creating and maintaining the register will require resources. National compilers must also avoid either double counting or over recording securities holdings. For instance, entities might consolidate, in their reports, holdings of subsidiaries who report separately to the national compiler (double counting) or holdings of subsidiaries located outside the resident economy (over recording). Essentially, compilers should be clear as to whose holdings are covered by the survey respondent. Moreover, compilers should be alert to the possibility that entities not familiar with completing portfolio investment survey forms may have difficulty in reporting the required information and may require advice. Further information on compiling a register, including advice on choosing the reporting unit, is provided in chapter 4, section IV.

A “custodian”survey

33. A “custodian” survey focuses primarily on those financial institutions that hold nonresident securities on behalf of end-investors. In other words, the survey population is not the same as the population of domestic investors. This type of collection system should provide good coverage when domestic residents mainly hold their nonresident securities at domestic custodians. Because the number of survey respondents will be fewer than the number of end-investors, the difficulties inherent in identifying, and maintaining, the appropriate reporting population are much reduced compared with an end-investor survey.

34. If, however, good quality data are to be compiled from a survey of custodians, the national compiler must be aware of the potentially complex relationships between end-investors and custodians, and among custodians. Because of the hierarchy of custodians (the investor may deal with a primary custodian who in turn deals with a global custodian⁷), the possibility that a security holding may be double counted is ever present. This possibility should not be underestimated. Essentially, national compilers should understand the structure of the custodian business, be clear in their own minds who should report, and write very clear instructions on the survey form. National compilers are also advised to develop flowcharts (examples are provided in appendix VI).

35. Custodians may encounter difficulties detailing the geographic breakdown of nonresident security holdings, distinguishing between resident and nonresident holdings (unless they have a reason, such as tax purposes, to make such a distinction), and valuing some securities at market value. In other words, they may store information in a format that makes it difficult for them to extract the data required by the national compiler. Hence, it is essential that those countries undertaking a custodian survey conduct pre-survey consultations with domestic custodians. Quality control checks are essential (see chapter 4, section VI).

Combined custodian/end-investor survey

36. Surveys of only domestic custodians would miss any securities owned by domestic residents and held directly with nonresident custodians or in self custody. A survey of domestic custodians who hold nonresident securities owned by domestic residents, and of end-investors who hold nonresident securities outside of domestic custodians, would provide more complete coverage. Countries that use combined custodian/end-investor surveys have adopted the approach of collecting data primarily from custodians (and from end-investors only when the end-investors do not use domestic custodians [see this chapter’s section III on country experience]). These countries have found that the vast majority of nonresident securities owned by residents are entrusted to domestic custodians.

37. The caveat in this approach is that there is a clear danger of double counting; domestic investors might erroneously report securities deposited with domestic as well as nonresident custodians. There is also a clear danger of undercounting; the compiler might not be able to compile and maintain a comprehensive register of end-investors who hold nonresident securities outside of domestic custodians. Hence, compilers will need to take particular care in deciding upon the respondent population,⁸ as well as writing very clear reporting instructions. In some countries, national compilers require the end-investor (custodian) to indicate the name of the custodian (end-investor) to whom (on whose behalf) nonresident securities have been entrusted. This helps reduce under-or double counting. However, it should be recognized that custodians, in particular, may be very reluctant to provide such information. Quality control checks are essential (see chapter 4, section VI).

Security-by-security basis

39. Collecting data security-by-security, rather than on an aggregate basis, has the potential of providing more information.⁹ This approach not only provides the building blocks required to construct the position data—geographically allocated—but can also provide additional information, for instance, on currency attribution, industry/sector of the debtor, yields on securities, etc. It could also help identify direct investment securities.¹⁰ But the national compiler will probably need to acquire a database that provides information on individual securities, so that securities can be correctly allocated and valued.¹¹ Before doing so, the compiler needs to decide whether the extra information is required.

40. The decision whether to collect data security-by-security will also depend on the costs involved, which in turn relate to how frequently the national survey will be repeated. A one-time security-by-security survey is almost certain to be more expensive and to take longer to conduct. Given the considerable amount of information to be collected, this type of survey probably needs to be conducted by electronic means, which involves developing compatible software to receive and process the information from survey respondents and to provide a link between the reported data and the security database.

Aggregate basis

41. In contrast, an aggregated survey relies on the survey respondent to aggregate, allocate, and value securities, thus entailing relatively less involvement from national compilers. (However, if the security-by-security survey is repeated on a periodic basis—for instance, once a year—experience suggests that the marginal costs fall while the benefits of extra detail are maintained.)

42. There is also the issue of data quality. Those conducting an end-investor survey regard it as reasonable to expect the data reported on an aggregate security basis to be of good quality, provided the national compiler ensures that survey respondents are fully aware of the survey requirements, for instance, regarding market valuation. This is because end-investors are the best informed about the size, composition, and value of their portfolios, and there is a direct link with their balance sheet data. Hence, end-investor surveys tend to be associated with an aggregated approach.

43. In contrast, to ensure that (1) double counting is kept to a minimum, (2) securities are correctly valued, and (3) quality is maintained, the evidence from those countries collecting data from custodians is that it is advisable to collect data at the individual security level. The U.S. experience is that initial reports from custodians contain frequent, serious inaccuracies; in their latest survey (reference date end-March 1994), the U.S. compilers estimated that relying on these initial data would have led to an overestimation of U.S. holdings of nonresident securities by a factor of between two to four times. Detailed security information helped the U.S. compilers identify possible sources of error and was required to ensure good quality data.

44. Nonetheless, aggregate custodian surveys are conducted. A key factor in ensuring success appears to be the experience of the national statistical agency in organizing custodian surveys. The greater the experience, the deeper the knowledge. The more frequent the survey, the more reliable the data from an aggregate security survey of custodians. (See section III on country experience.)

Austria

48. It became evident in Austria from the mid-1980s that investment in nonresident securities by Austrian residents was growing rapidly. (Between 1985 and 1994, holdings in domestic currency terms increased almost fourfold.) In addition, increasing diversification of portfolio investment holdings became evident, for instance, in terms of currency, residency of issuer, and residual maturity, notably in the nonbank sector. Thereby, policymakers required information on these developments, and it was clear that data based on accumulated flows information from domestic banks were no longer sufficient to meet their needs. Subsequently, in 1988 the Oesterreichische National Bank (ONB) started preparatory work on a new collection system by investigating the custodian market in Austria. In 1991 it introduced a monthly reporting system.

Japan

49. The Ministry of Finance and the Bank of Japan recently undertook the most comprehensive review of the compilation of balance of payments statistics in 30 years to comply, as closely as possible, with the recommendations of BPM5. One significant change, as a consequence of this review, was a methodological improvement of international investment position statistics. Previously, position data were compiled by cumulating flow data, with some partial adjustment to reflect disinvestments and price changes. This was done especially for direct and portfolio investments, which comprise a major part of assets and liabilities. However, concern mounted that the recorded outstanding amounts would, over a long period, deviate significantly from the true position. Thus, to obtain a more reliable picture of Japan’s external assets and liabilities, the authorities introduced a position data reporting system for portfolio and direct investment. They collected the first data with reference to end-December 1995, and they intend to conduct a survey of portfolio investment annually, with data for the year ending December 31, to be reported within one month, that is, end-January of the following year.

Netherlands

50. The Nederlandsche Bank began a survey covering Dutch holdings of nonresident securities in 1986, to complement a survey on nonresident holdings of Dutch securities begun a year earlier. The objective of the annual year-end survey is to collect data for the compilation of statistics on foreign claims (and liabilities) of the Netherlands, to check balance of payments transactions data, and to meet international obligations. Data are to be reported by the following March 31.

United States

51. U.S. position data on nonresident securities owned by domestic residents had been estimated by cumulating transactions data and revaluing to end-period market prices. With the rapid growth of U.S. holdings of nonresident securities in the 1990s, concern mounted that significant errors in the estimated position data existed. A position-level survey was thus undertaken with a reference date of March 1994, which collected security-by-security data from both custodians and end-investors.

Austria

52. The ONB’s initial investigations revealed that domestic residents held over 90 percent of their nonresident securities with domestic custodians. Thereupon, ONB introduced a survey primarily targeted at custodians, who report on a monthly frequency their own nonresident securities and those of their clients. (Data on nonresident ownership of domestic securities are also collected to measure portfolio liabilities.) The reporting population consists of primary, not global, custodians (those closest to the end-investor) to avoid double counting. Nonresident securities owned by the central bank are included. End-investors report data only if the nonresident securities they own are entrusted directly to nonresident institutions and exceed 1 million Austrian schillings at the end of the reporting period. These data are provided only annually, in view of their relative insignificance.

Japan

53. The Japanese reporting system for portfolio position data relies both on custodians and end-investors, reflecting the institutional arrangements through which portfolio investment transactions are normally conducted. Under Japanese foreign exchange trade control laws, portfolio investments are normally conducted or intermediated by foreign exchange banks, designated securities companies, and pre-authorized institutional investors such as insurance companies and investment trust companies. Other investors have to submit prior notification for purchases of nonresident securities of 100 million yen or more, unless they transact through the designated securities companies. However, such transactions are very rare because of transactions and other costs. The following table presents the data collection framework.

Framework of Data Collection in Japan¹

¹For the avoidance of double counting, only the data in the “shaded area” are aggregated to produce total Japanese portfolio investments.

²Disaggregated into public sector, banking sector, and other sectors.

Framework of Data Collection in Japan¹

Banks	Securities Companies	Pre-authorized Investors
Deposited overseas in respondent’s name	Deposited overseas in respondent’s name	Deposited overseas in respondent’s name
(1) own account (2) held on custody for clients2	(1) own account (2) held on custody for clients - public sector - banking sector - other sector
Deposited in other domestic banks		Deposited in domestic banks
Deposited in domestic securities companies		Deposited in domestic securities cos.

¹For the avoidance of double counting, only the data in the “shaded area” are aggregated to produce total Japanese portfolio investments.

²Disaggregated into public sector, banking sector, and other sectors.

View Table

Framework of Data Collection in Japan¹

Banks	Securities Companies	Pre-authorized Investors
Deposited overseas in respondent’s name	Deposited overseas in respondent’s name	Deposited overseas in respondent’s name
(1) own account (2) held on custody for clients2	(1) own account (2) held on custody for clients - public sector - banking sector - other sector
Deposited in other domestic banks		Deposited in domestic banks
Deposited in domestic securities companies		Deposited in domestic securities cos.

¹For the avoidance of double counting, only the data in the “shaded area” are aggregated to produce total Japanese portfolio investments.

²Disaggregated into public sector, banking sector, and other sectors.

Netherlands

54. The Netherlands collect data from about 40 of the most important Dutch custodian banks on their own account holdings and those of their clients. Nonetheless, because of the significant investment in nonresident securities by Dutch institutional investors, the authorities have found it necessary to expand their data collection. Therefore, they also derive total holdings of portfolio investments by pension funds and insurance companies from their annual balance sheets. The difference between these total holdings and those reported by custodians as being held on behalf of pension funds and insurance companies is regarded as investments held with nonresident custodians. The compilers also cover a few important investment companies that are known to hold their nonresident securities with nonresident custodians.

United States

55. The United States collects data primarily from custodians. End-investors are surveyed only when they do not use a U.S.-based global custodian. End-investors using U.S.-based custodians are required to identify their custodians and specify the amounts placed with each custodian. Similarly, primary custodians who use U.S.-based global custodians also identify these custodians and report the amounts placed with them; the U.S.-based global custodians report the data. Nonresident securities owned by the central bank are included in the data reported.

Limitations in approaches to coverage

56. Each of the systems faces potential problems of coverage. In the primarily custodian-based systems, that is, Austria, the Netherlands, and the United States, the potential weakness is that holdings may be missed if residents do not hold nonresident securities with domestic custodians. The Dutch compilers cover this problem through the use of published annual accounts of the major investing institutions, although they acknowledge problem areas among nonfinancial companies and the household sector. Austria and the U.S cover this problem by requiring large investors who entrust securities directly with foreign-based financial institutions to report their holdings directly to the compiler. Nonetheless, the extent to which residents entrust securities with nonresident custodians unknown to the national compiler creates a problem of coverage, although the U.S compilers believe that any amounts missed are likely to be relatively small.

57. For the system in Japan, the important issue is ensuring that those resident investors who are pre-authorized to transact and hold nonresident securities are included in any register of entities to survey. Such a register is kept for regulatory purposes and is used to determine the survey respondent population.

Austria

58. In its initial investigations, the ONB discovered that the banks identify securities by using the security codes of either the Austrian numbering agency (Oesterreichische Kontrolbank) or the German numbering agency (Wertpapiersammelstelle Frankfurt). These codes are linked to the International Securities Identification Number (ISIN) code system (see appendix VII for more information on the ISIN code system). By requiring custodians to report a security identification code and the par value for each security held, the Austrian compilers can, with the use of commercial security databases, ensure that securities are correctly geographically allocated and revalued to end-period market prices in the international investment position (as well as the balance of payments). Also, as the ONB has collected detailed information, it can reconcile stocks and transactions data, recalculate historical time series aggregates, if need be, and ensure that data are consistent with international concepts and principles.

59. Although the purchase of commercial security databases incurs significant costs, the Austrian compilers have found that the benefits of reliable and frequent data outweigh the costs. Furthermore, they have discovered that the custodians prefer security-by-security reporting. The central bank bears the social costs of the compilation of aggregates and the costs of development and running of the commercial security database, making the security-by-security reporting cost-effective for the custodians.

Japan

60. The Japanese compilers require reporters to provide aggregate data, at market price valuation, on equities, long- and medium-term bonds and notes, money market securities, and warrants—by country and currency. The requirement to provide aggregate data, rather than security-by-security data, reflects perhaps the scale of Japanese investment in nonresident securities, combined with the timeframe in which data are to be finalized and the integrated nature of the new reporting system. The Japanese compilers regard the quality of data produced as high, because they check the data against data in other reports; moreover, they contact respondents to discuss data that appear unusual or inconsistent.

Netherlands

61. The compilers in the Netherlands require custodians to report aggregate data on their own nonresident securities and on those of their clients. The survey covers equities and bonds, geographically allocated by the country of nonresident. To obtain a geographic allocation of the holdings of nonresident securities with nonresident custodians, the compilers approach a few important pension funds and insurance companies. Although experience from other countries, as noted above, suggests that it may be very difficult to collect accurate aggregate data from custodians without a regular, ongoing data collection program, Dutch compilers have extensive and in-depth experience collecting data by this method and are convinced their data are of good quality. Indeed, by comparing data reported by custodians against that reported in the annual reports—which ensures good coverage—an independent check is being provided, assuming that the definitional and conceptual approaches are consistent.

United States

62. The United States collected data for its 1994 survey on a security-by-security basis. The compilers requested, from survey respondents, information almost identical to that specified in appendix II of this Survey Guide. To ensure that the securities were correctly geographically allocated and valued, the survey respondents provided a security identification code number for each security reported. Any of the 33 most commonly used coding systems, for instance, ISIN, CUSIP, SEDOL, etc., could be used (see annex B to the model form in appendix II). It is not unusual for commonly traded securities to have codes allocated by ten different coding systems. Survey respondents were encouraged to use ISIN codes to identify securities, but in the 1994 survey only 14 percent of all securities were so identified.

63. To assist in the process of identifying and valuing securities, the U.S. compilers purchased commercial security databases that covered all the commonly used coding systems and provided pricing information on all securities that had traded within the previous year on any of the world’s 87 largest non-U.S. securities exchanges. (General information on securities databases is available in appendix VII.) These databases provided key information that was not always supplied by survey respondents, helped resolve conflicting information, and confirmed the information reported. Nonetheless, a major problem for the compilers was revaluing, to end-period market prices, the rarely traded securities, particularly equity securities, reported unpriced by custodians. These securities accounted for approximately 15 percent of the total number of securities reported.