Abstract
This study examines the relationship between domestic saving and investment and measures the international mobility of capital in the United States. The long-run model, “with” and “without” structural breaks, is estimated using several single-equation and system estimators to assess the robustness of results and take an exhaustive account of the methodological and measurement issues. The results provide dominant support for the long-run relationship between domestic saving and investment. The estimates of the slope parameter on saving above zero and the dominant support for cointegration between saving and investment across estimators vindicate the validity of intertemporal budget constraint and suggest the sustainability of current account deficits. The numerical magnitude of the slope parameter on saving is consistently low across estimators. The results showing the low slope parameter on saving resonate with the observed high mobility of capital. The estimates of the model with structural breaks reinforce the dominant support for the long-run relationship between domestic saving and investment. The inclusion of these structural breaks in the model generally reduces the numerical magnitude of the slope parameter on saving and suggests the high mobility of capital.
1 Introduction
The international mobility of capital remains an area of unresolved controversy in the open economy macroeconomics. The Mundell-Fleming model (Mundell 1962, 1963; Fleming 1962) postulates the perfect mobility of capital and frictionless integration of international financial markets. The influential paradigm pioneered by Feldstein and Horioka (1980) provides a contrary dimension to current account (saving minus investment) and suggests the imperfect mobility of capital and the near-autarkic behaviour of international capital markets. Feldstein and Horioka (FH) find high long-run correlations between domestic saving and investment for the OECD countries, and they interpret these results in terms of the immobility of capital and imperfect integration of international capital markets. The high saving-investment (SI) correlations (small size of current accounts) in the wake of high mobility of capital, as manifested by large capital flows, competitive returns on financial assets and persistent current account imbalances in the OECD countries, marked an unresolved puzzle. Obstfeld and Rogoff (2000) characterise high SI correlations and high mobility of capital as one of the six major puzzles in the international macroeconomics. The micro-founded intertemporal optimization approach to current account that came into vogue almost contemporaneously with the FH strand, since the early-1980s, generally accepts the findings of numerically high and statistically significant SI correlations in FH strand, but it develops several theoretical channels to explain these correlations in the wake of high mobility of capital; see Singh (2007) for a survey. It views high SI correlations as the corollary of current account solvency constraint, rather than as an index of capital immobility. The intertemporal budget constraint may not allow the countries to run high and perpetual current account deficits, and the solvency constraint requires the long-run relationship between domestic saving and investment. The SI correlations, thus, tend to be high, regardless of the degree of capital mobility and financial openness of the economy.
The econometric methodology remains central to the empirical findings and, as such, a part of the FH puzzle could be ascribed to the methodological and measurement problems surrounding the SI correlations. The endogeneity of saving, omitted variables model mis-specification bias, serial-correlation of residuals and the heteroscedasticity of residual variance could induce bias in SI correlations, as the unobserved and unspecified factors that affect investment could also contemporaneously affect the behaviour of saving. [1] The time-averaged SI series used in FH cross-sectional regressions may induce bias in SI correlations and possibly lead to the rejection of capital mobility. The endogeneity of regressors makes the standard least squares estimates biased and inconsistent, and overturns the statistical inference. The efficiency of the instrumental variables (IV) or generalised method of moments estimators, commonly used to alleviate endogeneity, hinges heavily on the quality (weak or strong) and validity (orthogonality) of instruments. The instruments that are weakly related to endogenous regressors (weak instruments) and are non-orthogonal to the Gaussian disturbances (invalid instruments) can still produce biased and inconsistent estimates. The weak instruments may yield biased two-stage least squares (2SLS) estimates even in large samples (Bound, Jaeger, and Baker 1995; Staiger and Stock 1997). When several regressors in a model are instrumented, then the validity requirements for the instruments used for endogenous regressors become even more stringent (Staiger and Stock 1997). It is, in fact, difficult to find appropriate instruments that are strongly correlated with endogenous regressors, but are uncorrelated with the Gaussian disturbances. The paradigm shift in time-series econometrics since the late-1980s fashioned the use of several optimal estimators developed in both single-equation and vector autoregression (VAR)-based system settings. These estimators resolve the problems of “spurious regression,” serial-correlation and long-run endogeneity, and provide efficient parameter estimates.
Most time-series studies examining SI correlations have implicitly assumed a temporally stable and time-invariant parameter vector and, thus, have estimated the long-run model without allowing structural breaks in the cointegrating vector (Miller 1988; Leachman 1991; Coakley, Kulasi, and Smith 1996; Jansen 1996; Coiteux and Olivier 2000; Levy 2000; De Vita and Abbott 2002; Caporale, Panopoulou, and Pittis 2005; Nell and Santos 2008). The financial markets are vulnerable to the speculative (systematic or stochastic) expectations (rational or irrational) of international investors and, as such, are susceptible to the structural breaks and regime switches. The possibilities of structural breaks become particularly pronounced when the relationship between the model series is examined over a longer time-horizon. The structural breaks reduce the power of cointegration tests and weaken the robustness of statistical evidence obtained from the standard one-regime models with time-invariant parameters. A few studies that attempt to account for structural breaks (Sarno and Taylor 1998a, 1998b; Evans, Kim, and Oh 2008) employ the standard estimators/tests that assume a single structural break and/or implicitly rely on the assumption that the post-breakdown periods are relatively long and on the asymptotics in which the length goes to infinity with the sample size. The cointegration breakdowns could occur even over the short time periods such as at the end of the sample.
This study examines the relationship between domestic saving and investment and measures the international mobility of capital in the United States. The persistent current account deficits, reserve-currency characteristic of the U.S. dollar and the commonly observed high preferences for the U.S. financial assets are among the key catalysts that contributed to the mobility of capital into the U.S. The episodes of economic and financial crises ranging from the Great Depression of the early-1930s to the recent global financial crises of the late-2000s consistently suggest that the economic states (booms and recessions) of the goods and financial markets in the U.S. have significant bearing on the states of the goods and financial markets in the world economy. The evidence obtained from the U.S., as such, would be a useful approximation to the international mobility of capital and the integration of financial markets. The novelty of the study merits attention on two counts. First, most studies estimating FH model have drawn conclusions from the estimates of a single or select estimators and such a reliance could lead to biased assessment in terms of both statistical inference and magnitudes of the long-run parameters. The study estimates the model using several single-equation and system estimators to assess the robustness of results and take an encompassing account of the methodological and measurement issues. Second, the study estimates the model “with” and “without” structural breaks to discern the possible dispersion in the magnitude of the slope parameter on saving across regimes. The remainder of the study is structured as follows. Section 2 specifies the model. Section 3 presents the empirical results. Section 4 sums up the conclusions.
2 The Model
The reduced-form bi-variate FH model is estimated to examine the long-run relationship between domestic saving and investment and measure the international mobility of capital.
In model [1], the saving, S, is measured in terms of the gross saving, investment, I, in terms of the gross capital formation (gross fixed capital formation plus inventories) and output, Y, in terms of the gross domestic product (GDP); all at current prices. The ratios of saving,
The macroeconomic identity,
3 Empirical Results
3.1 Unit Root Tests
The unit root tests are first performed to examine the time-series properties of the model series. Such an analysis assumes particular importance for the FH model, as, prima facie, it may seem puzzling to recognise the I(d) property of the ratio of two possibly I(d) series of saving (and investment) and GDP; where the order of integration
The results suggest that the augmented Dickey-Fuller (ADF) (Dickey and Fuller 1981) test does not reject the null hypothesis of a unit root in the level series of both saving and investment in the model with drift and no trend (Table 1). The ADF test rejects the null for saving, but not for investment, in the model with drift and trend. The ADF test rejects the null hypothesis in the differenced series of (i) both saving and investment in the model with drift and (ii) only saving in the model with drift and trend. The Phillips-Perron (PP) (Phillips and Perron 1988) test rejects the null hypothesis for the level series of (i) investment, but not saving, in the model with drift and (ii) both saving and investment in the model with drift and trend. The PP test rejects the null for the differenced series of saving and investment in the model estimated with drift as well as with drift and trend. The KPSS (Kwiatkowski, Phillips, Schmidt, and Shin 1992) test rejects the contrary null hypothesis of no unit root in the level series of saving and investment in the model estimated with drift, but not with trend. The KPSS test does not reject the null in the differenced series. The ADF and PP tests have low power, while the KPSS test has a tendency to over-reject the null hypothesis in small samples. The asymptotically powerful DF-GLS, PT, DF-GLSu and QT tests (Elliott, Rothenberg, and Stock 1996; Elliott 1999), based on the generalised least squares (GLS), are carried out to cross-examine the evidence and assess the robustness of results. While the evidence obtained from the GLS-based point optimal tests is somewhat mixed, these tests generally point towards I(1) properties of the model series (Table 1).
Series | Conventional Tests | GLS-Based Point Optimal Tests | |||||
H0: Unit root | H0: No Unit root | H0: Unit root | |||||
ADF(k) | PP [lw=4] | KPSS [lw=4] | DF-GLS(k) | PT (k) | DF-GLSu(k) | QT(k) | |
Level series | |||||||
Model I: Drift and No Trend | |||||||
−2.83 (2) | −3.96* | 0.528** | −3.94* (1) | 1.26 (1) | −3.91** (1) | 2.34 (1) | |
−0.77 (2) | −1.70 | 1.074* | −0.24 (2) | 17.47* (2) | −0.89 (2) | 19.39** (2) | |
Model II: Drift and Trend | |||||||
−3.15 (5) | −4.54* | 0.062 | −5.06* (1) | 3.13 (1) | −5.24** (1) | 1.68 (1) | |
−3.45** (5) | −3.88** | 0.141 | −4.84* (1) | 4.07 (1) | −4.98** (1) | 2.23 (1) | |
Differenced series | |||||||
Model I: Drift and No Trend | |||||||
−3.20** (10) | −9.95* | 0.051 | −3.00* (1) | 3.86* (1) | −6.98** (1) | 2.13 (1) | |
−8.31* (1) | −9.58* | 0.059 | −1.27 (2) | 17.31* (2) | −4.20** (2) | 4.26 (2) | |
Model II: Drift and Trend | |||||||
−3.16 (10) | −9.99* | 0.044 | −5.26* (1) | 5.78* (1) | −6.83** (1) | 2.16* (1) | |
−8.27* (1) | −9.86* | 0.040 | −1.82 (10) | 245.56* (10) | −1.69 (10) | 101.25** (10) |
The one-regime unit root tests become mis-specified and are not very informative of non-stationarity in the presence of structural breaks in the underlying series. These tests are biased towards non-rejection of the null hypothesis of a unit root, if the underlying series contains a structural break (Perron 1989).
The stationary series may erroneously appear to be non-stationary due to the false non-rejection of the null hypothesis. Perron (1989) provides a test for the null hypothesis of a unit root in the presence of a exogenously determined structural break in the series at the known location. The estimates from the Perron (1989) test, however, could be biased in favour of the rejection of the null hypothesis, as the break-point is not treated as data-dependent and unknown under the alternative hypothesis (Zivot and Andrews 1992). The study uses the endogenous structural break unit root tests of Zivot and Andrews (1992), Lumsdaine and Papell (1997) and Lee and Strazicich (2003, 2004) to test the null hypothesis of a unit root and determine the break-points endogenously from the data. These tests involve the estimation of the model for different break dates using the recursive (rolling or sequential) approach, and then performing the grid-search to locate the most significant break-point,
The Zivot-Andrews test tests the joint null hypothesis of a unit root with no structural break against the alterative hypothesis of a one-time break in the series. It sets
Series | One structural break | Two structural breaks | |||
Zivot-Andrews | Lumsdaine-Papell | Lee-Strazicich | Lumsdaine-Papell | Lee-Strazicich | |
Level series | |||||
Model I: Drift and No Trend | |||||
−5.14 (0) | −5.14 (0) | −4.31* (0) | −5.51 (1) | −4.42 (1) | |
[1988] | [1987] | [1984] | [1987, 1995] | [1984, 1992] | |
−4.10 (3) | −4.10 (3) | −4.37* (3) | −4.23 (3) | −4.70 (3) | |
[1985] | [1984] | [1985] | [1960, 1984] | [1985, 1999] | |
Model II: Drift and Trend | |||||
−5.30 (0) | −5.28 (0) | −4.05 (0) | −6.02** (0) | −5.78** (9) | |
[1990] | [1989] | [1989] | [1976, 1994] | [1974, 1989] | |
−4.13 (3) | −4.10 (3) | −4.39 (9) | −4.94 (3) | −5.49 (3) | |
[1985] | [1984] | [1983] | [1977, 1996] | [1983, 1995] | |
Differenced series | |||||
Model I: Drift and No Trend | |||||
−6.57* (3) | −6.57*(3) | −2.79 (0) | −7.10* (3) | −3.32 (0) | |
[1982] | [1981] | [1979] | [1975, 1993] | [1981, 1989] | |
−8.43* (1) | −8.43* (1) | −2.17 (0) | −8.73* (1) | −2.90 (1) | |
[1994] | [1993] | [1985] | [1970, 1993] | [1982, 1989] | |
Model II: Drift and Trend | |||||
−6.62* (3) | −6.60* (3) | −3.44 (0) | −7.83* (3) | −4.49 (6) | |
[1996] | [1995] | [1966] | [1975, 1993] | [1966, 1981] | |
−8.52* (1) | −8.53* (1) | −3.59 (3) | −9.26* (1) | −4.79 (10) | |
[1994] | [1993] | [1968] | [1981, 1999] | [1968, 1983] |
3.2 Tests for Cointegration and the Long-Run Estimates
Most time-series studies have implicitly assumed a temporally stable and time-invariant parameter vector and, thus, have estimated the long-run model without allowing structural breaks in the cointegrating vector (Miller 1988; Leachman 1991; Coakley, Kulasi, and Smith 1996; Jansen 1996; Coiteux and Olivier 2000; Levy 2000; De Vita and Abbott 2002; Caporale, Panopoulou, and Pittis 2005; Nell and Santos 2008). The study first undertakes the base-line analysis and estimates the long-run model in a standard one-regime setting without structural breaks. The model is estimated using (i) the OLS-based estimator of Engle and Granger (OLSEG) (1987), generalized method of moments (GMM) estimator of Hansen (1982), dynamic OLS (DOLS) estimator of Saikkonen (1991) and Stock and Watson (1993), non-linear least squares (NLLS) estimator of Phillips and Loretan (1991) and the fully-modified OLS (FMOLS) estimator of Phillips and Hansen (1990) in a single-equation setting and (ii) the maximum-likelihood system estimator of Johansen (1991) and the over-parameterized level-VAR estimator of Toda and Yamamoto (1995) in a VAR-based system setting. The use of several estimators is intended to assess the robustness of results across methodologies.
3.2.1 Standard OLSEG and RLS Estimates
The two-step OLSEG estimator sequentially involves the estimation of a static regression model in levels,
The figures in round parentheses are the t-ratios and in square brackets the 5 % critical values for the null hypothesis of a unit root in ε(t) for the ADF, PP (Davidson and Mackinnon 1993) and CRDW (Sargan and Bhargava 1983) tests. Both ADF and PP tests consistently reject
The equilibrium residuals measure the distance between the actual and forecast series of the model. Another way to test cointegration is to map the movements in the residual process of a cointegrating regression (Xiao 1999; Xiao and Phillips 2002). If the given series are cointegrated, then the residuals of a cointegrating regression should be stable with long-run movements within some critical bounds. Xiao (1999) and Xiao and Phillips (2002) argue that the cumulative sum (CUSUM) of recursive residuals test of Brown, Durbin, and Evans (1975) can be applied to the residuals of a regression to directly test the null hypothesis of cointegration. The study estimates the model using recursive least squares (RLS) to map the time-profile of the disequilibrium residuals and trace the temporal-trajectories of the model parameters. The time-trajectories of both CUSUM and CUSUM of squared residuals (CUSUMSQ) remain well within the 5 % critical bounds (Figure 1). The sequential F statistics are scaled by 5 % critical value such that the value larger than unity implies rejection and that less than unity the acceptance of the null hypothesis of model stability. The F statistics remain well below the critical unity grid and suggest the temporal stability of the recursive residuals, except for two spikes that cross the critical unity grid. The estimated residuals normally remain within the critical boundaries (Figure 1). The intercept, α, and slope, β, parameters move within the standard error confidence bands and seem temporally stable.
3.2.2 Optimal Single-Equation Estimates
The standard OLS estimates become biased and inefficient in the presence of non-orthogonality of regressors and serial-correlation of residuals. The “super-consistency” property of OLS indeed allows one to omit I(0) regressors from the cointegrating model and asymptotically ignore the problems of endogeneity and serial-correlation.
[4] In small samples, however, the OLS estimates remain biased and have inferential problems for the significance of long-run parameters. The bias is often substantial (Banerjee, Dolado, Galbraith, and Hendry 1993; Inder 1993) and the t statistics of cointegrating coefficients are generally not valid for statistical inference. The commonly used method to alleviate endogeneity is the IV or GMM estimator of Hansen (1982). The GMM has desirable properties in large samples. The standard moment conditions in a linear regression,
If the orthogonality condition is not satisfied such that
An optimal alternative to using GMM is to introduce an explicit AR(1) specification for X(t) along with the stochastic model for the relationship between Y(t) and X(t). The triangular representation of the cointegrated system of Phillips (1991), with I(1) series of Y(t) and X(t) and I(0) series of μ(t), suggests that the OLS estimator of β is consistent, but not generally fully efficient,
The asymptotic distribution of OLS estimator depends on various nuisance parameters engendered by serial-correlation in μ(t) and by correlation between μ(t) and innovation term for ΔX(t) in model [4]. The μ(t) and η(t) are cross-correlated not only contemporaneously, but also at various leads and lags. Phillips (1991) suggests using the following representation for μ(t),
The
Since
The
Phillips and Hansen (1990) develop the fully-modified OLS (FMOLS) estimator to resolve the problems of endogeneity-bias and serial-correlation, and obtain the efficient estimates of the model parameters. The FMOLS estimator starts with the standard OLS regression and then, analogous to the Phillips-Perron (Phillips and Perron 1988) unit root test, makes a non-parametric correction to account for the endogeneity-bias and serial-correlation that may show up in the OLS residuals. The estimates of the long-run parameters and the associated t-statistics are, thus, adjusted to correct for the bias arising from the endogeneity of regressors and serial-correlation of residuals. The FMOLS estimator is super-consistent and is asymptotically both unbiased and normally distributed (Park and Phillips 1988; Phillips and Hansen 1990; Hansen and Phillips 1990). The t-statistics of the long-run coefficients are asymptotically normally distributed, and the standard limiting distributions can be used to perform the statistical inference in the FMOLS estimates.
The DOLS and NLLS estimations require the determination of the optimal lags and leads structures of the dynamic regressors. The Akaike information criterion (AIC) pointed towards an unduly large lags and leads structure of k{–9, 0, +9}, while the Schwarz information criterion (SIC) suggested a relatively parsimonious structure of k{–4, 0, +4} for the DOLS estimator. The model with DOLS is estimated using the parsimonious lags and leads structure of k{–4, 0, +4} as suggested by SIC. The AIC and SIC consistently suggested k{–1, 0, +1} for the model with NLLS, and the Ljung-Box portmanteau (LB-Q) statistic (Ljung and Box 1978) did not reject the null hypothesis of no serial-correlation in the model residuals. The NLLS estimation is, therefore, carried out using k{–1, 0, +1}. The estimations are also carried out using the higher lags and leads structure of k{–5, 0, +5} for the DOLS and k{–2, 0, +2} for the NLLS estimator so as to assess the robustness of results. The standard errors of the parameters from DOLS and NLLS estimators are adjusted using the heteroscedasticity and autocorrelation consistent (HAC) estimator of Newey and West (1987). The GMM, DOLS, FMOLS and NLLS estimates of the model consistently suggest the positive and significant long-run effects of domestic saving on investment (Table 3). The J-statistics in GMM do not reject the null hypothesis of no correlation between the regressors and the residual term. The long-run slope parameter on saving ranges between 0.31 and 0.37, and the t-ratios reject both
Regressor | Dependent variable: | |||||||
GMM | DOLS | FMOLS | NLLS | |||||
GMM1 | GMM2 | k{–4, 0, +4} | k{–5, 0, +5} | lw=1 | lw=4 | k{–1, 0, +1} | k{–2, 0, +2} | |
Constant | 14.17 | 14.27 | 13.03* | 13.30* | 13.66* | 13.82* | 14.19* | 14.57* |
(9.60) | (10.29) | (19.78) | (21.57) | (13.18) | (10.81) | (7.42) | (6.27) | |
0.32* | 0.31* | 0.37* | 0.36* | 0.35* | 0.34* | 0.33* | 0.31* | |
(4.24) | (4.44) | (10.37) | (10.82) | (6.28) | (4.97) | (3.34) | (2.59) | |
J=1.04 | J=0.0023 | – | – | – | – | |||
[0.31] | [0.96] | (10.76) | (6.30) | |||||
Null Hypothesis | t-ratios | |||||||
4.24* | 4.44* | 10.37* | 10.82* | 6.28* | 4.97* | 3.34* | 2.59* | |
−9.01* | −9.88* | −17.66* | −19.24* | −11.66* | −9.65* | −6.78* | −5.76* | |
New CUSUM and MOSUM Tests | ||||||||
0.7191 | 0.8229 | 0.9221 | 1.0042 | 1.0416 | 0.9757 | 0.5883 | 0.5594 | |
Bandwidth | ||||||||
Parameter | ||||||||
h=0.1 | 1.1140 | 1.1467 | 1.1586 | 1.2006 | 0.9696 | 0.9911 | 0.6885 | 0.8052 |
h=0.2 | 1.2489 | 1.3636 | 1.6010 | 1.4374 | 1.2315 | 1.2301 | 0.7870 | 0.8592 |
h=0.3 | 1.1464 | 1.0646 | 1.4535 | 1.2652 | 1.1598 | 1.1144 | 0.6633 | 0.6293 |
h=0.4 | 0.9561 | 0.8229 | 0.9007 | 1.0656 | 0.9984 | 0.9409 | 0.7762 | 0.7461 |
h=0.5 | 0.6974 | 0.8212 | 0.8654 | 0.9425 | 0.9491 | 0.8871 | 0.5958 | 0.5197 |
The conventional CUSUM and CUSUMSQ tests used to map the time-profile of the disequilibrium residuals become biased and inefficient when the model residuals are autocorrelated and regressors are characterised by endogeneity. Xiao and Phillips (2002) use the FMOLS estimator to resolve the problem of serial-correlation and endogeneity. They construct the cumulative sum (CSn) and moving sum (MSn) test statistics to test the direct null hypothesis of “cointegration” against the alternative hypothesis of “no cointegration” among I(1) variables. The study uses the optimal residuals obtained from all the GMM, DOLS, FMOLS and NLLS estimates of the long-run model to construct the cumulative sum,
3.2.3 Maximum-Likelihood System Estimates
The maximum-likelihood (ML) system estimator of Johansen (1991) estimates the kth order vector autoregression model and takes a system-based account of endogeneity.
Model [8] can be reparameterized as
The
The asymptotic λ-trace and the λ-trace adjusted for small-sample (Johansen 2000, 2002) consistently reject the null hypothesis of r=0 (but not r ≤ 1) and suggest the presence of one cointegrating vector (Table 4). The long-run parameter on S/Y is statistically significant and dimensionally consistent with the estimates obtained from the single-equation estimators. The
Panel I: | |||||||||
Eigenvalues | λ−trace Rank Test | λ−max Rank Test | |||||||
H0 | H1 | λ−trace | λ−trace@ | 95 % CV | H0 | H1 | λ−max | 95 % CV | |
0.248 | r=0 | r≥1 | 17.10** | 16.45** | 15.41 | r=0 | r=1 | 16.54** | 14.07 |
0.010 | r≤1 | r=2 | 0.56 | 0.54 | 3.84 | r≤1 | r=2 | 0.56 | 3.76 |
Long-run parameters of the first cointegrating vector normalised on | |||||||||
LR Test of: | 95 % χ2 value | ||||||||
1 | −0.295 | ||||||||
Exclusion Restrictions | 15.19* (0.00) | 5.66** (0.02) | 3.84 | ||||||
Weak Exogeneity | 14.58* (0.00) | 6.09* (0.01) | 3.84 | ||||||
Panel II: Lag Structures and the λ-trace and λ-max Rank Tests: A Sensitivity Analysis | |||||||||
k | Eigenvalues | λ−trace Rank Test | λ−max Rank Test | ||||||
H0 | H1 | λ−trace | λ−trace@ | H0 | H1 | λ−max | |||
k=1 | 0.318 | r=0 | r≥1 | 24.94** | 24.58** | r=0 | r=1 | 22.61** | |
0.039 | r≤1 | r=2 | 2.33 | 2.32 | r≤1 | r=2 | 2.33 | ||
k=3 | 0.219 | r=0 | r≥1 | 14.34 | 13.54 | r=0 | r=1 | 14.10** | |
0.004 | r≤1 | r=2 | 0.25 | 0.24 | r≤1 | r=2 | 0.25 | ||
k=4 | 0.233 | r=0 | r≥1 | 14.99 | 13.75 | r=0 | r=1 | 14.84** | |
0.003 | r≤1 | r=2 | 0.14 | 0.14 | r≤1 | r=2 | 0.14 | ||
k=5 | 0.139 | r=0 | r≥1 | 8.29 | 8.29 | r=0 | r=1 | 8.26 | |
0.001 | r≤1 | r=2 | 0.03 | 0.03 | r≤1 | r=2 | 0.03 |
3.3 Tests for Granger Non-Causality
3.3.1 Vector Error-Correction Estimates
The estimates of the cointegrating model show only the steady-state equilibrium relationship and do not provide any information on the short-run dynamics. The Granger Representation Theorem (Engle and Granger 1987) postulates that if y(t) and x(t) sequences of a given regression in levels,
The z(t–1) is the lagged error-correction term and
Regressor | Dependent Variable | |||||||||
Model I | Model II | Model III | Model IV | Model V | ||||||
[OLSEG–Based z(t–1)] | [DOLS–Based z(t–1)] | [FMOLS–Based z(t–1)] | [NLLS–Based z(t–1)] | [ML-Based z(t–1)] | ||||||
Constant | 7.45* | 3.53*** | 8.07* | 3.92*** | 8.47* | 4.21*** | 8.72* | 4.40*** | 8.96* | 4.62*** |
(0.00) | (0.07) | (0.00) | (0.06) | (0.00) | (0.06) | (0.00) | (0.07) | (0.00) | (0.07) | |
0.70** | 0.55 | 0.71** | 0.56 | 0.707** | 0.57 | 0.70** | 0.569 | 0.68** | 0.565 | |
[4.33] | [1.96] | [4.57] | [2.10] | [4.59] | [2.19] | [4.50] | [2.23] | [4.22] | [2.25] | |
(0.01) | (0.14) | (0.01) | (0.12) | (0.01) | (0.11) | (0.01) | (0.11) | (0.02) | (0.11) | |
−0.75** | −0.60** | −0.73** | −0.59** | −0.710** | −0.58** | −0.69** | −0.573** | −0.66** | −0.556*** | |
[4.56] | [3.23] | [4.56] | [3.20] | [4.47] | [3.11] | [4.35] | [3.01] | [4.07] | [2.81] | |
(0.01) | (0.04) | (0.01) | (0.04) | (0.01) | (0.04) | (0.01) | (0.05) | (0.02) | (0.06) | |
z(t–1) | −0.60* | −0.29*** | −0.62* | −0.31** | −0.63* | −0.32** | −0.63* | −0.33** | −0.62* | −0.33*** |
(0.00) | (0.06) | (0.00) | (0.05) | (0.00) | (0.05) | (0.00) | (0.05) | (0.00) | (0.06) | |
Model Adequacy Statistics | ||||||||||
0.24 | 0.17 | 0.25 | 0.18 | 0.25 | 0.18 | 0.25 | 0.18 | 0.26 | 0.18 | |
DW | 2.11 | 1.95 | 2.12 | 1.95 | 2.11 | 1.95 | 2.11 | 1.94 | 2.11 | 1.94 |
LB-Q (14) | 15.21 | 25.74 | 15.29 | 26.00 | 15.11 | 25.85 | 14.82 | 25.49 | 14.13 | 24.50 |
(0.36) | (0.03) | (0.36) | (0.03) | (0.37) | (0.03) | (0.39) | (0.03) | (0.44) | (0.04) | |
Skewness | −0.27 | 0.05 | −0.24 | 0.07 | −0.23 | 0.07 | −0.21 | 0.08 | −0.19 | 0.09 |
(0.42) | (0.87) | (0.46) | (0.84) | (0.50) | (0.82) | (0.52) | (0.81) | (0.56) | (0.80) | |
Kurtosis (excess) | −0.17 | −0.35 | −0.13 | −0.33 | −0.11 | −0.31 | −0.09 | −0.30 | −0.08 | −0.28 |
(0.80) | (0.62) | (0.85) | (0.63) | (0.88) | (0.65) | (0.89) | (0.67) | (0.91) | (0.69) | |
Jarque-Bera | 0.74 | 0.31 | 0.61 | 0.30 | 0.51 | 0.28 | 0.45 | 0.27 | 0.37 | 0.25 |
(0.69) | (0.85) | (0.74) | (0.86) | (0.77) | (0.87) | (0.80) | (0.87) | (0.83) | (0.88) |
The results remain consistent across all the VECMs, and suggest the significant effects of lagged disequilibrium (Model I to Model V, Table 5). The VECMs reinforce the evidence obtained from the long-run models and point towards the equilibrium relationship between domestic saving and investment. The t-statistics reject
The impulse response and variance decomposition analyses are carried out on all the VECMs (Model I to Model V) to trace the time-profile and map the response trajectories of
Period | DGDIRATE | DGDSRATE | ||||
S.E. | DGDIRATE | DGDSRATE | S.E | DGDIRATE | DGDSRATE | |
1 | 0.89 | 100.00 | 0.00 | 0.99 | 60.31 | 39.69 |
2 | 0.92 | 94.84 | 5.16 | 1.02 | 61.57 | 38.43 |
3 | 0.96 | 87.78 | 12.22 | 1.07 | 56.48 | 43.52 |
4 | 0.96 | 87.71 | 12.30 | 1.08 | 56.47 | 43.53 |
5 | 0.96 | 87.54 | 12.46 | 1.08 | 56.25 | 43.75 |
10 | 0.96 | 87.50 | 12.50 | 1.08 | 56.23 | 43.77 |
3.3.2 Over-Parameterized Level-VAR Estimates
The VECM approach to testing Granger non-causality builds sequentially on the (i) pretesting of model series for unit root(s), (ii) determining of cointegration rank and (iii) eventually the testing of zero-restrictions on the parameters of the short-run dynamic and long-run lagged disequilibrium regressors. The unit root tests used to determine the I(d) properties of the model series are known to have low power in small samples. The subsequent cointegration tests conditioned on the unit root pre-tests and that non-causality tests conditioned sequentially on the cointegration pre-tests could potentially contain a pre-test bias. The over-parameterized level-VAR estimator of Toda and Yamamoto (1995) (TY) by-passes all the pretesting requirements and is applicable to all VAR systems characterised by a stationary (around a deterministic trend) or integrated or cointegrated process of an arbitrary order. The TY estimator involves the (i) estimation of the level-VAR model of order k augmented artificially with extra dmax lags, and then the (ii) testing of zero-restrictions on the parameters of first k lagged (but not all lagged) regressors to draw long-run Granger causal inference; where dmax is the maximal order of integration of the system series. The VAR lag-length, k, can be determined using the usual model selection criteria and lag-length selection tests.
The
Under the null hypothesis of zero-restrictions on the parameters of each variable in vector
The joint F statistics computed for first k=2 lags [in the model with q=[k + dmax]=[2 + 0]=2 lags] reject the null hypothesis of zero restrictions on the parameters of the lagged regressors of (i)
Regressor | Dependent variable | |||||
Maximal order of integration (dmax) of the model series | ||||||
dmax=0 | dmax=1 | dmax=2 | ||||
VAR Lags: q=[k+dmax]=[2+0]=2; | VAR Lags: q=[k+dmax]=[2+1]=3; | VAR Lags: q=[k+dmax]=[2+2]=4; | ||||
Zero-Restrictions for first k=2 Lags | Zero-Restrictions for first k=2 Lags | Zero-Restrictions for first k=2 Lags | ||||
8.03* (0.00) | 5.01* (0.01) | 8.94* (0.00) | 0.15 (0.86) | 7.48* (0.00) | 0.11 (0.90) | |
4.96* (0.01) | 58.70* (0.00) | 0.59 (0.56) | 14.91* (0.00) | 0.64 (0.53) | 14.37* (0.00) | |
VAR Lags: q=[k+dmax]=[3+0]=3; | VAR Lags: q=[k+dmax]=[3+1]=4; | VAR Lags: q=[k+dmax]=[3+2]=5; | ||||
Zero-Restrictions for first k=3 Lags | Zero-Restrictions for first k=3 Lags | Zero-Restrictions for first k=3 Lags | ||||
6.03* (0.00) | 0.96 (0.42) | 5.25* (0.00) | 0.34 (0.80) | 4.85* (0.01) | 0.48 (0.70) | |
2.37*** (0.08) | 40.01* (0.00) | 0.55 (0.65) | 14.87* (0.00) | 0.79 (0.51) | 13.93* (0.00) |
3.4 Structural Breaks
The model estimated in a one-regime and parameter-invariant setting provides useful information so long as there are no structural breaks in the relationship among variables. The long-run relationship and model parameters, however, may change either suddenly at a given date or smoothly over time. If a change occurs in the population regression function during the sample period, then the regression over the full-sample would estimate the relationship that holds “on average” in that the regression estimates would combine two different periods. The “average” regression function in the model with structural break can be quite different from the true regression function at the end of the sample, depending on the location and magnitude of the break-point. The structural breaks reduce the power of standard cointegration tests and weaken the robustness of statistical evidence obtained from one-regime models without structural breaks. This section allows structural breaks in the cointegrating vector and cross-examines the evidence obtained from the standard base-line model without structural breaks. The analysis is carried out using the (i) standard tests for model instability (Hansen 1992; Quandt 1960; Andrews 1993, 2003; Andrews and Ploberger 1994), (ii) test for cointegration with one structural break (Gregory and Hansen 1996), (iii) tests for multiple structural breaks (Bai and Perron 1998, 2003; Kejriwal and Perron 2008, 2010), (iv) test for cointegration with multiple structural breaks (Johansen, Mosconi, and Nielsen 2000), and the (v) new tests for cointegration breakdowns over the short time periods (Andrews and Kim 2006).
3.4.1 Standard Tests for Model Instability
The parameter instability tests of Hansen (1992) test the null hypothesis of constant parameters against the alternative that the parameters follow a martingale. Hansen (1992) develops the test statistics denoted as (i) L to test the stability of the individual coefficients of a regression model, (ii) Lc to test the joint stability of the regression coefficients and (iii)
The figures in round parentheses in model [14] are the t-ratios and in curly brackets are the p-values. The L statistics generally reject the null hypothesis of stability and point towards the instability of both intercept and slope parameters. The Lc statistic weakly rejects the joint null hypothesis of model stability, while the
A limitation of the tests developed in Hansen (1992) is that these tests simply test the null hypothesis of constancy, and do not provide any information on the timing of structural break. Similarly, the limitation of the conventional Chow test (Chow 1960) alternative is that it tests the structural break only at the known location.
[7] Nevertheless, if the break date is unknown, then the recursive Chow test can be used to test for structural break at every possible point and grid-search the break-point from some bounded space,
Andrews (1993, 2003) and Andrews and Ploberger (1994) derive the asymptotic (large sample) null distributions and provide the critical values for the Supremum Wald, LM and LR test statistics. The asymptotic critical values depend on the (i) number of parameters, k, that are allowed to break (change) under the alternative hypothesis and the (ii) sub-sample interval over which the test statistics are computed. Andrews (1993) considers the parametric model indexed by the parameters (
The
The sub-sample interval used to perform grid-search for the break-point is bounded between the trimming parameters (
The Andrews-Quandt (AQ) and Andrews-Ploberger (AP) tests used in the study are based on the Lagrange Multiplier tests for the null hypothesis of no structural break against the alternative of a one-time unknown break in the parameters of the linear regression. The AQ test uses the supremum (maximum) of the LM statistics as the test statistic, while the AP test uses the geometric mean. A series of LM statistics are computed for each of the possible break-point and the grid-search is performed over the trimmed region of the sample space to locate the break-point,
3.4.2 Test for Cointegration with One Structural Break
The OLS-based estimator of Gregory and Hansen (OLSGH) (1996) is one of the most commonly used estimators to detect structural breaks in the cointegrating vector. The OLSGH is the direct extension of the residual-based OLSEG and it allows one-time structural break, via dummy variable, in either intercept or both intercept and slope parameter. The break date is unknown, a priori, and is determined endogenously by the model. The first step in OLSGH involves the estimation of a set of static regression models augmented with (i) intercept dummy to account for the level shift (Model I), (ii) intercept dummy and a linear trend to assess the level shift with trend (Model II) and (iii) both intercept and slope dummies (entire coefficient vector) to determine the regime shift (Model III) in the cointegrating vector.
Model I: Constant; Level Shift:
Model II: Constant and Trend; Level Shift with trend:
Model III: Constant and Slope; Regime Shift:
The DU denotes the dummy variable that takes value 0 if it is below and value 1 if it is above the unknown break-point, and {·} is the integer part. The unknown regime shift parameter
Model | Minimized GH-ADF* | Break Year | Critical Values | |
0.01 | 0.05 | |||
Model I: C | –4.69** | 1996 | –5.13 | –4.61 |
Model II: C/T | –4.39 | 1957 | –5.45 | –4.99 |
Model III: C/S | –4.77 | 1996 | –5.47 | –4.95 |
The FH model [1] is now augmented with the intercept and interaction slope dummies, and is re-estimated to examine the shifts in the intercept and slope parameters.
The dummy-augmented FH model [18] transforms to the standard bi-variate model,
The standard OLS and optimal DOLS and FMOLS estimates of the dummy-augmented model [18] point towards the upward shift in the intercept and the downward shift in the slope parameter on saving (Table 10). The magnitude of the slope parameter on saving,
Regressor | Dependent variable: | ||||||||
OLS | GMM | DOLS | FMOLS | NLLS | |||||
GMM1 | GMM2 | k{–4, 0, +4} | k{–5, 0, +5} | lw=1 | lw=4 | k{–1, 0, +1} | k{–2, 0, +2} | ||
Constant | 10.24* | 12.53* | 11.71* | 13.56* | 14.57* | 11.37* | 11.24* | 11.19* | 11.65* |
(8.58) | (6.13) | (6.01) | (18.20) | (21.42) | (8.87) | (7.21) | (5.30) | (4.34) | |
0.52* | 0.40* | 0.44* | 0.35* | 0.30* | 0.46* | 0.47* | 0.47* | 0.45* | |
(8.65) | (3.95) | (4.48) | (8.81) | (8.41) | (6.96) | (5.81) | (4.49) | (3.37) | |
DUt | 3.92** | 2.42 | 3.05 | 2.48** | 2.83** | 3.63 | 3.82 | 1.50 | 2.01 |
(2.44) | (1.05) | (1.28) | (2.03) | (2.11) | (1.29) | (1.11) | (1.22) | (1.28) | |
–0.18** | –0.11 | –0.13 | –0.16** | –0.20** | –0.17 | –0.18 | –0.07 | –0.10 | |
(–2.01) | (–0.93) | (–1.04) | (–2.11) | (–2.22) | (–0.98) | (–0.85) | (–1.00) | (–1.10) | |
J=2.78 | J=4.45 | ||||||||
[0.43] | [0.22] | (7.68) | (5.57) | ||||||
Null Hypothesis | t-ratios | ||||||||
8.65* | 3.95* | 4.48* | 8.81* | 8.41* | 6.96* | 5.81* | 4.49* | 3.37* | |
–8.05* | –5.88* | –5.78* | –16.68* | –19.96* | –8.20* | –6.66* | –4.98* | –4.12* | |
Dummy value | Shift in Intercept Parameter: | ||||||||
DUt=0 | 10.24 | 12.53 | 11.71 | 13.56 | 14.57 | 11.37 | 11.24 | 11.19 | 11.65 |
DUt=1 | 14.16 | 14.95 | 14.76 | 16.04 | 17.40 | 15.00 | 15.06 | 12.69 | 13.66 |
Dummy value | Shift in Slope Parameter: | ||||||||
DUt=0 | 0.52 | 0.40 | 0.44 | 0.35 | 0.30 | 0.46 | 0.47 | 0.47 | 0.45 |
DUt=1 | 0.34 | 0.29 | 0.31 | 0.19 | 0.10 | 0.29 | 0.29 | 0.40 | 0.35 |
3.4.3 Tests for Multiple Structural Breaks
The standard tests for model instability preclude the possibilities of multiple breaks in the model parameters. Bai and Perron (BP) (1998, 2003) consider the linear model and use the dynamic programming algorithm to determine m number of unknown breaks and implied m+1 number of regimes. The BP statistics are the generalization of the single-break test statistics of Andrews (1993, 2003) and are robust to the serial-correlation and heterogeneity of residuals under the null hypothesis. Kejriwal and Perron (KP) (2008, 2010) allow I(1) as well as I(0) regressors in the cointegrating model, and derive the limiting distribution of the Sup-Wald test under the mild conditions on the errors and regressors for a variety of testing problems. Kejriwal and Perron (2008) show that if the coefficients of integrated regressors are allowed to change, then the estimated break fractions are asymptotically dependent so that the confidence intervals need to be constructed jointly. If, however, only the intercept and/or the coefficients of the stationary regressors are allowed to change, then the estimates of break dates are asymptotically independent as in the stationary case analyzed by Bai and Perron (1998, 2003). The structural breaks can take place in the form of the changes in either intercept or slope of the cointegrating vector. Kejriwal and Perron (2008, 2010) suggest the use of the linear DOLS estimator of Saikkonen (1991) and Stock and Watson (1993) to resolve the problem of endogeneity of regressors and serial-correlation of residuals. The KP results are valid, under very weak conditions, when the potential endogeneity of non-stationary regressors is accounted for via an increasing sequence of lags and leads of the first-differenced dynamic regressors in DOLS. They show that the limiting distributions of the tests based on DOLS are the same as those obtained with the static regression under strict exogeneity.
Both BP and KP suggest three tests for testing multiple breaks. The first test is the Sup-Wald test for the null hypothesis of no structural break
The study estimates the DOLS model [6] using the lags-leads structure of k={–4, 0, +4} for the first-differenced I(0) regressors. Both intercept and slope parameters are allowed to change across regimes. The coefficients of the lagged, contemporaneous and lead I(0) regressors are not allowed to break and, thus, are considered fixed and invariant over time. The inclusion of I(0) regressors whose coefficients are not allowed to change does not alter the limit distribution. The DOLS estimation is also carried out using one lower, k={–3, 0, +3}, and one higher, k={–5, 0, +5}, lags and leads structures of the I(0) regressors so as to assess the robustness of results. The results obtained from the alternative model structures were generally consistent in terms of the number and locations of the break-points.
[8] The results obtained from the multiple structural break tests performed on the model with k={–4, 0, +4} suggest that the F test rejects the null hypothesis of no structural break
Panel I: Sup-F test for zero versus an unknown number of structural breaks | ||||||
Sup-F(m) Statistics; | ||||||
Sup-F(1|0) | Sup-F(2|0) | Sup-F(3|0) | Sup-F(4|0) | Sup-F(5|0) | UDmax(L) | |
Test statistics | 19.98* | 15.03* | 14.87* | 30.33* | 34.6* | 34.6* (5) |
Significance level | Critical values | |||||
1 % | 17.67 | 14.73 | 12.21 | 10.77 | 8.82 | 17.67 |
5 % | 14.30 | 12.11 | 10.41 | 9.19 | 7.64 | 14.47 |
10 % | 12.36 | 11.01 | 9.60 | 8.45 | 6.96 | 12.64 |
Panel II: Sequential | ||||||
Sup-F(1|0) | Sup-F(2|1) | Sup-F(3|2) | Sup-F(4|3) | Sup-F(5|4) | ||
Test statistics | 19.98* | 30.06* | 44.61* | 121.33* | 173.01* | |
Significance level | Critical values | |||||
1 % | 19.04 | 19.35 | 19.90 | 19.99 | 20.01 | |
5 % | 15.65 | 16.61 | 17.12 | 17.66 | 17.85 | |
10 % | 14.26 | 15.02 | 15.64 | 16.02 | 16.51 | |
Number of breaks=5 (based on Sup-F statistics); Minimised BIC(L)=–3.33 (5); Minimised LWZ(L)=–1.34 (5); Minimised residual sum of squares=0.10 (5). | ||||||
Panel III: Break-Points and the 95 % Lower and Upper Confidence Bands | ||||||
m=1 | m=2 | m=3 | m=4 | m=5 | ||
Break year | 1965 | 1970 | 1975 | 1982 | 1993 | |
95% confidence Band | [1964–1966] | [1969–1971] | [1974–1976] | [1981–1983] | [1992–1994] |
3.4.4 Test for Cointegration with Multiple Structural Breaks
The break years (1965, 1970, 1975, 1982, 1993) suggested by the multiple structural break tests are now used to set the exogenous break dummies in the VAR model. The ML estimator of Johansen, Mosconi, and Nielsen (2000) is used to estimate the VAR model augmented with such exogenous break dummies, and test the null hypothesis of no cointegration in the presence of multiple structural breaks. The ML estimator is useful to determine the number of cointegrating vectors in the presence of breaks at the known points in time. The estimation is carried out sequentially in that the model is first estimated with one structural break in level corresponding to 1965 (Model I) followed by the model with two (1965, 1970; Model II), three (1965, 1970, 1975; Model III), four (1965, 1970, 1975, 1982; Model IV) and five (1965, 1970, 1975, 1982, 1993; Model V) structural breaks in level (Table 12). Such sequential analysis is intended to discern the possible sensitivity of results to the inclusion of an additional break. The results are sensitive to the use of lag structures and the inclusion of exogenous breaks in the VAR model. [9] Both asymptotic λ-trace and λ-trace adjusted for small-sample consistently do not reject the null hypothesis of no cointegration in the models estimated with one (Model I) and two (Model II) structural breaks in level, but reject the null hypothesis in the models estimated with three (Model III), four (Model IV) and five (Model V) structural breaks in level (Table 12). The slope parameter on saving is consistently low across all the models, reinforcing the high mobility of capital.
Model | Eigenvalues | λ-trace Test | λ-trace Test@ | ||||
H0: r=0 | H0: r≤1 | H0: r=0 | H0: r≤1 | H0: r=0 | H0: r≤1 | ||
Model I | 0.263 | 0.088 | 23.06 | 5.34 | 22.18 | 5.11 | |
Model II | 0.269 | 0.087 | 23.50 | 5.29 | 22.60 | 5.06 | |
Model III | 0.294 | 0.132 | 28.45** | 8.22** | 27.38** | 7.83** | |
Model IV | 0.380 | 0.230 | 42.85** | 15.17** | 41.21** | 14.48 | |
Model V | 0.418 | 0.255 | 48.52** | 17.10** | 46.71** | 16.36 | |
Model | Long-run parameters of the first cointegrating vector normalized on I/Y | ||||||
I/Y | S/Y | Constant (1965) | Constant (1970) | Constant (1975) | Constant (1982) | Constant (1993) | |
Model I | 1 | –0.18 | 0.78 | ||||
[1.07] | [1.44] | ||||||
(0.30) | (0.23) | ||||||
Model II | 1 | –0.22 | 0.66 | –0.09 | |||
[1.50] | [0.62] | [0.01] | |||||
(0.22) | (0.43) | (0.92) | |||||
Model III | 1 | –0.35*** | 0.54 | –0.50 | –0.09 | ||
[3.26] | [0.56] | [0.32] | [0.01] | ||||
(0.07) | (0.45) | (0.57) | (0.91) | ||||
Model IV | 1 | 0.04 | 0.91 | –0.33 | –1.19 | 2.64* | |
[0.01] | [1.19] | [0.10] | [1.47] | [6.02] | |||
(0.91) | (0.28) | (0.76) | (0.24) | (0.01) | |||
Model V | 1 | 0.13 | 0.90 | –0.13 | –1.17 | 2.51* | 0.41 |
[0.13] | [1.38] | [0.02] | [1.67] | [7.59] | [0.33] | ||
(0.71) | (0.24) | (0.90) | (0.20) | (0.01) | (0.56) |
3.4.5 New Tests for Cointegration Breakdowns over the Short Time Periods
The standard structural break estimators rely on the assumption that the post-breakdown periods are relatively long and on the asymptotics in which the length goes to infinity with the sample size. The power of the standard structural break estimators may tend to decline as the break-point moves towards the end of the sample space. These estimators could be inadequate to take an efficient account of structural breaks that may occur over the short time period and at the end of the sample. The possibilities of these short and end-of-sample breaks in the cointegrating vector become particularly pronounced for the models examining the current account imbalances and the associated SI correlations and capital flows. Apart from economic fundamentals, the capital flows are conditioned by the speculative (systematic or stochastic) expectations (rational or irrational) of international investors. Andrews and Kim (AK) (2006) develop the new cointegration breakdown tests that are efficient in the presence of short and end-of-sample breaks in the cointegrating vector. These tests are asymptotically valid when the length, m, of post-breakdown period is fixed, as the total sample size, T+m, goes to infinity. The AK tests build on the estimation of the model represented by
The regressors for all time periods,
Under the alternative hypothesis, the model is a well-specified cointegrating regression for all
The
The study performs
Estimator | H0: Cointegration Prevails for the Full Sample From 1948 To 2007 | |||||||
Pa | Pb | Pc | Ra | Rb | Rc | |||
1948–1981 | 1948–1994 | H1: Cointegration Breaks Down During: 1982–2007; m=26 | ||||||
OLS | 0.72 | 0.52 | 94.78* | 34.37 | 23.07 | 20342.80* | 4020.62 | 565.31 |
(0.00) | (0.11) | (0.33) | (0.00) | (0.33) | (0.78) | |||
FMOLS | 0.58 | 0.46 | 51.48 | 27.99 | 22.97 | 8720.78 | 2127.02 | 199.72 |
(0.56) | (1.00) | (0.44) | (0.44) | (0.78) | (1.00) | |||
FIML | –0.77 | 0.36 | 654.12* | 23.19 | 24.63 | 152179.42* | 387.52 | 277.05 |
(0.00) | (1.00) | (0.67) | (0.00) | (1.00) | (1.00) | |||
1948–1984 | 1948–1995 | H1: Cointegration Breaks Down During: 1985–2007; m=23 | ||||||
OLS | 0.65 | 0.52 | 64.26* | 29.95 | 19.00 | 10036.99* | 2956.23 | 457.04 |
(0.00) | (0.13) | (0.13) | (0.00) | (0.40) | (0.53) | |||
FMOLS | 0.59 | 0.47 | 46.34 | 24.60 | 18.73 | 6374.16 | 1779.75 | 189.65 |
(0.33) | (0.60) | (0.27) | (0.33) | (0.60) | (0.93) | |||
FIML | –0.47 | 0.39 | 331.50 | 19.80 | 20.22 | 56778.74* | 604.54 | 261.46 |
(0.33) | (1.00) | (1.00) | (0.00) | (1.00) | (0.83) | |||
1948–1989 | 1948–1998 | H1: Cointegration Breaks Down During: 1990–2007; m=18 | ||||||
OLS | 0.44 | 0.51 | 15.56 | 20.49 | 14.33 | 836.48 | 1810.63 | 417.10 |
(0.60) | (0.16) | (0.24) | (0.52) | (0.12) | (0.44) | |||
FMOLS | 0.29 | 0.46 | 17.85 | 16.84 | 14.83 | 137.50 | 1130.52 | 168.70 |
(0.40) | (0.56) | (0.48) | (0.76) | (0.44) | (0.48) | |||
FIML | 0.01 | 0.35 | 65.43** | 14.42 | 16.95 | 3732.28* | 260.28 | 112.39 |
(0.05) | (0.73) | (0.64) | (0.00) | (0.55) | (0.77) | |||
1948–1994 | 1948–2000 | H1: Cointegration Breaks Down During: 1995–2007; m=13 | ||||||
OLS | 0.52 | 0.50 | 18.55 | 15.01 | 7.03 | 1252.04*** | 980.94 | 338.33 |
(0.26) | (0.29) | (0.69) | (0.09) | (0.14) | (0.17) | |||
FMOLS | 0.46 | 0.44 | 12.54 | 10.24 | 5.02 | 788.71 | 605.37 | 143.79 |
(0.34) | (0.60) | (0.89) | (0.17) | (0.26) | (0.63) | |||
FIML | 0.36 | 0.31 | 6.16 | 4.80 | 4.32 | 254.15 | 115.10 | 39.02 |
(0.81) | (0.94) | (0.78) | (0.41) | (0.78) | (0.91) | |||
1948–1999 | 1948–2003 | H1: Cointegration Breaks Down During: 2000–2007; m=8 | ||||||
OLS | 0.50 | 0.46 | 14.70 | 10.61 | 6.11 | 386.47* | 266.28*** | 129.98 |
(0.13) | (0.20) | (0.38) | (0.00) | (0.07) | (0.16) | |||
FMOLS | 0.45 | 0.43 | 10.05 | 8.22 | 3.88 | 248.08 | 193.58 | 55.75 |
(0.38) | (0.42) | (0.73) | (0.11) | (0.18) | (0.44) | |||
FIML | 0.32 | 0.33 | 3.98 | 4.12 | 2.88 | 56.89 | 62.50 | 16.03 |
(0.67) | (0.64) | (0.64) | (0.57) | (0.55) | (0.76) | |||
1948–2000 | 1948–2003 | H1: Cointegration Breaks Down During: 2001–2007; m=7 | ||||||
OLS | 0.50 | 0.46 | 12.54*** | 9.53 | 5.29 | 267.69* | 198.86*** | 98.36 |
(0.09) | (0.19) | (0.43) | (0.00) | (0.06) | (0.11) | |||
FMOLS | 0.44 | 0.43 | 8.16 | 7.23 | 3.15 | 167.01 | 145.09 | 42.92 |
(0.38) | (0.45) | (0.74) | (0.13) | (0.17) | (0.43) | |||
FIML | 0.31 | 0.33 | 2.85 | 3.33 | 2.24 | 33.93 | 47.77 | 13.11 |
(0.68) | (0.64) | (0.84) | (0.57) | (0.52) | (0.75) | |||
1948–2001 | 1948–2004 | H1: Cointegration Breaks Down During: 2002–2007; m=6 | ||||||
OLS | 0.50 | 0.44 | 12.08*** | 7.70 | 5.28 | 189.70** | 117.56*** | 76.13*** |
(0.06) | (0.16) | (0.39) | (0.02) | (0.08) | (0.08) | |||
FMOLS | 0.43 | 0.39 | 7.68 | 5.42 | 3.10 | 117.20 | 78.32 | 35.49 |
(0.29) | (0.47) | (0.65) | (0.12) | (0.20) | (0.43) | |||
FIML | 0.33 | 0.31 | 3.20 | 2.62 | 2.09 | 37.18 | 25.23 | 12.28 |
(0.54) | (0.61) | (0.70) | (0.39) | (0.54) | (0.72) | |||
1948–2002 | 1948–2004 | H1: Cointegration Breaks Down During: 2003–2007; m=5 | ||||||
OLS | 0.48 | 0.44 | 10.77** | 7.62*** | 5.28 | 114.74** | 79.74*** | 53.30*** |
(0.04) | (0.10) | (0.25) | (0.02) | (0.06) | (0.06) | |||
FMOLS | 0.44 | 0.39 | 7.84 | 5.41 | 3.04 | 82.25 | 54.78 | 26.75 |
(0.16) | (0.33) | (0.55) | (0.10) | (0.18) | (0.27) | |||
FIML | 0.31 | 3.10 | 2.49 | 1.80 | 27.40 | 19.81 | 10.60 | |
0.33 | (0.50) | (0.52) | (0.67) | (0.31) | (0.44) | (0.56) | ||
1948–2003 | 1948–2005 | H1: Cointegration Breaks Down During: 2004–2007; m=4 | ||||||
OLS | 0.46 | 0.42 | 8.53** | 6.04 | 5.08 | 55.54** | 37.86 | 31.01 |
(0.04) | (0.19) | (0.23) | (0.04) | (0.11) | (0.11) | |||
FMOLS | 0.43 | 0.36 | 6.72 | 3.91 | 3.03 | 42.67*** | 22.71 | 16.45 |
(0.13) | (0.40) | (0.42) | (0.09) | (0.23) | (0.28) | |||
FIML | 0.33 | 0.28 | 3.23 | 1.87 | 1.74 | 17.87 | 8.17 | 7.26 |
(0.38) | (0.48) | (0.54) | (0.28) | (0.52) | (0.48) | |||
1948–2004 | 1948–2005 | H1: Cointegration Breaks Down During: 2005–2007; m=3 | ||||||
OLS | 0.44 | 0.42 | 5.12 | 4.39 | 3.73 | 18.95 | 15.72 | 12.78 |
(0.16) | (0.20) | (0.25) | (0.13) | (0.16) | (0.20) | |||
FMOLS | 0.39 | 0.36 | 3.83 | 2.95 | 2.35 | 13.19 | 9.30 | 6.68 |
(0.29) | (0.36) | (0.36) | (0.18) | (0.27) | (0.38) | |||
FIML | 0.31 | 0.28 | 1.99 | 1.56 | 1.47 | 5.09 | 3.28 | 2.91 |
(0.37) | (0.44) | (0.44) | (0.38) | (0.48) | (0.50) |
The stylized evidence provides dominant support for the presence of high mobility of capital in the U.S. Such support is consistent with the observed dynamics of current account and the implied financing of domestic investment through foreign saving. The support for low SI correlations and high mobility of capital resonates with the findings of the studies by Schmidt (2003) and Hoffmann (2004) for the OECD countries including the U.S. The support for cointegration between saving and investment found in the study is also consistent with the studies finding support for cointegration (Levy 2000; De Vita and Abbott 2002; Nell and Santos 2008), but inconsistent with the studies providing no (Gulley 1992; Byrne, Fazio, and Fiess 2009) or mixed (Miller 1988; Moreno 1997) support for cointegration between saving and investment. A number of factors seem catalytic to the U.S. current account deficits and implied reliance of domestic investment on foreign saving. First, the level of domestic saving remains inadequate to finance the high levels of investment and such inadequacy induces the need for borrowing from the world financial markets. Second, the current account surpluses in the rest of the world are commensurately reflected in the current account deficits in the U.S., given the reserve-currency characteristic of the U.S. dollar. Third, the international investors seem to speculate low exchange rate risks and prefer the higher holdings of the U.S. financial assets in their asset portfolios. Obstfeld (2010) observes that some recent theories of the global imbalances stress the U.S. as a source of high quality assets that the rest of the world’s savers crave. Fourth, the high-valued U.S. currency provides incentive for borrowings from the international financial markets.
4 Conclusions
This study has examined the relationship between domestic saving and investment and measured the international mobility of capital in the United States. The long-run model, “with” and “without” structural breaks, is estimated using several single-equation and system estimators to assess the robustness of results and take an exhaustive account of the methodological and measurement issues. The results provide dominant support for the long-run relationship between domestic saving and investment. The estimates of the slope parameter on saving above zero and the dominant support for cointegration between saving and investment across estimators vindicate the validity of intertemporal budget constraint and suggest the sustainability of current account deficits. The numerical magnitude of the slope parameter on saving is consistently low across estimators. The estimates of the model with structural breaks reinforce the dominant support for the long-run relationship between domestic saving and investment. The inclusion of these structural breaks in the model generally reduces the numerical magnitude of the slope parameter on saving and suggests the high mobility of capital. The results showing the low slope parameter on saving resonate with the observed high mobility of capital and reflect the globalization of financial markets. The numerically low magnitude of the slope parameter on saving in the wake of observed high mobility of capital is consistent with the theoretical predictions of the model.
The internal need for borrowing from the world financial markets arising from the domestic resource gap (saving minus investment) in the U.S. seems equally-whelmed by the external supply of funds by the international investors to purchase the U.S. assets. The international investors seem to speculate low exchange rate risks and predict the higher exchange-rate-adjusted rates of returns and, as such, prefer the higher holdings of the U.S. assets in their asset portfolios. The reserve-currency characteristic of the U.S. dollar provides an added dimension to the demand for the U.S. currency. The mutually-reinforcing internal demand for and the external supply of loanable funds appear to have played a catalytic role in conditioning the mobility of capital into the U.S. While the capital inflows relinquish the domestic saving and binding financing constraints on domestic investment, these inflows lead to the appreciation of exchange rate, which, in turn, reduces the competitiveness of exports and accentuates the current account deficits. The gains of financial openness remain surrounded by several risks arising from the sudden stops and stochastic reversals of particularly the high-resolution and speculative capital inflows. Such sudden stops and stochastic reversals of the short-term and speculative capital inflows tend to induce financial instability and lead to the spiral of financial and economic calamities in the domestic as well as global economy. This underlines the need to stimulate domestic saving to finance the higher levels of domestic investment and reduce the magnitudes of current account deficits and external debt. The domestic saving could be accelerated through the reduction in budget deficits and the provision of saving incentives to the household and private corporate business sectors. The depreciation of the domestic currency (the U.S. dollar) against the currencies of particularly the major trading partners and/or international lenders would be useful to stimulate the net exports (exports minus imports) as well as to reduce the effective magnitude of the external debt denominated in the U.S. currency.
Acknowledgement
This paper is based on the research project undertaken under the Griffith University Research Grant (GURG), Griffith Business School, Griffith University, Australia. I am grateful to the Griffith Business School for the research grant for the project. I am also grateful to the Editor and an anonymous Referee of the journal for very useful comments and suggestions on the paper. I am, however, solely responsible for any errors and omissions that may remain in the paper.
Appendix
Model Structure | Dependent Variable: | ||
Constant | |||
k | DOLS Estimates | ||
k{–1, 0, +1} | 13.43* (10.42) | 0.36* (5.45) | |
k{–2, 0, +2} | 13.10* (12.08) | 0.37* (6.67) | |
k{–3, 0, +3} | 13.02* (15.03) | 0.37* (8.21) | |
lw | FMOLS Estimates | ||
lw=2 | 13.82* (11.77) | 0.34* (5.42) | |
lw=3 | 13.88* (11.11) | 0.34* (5.04) | |
lw=5 | 13.82* (10.71) | 0.34* (4.92) | |
lw=8 | 13.94* (11.03) | 0.33* (4.93) | |
k | NLLS Estimates | ||
k{–3, 0, +3} | 14.17* (6.61) | 0.32* (2.93) | 0.18* (3.88) |
k{–4, 0, +4} | 12.89* (8.68) | 0.38* (4.94) | 0.06 (1.45) |
k{–5, 0, +5} | 12.79* (12.19) | 0.38* (6.85) | –0.006 (–0.15) |
Regressor | Dependent Variable | |||||
Maximal Order of Integration (dmax) of the Model Series | ||||||
dmax=0 | dmax=1 | dmax=2 | ||||
VAR Lags: q=[k+dmax]=[4+0]=4; | VAR Lags: q=[k+dmax]=[4 + 1]=5; | VAR Lags: q=[k+dmax]=[4+2]=6; | ||||
Zero-Restrictions for first k=4 Lags | Zero-Restrictions for first k=4 Lags | Zero-Restrictions for first k=4 Lags | ||||
4.69* (0.00) | 0.97 (0.43) | 3.70* (0.01) | 0.55 (0.70) | 4.69* (0.00) | 0.82 (0.52) | |
2.00 (0.11) | 27.82* (0.00) | 0.70 (0.59) | 13.70* (0.00) | 1.13 (0.35) | 12.95* (0.00) | |
VAR Lags: q=[k+dmax]=[5+0]=5; | VAR Lags: q=[k+dmax]=[5+1]=6; | VAR Lags: q=[k+dmax]=[5+2]=7; | ||||
Zero-Restrictions for first k=5 Lags | Zero-Restrictions for first k=5 Lags | Zero-Restrictions for first k=5 Lags | ||||
3.20** (0.02) | 0.68 (0.64) | 3.84* (0.01) | 0.67 (0.65) | 3.81* (0.01) | 0.83 (0.54) | |
1.44 (0.23) | 20.54* (0.00) | 1.41 (0.24) | 12.42* (0.00) | 1.54 (0.20) | 10.50* (0.00) |
References
Andrews, D. W. K. 1993. “Tests for Parameter Instability and Structural Change with Unknown Change Point.” Econometrica July 61 (4):821–56.10.2307/2951764Search in Google Scholar
Andrews, D. W. K. 2003. “Tests for Parameter Instability and Structural Change with Unknown Change Point: A Corrigendum.” Econometrica January 71 (1):395–7.10.1111/1468-0262.00405Search in Google Scholar
Andrews, D. W. K., and J. -Y. Kim. 2006. “Tests for Cointegration Breakdown over a Short Time Period.” Journal of Business and Economic Statistics October 24 (4):379–94.10.1198/073500106000000297Search in Google Scholar
Andrews, D. W. K., and Ploberger, W.. 1994. “Optimal Tests when a Nuisance Parameter is Present only Under the Alternative.” Econometrica November 62 (6): 1383–414.10.2307/2951753Search in Google Scholar
Bai, J., and P. Perron. 1998. “Estimating and Testing Linear Models with Multiple Structural Changes.” Econometrica January 66 (1):47–78.10.2307/2998540Search in Google Scholar
Bai, J., and P. Perron. 2003. “Computation and Analysis of Multiple Structural Change Models.” Journal of Applied Econometrics January/February 18 (1):1–22.10.1002/jae.659Search in Google Scholar
Banerjee, A., J. J. Dolado, J. W. Galbraith, and D. F. Hendry. 1993. Co-integration, Error-correction, and the Econometric Analysis of Non-stationary Data, Advanced texts in Econometrics. Oxford: Oxford University Press.10.1093/0198288107.001.0001Search in Google Scholar
Ben-David, D., R. L. Lumsdaine, and D. H. Papell. 2003. “Unit Roots, Postwar Slowdowns and Long-Run Growth: Evidence from Two Structural Breaks.” Empirical Economics April 28 (2):303–19.10.3386/w6397Search in Google Scholar
Bound, J., D. Jaeger, and R. Baker. 1995. “Problems with Instrumental Variables Estimation When the Correlation between the Instruments and the Endogenous Explanatory Variable is Weak.” Journal of the American Statistical Association 90 (430):443–50.10.1080/01621459.1995.10476536Search in Google Scholar
Brown, R. L., J. Durbin, and J. M. Evans. 1975. “Techniques for Testing the Constancy of Regression Relationships over Time.” Journal of the Royal Statistical Society Series B 37 (2):149–92.10.1111/j.2517-6161.1975.tb01532.xSearch in Google Scholar
Byrne, J. P., G. Fazio, and N. Fiess. 2009. “The Global Side of the Investment-Saving Puzzle.” Journal of Money, Credit & Banking 41 (5):1033–40.10.1111/j.1538-4616.2009.00244.xSearch in Google Scholar
Carstensen, K. 2006. “Stock market Downswing and the Stability of European Monetary Union Money Demand.” Journal of Business and Economic Statistics October 24 (4):395–402.10.1198/073500106000000369Search in Google Scholar
Caporale, G. M., E. Panopoulou, and N. Pittis. 2005. “The Feldstein-Horioka Puzzle Revisited: A Monte Carlo Study.” Journal of International Money and Finance 24 (7):1143–9.10.1016/j.jimonfin.2005.08.003Search in Google Scholar
Cavaliere, G. 2005. “Limited Time Series with a Unit Root.” Econometric Theory October 21 (5):907–45.Search in Google Scholar
Chow, G. C. 1960. “Tests of Equality between Sets of Coefficients in Two Linear Regressions.” Econometrica July 28 (3):591–605.10.2307/1910133Search in Google Scholar
Coakley, J., F. Kulasi, and R. Smith. 1996. “Current Account Solvency and the Feldstein-Horioka Puzzle.” Economic Journal 106 (436):620–7.10.2307/2235567Search in Google Scholar
Coiteux, M., and S. Olivier. 2000. “The Saving Retention Coefficient in the Long Run and in the Short run: Evidence from Panel Data.” Journal of International Money and Finance 19 (4):535–48.10.1016/S0261-5606(00)00014-0Search in Google Scholar
Davidson, R., and J. G. MacKinnon. 1993. Estimation and Inference in Econometrics. New York: Oxford University Press.Search in Google Scholar
De Vita, G., and A. Abbott. 2002. “Are saving and Investment Cointegrated? An ARDL Bounds Testing Approach.” Economics Letters 77 (2):293–9.10.1016/S0165-1765(02)00139-8Search in Google Scholar
Dickey, D. A., and W. A. Fuller. 1981. “Likelihood Ratio Statistics for Autoregressive Time Series with A Unit Root.” Econometrica July 49 (4):1057–72.10.2307/1912517Search in Google Scholar
Elliott, G. 1999. “Efficient Tests for a Unit Root When the Initial Observation Is Drawn from Its Unconditional Distribution.” International Economic Review August 40 (3):767–84.10.1111/1468-2354.00039Search in Google Scholar
Elliott, G., T. J. Rothenberg, and J. H. Stock. 1996. “Efficient Tests for an Autoregressive Unit Root.” Econometrica July 64 (4):813–36.10.3386/t0130Search in Google Scholar
Enders, W. 2004. Applied Econometric Time Series, 2nd ed. U.S.A.: John Wiley & Sons.Search in Google Scholar
Engle, R. F., and C. W. J. Granger. 1987. “Cointegration and Error Correction: Representation, Estimation, and Testing.” Econometrica March 55 (2):251–76.10.2307/1913236Search in Google Scholar
Evans, P., B. -H. Kim, and K. -Y. Oh. 2008. “Capital Mobility in Saving and Investment: A Time-Varying Coefficients Approach.” Journal of International Money and Finance September 27 (5):806–15.10.1016/j.jimonfin.2008.04.005Search in Google Scholar
Feldstein, M., and C. Horioka. 1980. “Domestic Saving and International Capital Flows.” The Economic Journal June 90 (358):314–29.10.3386/w0310Search in Google Scholar
Fleming, J. M. 1962. “Domestic Financial Policies Under Fixed and Under Floating Exchange Rates.” International Monetary Fund Staff Papers November 9 (3):369–80.10.2307/3866091Search in Google Scholar
Gregory, A. W., and B. E. Hansen. 1996. “Residual-Based Tests for Cointegration in Models with Regime Shifts.” Journal of Econometrics January 70 (1):99–126.10.1016/0304-4076(69)41685-7Search in Google Scholar
Gulley, O. D. 1992. “Are Saving and Investment Cointegrated? Another Look at the Data.” Economics Letters 39 (1):55–8.10.1016/0165-1765(92)90101-4Search in Google Scholar
Hansen, B. E. 1992. “Testing for Parameter Instability in Linear Models.” Journal of Policy Modeling August 14 (4):517–33.10.1016/0161-8938(92)90019-9Search in Google Scholar
Hansen, B. E. 1997. “Approximate Asymptotic P Values for Structural-Change Tests.” Journal of Business & Economic Statistics January 15 (1): 60–7.10.2307/1392074Search in Google Scholar
Hansen, B. E. 2001. “The New Econometrics of Structural Change: Dating Breaks in U.S. Labor Productivity.” The Journal of Economic Perspectives Autumn 15 (4):117–28.10.1257/jep.15.4.117Search in Google Scholar
Hansen, B. E., and P. C. B. Phillips. 1990. “Estimation and Inference in Models of Cointegration.” In Advances in Econometrics, Volume 8, edited by T. B. Fomby and G. F. Rhodes, Jr225–48. CT: JAI Press.Search in Google Scholar
Hansen, H., and S. Johansen. 1993. “Recursive Estimation in Cointegrated VAR-Models, Manuscript.” Institute of Mathematical Sciences, University of Copenhagen.Search in Google Scholar
Hansen, H., and S. Johansen. 1999. “Some Tests for Parameter Constancy in Cointegrated VAR-Models.” The Econometrics Journal December 2 (2):306–33.10.1111/1368-423X.00035Search in Google Scholar
Hansen, L. P. 1982. “Large Sample Properties of Generalized Method of Moments Estimators.” Econometrica July 50 (4):1029–54.10.2307/1912775Search in Google Scholar
Hoffmann, M. 2004. “International Capital Mobility in the Long Run and the Short Run: Can We Still Learn from Saving–Investment Data?” Journal of International Money and Finance 23 (1):113–31.10.1016/j.jimonfin.2003.08.006Search in Google Scholar
Inder, B. 1993. “Estimating Long-Run Relationships in Economics: A Comparison of Different Approaches.” Journal of Econometrics May-June 57 (1–3):53–68.10.1016/0304-4076(93)90058-DSearch in Google Scholar
Jansen, W. J. 1996. “Estimating Saving-Investment Correlations: Evidence for OECD Countries Based on An Error Correction Model.” Journal of International Money and Finance 15 (5):749–81.10.1016/0261-5606(96)00034-4Search in Google Scholar
Jarque, C. M., and Bera, A. K.. 1980. “Efficient Tests for Normality, Homoscedasticity and Serial Independence of Regression Residuals.” Economics Letters 6 (3): 255–59.10.1016/0165-1765(80)90024-5Search in Google Scholar
Johansen, S. 1991. “Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models.” Econometrica November 59 (6):1551–80.10.2307/2938278Search in Google Scholar
Johansen, S. 2000. “A Bartlett Correction Factor for Tests on the Cointegrating Relations.” Econometric Theory October 16 (5):740–78.10.1017/S0266466600165065Search in Google Scholar
Johansen, S. 2002. “A Small Sample Correction for the Test of Cointegrating Rank in the Vector Autoregressive Model.” Econometrica September 70 (5):1929–61.10.1111/1468-0262.00358Search in Google Scholar
Johansen, S., R. Mosconi, and B. Nielsen. 2000. “Cointegration Analysis in the Presence of Structural Breaks in the Deterministic Trend.” The Econometrics Journal December 3 (2):216–49.10.1111/1368-423X.00047Search in Google Scholar
Kejriwal, M. 2008. “Cointegration with Structural Breaks: An Application to the Feldstein-Horioka Puzzle.” Studies in Nonlinear Dynamics & Econometrics 12 (1) Article 3:March:1–37.10.2202/1558-3708.1467Search in Google Scholar
Kejriwal, M., and P. Perron. 2008. “The Limit Distribution of the Estimates in Cointegrated Regression Models with Multiple Structural Changes.” Journal of Econometrics September 146 (1):59–73.10.1016/j.jeconom.2008.07.001Search in Google Scholar
Kejriwal, M., and P. Perron. 2010. “Testing for Multiple Structural Changes in Cointegrated Regression Models.” Journal of Business & Economic Statistics October 28 (4):503–22.10.1198/jbes.2009.07220Search in Google Scholar
Kremers, J. J. M., N. R. Ericsson, and J. J. Dolado. 1992. “The Power of Cointegration Tests.” Oxford Bulletin of Economics & Statistics August 54 (3):325–48.10.1111/j.1468-0084.1992.tb00005.xSearch in Google Scholar
Kwiatkowski, D., P. C. B. Phillips, P. Schmidt, and Y. Shin. 1992. “Testing the Null Hypothesis of Stationarity Against the Alternative of a Unit Root: How Sure Are We That Economic Time Series Have a Unit Root?” Journal of Econometrics October-December 54 (1–3):159–78.10.1016/0304-4076(92)90104-YSearch in Google Scholar
Leachman, L. L. 1991. “Saving, Investment and Capital Mobility among OECD Countries.” Open Economies Review 2 (2):137–63.10.1007/BF01886897Search in Google Scholar
Lee, J., and M. C. Strazicich. 2003. “Minimum Lagrange Multiplier Unit Root Test with Two Structural Breaks.” The Review of Economics and Statistics November 85 (4):1082–9.10.1162/003465303772815961Search in Google Scholar
Lee, J., and Strazicich, M.C.. 2004. “Minimum LM Unit Root Test with One Structural Break.” Working Paper, No 04-17, Department of Economics, Appalachian State University.Search in Google Scholar
Levy, D. 2000. “Investment–Saving Co-movement and Capital Mobility: Evidence from Century Long U.S. Time Series.” Review of Economic Dynamics January 3 (1):100–36.10.1006/redy.1999.0060Search in Google Scholar
Liu, J., S. Wu, and J. V. Zidek. 1997. “On Segmented Multivariate Regression.” Statistica Sinica April 7 (2):497–525.Search in Google Scholar
Ljung, G. M., and G. E. P. Box. 1978. “On a Measure of Lack of Fit in Time Series Models.” Biometrika August 65 (2):297–303.10.1093/biomet/65.2.297Search in Google Scholar
Lumsdaine, R. L., and D. H. Papell. 1997. “Multiple Trend Breaks and the Unit-Root Hypothesis.” The Review of Economics and Statistics May 79 (2):212–18.10.1162/003465397556791Search in Google Scholar
Miller, S. M. 1988. “Are Saving and Investment Co-integrated?” Economics Letters 27 (1):31–4.10.1016/0165-1765(88)90215-7Search in Google Scholar
Moreno, R. 1997. “Saving-Investment Dynamics and Capital Mobility in the US and Japan.” Journal of International Money and Finance 16 (6):837–63.10.1016/S0261-5606(97)00040-5Search in Google Scholar
Mundell, R. A. 1962. “The Appropriate Use of Monetary and Fiscal Policy for Internal and External Stability.” International Monetary Fund Staff Papers March 9 (1):70–9.10.2307/3866082Search in Google Scholar
Mundell, R. A. 1963. “Capital Mobility and Stabilisation Policy under Fixed and Flexible Exchange Rates.” Canadian Journal of Economics and Political Science November 29 (4):475–85.10.2307/139336Search in Google Scholar
Nell, K. S., and L. D. Santos. 2008. “The Feldstein–Horioka Hypothesis versus the Long-Run Solvency Constraint Model: A Critical Assessment.” Economics Letters 98 (1):66–70.10.1016/j.econlet.2007.04.007Search in Google Scholar
Newey, W. K., and K. D. West. 1987. “A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix.” Econometrica May 55 (3):703–8.10.2307/1913610Search in Google Scholar
Ng, S., and P. Perron. 2001. “Lag Length Selection and the Construction of Unit Root Tests with Good Size and Power.” Econometrica November 69 (6):1519–54.10.1111/1468-0262.00256Search in Google Scholar
Nicolau, J. 2002. “Stationary Processes That Look Like Random Walks: The Bounded Random Walk Process in Discrete and Continuous Time.” Econometric Theory February 18 (1):99–118.10.1017/S0266466602181060Search in Google Scholar
Obstfeld, M., and K. Rogoff. 2000. “The Six Major Puzzles in International Macroeconomics: Is There A Common Cause?” In NBER Macroeconomics Annual 2000, edited by B. S. Bernanke and K. Rogoff, 339–90. The MIT Press, Cambridge, MA, U.S.A.: National Bureau of Economic Research.10.1086/654423Search in Google Scholar
Obstfeld, M. 2010. “The Immoderate World Economy.” Journal of International Money and Finance June 29 (4):603–14.10.1016/j.jimonfin.2010.01.006Search in Google Scholar
Park, J. Y., and P. C. B. Phillips. 1988. “Statistical Inference in Regressions with Integrated Processes: Part 1.” Econometric Theory December 4 (3):468–97.10.1017/S0266466600013402Search in Google Scholar
Perron, P. 1989. “The Great Crash, the Oil Price Shock, and the Unit Root Hypothesis.” Econometrica November 57 (6):1361–401.10.2307/1913712Search in Google Scholar
Phillips, P. C. B. 1991. “Optimal Inference in Cointegrated Systems.” Econometrica March 59 (2):283–306.10.2307/2938258Search in Google Scholar
Phillips, P. C. B., and B. E. Hansen. 1990. “Statistical Inference in Instrumental Variables Regression with I(1) Processes.” The Review of Economic Studies January 57 (189):99–125.10.2307/2297545Search in Google Scholar
Phillips, P. C. B., and M. Loretan. 1991. “Estimating Long-Run Economic Equilibria.” The Review of Economic Studies May 58 (195):407–36.10.2307/2298004Search in Google Scholar
Phillips, P. C. B., and P. Perron. 1988. “Testing for a Unit Root in Time Series Regression.” Biometrika June 75 (2):335–46.10.1093/biomet/75.2.335Search in Google Scholar
Phillips, P. C. B., and S. Ouliaris. 1990. “Asymptotic Properties of Residual Based Tests for Cointegration.” Econometrica January 58 (1):165–93.10.2307/2938339Search in Google Scholar
Quandt, R. E. 1960. “Tests of the Hypothesis that a Linear Regression System Obeys Two Separate Regimes.” Journal of the American Statistical Association June 55 (290): 324–30.10.1080/01621459.1960.10482067Search in Google Scholar
Saikkonen, P. 1991. “Asymptotically Efficient Estimation of Cointegration Regressions.” Econometric Theory March 7 (1):1–21.10.1017/S0266466600004217Search in Google Scholar
Sargan, J. D. 1958. “The Estimation of Economic Relationships Using Instrumental Variables.” Econometrica July 26 (3):393–415.10.2307/1907619Search in Google Scholar
Sargan, J. D., and A. Bhargava. 1983. “Testing Residuals from Least Squares Regression for Being Generated By the Gaussian Random Walk.” Econometrica January 51 (1):153–74.10.2307/1912252Search in Google Scholar
Sarno, L., and M. P. Taylor. 1998a. “Savings–Investment Correlations: Transitory Versus Permanent.” Manchester School 66 (S):17–38.10.1111/1467-9957.66.s.2Search in Google Scholar
Sarno, L., and M. P. Taylor. 1998b. “Exchange Controls, International Capital Flows and Saving-Investment Correlations in the UK: An Empirical Investigation.” Review of World Economics 134 (1):69–98.10.1007/BF02707579Search in Google Scholar
Schmidt, M. B. 2003. “U.S. Saving and Investment: Policy Implications.” Open Economies Review 14 (4):381–95.10.1023/A:1025312810428Search in Google Scholar
Sims, C. A. 1980. “Macroeconomics and Reality.” Econometrica January 48 (1):1–48.10.2307/1912017Search in Google Scholar
Singh, T. 2007. “Intertemporal Optimizing Models of Trade and Current Account Balance: A Survey.” Journal of Economic Surveys February 21 (1):25–64.10.1111/j.1467-6419.2007.00270.xSearch in Google Scholar
Staiger, D., and J. H. Stock. 1997. “Instrumental Variables Regression with Weak Instruments.” Econometrica May 65 (3):557–86.10.3386/t0151Search in Google Scholar
Stock, J. H., and M. W. Watson. 1989. “Interpreting the Evidence on Money-Income Causality.” Journal of Econometrics January 40 (1):161–81.10.3386/w2228Search in Google Scholar
Stock, J. H., and M. W. Watson. 1993. “A Simple Estimator of Cointegrating Vectors in Higher Order Integrated Systems.” Econometrica July 61 (4):783–820.10.2307/2951763Search in Google Scholar
Toda, H. Y., and T. Yamamoto. 1995. “Statistical Inference in Vector Autoregressions with Possibly Integrated Processes.” Journal of Econometrics March-April 66 (1–2):225–50.10.1016/0304-4076(94)01616-8Search in Google Scholar
Xiao, Z. 1999. “A Residual Based Test for the Null Hypothesis of Cointegration.” Economics Letters August 64 (2):133–41.10.1016/S0165-1765(99)00079-8Search in Google Scholar
Xiao, Z., and P. C. B. Phillips. 2002. “A Cusum Test for Cointegration Using Regression Residuals.” Journal of Econometrics May 108 (1):43–61.10.1016/S0304-4076(01)00103-8Search in Google Scholar
Yao, Y. -C. 1988. “Estimating the Number of Change-Points Via Schwarz Criterion.” Statistics & Probability Letters February 6 (3):181–9.10.1016/0167-7152(88)90118-6Search in Google Scholar
Zivot, E., and D. W. K. Andrews. 1992. “Further Evidence on the Great Crash, the Oil-Price Shock, and the Unit-Root Hypothesis.” Journal of Business & Economic Statistics July 10 (3):251–70.Search in Google Scholar
©2016 by De Gruyter