I. Introduction
2. Empirical studies on the relationship between saving and economic growth
2.1. Time-series and cross-country studies
t t
t S CA
I ≡ −
(5.1)
where, It and St are domestic investment and saving, respectively. CAt is the current account balance. They use annual data for 1970-2000. Data on saving and growth are from World Development Indicators and data on FDI inflows to Mexico are from the International Financial Statistics. They use the following vector autoregressive model
(VAR).
t S t S t
t t x x u
x =µ+β +Φ1 −1+...+Φ − + t = 1,…….., T (5.2) Where xt =
(
St′,Yt′,FDIt′)
, and Φ(i=1,....,s) are all matrices of coefficients and u is the error term. S is gross domestic saving. Y and FDI stand for GDP and FDI inflows,respectively. The Toda and Yamamoto test does not need the pre-testing for unit roots.
However, they test for unit roots using the ADF and PP tests. The unit root tests show that S, Y and FDI are integrated of order one.
They find that saving and GDP are cointegrated for Mexico. The results show that there is strong evidence of a causality running from domestic saving to GDP. There is also the reverse causality. The causality runs from FDI to GDP. The multivariate causality tests find that saving and FDI jointly Granger cause GDP. There is a one-way causality running
from saving to foreign direct investment. They argue that the uni-directional causality shows the increasing confidence of foreign investors when domestic saving is rising. They find that if foreign direct investment variable is excluded, results are affected. Therefore,
the omitted FDI variable may give misleading results of the true causality between saving
and growth.
Sinha and Sinha (1998) study the causal relationship between saving rate and economic growth for Mexico using annual data from 1960 to 1996. Data on saving are from Ortiz (1997). Real GDP data are from International Monetary Fund. The variables are public saving (PUBSAV), private saving (PVSAV) and GDP. All variables are transformed into logarithmic forms and growth rates are calculated by using the first difference of the logarithmic variables. ADF and multivariate cointegration tests are used to test for unit roots and cointegration, respectively. They use multivariate Granger causality tests and Johansen framework for cointegration tests. The vector error correction model is given by
n t
e w z
z t
a
y t i y t t
p
i y t iy y
oy
t , 1,2,...,
1
1 1
1 − + Γ ∆ +Ψ + =
+
=
∆ − −
−
∑
−∏
α (5.3)
where zt =(yt′,xt′)′,yt is an my x 1 vector of endogenous variables I(1) and wt is q x 1 vector of exogenous/deterministic variables I(0) variables.
The logarithms of PVSAV and GDP are found to be stationary in their first difference. PUBSAV is stationary in its level and it is excluded from the cointegration tests.
The results of cointegration tests find that there is a long-term relationship between private saving and economic growth in Mexico. Causality tests show that there is a causality running from growth of GDP to both private saving and public saving. There is no reverse causality running from either PVSAV or PUBSAV to growth of GDP in Mexico.
Alguacil et al. (2003) examine the relationship between saving and economic growth for Spain. The Solow growth model and the endogenous growth model find that higher saving precedes and causes economic growth. In contrast, the model of consumption with habit formation9 predicts that income growth can lead to a higher saving rate at least in the short-run.
Gross domestic saving is calculated by subtracting final consumption expenditure of both household and government consumption from GDP. Total saving is defined as private saving plus government saving. Data on GDP, saving, and GDP deflator are from OECD national accounts and World Development Indicators. Inward foreign direct
investment data are from the Bank of Spain. Annual data are for 1970-1999.
Their vector autoregressive (VAR) model is as follows.
t s t s t
i t X X u
X =α +β +φ1 −1 +...+φ − + t= 1,……,T (5.4)
In the above model, Xt is equalXt =(Sit′,Yt′,FDIt′)and,Si =ST,SH.t represents a deterministic time trend.φj(j=1,...,s)is a matrix of coefficients. S and Y are gross domestic saving and GDP, respectively. FDI stands for foreign direct investment. White noise error term is represented by u. The lag length is decided by using the Akaike
information criterion (AIC) and the Hanna-Quinn criterion (HQC). They use Toda and Yamamoto (1995) Granger non-causality tests.
9 The model of habit formation predicts that the individual’s consumption is not affected immediately by an unexpected income growth.
They find that there is evidence of a causality running from domestic saving to GDP. But, there is no reverse causality. FDI Granger causes GDP. But, there is no reverser causality. The results of the Multivariate Granger causality tests show that there is a Granger causality running from FDI and S to GDP. Thus, the results show the importance
of FDI to stimulate economic growth for Spain.
Mohan (2006) examines the relationship between saving and economic growth.
He studies how the relationship affects economies with different income levels. He tries to determine whether the level of GDP has an effect on the direction of causality between saving and economic growth. He uses data for 25 countries. The variables are gross domestic saving (GDS) and GDP. GDS is defined as GDP minus government consumption and private consumption. The logarithms of the two variables are taken. He uses first differences of the logarithms of GDS and GDP to test for Granger causality. Thus, growth rate of the variables are used.
Annual data from 1960 to 2001 are from the World Development Indicators. The
sample is divided into four subsets. The subsets are low-income, low-middle-income, upper-middle-income, and high-income countries. Each subset has five countries. ADF and the Johansen cointegration tests are used for unit root and cointegration tests, respectively.
If the variables are found to be cointegrated, he uses the vector error correction (VEC) model to test for causality. Otherwise, vector auto-regression (VAR) model is used to test
for causality.
Results of the ADF unit root tests indicate that logarithms of GDS and GDP are integrated of order one for 22 countries. For Egypt, Malaysia, and the USA, at least one of the variables is stationary in its level and thus, these countries are excluded from the Johansen cointegration tests. He finds that growth rate of GDP Granger causes growth rate of saving for 13 countries. There is a reverse causality for 5 countries among the 13 countries. Further, there is a unidirectional causality, which is from the growth rate of saving to the growth rate of GDP, for Indonesia and Singapore.
He finds that the level of GDP affects the direction of causality between saving and economic growth. For the high-income countries, the causality runs from the growth rate of saving to the GDP growth rate. For the upper-middle-income countries, there is a bi-directional causality. The causality runs from the growth rate of GDP to the growth rate of saving for the low-middle-income countries. Finally, he finds mixed results for the direction of causality for low-income countries.
Andersson (1999) test for the causality between saving and growth in the long-run and in the short-run. He uses annual data for real GDP and real gross saving for three countries. Data are for 1950-1997, 1952-1996, and 1950-1996 for USA, UK, and Sweden, respectively. These three countries are chosen because of two reasons. First, there are dissimilar trends of the saving rates for these countries after the Second World War.
Second, Andersson wants to compare the results of large and small open economies.
Gross saving is equal to fixed capital formation plus net exports. Data for the US are from the NIPA10 table. Data for UK and Sweden are from the OECD national accounts and Swedish Central Bank, respectively. The logarithms of the variables are used.
Thus, the first difference of the variables gives the growth rate. The results show that for the US, there is no cointegration between saving rate and GDP growth. Thus, the long-run Granger-causality is not performed for the US. For the UK and Sweden, the variables are cointegrated. Also, there is evidence of long-run Granger-causality between saving and GDP. The long-run causality for UK runs from GDP growth to saving rate. Also, there is the reverse causality. For Sweden, the causality runs from GDP growth to saving rate.
There is no evidence of the reverse causality.
The short-run causality between saving rate and GDP growth runs in both directions for the US and the UK, even though no long-run causality is found for the US.
For Sweden there is no statistically significant short-run causality in either direction between saving rate and GDP growth.
10 National Income and Product Accounts Tables (NIPA) are published by Bureau of Economic Analysis. It is an agency of the U.S. Department of Commerce.