An Investigation into the Discrepancy in Kenya s Balance of Payments Statistics

Transcription

1 MEFMI Macroeconomic & Financial Management Institute of Eastern and Southern Africa An Investigation into the Discrepancy in Kenya s Balance of Payments Statistics by Sheila Kaminchia Central Bank of Kenya May 2011 A Technical Paper Submitted in Partial Fulfillment of the Award of MEFMI Fellowship. i

2 Abstract The balance of payments (BoP) is a statistical statement in which data on economic transactions between one economy and the rest of the world are recorded. Compiling the BoP entails recording two entries - one credit and the other debit - for every transaction considered so that the sum of all entries in the statement is zero. Due to challenges in collecting requisite data on all BoP transactions, a net errors and omissions term, which is the statistical discrepancy, is introduced in BoP statements to balance out entries and ensure a zero sum result for all transactions as recorded. This paper investigates the statistical discrepancy in BoP statistics for Kenya for the period 1975 to 2009 and aims to identify the BoP items that most explain the discrepancy. A combination of principal components and time series analysis is used in the investigation. The study finds the statistical discrepancy to be most explained by financial account items and finds scope to improve on Kenya s BoP through better understanding of cross-border financial transactions, particularly payment arrangements that give rise to trade credits, and improving compatibility among the existing multiple sources of data. i

3 Acknowledgments I thank Dr. Anna Lennblad for her guidance, comments and suggestions on this paper and earlier drafts. I am grateful for the valuable input of Mr. Isaya Maana and Mr. Benjamin Avusevwa. I also thank MEFMI and the Central Bank of Kenya for their support during the Fellowship Programme. Any remaining errors in this study are mine. ii

4 Table of Contents Executive Summary... iv List of Tables and Figures... v Acronyms and Abbreviations.... vi 1. Introduction Literature Review Overview of the Balance of Payments Framework Practical Implementation of the Balance of Payments Framework Balance of Payments Statistics for Kenya Balance of Payments Compilation Practices in Kenya A Review of the Balance of Payments Statistics Methodology Theoretical Framework Model Specification The Data and Definitions Preliminary Analysis of the Data and Results Testing for Multicollinearity Data Compression using Principal Components Analysis Model Estimation and Results Analysis of Selected MEFMI Country Data Botswana Swaziland An Economic Appraisal of the Results Conclusions and Recommendations References Appendix Appendix Appendix Appendix Appendix iii

5 Executive Summary The Balance of Payments (BoP) is a source of information on a country s external sector. It is important that reported BoP data are sufficient enough in terms of quantity and quality so as not to misguide policy makers and/or frustrate empirical analysis. To improve on reported statistics, regular reviews of statistical systems are undertaken to correct imperfections in reported data as well as to adapt data systems to changes in economic structure. Such reviews can by guided by assessing the data generated by the systems for inconsistencies and gaps, and then tracing identified statistical problems back to the systems that generated the data. This paper investigates the statistical discrepancy in the BoP statistics for Kenya for the period 1975 to Compiling the BoP entails recording two entries - one credit and the other debit - for every transaction considered. The two entries recognize the giving and receiving sides of each transaction so that the sum of all entries is zero for each accounting period. The statistical discrepancy, which is the net errors and omissions term, is defined as the difference between all credit and all debit entries made in both the current account and in the capital and financial account. In this study, it is postulated that a linear combination of current account and capital and financial account items should explain the statistical discrepancy. Results from the statistical analysis are appraised to determine whether or not they are conclusive in the sense that an element of causality exists in the correlations identified. Principal components and time series analyses are used in the investigation. The statistical discrepancy is found to be explained most by portfolio investment assets in both equity and debt securities; direct investment abroad; other investment banks assets and liabilities; as well as income credit and debit; current transfers debit; and capital transfers credit. Lagged vectors associated with financial flows are also found to be significant in explaining the discrepancy, implying timing inconsistencies in recorded data. It is also determined that these results reflect unrecorded offshore transactions; valuation and timing discrepancies; holding gains and losses; unrecorded trade credits; and data handling errors. The study finds scope to improve on Kenya s BoP through better understanding of cross-border financial transactions, particularly payment arrangements that give rise to trade credits, and improving compatibility among the existing multiple sources of data for purposes of BoP compilation. iv

6 List of Tables and Figures List of Tables Table 1: Correlation Matrix for BoP Variables Table 2: Results from Auxiliary Regressions on BoP Variables Table 3: Analysis of Principal Components Table 4: Correlation Matrix - Principal Components (PC) and Statistical Discrepancy (NEO) Table 5: Unit Root Test Results Table 6: Granger Causality Test Results Table 7: Preferred Long-Run Estimation Results Table 8: Preferred Error Correction Estimation Results Table 9: Solved Short-Run Regression Results Table 10: Botswana Preferred Short-Run Estimates Table 11: Swaziland Preferred Short-Run Estimates Table 12: Quarterly Estimates of Trade Credits Table 13: Annual Estimates of Trade Credits List of Figures Figure 1: Distribution of Entries in Kenya s BoP Figure 2: Kenya s Statistical Discrepancy to Figure 3: Trends in Other Investment Liabilities Figure 4: Trends in Customs Data and Foreign Exchange Receipts and Payments Figure 5: Variance in the Balance of Payments Statistics by Item Figure 6: Trends in Private Liabilities and Interest Rates Figure 7: Trends in Direct and Portfolio Investment Flows Figure 8: Scatter Plot of Correlation Coefficients among BoP Variables Figure 9: Loadings of BoP Variables in Principal Components Figure 10: Loadings of BoP Variables in Selected Principal Components Figure 11: Movements in Selected Principal Components and their First Differences v

7 Acronyms and Abbreviations BoP BPM5 CBK CBK-FXS(M) CPI EAC ECT FATS GATS IIP IMF ITRS FDI KNBS KRA MNCs NEO NIPA PC PCA SITC UK UNCTAD US WTO Balance of Payments Balance of Payments Manual, fifth edition Central Bank of Kenya Central Bank of Kenya Monthly Foreign Exchange Survey Consumer Price Index East African Community Error Correction Term Foreign Affiliates Trade in Services General Agreement on Trade in Services International Investment Position International Monetary Fund International Transactions Reporting System Foreign Direct Investment Kenya National Bureau of Statistics Kenya Revenue Authority Multinational Corporations Net Errors and Omissions National Income and Product Accounts Principal Components Principal Components Analysis Standard International Trade Classification United Kingdom United Nations Conference on Trade and Development United States of America World Trade Organization vi

8 1. Introduction The System of National Accounts provides a framework through which the activities of an economy are systematically measured, recorded and subsequently analysed. Within national statistical accounts, domestic production and cross-border trade are linked through the balance of payments (BoP) framework. In this sense, a country s BoP statement provides information on external sector developments with reference to domestic economic activity. It is important, therefore, that reported BoP data are sufficient enough in terms of quantity and quality to truly reflect economic conditions, and changes in a country s trade and finance patterns overtime so as not to misguide policy makers and/or frustrate empirical analysis. To improve on reported statistics, regular reviews of national statistical systems are important and should be undertaken to correct imperfections in reported data as well as to adapt data systems to changes in the structure of the economy. One way to guide such reviews would be to assess the data generated by the systems for inconsistencies and gaps, and then trace identified statistical problems back to the systems that generated the data. It is against this background that the present study is undertaken. The study aims to formally identify the source of the statistical discrepancy in Kenya s BoP given recorded transactions. The study reviews the statistical systems supporting BoP compilation in Kenya vis-à-vis international compilation standards as articulated in the Balance of Payments Manual, fifth edition, (BPM5), which is published by the International Monetary Fund (IMF). This paper is structured as follows: a review of the literature on BoP compilation is presented in chapter 2. Chapter 3 contains background information on Kenya s BoP statistics including compilation practices. An analysis of the statistical discrepancy in Kenya s BoP is presented in chapter 4 and a similar analysis for two other MEFMI countries is presented in chapter 5. Chapter 6 appraises the results from the statistical analysis of chapter 4 from an economic perspective and chapter 7 concludes with recommendations. 1

9 2. Literature Review 2.1 Overview of the Balance of Payments Framework The balance of payments (BoP) is a statistical statement that systematically summarizes, for a specific time period, the economic transactions of an economy with the rest of the world (International Monetary Fund, 1993, para. 13, p. 6). The BoP is concerned with transactions between residents of the reporting economy and residents of the reporting economy s trading partners and is focused on transaction flows and not stocks. For BoP purposes, a transaction is defined as an economic flow that measures activity per unit of time and reflects the creation, transformation, exchange, transfer, or extinction of economic value (ibid. 1993, p. 6). Transactions of interest are those that entail provision and acquisition of goods; the exchange of financial assets through lending and borrowing; and the provision of services, labour and capital. BoP transactions are recorded in two main accounts, namely the current account and the capital and financial account. The classification of transactions into these two accounts of the BoP is meant to facilitate analysis. Transactions in real resources are classified separately from transactions in financial assets because the two types of transactions are undertaken for different purposes. When classifying transactions in the BoP, distinction is made among transactions that entail the generation of output from an economy s production processes (goods and services); the use of factors in the production process (income); the exchange of financial assets and liabilities that is directly related to trading of produced output; and autonomous financial investments (International Monetary Fund, 1993, para. 13, p. 6). Other considerations made when classifying transactions in the BoP relate to the compilation procedures of the System of National Accounts; the significance of each type of transaction for the reporting economy; and the practicability with which data can be sourced. Transactions recorded in the current account have a direct association with production and consumption of real resources in the domestic economy (link with gross domestic product). The current account thus consists of items that reflect the provision or acquisition of goods, services and factors of production (labour and capital) by the reporting economy to or from the rest of the world. Credit flows in the current account measure the proportion of the reporting economy s domestic product provided to the rest of the world. Debits flows measure 2

10 the acquisition of the rest of the world s gross domestic product and the use of non-resident factors of production by the reporting economy. In the current account, credits and debits are reported separately, that is, on gross basis (International Monetary Fund, 1993, para. 150, p. 38). Transactions recorded in the capital and financial account are not directly related to the processes of production and consumption. The capital account captures transfers that are associated with the provision or acquisition of capital assets (those that can produce a stream of services overtime) while the financial account covers transactions in external financial assets and liabilities and which show how an economy s cross-border transactions are financed. The capital and financial account thus measures net investment in/lending to or net borrowing from the rest of the world. To the extent that domestic savings is not matched by an increase in domestic capital accumulation, there will be an increase in private or official assets held with the rest of the world (International Monetary Fund, 1993, para. 560, p. 160). The item labelled Income in the current account provides a link between the current account and the capital and financial account. Returns earned on an economy s financial asset stock are recorded as investment income in the current account as are income payments for the use of foreign capital. The current account and capital and financial account relate as the BoP identity, which is given as Current Account Balance = - (Net Capital and Financial Account + Reserve Asset Transactions) The BoP identity states that the balance on the current account should exactly mirror the balance on the capital and financial account plus transactions in reserve assets. The identity requires that the net provision of real resources by an economy to the rest of the world matches a change in the country s net financial claims on the rest of the world. A surplus on the current account should therefore be reflected as an increase in net financial claims on nonresidents or as the acquisition of reserve assets. Conversely, a deficit on the current account implies that the net acquisition of real resources from the rest of the world must be paid for by either liquidating foreign assets or increasing financial liabilities to non-residents (International Monetary Fund, 1993, para. 557 and 558, p. 160). 3

11 Conceptually, therefore, the BoP identity provides for the sum of all entries in the statement to be zero. Compiling the BoP thus entails recording two entries - one credit and the other debit - for every transaction considered. The two entries are used to recognize the giving and receiving sides of every transaction so that the net effect of all transactions for each accounting period is zero. Following accounting conventions, credit entries denote a reduction in the assets of the reporting economy or an increase in the economy s liabilities while debit entries denote a reduction in liabilities or an increase in assets. For example, in the case of a transaction that involves the exchange of a good for a financial item, the current account of the exporting economy will be credited by the value of a good exported while the financial account will be debited by the value of the financial asset received in return. The Balance of Payments Manual, fifth edition, (BPM5) contains rules that guide the compilation of a complete and consistent BoP statement. For instance, the rules on valuing transactions ensure that no difference arises between the credit value and the debit value pertaining to a single transaction. The recommendation is for transactions to be valued at the actual price agreed upon by transactors (International Monetary Fund, 1993, para. 91, p. 26), which essentially is the market price that is established given the demand and supply conditions prevailing in the market. For financial transactions, either the price at which financial assets and liabilities are acquired or disposed of or nominal values of nontraded financial items should be applied (ibid.1993, para. 106, p. 28). For consistency, the pricing method used for a single transaction recorded in the BoP should apply to both entries. The rules on timing require that the two entries of a transaction be made simultaneously in order to ensure that both sides of the transaction are shown to have occurred on the same date (International Monetary Fund, 1993, para. 109, p. 30). It is recognised that some transactions may span more than one accounting period. The time of recording such transactions is therefore determined to be the time at which a change in ownership takes place. A change in ownership is considered to occur at the time the parties to a transaction record it in their books or accounts (ibid.1993, para. 113, p. 30). If the credit side of a transaction is recorded in one period and the corresponding debit is recorded in the next period, then asymmetries between the two main accounts will arise. However, due to imperfect information on BoP transactions and transactors, the time of recording may not correspond with the change of ownership. 4

12 Due to the diverse range of BoP transactions and the different sectors of the economy involved, data on current account and financial account transactions are commonly captured using a mix of data collection methods, which have different periodicity and timing. This practice departs from principles prescribed by international standards and leads to inconsistencies in the BoP. A separate item known as net errors and omissions is usually introduced in BoP statements to balance out entries thus ensuring a zero sum result for all transactions as recorded. Following from the BoP identity, the statistical discrepancy (that is, the net errors and omissions term, NEO) in the BoP can be defined as CAB = - (NKA + NEO) rearranging NEO = - (CAB + NKA)... (equation 1) where CAB represents the balance on the current account and NKA represented the balance on the capital and financial account including transactions in reserve assets. In equation (1), the NEO compensates for either under- or over-reporting in the capital and financial account given the entries made in the current account. For example, for a given transaction recorded in the current account, there must be recorded a financing transaction in the financial account of equal but opposite value. If the contra entry in the financial account is not of an equivalent value, the difference is taken up by the error term. The NEO, however, has its shortcomings. It represents the net effect of all recorded transactions so that its absolute value does not indicate the degree of misreporting since some reporting errors cancel out (International Monetary Fund, 1993, para. 148, p. 38). The discrepancy is, therefore, a rough measure of the net effect of transactions which fail to be captured by statistical systems and/or are erroneously recorded. As an example, suppose a resident is contracted to supply goods to someone in a foreign country in Period 1. Suppose that the source of information on exports is the customs authorities, who have valued the goods at 100 currency units, while the source of information on the financing side of the transaction is banking statistics, which shows a receipt of 110 currency units in Period 1. Given these sources of information, the exporter s country balance of payments will reflect the export value of 100 in the current account and the receipt of 110 in the financial account as an increase in foreign currency and deposits. The result of this valuation difference will be errors and omissions of +10 in Period 1 as shown below. 5

13 Example of Valuation Differences in BoP entries Statement with entries from different sources: Period 1 Current account Exports +100 Capital & Financial account Other investment Trade Credits Currency and Deposits -110 Net errors & omissions 10 Following from the previous example, suppose the resident exporter is contracted to supply the goods in Period 2. This time, assume that data on exports from the customs authorities shows a valuation of 100 currency units whereas data on the same transaction from the banking statistics shows a receipt of the same value of 100 currency units. The other difference now is that the customs data shows that the goods were exported in Period 2 while the banking statistics shows that the exporter received the money in Period 1 as an advance payment. The exporter s country balance of payments will reflect the receipt of 100 in Period 1 and an entry for exports in Period 2. The result will be errors and omissions of 100 in Period 1 and -100 in Period 2. If an entry was made for trade credits, then the trade credit entry would balance out the statement and no errors and omissions will arise as shown below. Example of Timing Differences in Balance of Payments entries Statement with no trade credit entries: Statement with trade credit entries: Period 1 Period 2 Period 1 Period 2 Current account Current account Exports +100 Exports +100 Capital & Financial account Capital & Financial account Other investment Other investment Trade Credits Trade Credits Currency and Deposits -100 Currency and Deposits -100 Net errors & omissions Net errors & omissions 0 0 The two examples above demonstrate the need to appraise different sources of information for BoP compilation for compatibility. If the two sources are to complement each other sufficiently enough to avoid errors and omissions in the final BoP statement, then it is necessary to introduce an adjustment item, for example, by explicitly accounting for trade credits of any other account receivable or payable. 6

14 The size of net errors and omissions can be estimated relative to the sum of all debit entries and the sum of all credit entries in the BoP. Calculated this way, the size of the discrepancy will depend on the size of transaction values captured in the BoP and correctness of recorded information. Sourcing information on the same transaction from different sources leads to discrepancies if data from the different sources does not match in terms of value and timing. The implication of this is that larger errors and omissions may occur if transactions are partially accounted for whereas net errors and omissions may be reduced by discarding partial information on transactions. There is also an implication to using only one source of data for BoP compilation; errors and omissions will be avoided only if the data provided by the only source is handled properly. These characteristics give benefit to using statistical systems that capture more complete data on BoP transactions. 2.2 Practical Implementation of the Balance of Payments Framework Although BoP compilation procedures vary across countries, there are some common practices. Customs authorities are relied upon to provide data on cross-border trade in goods, while enterprise surveys are commonly used to gather information on international trade in services, current transfers and exchange of financial assets. BoP data is also obtained using International Transactions Reporting Systems (ITRS) - mostly where foreign exchange markets are controlled; household expenditure surveys; sector surveys (such as tourism and immigration surveys) and administrative records (e.g. for government transactions, tax collections from non-residents, landing and stevedoring charges). See Appendix 1 Table A1 for a summary of BoP compilation practices in MEFMI 1 Member States. There are weaknesses inherent in these data capturing methods. The ITRS, for instance, is an indirect reporting system that captures transactions on a cash basis (that is, when payments are made) as opposed to when ownership change occurs. Respondents may also not be able to differentiate transactions for purposes of reporting in the ITRS leading to misclassification. The ITRS, however, has its benefits in that it involves a small number of respondents and provides timely data. Enterprise surveys, on the other hand, capture more accurate information directly from a larger sample of transactors. The quality of statistics collected through enterprise surveys, however, depends on the appropriateness of the technique used to determine the sample of enterprises to be surveyed, the technique used to design the 1 Macroeconomic and Financial Management Institute of Eastern and Southern Africa 7

15 questionnaire and process results and the quality of the business register (World Trade Organisation, 2006). The quality of statistics also depends on the rate of response of those included in the sample. A number of studies on discrepancies in country BoP statements have been undertaken with a view to improving on the accuracy of reported statistics. These studies show that the use of varied data collection methods creates data discrepancies within the BoP accounts. Klein and Makino (2000) studied the statistical discrepancy between the expenditure and income sides of the National Income and Product Accounts (NIPA) for the United States of America (US) and attributed statistical discrepancies within the NIPA accounts to the method of estimating each item within the accounts separately and independently. Barzyk and Laliberté (1992) also found that the use of a wide variety of data sources contributed to timing inconsistencies and data gaps in Canada s BoP statistics. Despite the discrepancy, Canada s data quality was found to be relatively higher than that of Australia and the US. This finding was based on the average calculated ratios of the statistical discrepancy to the sum of gross merchandise exports and imports for the period 1980 to The ratios were 2.2% for Canada, 2.8% for Australia, 3.9% for the US and 1.9% for the United Kingdom (UK). This relatively better quality was attributed to Canada s centralised statistical system and enabling legal framework, and the use of supplementary information from sources including administrative government records and data collected by its trading partners. Canada also sources micro-data which are used to validate specific transactions. Even with the common platform for data collection across countries, disparities arise in definition, valuation and coverage of BoP transactions, which in turn give way to inconsistencies and discrepancies in cross-country data (Lindner et al., 2001; UNCTAD, 2005a). A review of regional and global aggregates for BoP data for about 170 countries attributed the global discrepancy 2 to incomplete coverage of transactions, inaccurate/inconsistent recording of transactions, and different classification/timing of transactions (International Monetary Fund, 2007). Comparisons of bilateral BoP data have also been undertaken with a view to determining the source of asymmetries in partner-country data and improve the quality of individual country data. Bilateral comparisons are useful in 2 The global discrepancy is defined as the difference of the combined surpluses and deficits of the individual accounts, and the totals, for all countries. 8

16 identifying scope to harmonize concepts, definitions, and compilation methods across countries as well as exposing gaps and weaknesses in the reporting systems (Orford et al., 2007; Timmermann, 1997). Disparities among data on trade in goods largely stem from differences in the trade systems used by countries; differing ways of measuring goods and transaction types; timing differences that generate time lags between exports and corresponding imports; and the use of differing exchange rates (Lindner et al., 2001). Regarding Foreign Direct Investment (FDI) statistics, deviations from guidelines include the recording of FDI flows other than on a net basis, unconventional valuation of FDI stocks such as the summation of flows to arrive at stock data, which does not properly allow for among others re-investment and asset revaluation (UNCTAD, 2005a). A study for Portugal and Germany however shows that harmonization of theoretical concepts is not sufficient to ensure consistency and argues for a common approach to the practical application and interpretation of concepts and definitions (Timmermann, 1997). Actions taken to resolve BoP data problems are especially in areas that are gaining importance in international transactions such as autonomous international financial transactions, trade in services, migrant remittances, and FDI. The growing significance of these transactions in the global economy has in part also motivated the revision of the BPM5. Moreover, the coming into force of the WTO General Agreement on Trade in Services (GATS) has also increased the demand for high quality statistics on trade in services (World Trade Organisation, 2006; Bensidoun and Ünal-Kesenci, 2008). There are recommendations for improved measurement of services supplied by the foreign affiliates of Multinational Corporations (MNCs) within the new framework of Foreign Affiliates Trade in Services (FATS) statistics. This follows from the growing importance of services transactions conducted through enterprises affiliated to multinational companies. Unlike international trade in goods which has a well established administrative apparatus to support data collection, international service transactions are much more difficult to measure due to the intangible nature of services, which makes them more difficult to define (Lindner et al., 2001). Lipsey (2006) attributes the discrepancies in partner-country data related to sales of US affiliates to, among others, ambiguity in determining the residency of (particularly intangible) assets. This problem is exacerbated by the existence of special purpose enterprises. Lipsey 9

17 explains that determining the residence of a service provider and the consumer is key in measuring trade in services, the determination of which can change what is, on the face of it, a domestic transaction into an international transaction. Analytical problems also arise when reported assets are not associated with production in or use of factors of production of the reporting economy and that the output from these assets is only attributed to the reporting economy by statistical convention. Solutions to the problems of measuring the activities of MNCs while adhering to the territorial principle are offered in Stokrom and Roosendaal (2004) and Stokrom et al. (2006). The solutions hinge on enhanced communication between national statistical offices and respondent enterprises. This would facilitate adjustment to reported statistics in order to distribute value addition across the countries where MNCs are located while maintaining consistency within national statistics. Determining the residency and understanding the operations of economic agents is also an issue for FDI statistics. Hull (2002) studied FDI data for New Zealand and describes how various corporate financing decisions affect BoP statistics. The study demonstrates that differences between the economic purpose and the legal form of financial flows may misrepresent the various forms of financial flows in the BoP and the International Investment Position (IIP). UNCTAD (2005b) also notes that BoP statistics do not always reflect the true nature of corporate transactions and can give misleading and distorted information about corporate dealings. This gives rise to the need for increased understanding of the transactions of institutions involved through the collection of additional data on the activities of foreign affiliates and their parent companies. Wada and Oonishi (2003) reviewed FDI and IIP compilation practices in Japan. Disparities in the accumulated flow and stock data were observed and attributed mainly to differences in data coverage of both flows and stock data; time lags between the recording of flows and stock data; exchange rate changes; and other factors such as the use of book values for FDI stock data and market value for FDI flows data. For Japan, the use of book value was preferred to current price valuation as it was easier to obtain. 10

18 3. The Balance of Payments Statistics for Kenya 3.1 Balance of Payments Compilation Practices in Kenya The Kenya National Bureau of Statistics (KNBS) is the official compiler of Kenya s BoP. KNBS relies on enterprise surveys for data on services and financial transactions, covering among others households, airline companies, shipping companies and non-governmental organisations. Other sources of data include the monthly survey of commercial banks foreign exchange transactions [CBK-FXS(M)] administered by the Central Bank of Kenya (CBK), and administrative records. The use of administrative records is however constrained by inconsistent data compilation practices across Government Departments as well as accessibility issues (Ministry of Planning and National Development of the Republic of Kenya, 2003). Statistics on international trade in goods are obtained from the Kenya Revenue Authority (KRA). Coverage includes goods that enter or leave Kenya s economic territory whether dutiable or not (United Nations, 2007). Territorial elements include industrial free zones, commercial free zones, customs warehouses and premises designated for inward processing. The time of recording of trade statistics is not done according to change of ownership due to the difficulty of having simultaneous reporting by both exporters and importers (ibid., 2007). Statistics are therefore recorded at the time goods cross the customs border. Statistics on merchandise imports are compiled by country of origin and by country of consignment (not country of purchase). Statistics on merchandise exports are compiled by country of last known destination and by country of consignment (not country of sale) (United Nations, 2007). Goods are valued according to their fair market value. Imports are valued on C.I.F 3 basis while exports are valued on F.O.B 4 basis. The classification of trade statistics is in accordance with the United Nations Standard International Trade Classification (SITC) (United Nations, 2004). Adjustments are made to these statistics to arrive at goods classified on a BoP basis. For instance, statistics on newspapers and periodicals are netted out of customs trade data and added to services under computer and information services. Insurance 3 Cost, Insurance and Freight value includes the transaction value, and the cost of transportation and insurance to the frontier of the importing country or territory. 4 Free on Board value includes the transaction value and the cost of transportation and insurance to bring the merchandise to the frontier of the exporting country or territory. 11

19 and freight components of import values are also subtracted from customs import values and added back to services. Statistics on services, income and current transfers are sourced from enterprise surveys carried out annually by the KNBS; records of the Central Government and CBK; and CBK-FXS(M). CBK-FXS(M) is a reporting system through which commercial banks operating in Kenya report transactions that affect their net foreign asset position. The report only picks up on transactions settled through the domestic banking system. These transactions are recorded when payments are made. Before the CBK-FXS(M) was introduced, data on services and financial account items was sourced from CBK exchange control records. Data on government services and external financing are sourced from official records and are recorded when payments are due. Data on other capital and financial account items are predominantly sourced from enterprise surveys conducted by the KNBS, and the CBK-FXS(M). Figure 1 below shows that credit and debit values for both current and capital and financial account items portray the cyclical pattern in gross national income, which implies some consistency between the BoP statistics and national accounts data. Figure 1: Distribution of Entries in Kenya s BoP Source of data: International Monetary Fund (2010a) 12

20 3.2 A Review of the Balance of Payments Statistics This section provides a general overview of the developments in Kenya s balance of payments for the period 1975 to The statistics are presented in Appendix 2 Table A2. Between 1975 and 1979, Kenya s overall BoP was generally positive. The macroeconomic environment then was controlled (import and exchange controls were in effect) as policies were aimed at directing more investment to agriculture, manufacturing and tourism. During this period, the value of exports increased and was largely supported by commodity price booms in 1977 and Services, particularly travel and transportation, also increased during this period. The break-up of East African Community (EAC) in 1977 was followed by a rise in Kenya s import bill due to increased official imports to meet the capital equipment requirements of former EAC corporations (Kenya Economic Survey, 1977; Central Bank of Kenya, 1978). The resulting trade deficit was financed by official borrowing from the IMF as well as private financial inflows. Private financial inflows rose significantly as firms controlled by foreigners were encouraged to source for financing abroad. The statistical discrepancy during this period was positive but relatively small. The introduction of the structural adjustment programme in 1980 removed restraint on the outward transfer of investments and liberalised trade between Kenya and neighbouring countries (Central Bank of Kenya; 1981). The exchange and payments system was partially liberalized with tariffs replacing quantitative restrictions on imports. The value of exports declined in 1980/1981 on falling commodity prices, drought and reduced foreign demand due to recession in industrialized countries. Import values were however higher in the two years due to rising food imports and oil prices but declined thereafter following imposition of import bans and foreign exchange quotas in 1982 (Kenya Economic Survey, 1982). Subsequent removal of the bans in 1984 was followed by a rise in imports. Net receipts from services generally increased during the five-year period to Official and private financial inflows however declined particularly after During this period, the statistical discrepancy remained positive and was on average larger than in the previous five years. Between 1985 and 1989, the value of exports was, on average, flat as the country faced unfavourable movements in terms of trade and drought. This went against the shift from an inward-looking import-substitution regime to an export-oriented strategy during which time the government began offering incentives to export in the form of cash rebates and tax concessions (Central Bank for Kenya, 1987). During this period also, net receipts from 13

21 services and current transfers rose but slower than the rise in imports and income debits (indicating increased borrowing from abroad). The resulting deficit in the current account was financed by official borrowing and increased private financial inflows. The statistical discrepancy maintained it upward trend, peaking in Between 1990 and 1992, the value of exports of goods and services generally rose while the value of imports declined thereby narrowing the trade deficit. Recorded inward foreign direct investment also reduced as other financial flows worsened so that the financial account deteriorated to a deficit in In 1993, the Kenya shilling was devalued (and subsequently floated) to correct for the deterioration of the country s international export competitiveness, a development that was blamed on an overvalued nominal exchange rate as well as rising domestic inflation in Kenya relative to inflation in her trading partners. The current account was then liberalised in 1994, which allowed domestic exporters to retain all their foreign currency receipts. The financial account was also partially liberalised in 1995 allowing Kenyan residents to open and operate foreign currency accounts, borrow abroad without limit and make outward investments (Central Bank for Kenya, 1995). The restriction on transfer of equity shares was also abolished and non-residents were permitted to acquire up to 20 percent of issued share capital of companies listed on the Nairobi Stock Exchange; to invest in local money market instruments; and to repatriate their capital and income earned from such investments. Domestic interest rates and the Kenya shilling exchange rates were left to be determined by the market. The aim of these policies was to give market forces a greater role in determining the availability of foreign exchange domestically so as to attract foreign investments to the equity and money markets, and to encourage more vigorous private sector participation in the economy. The statistical discrepancy rose further peaking in 1992 and 1994 but dropped sharply in 1993 (Figure 2). Between 1995 and 1999, the value of imports surged as the export values of goods and services increased albeit less rapidly. The resultant deficit in the current account was financed by private financial inflows. The statistical discrepancy rose further and was largest in 1998 and

22 Figure 2: Kenya s Statistical Discrepancy to 2009 Source of data: International Monetary Fund (2010a) Since 2000, exports and imports of goods and services increased in tandem with domestic economic activity. Current transfers receipts also rose reflecting increased activity of nongovernmental organisations in the country, while income credits (mostly earnings on official foreign exchange reserves) rose faster than income payments. The current account deficit widened and was largely financed by capital transfers and private financial inflows. Since 2000, the statistical discrepancy has been both positive and negative, averaging US$ 200 million in absolute terms. To recap, there was a change in the conduct of business after liberalization in 1993/1994. Before, private sectors engaged in foreign trade were obliged to surrender foreign currency earnings to the Government through the Central Bank of Kenya (CBK) and to acquire foreign currency with which to settle import bills from the CBK. This meant that Kenya s current account was predominantly financed through government books meaning that any changes in the current account attributed to private sector activities during this period were reflected in financial liabilities of the general government, which in turn trended alongside the financial liabilities of the private sector (banks and other sectors). The tight co-movement between general government and private sector foreign financial liabilities however weakened after 1993/1994 as private sector foreign financing was delinked from government operations (see Figure 3). 15

23 Figure 3: Trends in Other Investment Liabilities In the liberalized environment, trading in foreign currency (cash or other foreign financial instrument) is by regulation preserved for commercial banks although foreign exchange bureaus may trade foreign cash at spot rates (Central Bank of Kenya, 2002). This means that receipts and payments related to balance of payments transactions can be analyzed by looking through commercial bank accounts. For instance, when a Kenyan exporter receives payment from abroad, the receipt is credited to the exporter s foreign currency account held with a bank locally. This translates into an increase in the receiving bank s foreign currency liabilities (vis-à-viz the exporter) until the exporter sells the foreign currency to the bank in exchange for local currency, in which case the receiving bank s foreign currency liabilities (vis-à-viz the exporter) decrease. Another example is that of transfers in the form of migrant remittances. A significant proportion of these receipts are channeled through commercial banks or money transfer agents (World Bank Group, 2010), sold to commercial banks or foreign exchange bureaus, and the Kenya shilling equivalent spent locally. Figure 4 below shows trends in foreign currency receipts and payments made through commercial banks in Kenya. The charts show that foreign currency flows associated with trade in goods made by commercial banks on behalf of their customers closely follows trade values reported by Kenya s customs authorities. 16

24 Figure 4: Trends in Customs Data and Foreign Exchange Receipts and Payments Immediately after liberalization of the foreign exchange market, the balance of payments was in surplus owing to a rise in export values of goods and services while goods imports declined. Thereafter, a significant widening of the current account deficit occurred, which corresponded with a rise in private financial liabilities. An increase in foreign private financing in the form of bank loans and other credit is observed in the data as well as a rise in direct and portfolio investment flows. Figure 5 below shows that the variance in the BoP statistics across the 35- year period under review is larger for goods, followed by services, current transfers, direct investment in Kenya and other investment liabilities. Figure 5: Variance in the Balance of Payments Statistics by Item 17

25 An anomaly in Figure 5 is the relative stability of income account items. It would be expected that increased foreign borrowing (in cash, credit or FDI) to finance external trade would raise the interest and dividend payable by the private sector and the government, unless the real cost of borrowing declined. Figure 6 below indicates that for the private sector, the latter scenario is the one alluded to by the BoP data for Kenya. Figure 6: Trends in Private Liabilities and Interest Rates Not all flows in the financial account of the balance of payments are directly related to the flow of goods and services in the current account. Autonomous financial transactions are also recorded in the financial account. As highlighted earlier, Hull (2002 p. 35) and UNCTAD (2005b p. 5) discuss the issue of round-tripping and how multinational firms balance between equity and debt financing in an effort to (among other reasons) reduce the amount of tax payable to the host government. In one such case, a non-resident parent company may initially loan money to a resident conduit firm and the conduit would then buy equity securities in the targeted resident enterprise. The impression left on the host economy s BoP by such corporate maneuvering is that the economy is attracting higher foreign financial inflows in the form of debt than equity. Another case to consider is that while an initial acquisition of equity in a resident company amounting to 10 percent or more of the resident company s issued capital constitutes FDI, subsequent incremental investment may not be categorized as FDI but rather as portfolio investment. 18

26 Whether or not the two practices presented in the preceding paragraph are prevalent in Kenya 5 is subject to investigation. Nonetheless, reviewing the annual reports of three FDI firms operating in Kenya, a common practice is to combine both foreign loans and foreign equity to finance operations. The East African Portland Cement Company Ltd. (an exporter of cement) and Kenya Airways Ltd. (a provider of air transportation services) favoured debt for capacity expansion and equity for strategic positioning/market seeking (among other considerations). A challenge with using foreign debt is to minimize foreign currency exposure and currency mismatch arising from revenues and scheduled debt payments being denominated in different currencies. For example, revenues of the East African Portland Cement Company Ltd (EAPCC) are mostly in Kenya shillings whereas their debt obligations are denominated in the Japanese Yen (EAPCC, 2010). While EAPCC is yet considering engaging financial derivatives to manage their currency mismatch problems, Kenya Airways uses foreign fuel derivatives (swaps and options) to minimize the impact of exchange rate and fuel price fluctuations on their operating costs (Kenya Airways, 2010). Another similarity among these firms is that they are FDI investors in other EAC enterprises that operate within their line of business. Another example is of a US-based firm that has more than 50 percent shareholding in and lends money to Kenya-based REA Vipingo Plantations Limited (an exporter of sisal fibres), which itself owns shares in and provides long-term debt financing to another firm located in neighbouring Tanzania (REA Vipingo, 2010). The point being made here is that one method of external finance may lead to the use of other foreign financial instruments for purposes of risk management or revenue enhancement. Figure 7 below shows that between 1994 and 2009, a sharp increase in FDI flows to Kenya was followed by an almost proportionate increase in portfolio investment flows into Kenya one to two years after the increase in FDI inflows. A similar trend is visible for investment outflows from Kenya. 5 It is at least known that the tax rates are a concern for firms operating in Kenya (World Bank Group, 2007). 19

27 Figure 7: Trends in Direct and Portfolio Investment Flows 20

28 4. Methodology 4.1 Theoretical Framework The statistical discrepancy in the BoP originates from a mismatch in recorded credit and debit items in the current account and in the capital and financial account. Conceptually, the balance on the current account and the balance on the capital and financial account must add up to zero. For the zero sum to materialize, the net errors and omissions term (NEO) must cover for the smaller (or weaker) balance on the two accounts. It is possible to identify the individual gross flows that most affect the net balance on the two main accounts. Variations in the gross values (exports and imports of goods and services) will be reflected in variations in the net balance on the current account, which in turn will be reflected in the statistical discrepancy, if variations in the offsetting financial flows do not match variations in these current account items. For example, if the NEO behaves more like imports, then it means that the offsetting financial flows are not adequately captured and that the NEO represents the omitted offsetting financial flows. Therefore, although the statistical discrepancy is a net term while individual current account items are gross terms, variations in the statistical discrepancy are still likely to reflect the stronger variation in the gross flows and represent weakness in the corresponding offset. A linear combination of current account, and capital and financial account items is therefore expected to provide some explanation for the statistical discrepancy 6. In the rest of this chapter, the statistical correlation between NEO and various BoP items will be investigated. The findings will later be subjected to an economic appraisal to determine whether or not they are conclusive in the sense that an element of causality exists in the correlations identified. 4.2 Model Specification Ordinary least squares was used to estimate a general linear model of the form yi = β0 + β1x1i + β2x2i + + Βkxki + εi, i = 1,..., n where yi represents the dependent variable (the statistical discrepancy or the net errors and omissions term), x1,, xk represent the explanatory BoP variables, i denotes observations, k denotes the number of explanatory variables, n denotes the sample size, β0... βk denote the coefficients of interest, and εi is a random disturbance. 6 Some recording errors may have cancelled out so the combination of the BoP items is not expected to explain 100 percent of the discrepancy. 21

29 4.2.1 The Data and Definitions Annual time series data was used for all test variables. The 23 BoP items listed in Appendix 2 Table A2 are the subject of analysis. The items to be analysed exclude totals, subtotals, and other investment assets of the Monetary Authority which has zero values throughout the sample period. The data sample runs from 1975 to While preliminary analysis was done on nominal values of the data, subsequent regression analysis was done using real variables. Export and import values were each divided, respectively, by export price index and import price index for Kenya to obtain real values. Export of services and financial liabilities were deflated by the consumer price index (CPI) for Kenya while import of services and financial assets were deflated by the average of the CPIs for the US and the UK (key trading partners of Kenya). The BoP and the CPI data were sourced from the December 2010 edition of the International Financial Statistics online statistical database. The data variables are defined as follows: Goods exports: this is the F.O.B value of movable goods for export by Kenyan residents to non-residents. The series mainly comprises general merchandise (mostly agricultural products and light manufactures) and includes repairs on goods and goods procured in ports by carriers. Goods imports: this is the F.O.B value of movable goods for import by Kenyan residents from non-residents. The series covers general merchandise, repairs on goods and goods procured in ports by carriers. For Kenya, the general merchandise imports category comprises mainly fuel, capital equipment and manufactured goods. Trade balance: the value of goods exports minus the value of goods imports. Services credit: covers international service transactions performed by Kenyan residents for non-residents. These transactions include transportation services for both passengers and goods by sea, air and road. The series also covers services acquired from Kenyan residents by non-resident travellers for their own use during their visits to Kenya of less than a year. The series also covers business services, government services, communication services, insurance services, financial services such as brokerage service fees and commissions, royalties and licence fees. 22

30 Services debit: these are services performed by non-residents for Kenyan residents. The category includes transportation services, which mainly comprises freight by sea but also includes other transportation services by air and road; travel debits which cover expenditures by travelling residents of Kenya abroad; government services; business services; communication services; insurance services; financial services such as brokerage service fees and commissions; royalties and licence fees. Income credit: covers transactions involving compensation of employees (wages, salaries, other benefits) paid by non-residents to resident Kenyan workers (that is, residents working in an economy other than Kenya) and income receipts on Kenyan residents external financial assets (direct investment income, portfolio investment income and other investment income). This data series mainly comprises investment income other than income earned on direct and portfolio investment. No income from direct investment abroad is recorded after Income debit: covers transactions involving compensation of employees paid by Kenyan residents to non-resident workers, and income payments on Kenyan residents external financial liabilities. The data series covers investment income (other than income from direct and portfolio investment), dividends and distributed branch profits, income from equity and bonds, and interest payments on foreign debt. Current transfers credit: these are offsets made for entries in the BoP that do not consist of the provision of a real resource or a financial item. The data series includes current transfers mainly to non-governmental organisations, workers remittances (beginning 2001) and program grants to the Government of Kenya. Current transfers debit: these are offsets made for entries in the BoP that do not consist of the receipt of a real resource or a financial item. The data series covers workers remittances, and transfers made by the Government of Kenya and by other sectors of the economy. Balance on services: the net balance on services, income and current transfers. Current account balance: the sum of the trade balance and the balance on services. 23

31 Capital account credit: involve transfers of funds linked to the acquisition of fixed assets, or cancellation of liabilities to creditors without any counterparts being paid in return. The data series mainly comprises project grants paid to the Kenya Government and migrant transfers. Capital account debit: involve transfers of funds linked to the disposal of fixed assets, or cancellation of liabilities to debtors without any counterparts being received in return. The data series mainly comprises migrant transfers. Balance on capital account: capital transfers credit minus capital transfers debit. Direct investment abroad: reflects a lasting interest of a resident entity in Kenya (direct investor) in an entity resident in another economy (direct investment enterprise) and covers initial and subsequent transactions between direct investors and direct investment enterprises. The data series mainly comprises direct investment transactions in the form of equity capital and reinvested earnings. Direct investment in Kenya: reflects a lasting interest of a resident entity in an economy other than Kenya (direct investor) in an entity resident in Kenya (direct investment enterprise) and covers initial and subsequent transactions between direct investors and direct investment enterprises. The data series mainly comprises liabilities to direct investors in the form of equity capital, reinvested earnings, and other forms of direct investment capital such as loans. Portfolio investment assets: covers transactions in which resident Kenyans acquire foreign equity and debt securities issued by non-resident Kenyans, excluding those covered under direct investment and reserve assets. The data series begins in 1993 and comprises transactions in debt securities (Treasury bills and bonds) and equity held by sectors other than the government, central bank and commercial banks. Portfolio investment liabilities: covers transactions in which non-resident Kenyans acquire equity and debt securities issued by residents of Kenya. The data series begins in 1993 and comprises bonds and notes held by the central bank and equity securities held by commercial banks. 24

32 Other investment assets: covers trade credits, loans, currency and deposits, other accounts receivable and payable extended by residents of Kenya to non-residents. Other investment liabilities: covers trade credits, loans, currency and deposits, other accounts receivable and payable extended by non-residents to residents of Kenya. Balance on financial account: the sum of direct investment assets net of liabilities, portfolio investment assets net of liabilities and other investment assets net of liabilities. Net errors and omissions: balancing item that ensures a zero sum result for all transactions recorded in the BoP. Overall balance: the sum of the balance on the current account, the balance on the capital account and the balance on the financial account. Analysis of this data in the following sections was done using EViews and PcGive econometric packages and MS Excel Preliminary Analysis of the Data and Results In order to estimate a general linear model, the test data was first analyzed to determine the most suitable form it should take in the estimable equation. Correlation analysis was first applied in order to explore the relationships among the variables. Figure 8 and Table 1 show the correlation coefficients among the variables. Goods exports and imports correlate more strongly with income credit, income debit, services credit, services debit, capital transfers debit, direct investment in Kenya, other investment general government liabilities and other investment other sectors liabilities. Services credit and services debit trend most closely with current transfers credit, current transfers debit, capital transfers debit, direct investment in Kenya, other investment general government liabilities and other investment other sectors liabilities. Income credit and income debit have the strongest correlation with services credit, services debit, current transfers debit, capital transfers debit, direct investment in Kenya, other investment general government liabilities and other investment other sectors liabilities. Portfolio investments returned relatively higher correlation coefficients with income debit, services credit, services debit, current transfers debit and capital transfers credit. 25

33 Figure 8: Scatter Plot of Correlation Coefficients among BoP Variables In general, correlation coefficients are higher with regard to current account items than with financial account items. This implies that the trends in the current account items are not very similar to the trends in financial account items. This is not unexpected since transactions in the current account are recorded on a gross basis whereas entries in the financial account are recorded on a net basis. It is noteworthy that the ratio of goods and services imports to current and capital transfers receipts declined steadily from an average of 18 in the late 1970s to 4 by This indicates that, over the sample period, a fair amount of current account transactions was increasingly financed through the financial account as opposed to through the capital account and other grants. 26

34 Table 1: Correlation Matrix for BoP Variables GDSEX GDSIM SERVCR SERVDR INCCR INCDR CURRTRANSCR CURRTRANSDR CAPACCCR CAPACCDR GDSEX 1.0 GDSIM SERVCR SERVDR INCCR INCDR CURRTRANSCR CURRTRANSDR CAPACCCR CAPACCDR DIABROAD DIKENYA PIESASSETS PIESLIAB PIDSASSETS PIDSLIAB OIMALIAB OIGGASSETS OIGGLIAB OIBANKSASSETS OIBANKSLIAB OIOSASSETS OIOSLIAB Table 1 (cont d) DIABROAD DIKENYA PIESASSETS PIESLIAB PIDSASSETS PIDSLIAB OIMALIAB OIGGASSETS OIGGLIAB OIBANKSASSETS OIBANKSLIAB OIOSASSETS OIOSLIAB DIABROAD 1.0 DIKENYA PIESASSETS PIESLIAB PIDSASSETS PIDSLIAB OIMALIAB OIGGASSETS OIGGLIAB OIBANKSASSETS OIBANKSLIAB OIOSASSETS OIOSLIAB Testing for Multicollinearity Given the high correlation among the variables, applying regression analysis on the data will encounter the problem of multicollinearity. This problem makes it difficult to isolate the effects of each explanatory variable on the dependent variable and yields coefficient estimates that are imprecise (Greene, 1997). High correlation among explanatory variables is however not a sufficient condition for multicollinearity (Sufian, 2005). To ascertain that the dataset has this problem, each explanatory variable was regressed on a constant and all other explanatory variables (see Table 2 for results). A high coefficient of determination (R 2 ) in these regressions is indicative of multicollinearity. The highest R 2 obtained was 0.99 and more, for goods exports, goods imports, services credit, services debit and income debit. The lowest R 2 obtained was 0.61 for other investment monetary authority liabilities, 0.63 for direct investment abroad and 0.64 for other investment other sectors assets. F-statistics and associated probability values were also 27

35 calculated for each of these equations to test for the degree of multicollinearity. The calculated F-statistics follow an F distribution with k-2 and n-k+1 degrees of freedom and are defined as F k 2, n k 1 R 2 k R n k 1 where k denotes the number of explanatory variables including the intercept, and n denotes the sample size. The F-test was applied at the 5 percent level of significance; if the F-statistic is significant as indicated by a p-value of less than 0.05, then the associated explanatory variable is determined to be collinear with the other explanatory variables, whereas if the F- statistic is not significant, then the associated explanatory variable is determined not to be collinear with the other explanatory variables. The p-values in Table 2 were calculated using MS Excel. Table 2: Results from Auxiliary Regressions on BoP Variables R 2 F-Statistic p-value Goods Exports Goods Imports Services credit Services debit Income credit Income debit Current Transfers credit Current Transfers debit Capital Account credit Capital Account debit Direct Investment Abroad Direct Investment in Kenya Portfolio Investment in equity securities assets Portfolio Investment in equity securities liabilities Portfolio Investment in debt securities assets Portfolio Investment in debt securities liabilities Other Investment Monetary Authority liabilities Other Investment General Government assets Other Investment General Government liabilities Other Investment Banks assets Other Investment Banks liabilities Other Investment Other Sectors assets Other Investment Other Sectors liabilities The F-statistics were found to be significant at the 5 percent level except for direct investment aboard, other investment monetary authority liabilities, other investment general government assets, other investment banks liabilities and other investment other sectors assets. Results 28

36 based on Klein s Rule of Thumb test, which suggests that multicollinearity may be a problem only if the R 2 obtained from an auxiliary regression is greater than the overall R 2 of the regression on the dependent variable (the statistical discrepancy), indicates a problem among current account items. The overall R 2 of the regression on the dependent variable is One way to determine the degree of collinearity among the explanatory variables is to use the eigenvalues of the correlation matrix of the explanatory variables. If the ratio of the highest eigenvalue to the lowest eigenvalue is between 100 and 1000, then there is moderate to strong multicollinearity. If this ratio exceeds 1000, then multicollinearity is severe. The maximum eigenvalue obtained from the correlation matrix of the explanatory BoP variables is 4696 whereas the lowest eigenvalue is The calculated ratio is therefore greater than 1000 indicating the presence of strong multicollinearity in the dataset. The simplest way to correct for multicollinearity is to drop variables from the model or to increase the sample size. Alternatively, ridge regression or principal components analysis (PCA) can be applied (Greene, 1997). The first two remedies are not suitable in the present case since there is no scope to increase the sample size whereas all variables in the sample are important to the study and may not be dropped. PCA is preferred since the desire is to also reduce the number of explanatory variables Data Compression using Principal Components Analysis PCA was applied to remove high correlation among the explanatory variables and to compress the dataset to a more manageable number of variables, thereby allowing further regression analysis. PCA discovers new variables called principal components (PCs) from a set of data, which account for majority of the variability in that dataset. In other words, PCA aims to explain relationships among several correlated variables in terms of a few, relatively independent factors (Kleinbaum and Kupper, 1978). The procedure for computing PCs is to first mean-adjust the variables in the original dataset, calculate the covariance matrix of the mean-adjusted data 7 and then calculate the eigenvectors and eigenvalues of the covariance matrix. The eigenvectors represent significant patterns or 7 The data has to be centered around zero so that the computed PCs all have a common origin. This procedure is done as a matter of convenience and does not change the distribution of the sample. 29

37 relationships in the dataset (sort of a line of best fit) while the eigenvalues indicate the strength of each corresponding relationship vis-à-vis other relationships present in the data. The larger the eigenvalue, the longer is the corresponding vector and the more concentrated data is around that vector. Finally, the PCs are constructed as linear functions of each variable in the mean-adjusted dataset, weighted by the corresponding coefficients in the eigenvectors (Smith, 2002). It is noteworthy that the sign of the weights has no meaning in the PC space. What is of interest when using PCA is to reduce the dimensionality of the data while retaining a fair amount of information content (Davies et al., 2005). The other desirable property of PCA is that the PCs are uncorrelated and suitable for use in regression analysis. The procedure described above was applied to the test variables. Table 3 shows the calculated eigenvectors and corresponding eigenvalues associated with each principal component. Out of the principal components constructed, 94 percent of the total variation in the data was found to be explained by the first component as indicated by the first R-square in Table 3. In Table 3, PC1 is the first principal component and represents the dominant relationship in the dataset. In PC1, the largest weights are given to goods imports followed by goods exports, services credit and other investment general government liabilities. The second principal component (PC2) gives most weight to services credit, goods imports, goods exports and current transfers credit. The third component (PC3) gives higher weight to goods exports, current transfers credit, other investment other sectors liabilities, and services credit. This finding is consistent with Figure 5 above which shows that current account items and other investment liabilities in the financial account exhibit most variance in the dataset. 30

38 Table 3: Analysis of Principal Components PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 PC10 PC11 PC12 PC13 PC14 PC15 PC16 PC17 PC18 PC19 PC20 PC21 PC22 PC23 Eigenvalue R-square Cumulative R-square Eigenvectors: GDSEX GDSIM SERVCR SERVDR INCCR INCDR CURRTRANSCR CURRTRANSDR CAPACCCR CAPACCDR DIABROAD DIKENYA PIESASSETS PIESLIAB PIDSASSETS PIDSLIAB OIMALIAB OIGGASSETS OIGGLIAB OIBANKSASSETS OIBANKSLIAB OIOSASSETS OIOSLIAB

39 Close inspection of the other principal components reveals that PC4 to PC10 are associated with current account items, direct investment and other investment items; PC11 to PC15 are associated more with income credits and debits, current transfers debits, portfolio investments debt securities liabilities, and other investment items; while PC16 to PC23 are associated more with current transfers debits and financial account items, particularly portfolio investment items. This is as depicted in Figure 9 below, which gives a graphical view of the weights given to each BoP item in the PCs. Figure 9: Loadings of BoP Variables in Principal Components Note that vectors 4 to 23 have small eigenvalues having been weighted down by either relatively small data scores of the BoP variables they represent as compared with the data scores for goods and services, and/or missing observations. Despite their small eigenvalues, PC6, PC9, PC12, PC14, PC17, PC19 and PC22 are sampled for inclusion in subsequent analysis because of their association with a range of financial account items. The compression of the 23 BoP variables into the few variables (or vectors) of interest achieved through PCA can be seen by comparing Figure 9 and Figure

40 Figure 10: Loadings of BoP Variables in Selected Principal Components Table 4 below shows that the selected principal components are independent of each other, and are weakly correlated with the statistical discrepancy (NEO), except for PC3. Table 4: Correlation Matrix Principal Components (PC) and Statistical Discrepancy (NEO) PC1 PC2 PC3 PC6 PC9 PC12 PC14 PC17 PC19 PC22 NEO PC PC PC PC PC PC PC PC PC PC NEO