## Drivers of Cross‑Border Banking in Sub‑Saharan Africa — wpiea2019146

## Source details

**Canonical URL:** [Drivers of Cross‑Border Banking in Sub‑Saharan Africa — wpiea2019146](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019146.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019146.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019146.pdf.json)

---

### I. Introduction: scope and questions
- Objective: document and analyze the pattern of growth of pan‑African banking, collect new quantitative evidence, and explore drivers of cross‑border investment in bank subsidiaries in SSA.
- Main empirical approach: apply a gravity model relating the annual value of a banking group’s investment in the net equity of its subsidiaries to a set of explanatory variables using data collected from PABs balance sheets and other sources (Orbis, Fitch).
- Key research questions:
  - Were SSA banking groups searching for yield, diversifying their portfolios, or seeking strategic presence?
  - What factors explain differences in cross‑border banking presence?
  - Are there significant differences between various banking groups?

### II. Emergence and recent evolution of pan‑African banks
- Growth and market structure:
  - The number of pan‑African banks (PABs) has increased significantly; seven groups from the region now dominate the landscape.
  - Ecobank (Togo and WAEMU) has operations in 33 SSA countries.
  - Standard Bank (South Africa) is the largest group by asset size.
  - Three Moroccan banks (Attijariwafa, BMCE/Bank of Africa, and GBCP) have large footprints particularly in francophone west and central Africa.
- Systemic importance of subsidiaries:
  - About 50 percent of the subsidiaries of Attijariwafa, BMCE/Bank of Africa, Ecobank, and Standard Bank are systemically important in their home countries.
  - Subsidiaries are systemic on the liability side (with more than 10 percent of total deposits) in 31 SSA countries.
  - Subsidiaries are systemic on the asset side (with assets greater or equal to 7 percent of GDP) in 20 SSA countries.
- Historical shift and expansion dynamics:
  - Cross‑border banking was historically dominated by European and U.S. groups; retrenchment of European groups after the global financial crisis coincided with expansion of SSA‑based groups, North African groups (Morocco), and other non‑European entrants.
  - Between 2006 and 2015 the number of subsidiaries of the 14 largest PABs more than tripled from 53 to 169.
  - Since 2015 PAB expansion has stalled (only Ecobank opened one additional subsidiary).

### III. Cross‑border investment patterns and drivers
- Aggregate investment and timing:
  - Between 2000 and 2015, pan‑African banking groups invested a total amount of about $6 billion (about $376 million per year) in cross‑border subsidiaries, including syndicated and bilateral loans for infrastructure financing.
  - Most investment occurred after 2007; since 2013 new investment declined significantly and turned negative in 2015.
- Concentration across groups:
  - Two groups—Ecobank and Standard/Stanbic—account for two‑thirds of total investment.
  - FirstRand group is the third largest contributor to investment.
  - Standard/Stanbic and UBA are the largest groups by assets, followed by First Rand and Nedbank.
  - Ecobank invested almost one‑fourth of its average group’s assets during the period in cross‑border subsidiaries, compared with less than 5 percent for the other groups.
- Recipient countries (largest by absolute inflows):
  - Nigeria: $1.6 billion
  - Kenya: $800 million
  - Namibia: $600 million
  - Ghana: $450 million
- Modes of expansion:
  - PABs expanded mainly through subsidiaries via acquisition of existing banks, including entities divested by retreating European groups.
  - Greenfield investments are less frequent; exceptions include Guarantee Trust Bank (acquisitions and greenfield) and United Bank for Africa (UBA) (mostly new entities).
  - Prevalence of subsidiaries over branches partly reflects regulators’ wish to minimize contagion.
- Correlates (surface evidence):
  - Investment significantly larger between countries sharing a common language.
  - Common border does not appear to have encouraged investment significantly on the surface.
  - Surface evidence shows no clear relation between investment and geographic distance; deeper analysis shows investment does decline with distance once other factors are accounted for.
  - Groups with higher profitability (average ROE over the period) appear to have invested less in dollar terms, but not in percent of their assets.

### IV. Network connectivity and governance features
- Ownership network structure:
  - Four major country clusters in ownership linkages: South Africa, Nigeria, Morocco, and Kenya.
  - Nigerian, Moroccan, and Kenyan banks serve as central conduits of “controlling” ownership linkages among African countries and sectors.
  - South African and Moroccan banks are the origin of most “non‑controlling” ownership linkages.
- Definitions preserved:
  - “Controlling ownership linkage”: equity holding > 50 percent by a parent company in a subsidiary (requires full consolidation in IFRS‑compliant reports).
  - “Non‑controlling ownership linkage”: equity holding < 50 percent (does not require consolidation).
- Non‑bank exposures and business model:
  - African banks hold significant ownership stakes in a wide variety of non‑bank and non‑financial entities, reflecting an atypical “universal” banking business model with potential moral hazard and possible distortions in lending practices.
- Regional connectivity patterns:
  - Nigeria, South Africa, Morocco, and Kenya are the main centers of connection among the region’s banking sectors.
  - A large share of cross‑border connectivity in South Africa and Morocco arises from investments outside the region; Nigeria and Kenya’s connectivity is driven largely by intra‑African ownership linkages.
  - Nigerian banks have the largest number of direct controlling ownership linkages, especially in banks (but also in corporations), primarily with the banking/corporate sectors of other African economies.
- Foreign banks’ residual presence:
  - Non‑African—mostly European—banking groups continue to have a significant presence in the region despite retrenchment.
  - On the liability side, the 7 foreign banks with the most widespread presence have systemic operations in 17 SSA countries.
  - On the asset side, they are systemic in 9 SSA countries.
  - In 8 SSA countries (Comoros, DRC, Eritrea, Ethiopia, Kenya, Sao Tome and Principe, South Sudan, and Zimbabwe) the 14 selected banks do not have operations deemed systemic.

### V. Empirical model, data, and measurement
- Gravity‑based specification:
  - Y_G,H,t = α_0 + α_G,H + ζ·S_G,H + β·X_G,t + γ·Z_G,t + χ·W_H,t + μ·Φ_G,H + ε_G,H,t
    - S_G,H: T×3 matrix — home GDP, host GDP, geodesic distance (kilometers).
    - X_G,t: group‑specific variables (total equity, assets, ROE).
    - Z_G,t: home country variables (real growth, inflation, trade openness, exchange rate volatility vis‑à‑vis the U.S. dollar).
    - W_H,t: host country variables (real GDP growth, inflation, trade openness, exchange rate volatility).
    - Φ_G,H: country‑pair variables (common border, common language, bilateral exchange rate volatility).
- Data scope and sources:
  - Sample: 8 pan‑African banking groups based in four countries (South Africa, Nigeria, Kenya, and Togo) investing in 41 SSA countries.
  - Total consolidated assets of groups range from $2.5 to $78 billion.
  - Dataset includes 1,632 observations for 102 banking subsidiaries over 2000–15.
  - Dependent variable: net investment flows defined as (a) net purchases of ownership shares and (b) group’s share in retained profits (after‑tax profits net of dividends); adjusted for valuation effects due to exchange rates.
  - Ownership and net equity sources: Bureau van Dijk Orbis; Fitch Connect; published financial statements.
- Measurement notes:
  - Value of group’s ownership in a subsidiary = ownership share × value of subsidiary’s net equity.
  - Ownership share = nominal value of shares held by entities controlled by the group / total nominal value of shares issued by the subsidiary.
  - Changes in ownership value due to share price movements are not considered; retained profits counted as new investment; group’s share of subsidiary losses not included as divestment unless shareholders inject new capital.
- Estimation strategies:
  - Averaged cross‑subsidiary (“BE”) estimation (panel “between”).
  - Panel estimates: Random effects (RE), Feasible GLS, System GMM (Arellano‑Bond) with lagged dependent variable.
  - Two annual exchange rate volatility measures: (i) standard deviation of monthly bilateral exchange rate vis‑à‑vis U.S. dollar; (ii) standard deviation of monthly bilateral exchange rate between home and host currencies.

### VI. Main empirical results — overview and group heterogeneity
- Overall explanatory power:
  - Gravity component not rejected; basic gravity accounts for about 6.6 percent of variance.
  - Adding other explanatory variables raises explained variance to 10.7 percent.
  - Panel data regression overall R2 ranges from 10 to 14 percent, depending on specification and sample.
  - Cross‑subsidiary (“BE”) regressions yield an R2 increase to about 40 percent (sample of observations below 100).
- Gravity and proximity:
  - Home and host country GDP coefficients: both significantly positive (host economy’s size coefficient less than unity, implying smaller countries receive proportionately more investment in percent of GDP).
  - Distance between capital cities: coefficient significantly negative.
  - Common border: significant in panel regressions (especially for Ecobank).
  - Common language: significant in pooled regressions and panel data for 2006‑onward.
- Search‑for‑yield evidence:
  - Group ROE and home country growth associated with lower investment flows (significant negative coefficients), consistent with search‑for‑yield motive.
  - Group ROE: significant negative coefficient overall; weaker after 2006; not visible for Ecobank and Standard/Stanbic in some specifications.
  - For smaller groups (excluding Ecobank and Standard/Stanbic), ROE coefficient remains negative and strongly significant.
  - Host inflation and exchange rate volatility: host inflation often negative; U.S. dollar exchange rate volatility at home significant negative coefficient.
- Diversification evidence:
  - Weaker than search‑for‑yield; visible primarily in panel GLS and cross‑subsidiary regressions for 2006‑onward.
  - Bilateral exchange rate volatility sometimes has a significant positive coefficient; the squared term of bilateral volatility often negative, implying exposure sought up to a point.
- Group‑specific patterns:
  - Ecobank: investment flows not significantly correlated with ROE; positive coefficients on home inflation and trade openness; dollar exchange rate volatility significant negative coefficients.
  - Standard/Stanbic: ROE has significant positive coefficient (consistent with reinvestment); net equity sometimes negative coefficient; investment positively correlated with South African inflation and rand volatility; not sensitive to host‑currency volatility.
  - Other groups: Orabank, KCB, and Standard/Stanbic appear to invest more cross‑border after controls; Guaranty Trust and UBA invest comparatively less.
  - Group size (equity): negatively correlated with cross‑border investment after 2006 (possibly driven by Standard/Stanbic); except for Ecobank, size positively correlated with investment in other groups.
- Host‑ and home‑country macro variables:
  - Home inflation and trade openness significant (signs vary by specification).
  - Host inflation and growth significant only in some specifications; host inflation generally negative but not significant for Ecobank and Standard/Stanbic.
  - Growth coefficient significant only in BE regressions with unexpected negative sign (driven by Ecobank investing in lower‑growth countries).
- Variance explained and limitations:
  - Explanatory variables jointly explain only 10–14 percent of variance in panel regressions; BE regressions explain ~40 percent but with small sample.
  - Results are correlations; causality not established.
  - Collinearity among exchange rate volatility measures may obscure distinct motives.

### VII. Stylized facts and key statistics
- Aggregate investment 2000–2015: about $6 billion (about $376 million per year).
- Timing: most investment after 2007; new investment declined significantly since 2013 and turned negative in 2015.
- Concentration: Ecobank and Standard/Stanbic account for two‑thirds of total investment.
- Recipient countries (absolute inflows): Nigeria $1.6 billion; Kenya $800 million; Namibia $600 million; Ghana $450 million.
- Systemic presence: subsidiaries systemic on liabilities (>10 percent of deposits) in 31 SSA countries; systemic on assets (assets ≥ 7 percent of GDP) in 20 SSA countries.
- Model fit examples:
  - Gravity component explains about 6.6 percent of variance in basic specification; additional variables raise to 10.7 percent.
  - Panel regression overall R2: 10 to 14 percent.
  - Cross‑subsidiary (“BE”) R2: about 40 percent (sample below 100 observations).

### VIII. Policy‑relevant implications and avenues for further research
- Strategic objectives and systemic risk:
  - The systemic importance of many PAB subsidiaries suggests gaining dominant presence in foreign markets was an important motivation for cross‑border expansion.
  - Home‑dominant groups (assets concentrated at home) imply contagion mainly from home to host; network‑dominant groups (assets spread across subsidiaries) imply higher cross‑border contagion risk.
- Supervisory considerations:
  - Typical structure—holding companies headquartered in one country with subsidiaries operating under host rules—creates significant supervisory burdens and underscores need for home‑host cooperation and information exchange.
  - Consolidated prudential supervision of entire groups is critical to ensure adequate capitalization and risk management and reduce international transmission of shocks.
- Competitiveness and inclusion:
  - PABs may serve underbanked populations, increase branch networks in host countries, and export innovative business models from home markets.
- Recommended further research:
  - Responsiveness of cross‑border banking flows to bilateral trade, nonbank FDI, and remittance flows.
  - Relationship between cross‑border investment and natural resource exports, commodity prices, world growth, and global interest rates.
  - Impact of regulatory and institutional environment in host and originating countries.
  - Determinants of the slowdown in investment flows after 2013.
  - Bilateral trade as a potential driver, given prior evidence of positive relationships between bilateral trade and cross‑border exposures.

*Source: WP/19/146 — Drivers of Cross‑Border Banking in Sub‑Saharan Africa (sections I–V; authors: Paul Mathieu, Marco Pani, Shiyuan Chen, and Rodolfo Maino).*

### Section 1

### Drivers of Cross-Border Banking in Sub‑Saharan Africa — Section 1

### I. Introduction: scope and questions
- Objective: document and analyze the pattern of growth of pan‑African banking, collect new quantitative evidence, and explore drivers of cross‑border investment in bank subsidiaries in SSA.
- Main empirical approach: apply a gravity model relating the annual value of a banking group’s investment in the net equity of its subsidiaries to a set of explanatory variables using data collected from PABs balance sheets and other sources (Orbis, Fitch).
- Key research questions:
  - Were SSA banking groups searching for yield, diversifying their portfolios, or seeking strategic presence?
  - What factors explain differences in cross‑border banking presence?
  - Are there significant differences between various banking groups?

### II. Emergence and recent evolution of pan‑African banks
- Recent pattern:
  - The number of pan‑African banks (PABs) has increased significantly in recent years; seven groups from the region now dominate the landscape.
  - Ecobank (Togo and WAEMU) has operations in 33 SSA countries.
  - Standard Bank (South Africa) is the largest group by asset size.
  - Three Moroccan banks (Attijariwafa, BMCE/Bank of Africa, and GBCP) have large footprints particularly in francophone west and central Africa.
- Systemic importance (subsidiaries of selected groups):
  - About 50 percent of the subsidiaries of Attijariwafa, BMCE/Bank of Africa, Ecobank, and Standard Bank are systemically important in their home countries.
  - Subsidiaries are systemic on the liability side (with more than 10 percent of total deposits) in 31 SSA countries.
  - Subsidiaries are systemic on the asset side (with assets greater or equal to 7 percent of GDP) in 20 SSA countries.
- Historical shift:
  - Cross‑border banking was historically dominated by European and U.S. groups; retrenchment of European groups after the global financial crisis coincided with the expansion of SSA‑based groups, North African groups (Morocco), and other non‑European entrants.
  - Between 2006 and 2015 the number of subsidiaries of the 14 largest PABs more than tripled from 53 to 169.
  - Since 2015 PAB expansion has stalled (only Ecobank opened one additional subsidiary).

### III. Cross‑border investment patterns and drivers
- Aggregate investment: 
  - Between 2000 and 2015, pan‑African banking groups invested a total amount of about $6 billion (about $376 million per year) in cross‑border subsidiaries, including syndicated and bilateral loans for infrastructure financing.
  - Most investment occurred after 2007; since 2013 new investment declined significantly and turned negative in 2015.
- Concentration across groups:
  - Two groups—Ecobank and Standard/Stanbic—account for two‑thirds of total investment.
  - FirstRand group is the third largest contributor to investment.
  - Standard/Stanbic and UBA are the largest groups by assets, followed by First Rand and Nedbank.
  - Ecobank invested almost one‑fourth of its average group’s assets during the period in cross‑border subsidiaries, compared with less than 5 percent for the other groups.
- Recipient countries (largest by absolute inflows):
  - Nigeria: $1.6 billion
  - Kenya: $800 million
  - Namibia: $600 million
  - Ghana: $450 million
- Recipient countries (by percent of GDP):
  - Namibia, Liberia, and Lesotho received significant investment inflows in percent of GDP (exact percent values not provided in source text).
- Correlates of investment:
  - Investment was significantly larger between countries sharing a common language.
  - A common border does not appear to have encouraged investment significantly on the surface.
  - On the surface there appears to be no clear relation between investment and geographic distance; however, the paper notes a deeper analysis (Section IV in the full paper) shows investment does decline with distance once other factors are accounted for.
  - Groups with higher profitability (average return on equity (ROE) over the period) appear to have invested less in dollar terms, but not in percent of their assets.
- Modes of expansion:
  - PABs have expanded mainly through subsidiaries via acquisition of existing banks, including entities divested by retreating European groups.
  - Greenfield investments are less frequent; exceptions include Nigeria’s Guarantee Trust Bank (acquisitions and greenfield) and United Bank for Africa (UBA) (mostly new entities).
  - The prevalence of subsidiaries over branches partly reflects regulators’ wish to minimize contagion.

### IV. Network connectivity and governance features
- Ownership network structure:
  - The network of ownership linkages among pan‑African banks exhibits four major country clusters: South Africa, Nigeria, Morocco, and Kenya.
  - Nigerian, Moroccan, and Kenyan banks serve as the central conduits of “controlling” ownership linkages among African countries and sectors.
  - South African and Moroccan banks are the origin of most “non‑controlling” ownership linkages.
- Non‑bank exposures and business model:
  - African banks have significant ownership stakes in a wide variety of non‑bank and non‑financial entities, reflecting an atypical “universal” banking business model with potential moral hazard and possible distortions in lending practices.
- Regional connectivity patterns:
  - Statistically, the banking sectors of Nigeria, South Africa, Morocco, and Kenya are the main centers of connection among the region’s banking sectors.
  - A large share of cross‑border connectivity in South Africa and Morocco arises from investments outside the region; Nigeria and Kenya’s connectivity is driven largely by intra‑African ownership linkages.
  - Nigerian banks have the largest number of direct controlling ownership linkages, especially in banks (but also in corporations), primarily with the banking/corporate sectors of other African economies.
- Foreign (non‑African) banks’ residual presence:
  - Non‑African—mostly European—banking groups continue to have a significant presence in the region despite retrenchment.
  - On the liability side, the 7 foreign banks with the most widespread presence have systemic operations in 17 SSA countries.
  - On the asset side, they are systemic in 9 SSA countries.
  - In 8 SSA countries (Comoros, DRC, Eritrea, Ethiopia, Kenya, Sao Tome and Principe, South Sudan, and Zimbabwe) the 14 selected banks do not have operations deemed systemic.

### V. Implications highlighted in Section I
- Strategic objectives: the systemic importance of many PAB subsidiaries suggests that gaining a dominant presence in foreign markets was an important motivation for cross‑border expansion.
- Supervisory considerations: the typical structure—holding companies headquartered in one country with subsidiaries operating under host rules—creates a significant burden on supervisors and underscores the need for home‑host cooperation and information exchange.
- Competitiveness and financial inclusion: prior studies and anecdotal evidence cited indicate PABs may serve underbanked populations, increase branch networks in host countries, and export innovative business models from home markets.

*Source: WP/19/146 — Drivers of Cross‑Border Banking in Sub‑Saharan Africa, Section I and II (authors: Paul Mathieu, Marco Pani, Shiyuan Chen, and Rodolfo Maino).*

### Section 2

### wpiea2019146 - Section 2

### Ownership linkages and pan-African banking networks
- South African banks have the largest corporate investments, both inside and outside the region; they have comparatively few direct controlling ownership linkages but are heavily connected to the rest of Africa via non-controlling interests.
- Definitions preserved:
  - A “controlling ownership linkage” is an equity holding of greater than 50 percent by a parent company in a subsidiary (requires full consolidation in IFRS-compliant reports).
  - A “non-controlling ownership linkage” is an equity holding of less than 50 percent (does not require consolidation).
- Moroccan banks:
  - Have the second largest number of controlling ownership linkages, evenly split between other African economies and non-African countries.
  - Serve as a conduit of African and non-African financial connections.
  - Have the second largest number of non-controlling ownership linkages (also evenly split between Africa and other regions).
- Kenyan banks:
  - Controlling ownership linkages are primarily with other African banking and asset management sectors.
  - Ownership interests are evenly divided among the corporate, banking, and asset management (“funds”) sectors.
  - Have virtually no ownership of foreign entities outside the region.
- Pan-African banks show significant common exposures, especially among groups based in Nigeria, Kenya, and Morocco; South Africa appears comparatively disconnected.

### Home-dominant versus network-dominant group structures and contagion implications
- Two illustrative group structures:
  - Home-dominant (e.g., Standard Bank/Stanbic group): cross-border subsidiaries numerous but account for a comparatively small share of group assets; assets concentrated in home country.
    - Implication: problems in subsidiaries are unlikely to have a significant impact on the home country; contagion risk runs mainly from home to host.
  - Network-dominant (e.g., Ecobank): home country parent accounts for a comparatively small part of group capital and assets, which are spread across cross-border subsidiaries.
    - Implication: higher risk of contagion, as problems in any part of the group could more easily spread through the parent to other countries hosting subsidiaries.
- Consolidated prudential supervision of the entire group is critical to ensure adequate capitalization and risk management and reduce the risk of international transmission of shocks, particularly given nascent cross-border supervisory collaboration in some jurisdictions.

### Motivations for cross-border expansion: search for yield versus diversification
- Two main motives distinguished:
  - Search for yield:
    - Occurs when profits obtainable in the host country exceed marginal profits from further expansion in the home country or existing countries of presence.
    - Capital flows to where its use is most productive.
  - Search for diversification:
    - A group invests to obtain returns weakly or negatively correlated with existing profit sources, improving the return/risk profile.
    - A group may invest in a country with lower expected returns if those returns are uncorrelated or negatively correlated with existing investments, thereby minimizing uncertainty of returns or achieving a desired risk-return balance.
- The study focuses on macroeconomic and group-specific variables to distinguish these motives.

### Specification of the gravity-based model
- Net investment flow Y_G,H,t between home country G and host country H at time t is explained by:
  Y_G,H,t = α_0 + α_G,H + ζ·S_G,H + β·X_G,t + γ·Z_G,t + χ·W_H,t + μ·Φ_G,H + ε_G,H,t
  - α_0 is a constant.
  - α_G,H (in fixed-effects specifications) is a fixed-effect term specific to the group/host country pair (G,H).
  - ζ is a vector of coefficients for the gravity component; S_G,H is a T×3 matrix: home country’s GDP, host country’s GDP, and geodesic distance (kilometers) between respective capitals.
  - β is a vector of coefficients for group-specific variables X_G,t (total equity, assets, return on equity).
  - γ is a vector of coefficients for home country-specific variables Z_G,t (real growth, inflation, openness to trade, exchange rate volatility vis-à-vis the U.S. dollar).
  - χ is a vector of coefficients for host country-specific variables W_H,t (real GDP growth, inflation, trade openness, exchange rate volatility).
  - μ is a vector of coefficients for country-pair variables Φ_G,H (common border and common language dummies; bilateral exchange rate volatility).
  - ε_G,H,t is a white noise error term.
- Gravity component proxies:
  - Economic mass proxied by nominal GDP (U.S. dollars) from World Bank Development Indicators.
  - Distance proxied by kilometers between financial capital cities from CEPII GeoDist or calculated via great circle formula.
  - Country-pair dummies: common language and common border (CIA World Factbook).

### Data used in the empirical analysis
- Sample scope:
  - 8 pan-Afrrican banking groups based in four countries: South Africa, Nigeria, Kenya, and Togo.
  - Investing in 41 countries in the SSA region.
  - Total consolidated assets of groups range from $2.5 to $78 billion.
  - Dataset includes 1,632 observations for 102 banking subsidiaries over the 16-year period 2000–15.
- Dependent variable:
  - Net investment flows of each of the 8 selected bank groups in the equity of their cross-border subsidiaries, by destination country and year, 2000–2015.
  - “Investment” variable defined as the sum of:
    - (a) the group’s net purchases (gross purchases net of any sales) of ownership shares in a given subsidiary; and
    - (b) the group’s share in any retained profits of the subsidiary (after-tax profits net of dividends).
  - Data adjusted for valuation effects due to changes in exchange rates.
  - Net equity of bank subsidiaries collected from Fitch Connect database (U.S. dollars) and integrated with published financial statements (local currency converted at end-of-period exchange rates).
- Groups included: Ecobank, FirstRand, Guaranty Trust, KCB, Nedbank, the Standard/Stanbic Group, Orabank, and United Bank for Africa (UBA).
- Notes on measurement:
  - Value of group’s ownership in a subsidiary = ownership share × value of subsidiary’s net equity.
  - Ownership share defined as ratio of nominal value of shares held by entities controlled by the group to total nominal value of shares issued by the subsidiary; sources: Bureau van Dijk Orbis and published financial statements.
  - Changes in value of ownership due to share price movements are not considered (no disbursement or withdrawal of funds).
  - Group’s share of subsidiary losses is not included as divestment; retained profits not distributed are considered new investment.

### Methodology and estimation strategies
- Estimation approaches:
  - Averaging data by subsidiary over entire sample period and running the equation on period average values (equivalent to panel “between” estimation, “BE”) to remove short-term noise.
  - Panel data estimates on the entire dataset using:
    - Random effects (RE).
    - Feasible Generalized Least Squares (GLS).
    - Systemic Generalized Method of Moments (GMM) based on Arellano-Bond, including the lagged dependent variable to capture dynamic/autocorrelation effects.
- Four types of explanatory variables included:
  - Gravity model and country-pair variables.
  - Banking group variables (total assets, total equity, ROE) from Fitch Connect.
  - Host country macroeconomic variables (real GDP growth, inflation, trade openness, exchange rate volatility) from World Bank WDI and IMF IFS.
  - Home country macroeconomic variables (same sources).

_Italic: Source: wpiea2019146 - Section 2 (https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019146.pdf)_

### Section 3

### wpiea2019146 - Section 3

### Methodology and empirical setup
- Investment decisions are modeled using period averages to focus on medium- and long-term determinants and to reflect expectations under rational expectations.
- Averaged cross-subsidiary (“BE”) basic equation: ,,,,,ijijijiijij YS   XZWαα ζ   β  γ  χ  ε=++ + ++ +  (where i and j identify a banking group and a host country).
- Estimations include pooled regressions over a sixteen-year sample and a shorter period starting in 2006.
- Regressions run separately for Ecobank, Standard/Stanbic, and the other six groups aggregated.
- Two measures of annual exchange rate volatility used:
  - (i) standard deviation of the monthly bilateral exchange rate vis-à-vis the U.S. dollar, computed every year;
  - (ii) standard deviation of the monthly bilateral exchange rate between the home and host country currencies, computed every year.
- More parsimonious specifications estimated by removing variables with statistically insignificant coefficients and that reduce sample size.

### Main empirical results — overview
- The econometric analysis does not identify causation but shows correlations consistent with:
  - An investment strategy based primarily on transaction cost considerations.
  - A secondary role for search-for-yield and diversification motives.
- The gravity model (size and distance factors) is not rejected and:
  - Accounts for about 6.6 percent of the variance in the basic specification.
  - Adding other explanatory variables raises the “explained” variance to 10.7 percent.
- Overall R2 of the panel data regression ranges from 10 to 14 percent, depending on specification and sample.
- Cross-subsidiary (“BE”) regressions yield an R2 increase to about 40 percent (noting the sample of observations falls below 100 in this format).

### Gravity factors and proximity
- Home and host country GDP coefficients: both significantly positive (host economy’s size coefficient less than unity, implying smaller countries receive proportionately more investment in percent of GDP).
- Distance between capital cities: coefficient significantly negative.
- Common border and common language:
  - Common border significant in panel data regressions (especially for Ecobank).
  - Common language significant in pooled regressions and in panel data for the period from 2006.
- Cross-subsidiary regression: home country GDP not statistically significant (possible collinearity).

### Search-for-yield evidence
- Group profitability (ROE) and home country growth are associated with lower investment flows (significant negative coefficients), suggesting cross-border investment may boost dissatisfactory domestic returns.
- Group ROE:
  - Significant negative coefficient overall (consistent with search-for-yield).
  - Weaker after 2006 and not visible for Ecobank and Standard/Stanbic in some specifications.
  - For smaller groups (excluding Ecobank and Standard/Stanbic), the coefficient on group ROE remains negative and strongly significant.
- Inflation and exchange rate volatility:
  - Host-country inflation often displays a significant negative coefficient in specifications including bilateral exchange rate volatility.
  - Volatility of the nominal exchange rate vis-à-vis the U.S. dollar has a significant negative coefficient, especially in the home country (suggesting reluctance to expand across borders when exchange rate risk is higher).
  - Bilateral exchange rate volatility between home and host currencies sometimes has a significant positive coefficient (see diversification motives section), but is not consistently significant.

### Diversification evidence
- Signs of diversification motive are weaker and visible primarily in panel GLS estimates and cross-subsidiary regressions for 2006–onward.
- Bilateral exchange rate volatility shows a significant positive coefficient in some specifications, possibly indicating a propensity to gain exposure to volatility uncorrelated with group-wide profit drivers.
- The squared term of bilateral volatility has a significant negative coefficient, suggesting exposure is sought only up to a point (very high volatility reduces investment).

### Group-specific patterns
- Ecobank:
  - Annual investment flows not significantly correlated with group profitability (ROE).
  - Significant positive coefficient on home country inflation and trade openness.
  - Dollar exchange rate volatility (home and host) significant negative coefficients.
- Standard/Stanbic:
  - Group ROE has a significant positive coefficient (consistent with reinvestment of profits).
  - Net equity exhibits a negative coefficient despite both net equity and investment increasing until 2012 and declining thereafter.
  - Investment positively correlated with inflation in South Africa and volatility of the South African rand; not sensitive to host-country currency volatility (consistent with large presence within rand currency area, SADC).
- Other groups:
  - Orabank, KCB, and Standard/Stanbic appear to invest more cross-border after controlling for other factors.
  - Guaranty Trust and UBA invest comparatively less.
- Group size (measured by equity):
  - Negatively correlated with cross-border investment in the period after 2006 (possibly driven by Standard/Stanbic).
  - Except for Ecobank, size appears to have a significant positive correlation with investment in other groups.

### Host- and home-country macro variables
- Home-country variables:
  - Inflation and trade openness show significant coefficients.
  - Home inflation: significant with opposite signs across specifications — positive in short term (panel and BE regressions) and negative in the long term — possibly reflecting persistently higher domestic profit opportunities.
  - Trade openness: positive in panel regressions; unexpectedly negative in BE regressions (may reflect correlation with home country size).
- Host-country variables:
  - Only inflation and growth appear statistically significant and only in some specifications (not in BE estimates).
  - Host inflation generally negative, but not significant for Ecobank and Standard/Stanbic.
  - Growth coefficient significant only in BE regressions and with an unexpected negative sign (driven by Ecobank investing in lower-growth countries).

### Variance explained and limitations
- Explanatory variables jointly explain only 10–14 percent of variance in panel regressions.
- Cross-subsidiary (“BE”) regressions explain a larger share (~40 percent), though sample size is much smaller (below 100 observations).
- Results should be interpreted as correlations; causality is not established.
- Possible collinearity among exchange rate volatility measures may obscure distinct motives (search-for-yield vs diversification).

### Stylized facts and key statistics
- From 2000 to 2015, pan-African banking groups invested about $6 billion in cross-border subsidiaries (including syndicated and bilateral loans for infrastructure financing).
- Since 2013 new investment has declined significantly and turned negative in 2015 (decline in net equity due to losses not offset by new capital injections).
- Gravity component explains about 6.6 percent of variance in the basic specification; with additional variables, explained variance rises to 10.7 percent.
- Panel data regression overall R2 ranges from 10 to 14 percent.
- Cross-subsidiary (“BE”) R2 about 40 percent (sample of observations below 100).

### Policy-relevant implications and avenues for further research
- Cross-border expansion in SSA banking is closely linked to transaction cost determinants (geographic and cultural proximity) and shows heterogeneous strategies across banking groups.
- Evidence consistent with both search-for-yield and diversification motives, varying by group and specification.
- Further research recommended on:
  - Responsiveness of cross-border banking flows to bilateral trade, nonbank FDI, and remittance flows.
  - Relationship between cross-border investment and natural resource exports, commodity prices, world growth, and global interest rates.
  - Impact of the regulatory and institutional environment in host and originating countries.
  - Determinants of the slowdown in investment flows after 2013.
- Bilateral trade should be considered as a potential driver, given prior research showing positive relationships between bilateral trade and cross-border exposures.

*Source: wpiea2019146 - Section 3*

### Section 4

### wpiea2019146 - Section 4

### Panel Regression Results (REGLSGMM / Panel)
- Sample / model notes:
  - Observations: 1,190; Number of banks: 100 (column (1) and (2)), 96 (column (3)).
  - Robust standard errors in parentheses. Hansen P-value: 1.000.
- Selected coefficient estimates (coefficient (standard error); significance):
  - ln GDP (host) (contemporary): 0.896*** (0.261); 0.620*** (0.224); 0.771** (0.385)
  - ln GDP (home) (contemporary): 1.578*** (0.549); 1.651*** (0.406); 1.679*** (0.563)
  - ln distance: -0.715*** (0.169); -0.817*** (0.190); -0.763*** (0.160)
  - language dummy: 1.207* (0.689); 1.032* (0.547); 0.855 (0.767)
  - border dummy: 1.538** (0.762); 2.445*** (0.701); 1.709** (0.750)
  - ln equity (group): 0.008 (0.351); 0.013 (0.225); -0.201 (0.357)
  - ROE (group): -0.057*** (0.016); -0.047*** (0.016); -0.065*** (0.017)
  - inflation (home): 0.433** (0.205); 0.338** (0.161); 0.238 (0.211)
  - inflation (squared) (home): -0.014 (0.009); -0.011* (0.006); -0.008 (0.009)
  - trade openness (home): 0.083*** (0.025); 0.082*** (0.020); 0.071*** (0.026)
  - exchange rate volatility (home): -0.972* (0.522); -1.404*** (0.399); -0.713 (0.550)
  - ln investment (lagged): -0.010 (0.015) (reported in one specification)
- Model fit / diagnostics:
  - r 2_w: 0.114; r 2_b: 0.106; r 2_o: 0.108 (reported aggregate)
  - F-test P-Value: 0.000

### Regression Results - Panel Data Grouped by Subsidiary (Table 2)
- Models reported: OLS, GLS, GL S, GMM across columns (1)–(6).
- Selected coefficient estimates (coefficient (standard error); significance) — examples:
  - lhost_ngdp: 0.345*** (0.101); 0.358*** (0.099); 0.390*** (0.073); 0.384*** (0.075); 0.417*** (0.098); 0.381*** (0.090)
  - lhome_ngdp: -0.515** (0.230); -0.523** (0.226); -0.367** (0.151); -0.371** (0.155); -0.220 (0.259); -0.185 (0.263)
  - ldistance: 0.006 (0.234); -0.029 (0.227); -0.293** (0.131); -0.348** (0.136); -0.270 (0.195); -0.310 (0.195)
  - group_roe: -0.053*** (0.007); -0.054*** (0.007); -0.041*** (0.006); -0.043*** (0.007); -0.059*** (0.010); -0.062*** (0.010)
  - host_inflation: 0.011*** (0.003) across multiple columns
  - home_trade: 0.025*** (0.007); 0.018** (0.007); 0.019*** (0.006); 0.018** (0.007); 0.034*** (0.008); 0.026*** (0.008)
  - lhome_usdxravgma: 0.318** (0.158); 0.341*** (0.113); 0.270* (0.152) (selected columns)
- Sample sizes and diagnostics:
  - Observations: 477, 476, 474, 473, 303, 303 (by column)
  - Number of banks: 80, 80, 77, 77, 57, 57
  - F-test P-Value: 0.000 (reported)
  - R-squared reported as 0.145 and 0.148 in some specifications
  - Hansen P-value: 1.000 (reported in GMM columns)

### OLS Regression Results (Table 3) — Grouped by Subsidiaries and Other Aggregations
- Multiple OLS specifications across groupings and averages (columns (1)–(7)).
- Selected coefficients and statistics (coefficient (standard error); significance):
  - lhost_ngdp: 0.366*** (0.140); 0.345*** (0.101); 0.358*** (0.099); 0.373*** (0.105); 0.001 (0.351); 0.515*** (0.134); 0.346 (1.645)
  - lhome_ngdp: -2.257*** (0.771); -0.515** (0.230); -0.523** (0.226); -1.384*** (0.410); -0.709** (0.283); -0.576 (0.380)
  - lgroup_equity: 1.023** (0.412); 0.193* (0.103); 0.210** (0.104); 0.600*** (0.225); 0.331*** (0.110); 0.982*** (0.241); -4.279*** (0.566)
  - group_roe: -0.071*** (0.013); -0.053*** (0.007); -0.054*** (0.007); -0.058*** (0.010); -0.049*** (0.009); 0.136*** (0.040); -0.072*** (0.022)
  - home_trade: 0.040* (0.022); 0.025*** (0.007); 0.018** (0.007); 0.009 (0.009); 0.020** (0.009)
- Observations and fit:
  - Observations: 173; 477; 476; 477; 477; 585; 7 (varies by column)
  - Number of banks: 41; 80; 80; (others reported)
  - R-squared examples: 0.323; 0.145; 0.148; 0.170
  - F-test P-Value: 0.000 (reported)

### Panel Regression Results—Selected Groups (Table 4)
- Subsamples reported: all groups; Ecobank; Standard/Stanbic; Other groups; All groups except Ecobank; All groups except Standard/Stanbic.
- Selected coefficient estimates (coefficient (standard error); significance):
  - ln GDP (host) (contemporary): 1.027*** (0.270); 0.759 (0.534); -1.558*** (0.497); 0.714* (0.419); 0.924*** (0.335); 0.900*** (0.320)
  - ln GDP (home) (contemporary): 1.640*** (0.551); 5.305*** (1.466); 8.781*** (1.692); -0.259 (0.637); 1.841*** (0.679); 0.309 (0.526)
  - ln distance: -0.686*** (0.186); -0.759*** (0.227); 2.645*** (0.878); -0.683*** (0.181); -0.637*** (0.203); -0.755*** (0.147)
  - ln equity (group): -0.042 (0.347); -1.573 (1.048); -1.268*** (0.261); 1.343*** (0.417); -0.342 (0.384); 1.186*** (0.379)
  - ROE (group): -0.058*** (0.016); -0.004 (0.036); 0.229 (0.204); -0.050** (0.020); -0.049*** (0.019); -0.060*** (0.017)
  - inflation (home): 0.504** (0.206); 1.894*** (0.483); -0.084 (0.807); 0.625** (0.280); 0.321 (0.235); 0.716*** (0.249)
  - inflation (squared) (home): -0.015* (0.009); -0.069*** (0.019); 0.098** (0.044); -0.017 (0.012); -0.010 (0.011); -0.021** (0.011)
  - trade openness (home): 0.072*** (0.024); 0.078** (0.032); -0.201 (0.175); 0.065* (0.036); 0.097*** (0.035); 0.071*** (0.026)
- Sample sizes and diagnostics:
  - Observations range: 1,172; 437; 204; 531; 735; 968 (by column)
  - Number of banks: 97; 30; 15; 52; 67; 82 (by column)
  - F-test P-Value: 0.000 (reported)
  - r 2_w examples: 0.12; 0.18; 0.14; 0.13; 0.09; 0.14

### Cross-Subsidiary Results — Baseline and Period after 2006 (Tables 5–6)
- Period after 2006 (Table 6) — selected estimates (coefficient (standard error); significance):
  - ln GDP (host): -1.762 (3.199); -1.540 (3.229); 1.573** (0.652) (different specifications)
  - ln GDP (home): 0.290 (0.330); 0.256 (0.356); 0.328 (0.333); 0.483* (0.271)
  - ln distance: -0.722*** (0.231); -0.766** (0.320); -0.630*** (0.226); -0.577*** (0.217)
  - language dummy: 1.982* (1.158); 1.885 (1.291); 1.442 (0.912); 1.459* (0.772)
  - ROE (group): -0.234** (0.104); -0.202* (0.107); -0.205* (0.104); -0.149* (0.075)
  - exchange rate volatility (bilateral): 7.249** (3.502); 5.321 (4.762) (in some specifications)
  - exchange rate volatility (bilateral) (squared): -2.713** (1.247); -2.090 (1.633) (reported)
- Sample / fit:
  - Observations: 978; 610; 100; 101 (by column)
  - R-squared examples: 0.288; 0.274; 0.240; 0.226
  - F-test P-Value: 0.009; 0.019; 0.021; 0.001 (by column)
- Cross-subsidiary selected groups (Table 7) — examples:
  - All groups: ln GDP (host) (contemporary) 1.675** (0.822); ln distance -0.636*** (0.196); ln equity (group) -2.928*** (0.997); ROE (group) -0.211*** (0.079); rea l GDP growth (host) -1.455 (0.893)
  - Ecobank / Standard/Stanbic / Other group columns report many coefficients with larger standard errors and subgroup-specific estimates (see table for exact values).
  - Observations and R-squared: Observations 973; 216; 52; R-squared 0.390; 0.445; 0.388; 0.449 (where reported)
  - F-test P-Value examples: 0.000; 0.030; 0.659; 0.011

### Dependent Variables and Measurement (Annex I)
- Two dependent variables used:
  - Ownership:
    - Definition: the value of the group’s ownership share in the subsidiary (“ownership”), a stock variable measured at the end of each year.
    - Computation: Ownership = group’s percentage share of the subsidiary’s capital multiplied by the subsidiary’s net equity.
    - Data sources: group ownership shares from Orbis online database or groups’/subsidiaries’ annual reports; net equity from Fitch (preferably consolidated) or annual reports.
  - Investment:
    - Definition: net investment in the subsidiary, a flow variable measuring resources invested in subsidiary capital by any company of the group in a given year.
    - Components: (a) value of shares bought or sold by the group during the year (measured as average value at beginning and end of year); (b) change in value of the shares held by the group due to positive change in subsidiary net equity (retained earnings); adjustments made for valuation changes from bilateral exchange rate movements.
    - Treatment of retained earnings: treated as "passive investment" (increase counted as investment). Losses are not treated as disinvestment unless shareholders inject new capital (treated as net new investment).
- Relationship: equity and investment are closely, but not perfectly, related; equity is affected by investment, losses, and exchange rate changes.
- Table A1 — data sources by group for Ownership Share (S), Net Equity (E), Ownership Value (V = S*E):
  - Examples:
    - Ecobank: Ownership Share — Orbis; Annual Reports. Net Equity — Fitch; Annual Reports. Ownership Value — computed.
    - First Rand: Orbis; Annual Reports (ownership); Fitch (net equity); computed (ownership value).
    - UBA: Annual Reports; Orbis; data collected from the country team (ownership); Annual Reports (net equity); computed (ownership value).
- Table A2 — Data definitions and sources (selected indicators and sources as reported):
  - share (Group share in subsidiary): Bureau van Dijk Orbis database; Published financial statements.
  - equity (Subsidiary net equity): Fitch Connect database; Published financial statements.
  - ownership (Group ownership in subsidiary): IMF staff calculations.
  - investment (Group investment in subsidiary): IMF staff calculations.
  - group_asset (Group total asset): Fitch Connect database.
  - group_equity (Group total equity): Fitch Connect database.
  - group_roa / group_roe: Fitch Connect database.
  - ngdp (Nominal GDP in U.S. dollars): World Bank World Development Indicators database.
  - inflation / cpi / tax / trade: World Bank World Development Indicators database.
  - institution (Rule of law): World Bank Worldwide Governance Indicators database.
  - usdxravg (U.S. dollar exchange rate (avg)) and usdxreop (U.S. dollar exchange rate (eop)): IMF International Financial Statistics database.

*Source: Authors’ calculations (wpiea2019146 - Section 4).*

### Section 5

### wpiea2019146 - Section 5

### Data definitions and sources
- IMF International Financial Statistics database: bi_nxr — Nominal bilateral exchange rate. "Calculated as the annual average host country currency per home country currency exchange rates."
- IMF International Financial Statistics database; IMF staff calculations: bi_rxr — Real bilateral exchange rate. "The real bilateral exchange rate is defined as the ratio of host country price level to home country price level, where home country price level is converted into host country currency units via nominal bilateral exchange rate. The calculation is based on the two countries' consumer prices index (2010 = 100) and nominal bilateral exchange rate."
- World Bank World Development Indicators database; IMF International Financial Statistics database; IMF staff calculations: distance — Capital cities distance. "The distance between capital cities in kilometers calculated following the great circle formula." (CEPII GeoDist database)
- CEPII GeoDist database: colony — Common colonial relationship. "The dummy variable that identifies whether the home countries and the host countries share common colonial relationship or not. If equals to 1, then there is a common colonial relationship."
- CIA World Factbook: language — Common language. "The dummy variable that identifies whether the home countries and the host countries share common language or not. The language is restricted to the use of English, Frence or Portuguese as official languages. If equals 1, then there is a common language."
- CIA World Factbook: border — Common border. "The dummy variable that identifies whether the home countries and the host countries share common land border or not. If equals to 1, then there is a common land border."

### Micro-foundations of the gravity model applied to cross-border banking
- Theoretical basis: Martin and Rey (2004) provide a micro-foundation applying the gravity model of trade to cross-border banking.
- Setup:
  - Each country i (0≤푖 ≤푁) is populated by risk averse agents endowed with a given amount of tradable goods (the numeraire) and a risky project.
  - In period 1 agents consume part of their endowment and trade with other agents.
  - Agents in country i pay p_j = (1 + τ_ij) for a share τ_ij of a project run in country j.
  - In period 2 each project in country j pays dividends d_j in only one of L possible states of nature, and 0 in all other states. The number of states of nature is larger than the number of traded assets and then the total number of projects; hence, markets are incomplete.
- Representative agent maximization problem (as presented):
  - The source presents the maximization problem in formal notation (equation (1) in the text), with σ representing the elasticity of substitution between assets.
- Agent h_i’s demand for shares in projects in country j (equation (2) in the text):
  - d^h_j i x_j = ... (presentation as in source): the formula includes σ, p_j, κ, τ_ij and other terms as specified in the source text.

### Aggregate asset holdings and the gravity equation
- Aggregate solution (Aviat and Coeurdacier, 2007): Asset_ij is derived from the maximization problem and aggregate asset holdings of country i from country j are given in the form presented in the source (equation (3)), involving p_j, n_j, d_j, σ, κ, τ_ij and aggregation across agents.
- Gross returns on assets in country j: R_j = d_j / p_j (as assumed in the text).
- Gravity equation (logarithmic form, equation (4) in the text):
  - log(Asset_ij) = α + log(n_i n_j) + (1−σ) log(R_j) + τ_ij σ + ... (presented in the source as the gravity equation in logarithmic terms).
- Interpretation of terms:
  - α = log(κ) (a constant).
  - Second term: combined market sizes of both countries.
  - Third term: effect of trading costs in financial markets (including distance-related transaction costs).
  - Additional term (not strictly "gravity"): interpreted as a "return chasing" component (Aviat and Coeurdacier, 2007), equivalent to the "search for yield" discussed in the study.

### Network connectivity findings
- Highest (direct and indirect) ownership network connections: Nigeria, South Africa, Morocco, and Kenya.
- Note on foreign ownership: When foreign ownership linkages are excluded, South Africa and Morocco’s estimated connectivity falls—suggesting these countries’ networks are driven more by foreign ownership interests than those of Nigeria and Kenya.
- Definition: "A foreign ownership linkage denotes an equity holding by an African-domiciled parent company in a non-African domiciled subsidiary. All eigenvector centrality scores were computed using both controlling and non-controlling ownership linkages."
- Visual/summary metrics presented in the source (figure labels and category breakdowns retained in the original): counts and breakdowns by entity type (Industrial company, Financial company, Mutual & Pension Fund/Nominee/Trust/Trustee, Insurance company, Bank) and numerical labels appearing in the original figures (e.g., 114, 24, 14, 12, 11, 10, 8, 5, 44, 149, 47, 41, 26, 13, 11). (These numeric labels are reported in the source figures and captions.)

### References cited in the section (select)
- Martin, Philippe; and Rey, Helene, 2004, “Financial Super-Markets: Size Matters for Asset Trade,” Journal of International Economics 64, pp. 335–61.
- Aviat A.; Coeurdacier, N., 2007, The Geography of Trade in Goods and Asset Holdings, Journal of International Economics 71, pp. 22–51.
- Additional cited works include empirical and theoretical studies on cross-border banking, gravity in banking, and financial integration (Ananchotikul et al., 2015; Arena et al., 2007; Buch, 2005; Bruno and Shin, 2013; Niepmann, 2015; Reinhardt and Riddiough, 2015; among others as listed in the source).

*Source: wpiea2019146 - Section 5*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019146.pdf_