Reverse Efficiency Spillovers from Host Country Banks to Foreign Banks: Evidence from Emerging Market Bank Subsidiaries in Developed Markets

Abstract

Extant research on Foreign Direct Investment (FDI) spillovers mainly focuses on how local firms benefit from foreign firms. Our study examines the reverse spillovers in multinational banks. More specifically, we study how emerging-market multinational banks (EMMNBs) can receive reverse efficiency spillovers from local banks when investing in developed markets. We hypothesize that, when operating in developed markets, EMMNB subsidiaries receive positive spillovers from local banks by means of absorbing advanced technological knowledge and managerial expertise. By contrast, EMMNB subsidiaries receive negative spillovers from other EMMNB subsidiaries from the same home country because they compete for similar markets and resources. Empirical analyses, based on a worldwide panel dataset of emerging-market bank subsidiaries in developed markets from 2000 to 2014, support our hypotheses. Our findings enrich the FDI spillover literature as we examine how EMMNBs learn from local banks in developed countries, and extend the learning theory by incorporating the multinational banks context.

Appendix: Bank efficiency estimation (SFA)

This study has adopted the intermediation approach to estimate bank efficiency (Sealey & Lindley, 1977). This approach is typically employed when banks are relatively independent entities (Bos & Kool, 2006). This method views banks as financial intermediaries, which convert input into valuable output. Thus, bank efficiency measures the extent to which banks use input to produce increased output.

Following the bank-efficiency literature (Chortareas et al., 2013; Gaganis & Pasiouras, 2013; Lozano-Vivas & Pasiouras, 2010), we assume that banks have two types of output: loans (Q₁) and other earning assets (Q₂), which include investment securities, inter-bank funds, and other investments. Consistent with numerous studies of bank efficiency (Duygun et al., 2014; Gaganis & Pasiouras, 2013; Goddard et al., 2014), we have selected the following input prices: the cost of labor (W₁), which is calculated as the ratio of personnel expenses to total assets; the cost of borrowed funds (W₂), which is calculated as the ratio of interest expenses to deposits; and the cost of physical capital (W₃), which is calculated by dividing the overhead expenses (other than personnel expenses) by the fixed assets.

The translog function is popular model for measuring bank efficiency, due to its advantages as a flexible-function model (Greene, 1980). The translog specification takes the following format:

$$ln\left( {TC/\left( {\alpha W_{{3}} } \right)} \right)_{it} = \beta_{0} + \beta_{{1}} ln\left( {Q_{{1}} /\alpha } \right)_{it} + \, \beta_{{2}} ln\left( {Q_{{2}} /\alpha } \right)_{it} + \beta_{{3}} ln\left( {W_{{1}} /W_{{3}} } \right)_{it} + \beta_{{4}} ln\left( {W_{{2}} /W_{{3}} } \right)_{it} + \beta_{{5}} \left( {lnQ_{{1}} /\alpha } \right)^{{2}}_{it} + \beta_{{6}} \left( {lnQ_{{2}} /\alpha } \right)^{{2}}_{it} + \beta_{{7}} \left( {ln\left( {Q_{{1}} /\alpha } \right) \times ln\left( {Q_{{2}} /\alpha } \right)} \right)_{it} + \beta_{{8}} \left( {lnW_{{1}} /W_{{3}} } \right)^{{2}}_{it} + \frac{1}{2}\beta_{{9}} \left( {lnW_{{2}} /W_{{3}} } \right)^{{2}}_{it} + \beta_{{{1}0}} \left( {lnW_{{1}} /W_{{3}} } \right) \, \times \, \left( {lnW_{{2}} /W_{{3}} } \right)_{it} + \beta_{{{11}}} \left( {lnW_{{1}} /W_{{3}} } \right) \, \times \, \left( {lnQ_{{1}} /\alpha } \right)_{it} + \beta_{{{12}}} \left( {lnW_{{1}} /W_{{3}} } \right)) \, \times \, \left( {lnQ_{{2}} /\alpha } \right)_{it} + \beta_{{{13}}} \left( {lnW_{{2}} /W_{{3}} } \right) \, \times \, \left( {lnQ_{{1}} /\alpha } \right)_{it} + \beta_{{{14}}} \left( {lnW_{{2}} /W_{{3}} } \right) \, \times \left( {lnQ_{{2}} /\alpha } \right)_{it} + v_{it} + u_{it}$$

(3)

Where i denotes banks; t denotes time periods; and TC, Q, and W refer to the total costs, output values, and input prices of the banks, respectively; β_k (k = 0, 1, 2, …, 14) are the parameters to be estimated, and u_it and v_it are the random error and inefficiency terms, respectively. Moreover, α refers to the total assets of the bank, output terms are expressed as a ratio of α, e.g. Q₂/α, to eliminate size-related heteroscedasticity and any potential bias arising from differences in scale. The normalization of input prices by either of the abovementioned variables, e.g. W₁/W₃, ensures price homogeneity (Berger & Mester, 1997). The frontier function is estimated using the Frontier 4.1 program (Coelli, 1996). This software estimates the frontier model as a one-step model by Battese and Coelli (1992), which uses maximum-likelihood estimation. Table 8 presents the descriptive statistics of variables in efficiency estimations.

In a robustness test (models 4–6 in Table 6), we have included a time trend in the translog specification:

$$ln\left( {TC/\left( {\alpha W_{{3}} } \right)} \right)_{it} = \beta_{0} + \beta_{{1}} ln\left( {Q_{{1}} /\alpha } \right)_{it} + \, \beta_{{2}} ln\left( {Q_{{2}} /\alpha } \right)_{it} + \beta_{{3}} ln\left( {W_{{1}} /W_{{3}} } \right)_{it} + \beta_{{4}} ln\left( {W_{{2}} /W_{{3}} } \right)_{it} + \frac{1}{2}\beta_{{5}} \left( {lnQ_{{1}} /\alpha } \right)^{{2}}_{it} + \beta_{{6}} \left( {lnQ_{{2}} /\alpha } \right)^{{2}}_{it} + \beta_{{7}} \left( {ln\left( {Q_{{1}} /\alpha } \right) \times ln\left( {Q_{{2}} /\alpha } \right)} \right)_{it} + \beta_{{8}} \left( {lnW_{{1}} /W_{{3}} } \right)^{{2}}_{it} + \frac{1}{2}\beta_{{9}} \left( {lnW_{{2}} /W_{{3}} } \right)^{{2}}_{it} + \beta_{{{1}0}} \left( {lnW_{{1}} /W_{{3}} } \right) \, \times \, \left( {lnW_{{2}} /W_{{3}} } \right)_{it} + \beta_{{{11}}} \left( {lnW_{{1}} /W_{{3}} } \right) \, \times \left( {lnQ_{{1}} /\alpha } \right)_{it} + \beta_{{{12}}} \left( {lnW_{{1}} /W_{{3}} } \right) \, \times \, \left( {lnQ_{{2}} /\alpha } \right)_{it} + \beta_{{{13}}} \left( {lnW_{{2}} /W_{{3}} } \right) \, \times \, \left( {lnQ_{{1}} /\alpha } \right)_{it} + \beta_{{{14}}} \left( {lnW_{{2}} /W_{{3}} } \right) \, \times \, \left( {lnQ_{{2}} /\alpha } \right)_{it} + \beta_{{{15}}} {\text{t }} + \beta_{{{16}}} {\text{t}}^{{2}} + \beta_{{{17}}} ln\left( {Q_{{1}} /\alpha } \right)_{it} \times {\text{t}}^{{}} + \beta_{{{18}}} ln\left( {Q_{{2}} /\alpha } \right)_{it} \times {\text{t}} + \frac{1}{2}\beta_{{{17}}} ln\left( {W_{{1}} /W_{{3}} } \right)_{it} \times {\text{t }} + \beta_{{{19}}} ln\left( {W_{{2}} /W_{{3}} } \right)_{it} \times {\text{t }} + v_{it} + u_{it} ,$$

(4)

Table 8 Variables in the bank-efficiency estimation