Journal Issue
Share
Chapter

4. Empirical Analysis of Portfolio Flows

Author(s):
Gavin Gray, and Tamon Asonuma
Published Date:
April 2011
Share
  • ShareShare
Show Summary Details

4.1. Cross-Sectional Analysis

We use a simple cross-sectional approach to provide a first pass at country-specific factors explaining flows to emerging countries during the recent boom. Our sample comprises 20 countries (mainly emerging markets) which are regarded as peer countries for South Africa in terms of GDP and their external openness. All series are in annual frequency in order to focus on current development of portfolio flow into EMs.6

We examine both flows normalized by GDP and each country’s share of total flows using the following specification:

where the dependent variable, Yi, is either the gross bond flow-to-GDP ratio, gross equity flow-to-GDP ratio, bond flow share, or equity flow share. xi is a set of country-specific factors and e is the random disturbance term. We use a weighted least squares regression rather than ordinal least square estimator (OLS) as the weighted regression enables us to capture the high degree of variability of gross flow-to-GDP ratios and flow shares.

The regression on equity flows confirms the importance of country fundamentals. Countries attract greater equity flows to the extent they are growing faster than the rest of the world, or (perhaps unsurprisingly) their economies are larger. Another intuitive result is a positive relationship between market capitalization (entering as a lag to avoid endogeneity problems) and the pace of flows.

We apply a partial weighted least square regression to assess whether equity flows to South Africa have exceeded levels implied by the model. In fact, the flows appear to be in line with expectations, as the results for South Africa do not deviate significantly from the regression line (Figure 1).

Figure 1.Relation of Estimated Residuals on Equity Flow Share with Share in World GDP

Note: Residuals shown are country’s residuals obtained from partial regressions. We apply simple regressions on equity flow share with other explanatory variables except one particular macro variable which we are interested in. Then we regress once again the obtained residuals with respect to the variable such as share in world GDP.

A similar approach shows that the size of the fiscal balance helps explain the bond flows-to-GDP ratio, although it loses significance in some specifications (Table 2). The intuition is that countries with larger deficits may turn to foreign investors for funding. As in the case of equity regressions, large EMs tend to receive a large share of total bond flows.

Table 1.Estimation Results for Equity Flow
Gross equityGross equityGross equityEquity flowEquity flow
flow/GDPflow/GDPflow/GDPshareshare
EstimationWeighted Reg.1Weighted Reg.1Weighted Reg.1Weighted Reg.1Weighted Reg.1
C (constant)−5.06E−2**−5.70E−2**−4.63E−2−36.31***−27.83***
γ1 (Capitalization - lag)1.91E-4*2.29E-42.17E-40.13***0.12***
γ2 (Diff. between country’s and world GDP growth rate)4.84E-3**5.72E-3*4.47E-33.62***2.63**
γ3 (Share in the world GDP)1.34E-2**1.42E-2**1.63E-2**10.51***12.13***
γ4 (Share in the world population)−8.50E-4−5.61E-4−1.10*−0.87
γ5 (Exchange rate volatility - lag)21.70*1.66*1.13735.34**3.11E+2
γ6 (CPI inflation rate - lag)−2.19E-3*−1.61E-3−2.09E-30.41−0.79
γ7 (Indicator for External Vulnerability)31.6312.92
Adj. R-squared0.660.640.620.860.88
Sample1919191919
Root MSE0.0160.0160.0175.224.82
Notes: ***, **, * show significance at 1%, 5%, and 10%. There are 20 countries in the sample: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Equity flow for China is not available.
Table 2.Estimation Results for Bond Flow
Gross bondGross bondGross bondBond flowBond flow
flow/GDPflow/GDPflow/GDPshareshare
EstimationWeighted Reg.1Weighted Reg.1Weighted Reg.1Weighted Reg.1Weighted Reg.1
C (constant)1.07E−26.98E−3−2.3E−2−1.64−2.07
γ1 (Spreads - lag)4.38E-49.64E-41.59E-36.62E-31.49E-2
γ2 (Diff. between country and world GDP growth rate)−3.26E-3−4.44E-3−3.85E-3−0.20−0.19
γ3 (Share in world GDP)−2.84E-32.00E-31.05E-33.84**3.82**
γ4 (Share in world population)−4.78E-3−5.24E-3
γ5 (Exchange rate volatility-lag)23.32**3.31**4.48**2.40E+22.57E+2
γ6 (CPI inflation rate - lag)−9.30E-3−1.01E-2−8.92E-3−0.18−0.16
γ7 (Indicator for External Vulnerability)3−6.21E-2−0.90
γ8 (Fiscal balance/GDP ratio)−3.17E-3***−3.24E-3**−3.74E-3**−7.42E-2−8.14E-2
Adj. R-squared0.610.570.560.390.32
Sample1717171717
Root MSE0.040.040.043.844.04
Notes: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Bond flows for China and Indonesia are not available.

Similar to bond flow share, we apply a partial weighted least square regression to see the level of bond flow that can be explained by its share in world GDP. The South African bond inflow is close to the level that can be justified by its share in world GDP (Figure 2).

Figure 2.Relationship of Estimated Residuals on Bond Flow Share with Its Share in World GDP

4.2. Panel Analysis

To capture the effect of global factors, we supplement the cross-sectional analysis with a panel data approach. A panel approach allows us to take better account of cross-country heterogeneity, while reducing the impact of omitted (missing or unobservable) time-varying variables that might be correlated with the explanatory variables. In this section, we use a sample of 2000:Q1 to 2010:Q3 in quarterly frequency, based on 20 countries, mainly EMs.

We estimate gross bond flow-to-GDP and gross equity flow-to-GDP using the following specification:

and apply the following regression for bond flow share and equity flow share:

where Yit is either gross bond flow-to-GDP ratio or gross equity flow-to-GDP ratio, while Yit is either bond flow share or equity flow share. αi is a country-specific constant which we regard as fixed, and αi captures fixed effects for South Africa. xit is a set of country-specific factors and zt is a set of country-invariant variables that capture global factors. εit and ɛit are random disturbance terms. The third term in equation (3), zt* αSA is included to capture the impact of global factors on South African inflow shares.7 As in the cross-sectional analysis, we also take advantage of weighted least square regressors which enables us to capture a high degree of variability of gross flow-to-GDP ratio and also flow shares.

A weighted regression confirms the importance of both global and country-specific factors (Tables 3 and 4). Two standard explanatory variables enter with the expected signs in explaining the equity flow share, such as the difference between the country’s and the world GDP growth rate, and exchange rate volatility entered as lag. Thus, the higher the growth rate relative to the world GDP growth rate, or the smaller the variation of exchange rate in the previous period, the larger the country’s share of equity flows.

Table 3.Estimation Results for Gross Equity Flow-to-GDP Ratio
Gross equityGross equityGross equityGross equityGross equity
flow/GDPflow/GDPflow/GDPflow/GDPflow/GDP1
EstimationWeighted Reg.2Weighted Reg.2Weighted Reg.2Weighted Reg.2Weighted Reg.
C (Constant)5.45***
γ0 (time trend)4.15E−2***−2.48E−22.26E−2
γ1 (U.S. stock price change)0.18***0.11***0.1***0.15***9.57E-2***
γ2 (VIX)7.50E−2***−0.1***−0.01***−9.15E-2***−8.69E-2***
γ1 (Capitalization - lag)2.58E-2**
γ2 (Diff. between country and world GDP growth rates)0.31***8.61E-25.11E-23.26E-22.40E-2
γ3 (Share in world GDP)−0.67***1.38***0.471.49−0.96***
γ4 (Exchange rate volatility - lag)3−1.9020.4**13.020.9−1.62
γ5 (CPI inflation rate - lag)−4.85E-28.2E-2**8.66E-2***0.21***−0.10***
γ6 (Indicator for External Vulnerability)42.08***3.40***2.85***3.22***−0.41
α1(Fixed effect - South Africa)−0.56−0.4−4.10***−2.48***
Average fixed effect50.45

(2.76)6
0.2

(0.39)7
−2.71

(-1.45)8
Adj. R-squared0.390.660.670.690.16
Sample506506506350506
Root MSE3.772.812.782.943.34
Note: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Equity flow for China is not available.
Table 4.Estimation Results for Equity Flow Share
EquityEquityEquityEquity
Flow shareflow shareflow shareflow Share1
EstimationWeighted Reg.2Weighted Reg.2Weighted Reg.2Weighted Reg.2
γ1 (Capitalization - lag)1.13***
γ2 (Diff. between country and world GDP growth rates)26.2***24.9***−7.29***25.0***
γ3 (Share in world GDP)90 9***20.437.7*20.5
γ4 (Exchange rate volatility - lag)3−884.2−1.81E+3***577.9−1.81E+3***
γ5 (CPI inflation rate - lag)−5.14**−0.62−6.8***−0.66
γ6 (Indicator for External Vulnerability)4−128.9**−93.9−128.9***−92.6
δ1 (impacts of U.S. stock price change on South African inflow)1.15
δ2 (impacts of VIX on South African inflow)0.62
α1(Fixed effect — South Africa)77.3−40.855.1
Average of fixed effect530.616.129.1
(312.9)6(44.3)7(312.7)6
Median of fixed effect18.719.018.4
Adj. R-squared0.400.550.580.55
Sample533533372533
Root MSE249.9216.092.3216.4
Notes: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Equity flow for China is not available.

The fixed effect for South Africa, which is remarkably higher than the average of fixed effects of emerging-market countries, reflects some capital market factors specific to South Africa, such as size of capital market (Figure 3).

Figure 3.Fixed Effect and Capitalization Ratio (Percent of GDP)

It plays a dominant role for equity flow for the entire period (2000–10), subsample period (2003–10), and also upswing period (2003–08). The difference between fixed effect of South Africa and the average of fixed effect of emerging-market countries explains 30 percent of the equity flow into South Africa throughout the last four quarters, equivalent to 0.5 percent of GDP in South Africa. It can be interpreted as indicating that South Africa has been clearly receiving larger equity flow than can be simply justified by macro fundamentals mostly due to South African-specific capital market structure.

Explanatory factors for both bond flows (normalized by GDP) and bond flow shares include the fiscal deficit, the difference between the country’s and global GDP growth rates. If the fiscal deficit of an emerging-market country deteriorates, the country tends to issue more bonds, which in turn absorbs more inflows through bond issuance. The higher the growth rate of the country relative to that of the world, the more inflows it receives. In tandem with these two factors, the bond flow share also tends to be negatively affected by inflation and measures of external vulnerabilities.

Some financial market elements specific to South Africa captured by a fixed effect, most probably liquidity of bond market, contributed to attracting more bond flows to other EMs for the most recent seven years (2003—10), particularly during the upswing period (2003—08) before the global financial crisis. By contrast, during the crisis period (2008—09) and subsequent recovery (2009—10), these impacts on inflow have been somewhat smaller compared with other EMs.

Table 5.Estimation Results for Equity Flow Share Under Different Sample Periods1
(1) Whole(2) Subsample(3) Upswing(4) Downswing(5) Most recent
periodsperiodperiodperiodperiod
2000:Q1–2003:Q1–2003:Q1–2008:Q3–2009:Q4–
Periods2010:Q32010:Q32008:Q22009:Q32010:Q3
γ1 (Capitalization - lag)
γ2 (Diff. between country and world GDP growth rates)24.9***−3.36−9.01*−2.70*−2.0*
γ3 (Share in world GDP)20.451.7***170.3***−12.0−17.7
γ4 (Exchange rate volatility – lag)2−1.81E+3***317.22. 29E+3***−6.06E+2**−43.8
γ5 (CPI inflation rate -lag)−0.62−3.364.494.56*2.2
γ6 (Indicator for External Vulnerability)3−93.9−188.6***−132.9**47.9*−23.0
δ1 (impacts of U.S. stock price change on South African inflow)
δ2 (impacts of VIX on South African inflow)
α1(Fixed effect —South Africa)77.3111.2***−41.0−29.520.1
Average fixed effect430.666.7−98.65.230.1
(312.9)5(90.3)6(−256.1)7(48.1)8(76.5)9
Median of fixed effect18.764.3−67.63.626.2
Adj. R-squared0.590.550.650.770.95
Sample5334132917349
Root MSE207.3100.598.020.57.4
Notes: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Equity flow for China is not available.
Table 6.Estimation Results for Gross Bond Flow/GDP
Gross bondGross bondGross bondGross bond
flow/GDPflow/GDPflow/GDPflow/GDP1
EstimationWeighted Reg.2Weighted Reg.2Weighted Reg.2Weighted Reg.2
C (constant)1.93
γ0 (time trend)1.69E-22.93E−2***
γ1 (U.S. Treasury Bill rate - nominal)0.10**−0.31*−0.20−0.27
γ2 (VIX)−8.38E−3−4.10E−2***−3.95E-2***−4.27E−2***
γ1 (Spreads - lag)−3.18E-20.13−2.38E-3−0.23***
γ2 (Diff. between country and world GDP growth rate)0.12***5.45E-2**4.68E-2*6.21E-2**
γ3 (Share in world GDP)0.190.230.360.37**
γ4 (Exchange rate volatility - lag)33.78−7.00−7.52−0.35
γ5 (CPI inflation rate - lag)−5.62E-2***−2.19E-2−2.31E-2−5.03E-2***
γ6 (Indicator for External Vulnerability)4−4.11E-20.460.530.64
γ7 (Fiscal balance/GDP ratio)−2.88E-2***−3.66E-2***−3.61E-2***−2.37E-2***
α1(Fixed effect — South Africa)1.811.29−3.89E-2
Average fixed effect2.17

(2.79)5
1.53

(1.96)6
Adj. R-squared0.100.210.210.08
Sample680680680680
Root MSE1.871.761.761.84
Notes: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Bond flows for China and Indonesia are not available.
Table 7.Estimation Results for Bond Flow Share
Bond flow shareBond flow shareBond flow share1
EstimationWeighted Reg.2Weighted Reg.2Weighted Reg.2
γ1 (Spreads - lag)−1.21E+3***−1.79E+3***−1.84E+3
γ2 (Diff. between country and world GDP growth rate)522.4***298.1***303.0***
γ3 (Share in world GDP)1.79E+3***−1.26E+3−1.17E+3
γ4 (Exchange rate volatility - lag)31.87E+5***1.23E+6***1.25E+5
γ5 (CPI inflation rate - lag)−375***−337.4***−338.2***
γ6 (Indicator for External Vulnerability)4−1.47E+3*−8.54E+3***−8.86E+3***
γ7 (Fiscal balance/GDP ratio)−118.5***−90.4***−87.55***
δ1 (impacts of U.S. stock price change on South African inflow)84.0
δ2 (impacts of VIX on South African inflow)19.0
α1(Fixed effect — South Afric)4.72E+3***5.40E+3
Average fixed effect6.50E+36.67E+3
(7.51E+3)5(7.23E+3)6
Median of fixed effect5.48E+35.23E+3
Adj. R-squared0.920.970.97
Sample447447447
Root MSE252816951693.5
Note: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Bond flows for China and Indonesia are not available.

We again use partial regressions to analyze whether South Africa has been receiving proportional bond flow in terms of two “pull factors” such as the difference between the country’s and world GDP growth rates, and fiscal balance-to-GDP ratio. Figure 4 shows that South Africa has received a higher share of bond flows than can be explained by the two main pull factors.

Figure 4.Relations of Estimated Residuals on Bond Flow Share with “Pull” Factors

Note: Residuals shown in the charts are country’s residuals obtained from partial regressions. We apply simple regressions on bond flow share with other explanatory variables except one particular macro variable which we are interested in. Then we regress once again the residuals obtained with respect to the variable such as difference between the country and the world GDP growth rates.

Table 8.Estimation Results for Bond Flow during Different Periods1
(1) Whole(2) Subsample(3) Upswing(4) Downswing(5) Most recent
periodsperiodperiodperiodperiod
Periods2000:Q1-2003:Q1-2003:Q1-2008:Q3-2009:Q4-
2010:Q32010:Q32008:Q22009:Q32010:Q3
γ1 (Spreads - lag)−1.79E+3***−97.8**−167.9**14.88.92
γ2 (Diff. between country and world GDP growth rate)298.1***−4.7614.510.5***−0.29
γ3 (Share in world GDP)−1.26E+3344.3***659.5***−242.1**5.69
γ4 (Exchange rate volatility- lag)21.23E+6***−5.95E+3***−5.99E+3***692.4*82.0
γ5 (CPI inflation rate - lag)−338.2***3.69−1.632.550.29
γ6 (Indicator for External Vulnerability)3−8.54E+3***−151.3−8.88−68.510.2
γ7 (Fiscal balance/GDP ratio)−90.4***−4.53**−10.3***1.720.07
δ1 (impacts of U.S. stock price change on South African inflow)
δ2 (impacts of VIX on South African inflow)
α1(Fixed effect — South Africa)4722***424***426.4*40.020.0
Average fixed effect4649922387.9156
(7508)5(244)6(32.7)7(327)818.7
Median of fixed effect5481244−67.692.216.9
Adj. R-squared0.970.620.650.910.92
Sample4473692476557
Root MSE1695245.0254.323.93.8
Note: ***, **, * show significance at 1%, 5%, and 10%. Our sample includes 20 countries: Argentina, Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Israel, Korea, Malaysia, Mexico, Peru, Poland, the Philippines, Russia, South Africa, Thailand, Turkey, and Ukraine. Bond flows for China and Indonesia are not available.

More surprisingly, during 2009:Q4–2010:Q3, its bond shares have not only been higher than values over the entire sample period (2000:Q1—2010:Q3), but deviated significantly from the amounts explained by macroeconomic fundamentals (regressed line), showing that bond flows during the recent periods have been extraordinarily high.8

    Other Resources Citing This Publication