Savings-Investment Dynamics and Time-Varying Capital Mobility in Emerging Markets in India, Brazil, Indonesia and South Africa between the years 1990-2024.

Authors: Manvi Marolla, Renee Pabari, Diya Nagpal

Abstract

This study examines the relationship between savings, investment, and capital mobility in India, Brazil, Indonesia, and South Africa within the Feldstein-Horioka framework during 1990-2024. The paper investigates whether financial globalisation has reduced the dependence of investment on domestic savings and whether this relationship has remained stable over time.

Using both linear and nonlinear econometric techniques, including FMOLS, DOLS, PMG-ARDL, Markov Switching Models, structural break analysis, threshold regression, and Kalman filtering, the study finds that the average long-run Feldstein-Horioka coefficient remains around 0.61, indicating moderate but incomplete capital mobility. The results further show that capital mobility changes across crises, policy shifts, and levels of trade openness. Major global shocks, including the Asian Financial Crisis, the Global Financial Crisis, and COVID-19, temporarily increased dependence on domestic savings.

Overall, the findings suggest that capital mobility in emerging economies remains dynamic, uneven, and institutionally constrained rather than fully determined by financial liberalisation alone.

Introduction

The relationship between savings, investment, and international capital mobility remains one of the most important debates in international macroeconomics. According to economic theory, financial globalisation should reduce the dependence of domestic investment on domestic savings by allowing countries to access foreign capital. However, Martin Feldstein and Charles Horioka (1980) found that savings and investment remained highly correlated even in relatively open economies, creating what became known as the Feldstein-Horioka puzzle.

This issue is particularly important for emerging economies such as India, Brazil, Indonesia, and South Africa, which have experienced liberalisation, financial reforms, and repeated economic shocks since the 1990s. Capital mobility in these economies may therefore be unstable and sensitive to crises, policy changes, and structural conditions.

This study examines whether financial globalisation has weakened the savings-investment relationship in these four emerging economies between 1990 and 2024. By combining long-run and time-varying econometric approaches, the paper finds that although capital mobility has increased over time, domestic savings continue to play a significant role in financing investment, especially during periods of economic instability.

Theoretical Framework and Literature Review:

The Savings-Investments Relationship

The national income identity: S-I = CA, explains the relationship between Savings (S), Investment (I) and Current Account (CA). In the case of perfect capital mobility, the investment mainly depends on external capital, and its correlation with savings weakens. Similarly, in a closed economy, the investment is closely linked with domestic savings.

The Feldstein and Horioka model:

(I/GDP)_it = α + β (S/GDP)_it

Where β is a coefficient that represents the level of dependence of investment on savings. If β=1, there is a strong correlation between both variables, leading to low capital mobility. And β=0 signifies high mobility. Feldstein and Horioka (1980) estimated β = 0.89 in 16 OECD countries, which led to decades of research.

Nonlinear and Time-Varying Capital Mobility

The linear models require the presence of a stable time-invariant β, and this is not applicable in the emerging markets where the regime keeps changing. The 1997 shock in India, the 1999 and 2002 crises in Brazil, the 1997 opening in Indonesia, and the 1994 opening in South Africa are all discrete transitions in the regimes that cannot be explained by the linear regressions.

The Markov Switching Model (MSM), which was derived by Hamilton (1989), formalises regime changing by permitting the savings-investment relation to be governed by two or more latent states, which change each other in a probabilistic manner. This is specifically appropriate to the four countries discussed. The discrete changes in the S-I relationship cannot be described by a single linear regression.

Threshold regression is also part of non-linear analysis; it examines whether a particular variable (trade openness) affects the intensity of the savings-investment nexus at a statistically significant breakpoint.

Literature Review:

The relationship between savings and investment has remained a major topic in international macroeconomics since the work of Martin Feldstein and Charles Horioka (1980), who found a strong correlation between domestic savings and investment in OECD economies. According to standard economic theory, countries with open financial markets should be able to finance investment through international capital flows, resulting in a weak savings-investment relationship. However, the high correlation observed by Feldstein and Horioka became known as the Feldstein-Horioka puzzle and generated extensive debate regarding the true extent of international capital mobility.

Several explanations have been proposed for the persistence of the puzzle. Some researchers argue that home bias, institutional rigidities, and policy restrictions continue to limit cross-border capital flows. Others suggest that macroeconomic factors such as current account targeting, demographic structure, and imperfect financial markets maintain a strong domestic savings-investment relationship even under globalisation.

In emerging economies, the savings-investment relationship is often influenced by financial liberalisation, external shocks, and structural reforms. Studies on countries such as India, Brazil, Indonesia, and South Africa suggest that capital mobility has increased over time, although dependence on domestic savings remains significant. Research also shows that financial crises and policy transitions can temporarily reverse financial integration and strengthen the savings-investment correlation.

Recent literature increasingly uses advanced panel econometric techniques such as FMOLS, DOLS, PMG-ARDL, and Common Correlated Effects estimators to account for endogeneity, cross-sectional dependence, and long-run dynamics. At the same time, nonlinear approaches such as Markov Switching Models and time-varying parameter models have become important in understanding how capital mobility changes across crises, reforms, and different economic regimes.

Building on this literature, the present study examines whether capital mobility in emerging economies remains stable over time and how structural factors such as trade openness and global shocks influence the savings-investment relationship between 1990 and 2024.

Methodology:

Data

Research design

This study uses a quantitative methods research design to examine the long-run savings investment relationship in emerging markets (India, Brazil, Indonesia, South Africa) across the time period of 1990-2024.

Data

The sample consisted of four countries- India, Brazil, Indonesia and South Africa. Participant countries were selected using a purposive sampling method, as they best depict the characteristics of emerging markets.

Data was collected from the World Development Indicators website. We measured the following key variables:

Investment	Gross Capital Formation (% of GDP)
Savings	Gross Domestic Savings (% of GDP)
Current Account Balance	Current Account Balance (% of GDP)
Trade	Trade Openness (% of GDP)
GDP Growth	Annual GDP Growth Rate (%)

The dataset is structured as a balanced panel dataset, where each country is observed across the entire time period of 35 years (1990-2024). Panel data analysis allows the study to capture both cross-country variation and time series dynamics in macroeconomic indicators.

Data Summary

Average Savings and Investment Ratios (1990–2024)

Country	Avg Savings (%GDP)	Avg Investment (%GDP)	S – I Gap
India	28.38	31.73	3.35
Brazil	18.43	18.45	0.02
Indonesia	31.01	29.69	1.32
South Africa	18.19	16.69	15.0

Data Analysis

To examine the relationship between savings and investment, correlation analysis was applied. Descriptive statistics such as mean, standard deviation, and frequency distributions were first calculated to summarise the data. Subsequently, econometrics was used to evaluate the research hypotheses.
All analyses were conducted using R, Python and MS Excel, including statistical computations and visualisation.

Econometric Approach

The estimation plan follows three stages:

The estimation of a baseline panel with Fixed Effects (FE) and Random Effects (RE) models, with the Hausman test to select a model
Long-term estimation with FMOLS and DOLS models to determine endogeneity and serial correlation. PMG-ARDL to estimate the heterogeneous behaviour of the short-run and CCE to adjust exogenous factors
Nonlinear and time-varying estimation of time-varying coefficient with Markov Switching Models, Bai-Perron structural break.
Kalman filter time-varying parameter models permit the evolution of β recursively, giving a more realistic view of the financial integration dynamics.
Pesaran CD test of cross-sectional dependence and CIPS unit root tests are pre-estimation diagnostics tests. The findings established that there is restricted cross-sectional dependence and some stationarity (savings and trade are I (1)) to support cointegration-based estimation of the long run.

Results:

Baseline Panel Regressions: FE and RE Models

The Fixed Effects and Random Effects model considers that GDP growth can be predicted by Savings, Investment, Current Account and Trade. Results are:

Variable	FE Coef.	FE p-value	RE Coef.	RE p-value
savings	0.3689	0.0015 **	0.3424	0.0015 **
investment	-0.1141	0.3243	-0.0481	0.6273
Current account	-0.3655	0.0103 *	-0.3146	0.0146 *
trade	-0.0321	0.1755	-0.0347	0.0990

Significance: ** p < 0.01, * p < 0.05, p < 0.10

The table explains that:

Savings: It has a strong effect (FE-0.36, RE-0.34) on GDP growth. High savings lead to higher growth of a country.
Investment: It doesn’t show any clear direct effect on GDP; this indicates that investment is linked with savings.
Current Account: It shows a significant negative effect on GDP. Current account balance, in case of a deficit, might decrease GDP growth.
Trade: It shows a weak direct growth effect on GDP. Although its importance as a structural modulator of the savings-investment relationship can be seen in the nonlinear analysis

In both models, investment displays an insignificant effect. It might indicate that, as savings are already explaining GDP, with moderate capital mobility, the investment becomes endogenous to savings. So, it is not an independent variable.

As it is seen that there is no systematic difference between FE and RE, with the help of the Hausman test, the RE model is preferred for its higher efficiency.

The panel regression results identify savings and current account balance as the primary statistically significant drivers of GDP growth in this sample. Investment and trade openness do not emerge as robust predictors in this specification, though investment warrants further investigation given the divergence in point estimates across the two models.

Long-run estimation methods: FMOLS, DOLS, PMG-ARDL and CCE

The results of all four long-run estimators are very consistent, which proves the strength of the savings investment relationship:

Table: FMOLS Country-Level Coefficients (p = 0.000 for all)

Country	Coefficient	t-statistic
Brazil	0.624	4.36
India	0.891	5.69
Indonesia	0.532	3.96
South Africa	0.479	3.72

Table: Comparison of Long-Run Estimators

Estimator	Long-Run Coefficient
FMOLS	0.632
DOLS	0.618
PMG-ARDL	0.598
CCE	0.608
Differenced OLS	0.422

FMOLS (Fully Modified OLS): It corrects for endogeneity and serial correlation in cointegrated panel models. In the long run, β is still 0.6. Even after adjusting for statistical problems, savings remain a strong driver of investment.
DOLS (Dynamic OLS): It incorporates leads and lags of differenced regressors to address endogeneity issues. β ≈ 0.6 in the long run is obtained after adding dynamics. Savings remain an important long-run determinant of investment.
PMG-ARDL (Pooled Mean Group Estimator): The PMG estimator allows short-run dynamics to differ across countries while constraining long-run coefficients to be identical.

Long-run: β ≈ 0.61

Short-run: β ≈ 0.42

Error correlation term: -0.917 (about 20% adjustment per year)

In the short run, countries can deviate by using foreign capital, but in the long run, savings become crucial. Approximately 20 per cent per year PMG error-correction speed suggests that it takes about five years to complete the adjustment process after a savings investment disequilibrium, a time scale that is in agreement with the multi-year capital flow cycles in the MSM and structural break analysis.

CCE (Common Correlated Effects): To control for unobserved common global shocks, the CCE estimator includes cross-sectional averages of the variables. It gives β ≈ 0.608, similar to FMOLS and DOLS. So, the result is not because of global factors but due to the genuine domestic structure between S and I.

The mean long-run coefficient among the estimators shows that there is moderate capital mobility, and this is β ≈ 0.61. Coefficients in the short-run (pooled β ≈ 0.422) are less yet significant.

Capital Mobility Interpretation

Country-Level Findings

Country	β Range	Interpretation
Brazil	0.59 – 0.62	Moderate mobility
India	0.84 – 0.89	Low mobility
Indonesia	0.50 – 0.53	Moderate mobility
South Africa	0.46 – 0.48	Moderate mobility

The coefficient of 0.84-0.89 in India implies that the capital mobility is very low and the domestic investment strongly relies on the local savings. Brazil, Indonesia and South Africa exhibit moderate mobility (β = 0.46-0.62) and are characterised by partial financial integration, where local and foreign capital have a role to play in investment.

Capital Mobility Index Table

Country	β	CMI	Interpretation
India	1.12	-0.12	Very Low Capital Mobility
Brazil	1.01	-0.01	Very Low Capital Mobility
Indonesia	0.96	0.04	Low Capital Mobility
South Africa	0.92	0.08	Low Capital Mobility

Negative CMI indicates very high dependence of investment on domestic savings. India and Brazil showcase this behaviour. South Africa and Indonesia have relatively higher CMI; the values are small, indicating low capital mobility and considerable dependence of investment on domestic savings.

Negative CMI values are unusual in theory; they are possible during empirical analysis because they consider the possibility of country-specific rigidities and constraints. They reflect on adjustment, domestic bias, or econometric artefacts.

Analysis of Time-Varying and Nonlinear

Instead of assuming the relationship is always the same, this approach allows it to change over time or under different conditions. Because in reality, economies are not constant.

Markov Switching Model

This model identifies different “states” in the data, like periods of high capital mobility and low capital mobility. It also tells us the chances of being in each state at any time.

In this study, MSM identifies two different regimes:

Regime 1: Low capital mobility and high β, caused by periods of financial restriction, home bias and capital reversal due to crisis.

Regime 2: High capital mobility and low β, seen in the periods of more open markets and foreign capital integration.

MSM Smoothed Regime Probabilities

It shows the probability of being in a particular state (high or low mobility) for each time period. It helps us clearly see when shifts happened.

Persistence of Regimes

The plots of probability indicate long intervals of dominance of one regime:

One regime in India and Brazil has close probabilities of trend and extends through long durations, showing strong persistence
This implies that the economy is anchored on structural factors, including policy regimes and institutional environments.

Sharp Transitions

The switches are, however, also abrupt in the graphs:

The regime transformation is characterised by sharp probabilities in the case of economic disruption in South Africa.
A dramatic shift in Indonesia occurred around the late 1990s when the state faced a serious economic crisis.

These changes affirm that regime changes are not gradual, but they are brought about by structural breaks. On the whole, the results reveal that capital mobility is dynamic and varies across regimes and give a more nuanced perspective on international financial integration.

Structural Break Analysis (Bai–Perron)

Country	Detected Break Years
Brazil	2014, 2021
India	1992, 2021
Indonesia	1997, 1999
South Africa	1993, 2009

India: 1992, 2021

In 1991, India saw new economic reforms, i.e., Liberalisation, Globalisation and Privatisation. It led to a more open market economy, which resulted in a rapid rise in GDP and increased capital mobility in the following years.
The year 2021 signifies the post-COVID-19 recovery phase, where digitalisation and policy played a huge role in stabilising the economy by giving easy access to investors, setting low borrowing rates and making the process completely digital.

Brazil: 2014, 2021

In 2014, Brazil experienced a deep recession and economic crisis, which led to a major fall in GDP.
Similar to India, 2021 marked a post-COVID recovery phase for Brazil.

Indonesia: 1997, 1999

1997 was the start of the Asian Financial Crisis. Indonesia was one of the countries that experienced a huge economic shock during the crisis.
1991 shows the recovery period of the country with the assistance of the IMF, structural reforms and political transformation.

South Africa: 1993, 2009

1993 marked a transition period for South Africa with the end of Apartheid and economic reforms.
The country was deeply affected by the Global Financial Crisis, leading to a recession because of its financial links with the world. The crisis led to a decline in exports of raw material, a fall in Foreign Direct Investment, unemployment and financial instability.

We can observe a cross-country pattern of structural breaks. Major breaks happened in Indonesia and South Africa due to external shocks (the Asian Financial Crisis and the Global Financial Crisis, respectively) as they possess relatively open capital accounts and high amounts of external exposure. India and Brazil, on the other hand, saw their main domestic reform-induced break (1992 and the recession after 2014) as an endogenous policy phenomenon, but not exogenous shocks.

Time-Varying Coefficient (Kalman Filter)

The Kalman filter shows a long-run downward trend in β throughout the sample, which is in line with gradual financial integration. The reversal of the capital flows in 2010-2011 after the GFC is visible through a sharp peak in the graph. The Capital Mobility Index (CMI = β-1) that changes with time confirms that all the countries are clustered near very low or low mobility, when full sampled average estimates of β are used.

Kalman filters are more volatile than smoothed estimates since they respond to every new data point without considering future information. The trend shows a gradual decline in β after 1990, indicating an increase in capital mobility. The Global Financial Crisis caused a temporary reversal of foreign capital during 2008-2011, leading to increased dependence of investments on domestic savings. Post 2012, the β value slowly declined, signalling the return of capital and recovery of capital mobility. The global capital mobility faced a small disruption during COVID-19 in 2020, causing short-term savings dependence. The cycle has been stabilising after 2020, as β became more stable. Across the countries, India and Brazil show sudden spikes in β during crisis, as they show higher dependence on foreign portfolio investments and short-term capital. While Indonesia and South Africa experience more gradual but long-lasting effects due to their commodity-export dependence.

Country	β	CMI	Interpretation
India	1.12	-0.12	Very Low Capital Mobility
Brazil	1.01	-0.01	Very Low Capital Mobility
Indonesia	0.96	0.04	Low Capital Mobility
South Africa	0.92	0.08	Low Capital Mobility

Threshold Regression: Openness to trade

This regression identifies the threshold value of trade openness to be at 57.43 % of GDP. Further, it divides the data into two regimes.

First is low trade openness (<57%), where β = 1.071 (very high) and R² = 0.844 (very strong fit). Here, investment is almost completely dependent on domestic savings, leading to very low capital mobility. Example: Closed economies that need to fund their whole investment with savings.

The second regime is high trade openness (>57%). Here β = 0.58 (low) and R²= 0.632 (weak fit). In this case, investment is less dependent on savings, and capital mobility is high. Example: Open economies have easier access to foreign funds, so they depend less on domestic capital.

If the country’s average trade ratio is close to the threshold value, like that of Indonesia (≈ 52%), then it shows moderate capital mobility and is in the transition phase. India (≈ 36%) and Brazil (≈ 25%) are below the threshold value, meaning a strong investment-savings link and low capital mobility. South Africa (≈ 50% on average, with a maximum value of 65.97) shows fluctuations. Sometimes it is close to the threshold, and sometimes greater than the threshold. It indicates more integration with the international market and higher capital mobility.

R² = 0.844 suggest low openness, and R² = 0.632 implies high openness of markets. Higher trade leads to more global integration, fuelling investments with international capital flows.

Metric	Regime 1 (Low Trade)	Regime 2 (High Trade)
Observations	126	14
Savings Coefficient	1.071	0.583
R-squared	0.844	0.632
Capital Mobility	Low	High

Discussion:

This study discusses the changing trend of capital mobility in reference to Feldstein-Horioka’s coefficient, to find the effect of globalisation in emerging economies. The evidence suggests an increase in mobility over time; however, it is not stable in nature. The β is estimated around 0.61 in the long run, which implies a weakening relationship between savings and investment, but this average value ignores regime dependence and time variation.

Considering the declining β from Kalman Filter results, globalisation has been driving capital mobility across countries. This trend is sometimes disturbed by crises, where we notice that correlation strengthens and investment turns towards domestic savings. This instability is clearly depicted in the form of low and high capital mobility regimes in the Markov Switching Model.

Additionally, threshold regression analysis suggests that capital mobility is dependent on structural factors like trade openness. Only when economies cross a certain threshold value does their investment-savings relationship weaken. Through this, we can be sure that the effect of globalisation alone is not efficient because other things, like policies and institutional support, come into play.

The findings also contribute to broader debates on financial globalisation. The results suggest that financial integration in emerging economies is not automatic or stable even after liberalisation. Instead, capital mobility remains sensitive to crises, institutional quality, and structural conditions, indicating that globalisation alone is insufficient to produce fully integrated capital markets.

In the end, findings suggest that capital mobility is not just a direct product of liberalisation, but it is understood better when we study it with other constraints like policy, institutions and structural shocks. The persisting nature of the Feldstein-Horioka puzzle does not mean financial integration is failing, but it actually allows us to believe that emerging economies behave in a careful and inconsistent way to protect themselves from huge external disturbances that could disrupt their growing economy.

Conclusion:

This paper examines whether capital mobility has increased with globalisation and if it has been stable over time in emerging economies like India, Brazil, Indonesia and South Africa. The results provide a clear answer that it has been increasing, but not in a stable way. The Feldstein-Horioka Puzzle still persists, limiting financial integration. It can be derived from multiple tests that capital mobility is highly sensitive to macroeconomic shocks and policies. This implies that stability also depends on strong domestic institutions, an efficient macroeconomic framework and resilience to external shocks.

Policymakers should prioritise strengthening domestic financial systems to handle the robust nature of international capital mobility and open market operations. The countries see good growth with more open markets. So maintaining an effective monetary and fiscal framework is essential to build investor confidence in the long run, and across regimes. Policies that support trade integration are important as they play a key role in sustaining mobility.

Finally, findings support greater coordination among economies and aligning their policies to achieve the best results, and also avoid global shocks in emerging economies. Sustainable capital mobility is a policy driven process that requires constant management, functional institutions and time-to-time investigation. So, it can be concluded that capital mobility is not a target achieved through liberalisation; it is a dynamic concept that requires continuous governance and evaluation.

References

Acknowledgment

We would like to express our sincere gratitude to International Institute of SDGs and Public Policy Research for providing guidance and academic support throughout this research work. We also extend our heartfelt appreciation to Ishu Verma, Shruti Bhamare, and Shashi Kulkarni for their valuable guidance, support, and contribution during different stages of the research process. Their insights and encouragement were important in the successful completion of this study.

Author

Shoo Phar Dhie

Kang Bakso