A. Is a Consumption Boom Responsible for the Decline in the Private Saving Rate in Colombia?

Most economists attribute the recent fall in private saving in Colombia to a consumption boom. The literature cites in support of this argument four main factors. First, Urrutia and López (1994, 1995) attribute the consumption boom to the relaxation of liquidity constraints brought on by capital inflows and financial liberalization. Casual evidence also suggests that capital inflows and financial liberalization triggered a boom in asset prices in Colombia, producing a positive wealth effect and a decline in saving. Second, Steiner (1995) argues that the monetary policy pursued at the beginning of the 1990s was responsible for the consumption boom. He argues that in 1992, when the monetary stance was modified because of the costs and limits of a two-year sterilization policy, interest rates were drastically reduced, international reserves began to be intermediated by the financial system, and credit to the private sector increased, leading to the consumption boom. Third, Echeverry (1996) argues that the consumption boom was caused by trade liberalization, which, combined with the appreciation of the peso experienced in the first half of the 1990s, reduced the relative price of durable goods. Fourth, Cárdenas and Escobar (1997), among others, explain that the increase in consumption was a reassessment of permanent income created by the oil bonanza and the expectations associated with the structural reforms.

However, attributing the recent fall in private saving mostly to a consumption boom is inconsistent with the facts presented by the National Accounts. Although in the 1990s private consumption increased, the increase had a relatively minor effect on private saving. This can be seen in Table 2, which shows five -year averages of the adjusted private saving rate, quantifies the changes in the saving rate as a percentage of GNP, and identifies the causes behind the changes. The data shows that in the first half of the 1990s, the private saving rate deteriorated 4.3 percent of GNP and private consumption increased 0.6 percent of GNP. Thus, the increase in private consumption only explains 14 percent of the decline in the private saving rate.

Table 2.

Accounting for the Change in Private Saving

Source: National Accounts and authors calculations. In the National Accounts saving is a current account measure. To have a capital account measure of saving consistent with consumption theory, the table adjusts the current account measure by net capital gains from revaluation of assets and liabilities because of inflation and changes in the real exchange rate. For a detailed derivation of the adjusted measures, see Arrau and Oks (1992) and Schmidt-Hebbel and Servén (1996). For the case of Colombia, see Lόpez (1997).

Table 2.

Accounting for the Change in Private Saving

Period	Adjusted Saving Rate as Percentage	Movements in Private Saving (Percentage of GNP)
	of GNP	Change	Explanations
			Consumption	Capital Gains	Disposable Income
					Tot.	GOS	TAX	Salaries	Other
1956-60	14.0	4.3	-3.3	0.6	0.4	0.7	0.1	-0.1	-0.3
1961-65	12.0	-2.0	2.3	-1.1	1.4	-2.7	0.6	3.6	-0.1
1966-70	13.0	1.0	-1.5	3.0	-3.5	-1.2	-3.1	0.7	0.1
1971-75	13.4	0.4	-1.9	-0.6	-0.9	3.7	-0.6	1.0	-5.0
1976-80	14.5	1.1	-2.4	0.5	-1.8	-3.8	-1.7	0.7	3.0
1981-85	13.6	-0.9	2.0	-0.8	1.9	-4.8	2.0	3.8	0.9
1986-90	16.7	3.1	-5.2	0.2	-2.3	-0.3	-2.5	-4.8	5.3
1991-95	12.4	-4.3	0.6	1.6	-5.4	-4.5	-1.9	1.2	-0.2

Source: National Accounts and authors calculations. In the National Accounts saving is a current account measure. To have a capital account measure of saving consistent with consumption theory, the table adjusts the current account measure by net capital gains from revaluation of assets and liabilities because of inflation and changes in the real exchange rate. For a detailed derivation of the adjusted measures, see Arrau and Oks (1992) and Schmidt-Hebbel and Servén (1996). For the case of Colombia, see Lόpez (1997).

View Table

Table 2.

Accounting for the Change in Private Saving

Period	Adjusted Saving Rate as Percentage	Movements in Private Saving (Percentage of GNP)
	of GNP	Change	Explanations
			Consumption	Capital Gains	Disposable Income
					Tot.	GOS	TAX	Salaries	Other
1956-60	14.0	4.3	-3.3	0.6	0.4	0.7	0.1	-0.1	-0.3
1961-65	12.0	-2.0	2.3	-1.1	1.4	-2.7	0.6	3.6	-0.1
1966-70	13.0	1.0	-1.5	3.0	-3.5	-1.2	-3.1	0.7	0.1
1971-75	13.4	0.4	-1.9	-0.6	-0.9	3.7	-0.6	1.0	-5.0
1976-80	14.5	1.1	-2.4	0.5	-1.8	-3.8	-1.7	0.7	3.0
1981-85	13.6	-0.9	2.0	-0.8	1.9	-4.8	2.0	3.8	0.9
1986-90	16.7	3.1	-5.2	0.2	-2.3	-0.3	-2.5	-4.8	5.3
1991-95	12.4	-4.3	0.6	1.6	-5.4	-4.5	-1.9	1.2	-0.2

Source: National Accounts and authors calculations. In the National Accounts saving is a current account measure. To have a capital account measure of saving consistent with consumption theory, the table adjusts the current account measure by net capital gains from revaluation of assets and liabilities because of inflation and changes in the real exchange rate. For a detailed derivation of the adjusted measures, see Arrau and Oks (1992) and Schmidt-Hebbel and Servén (1996). For the case of Colombia, see Lόpez (1997).

This is an important result. Since the recent fall of the private saving rate is not associated with significant increases in the share of private consumption in GNP, the changes in the private saving rate during the 1990s have had little effect on the level and composition of aggregate demand.³ Still, the increase in private consumption certainly played a role in the decline of private saving. It is thus important to understand the determinants of consumption in Colombia.

B. Understanding the Determinants of Private Consumption in Colombia

The regressions reported in Table 3 include as explanatory variables permanent income, wealth, taxes, permanent government consumption,⁴ a proxy of financial liberalization (M2/GNP), the real interest rate (r), and a dummy variable (D). The dummy variable is 1 in the 1990s and 0 in previous years.⁵ This variable should capture the effects on consumption of the reforms of the early 1990s, if they are not fully captured in the measure of future income and in the proxy of financial liberalization. Appendix I provides relevant information on the data (i.e., descriptive statistics and the correlation matrix of the regression variables).

Table 3.

Nondurables consumption: 1966–94^1/

^1/All variables are in logs and in real terms except m2/gnp.

The variables are defined as follows: y1p is permanent labor income; ypp is permanent private income; ypap is permanent private income adjusted by capital gains; gp is permanent public consumption; r is the real interest rate; D are step dummies that are 1 in the 1990s.The standard deviations are in parenthesis.

^**The coefficient is significant at the 1 percent level.

^*The coefficient is significant at the 5 percent level.

Table 3.

Nondurables consumption: 1966–94^1/

Variables	Eq.1	Eq.2	Eq.3	Eq.4	Eq.5	Eq.6
Constant	1,871 [0.517]	1,8 [0.563]	2,299 [0.59]**	2,064 [0.618]	3,825 [0.457]**	3,687 [0.446]**
Log(ylp)	0,623 [0.069]**	0,588 [0.078]**
Log(ypp)			0,708 [0.099]**	0,736 [0.115]**
Log(ypap)					0,586 [0.065]**	0,593 [0.074]**
Log(gp)	-0,124 [0.043]**	-0,11 [0.049]**	0,11 [0.037]**	0,1 [0.038]**	0,182 [0.031]**	0,178 [0.031]**
Log(tx)	0,0359 [0.028]	0,043 [0.029]	-0,03 [0.036]	-0,033 [0.038]	-0,014 [0.029]	-0,014 [0.03]
Log(w)	0,245 [0.062]**	0,266 [0.064]**	-0,032 [0.087]	-0,027 [0.086]	-0,138 [0.079]	-0,123 [0.079]
m2/gnp	0,024 [0.095]	0,015 [0.10]	0,23 [0.10]	0,197 [0.103]	0,391 [0.081]**	0,369 [0.079]**
r-1	-0,07 [0.071]	-0,086 [0.074]	0,038 [0.085]	0,034 [0.085]	0,169 [0.076]*	0,166 [0.077]*
D1990	0,019 [0.011]		0,017 [0.013]		0,014 [0.011]
D1991		0,008 [0.012]		0,019 [0.015]		0,013 [0.011]
R2	0,99	0,99	0,99	0,99	0,99	0,99
F[7,22]=	581.3**	520.77**	417.69**	414.57**	580.4**	569.85**
DW	1,17	1,12	1,4	1,39	1,87	1,86
RSS	0,004	0,0045	0,006	0,006	0,004	0,004

^1/All variables are in logs and in real terms except m2/gnp.

The variables are defined as follows: y1p is permanent labor income; ypp is permanent private income; ypap is permanent private income adjusted by capital gains; gp is permanent public consumption; r is the real interest rate; D are step dummies that are 1 in the 1990s.The standard deviations are in parenthesis.

^**The coefficient is significant at the 1 percent level.

^*The coefficient is significant at the 5 percent level.

View Table

Table 3.

Nondurables consumption: 1966–94^1/

Variables	Eq.1	Eq.2	Eq.3	Eq.4	Eq.5	Eq.6
Constant	1,871 [0.517]	1,8 [0.563]	2,299 [0.59]**	2,064 [0.618]	3,825 [0.457]**	3,687 [0.446]**
Log(ylp)	0,623 [0.069]**	0,588 [0.078]**
Log(ypp)			0,708 [0.099]**	0,736 [0.115]**
Log(ypap)					0,586 [0.065]**	0,593 [0.074]**
Log(gp)	-0,124 [0.043]**	-0,11 [0.049]**	0,11 [0.037]**	0,1 [0.038]**	0,182 [0.031]**	0,178 [0.031]**
Log(tx)	0,0359 [0.028]	0,043 [0.029]	-0,03 [0.036]	-0,033 [0.038]	-0,014 [0.029]	-0,014 [0.03]
Log(w)	0,245 [0.062]**	0,266 [0.064]**	-0,032 [0.087]	-0,027 [0.086]	-0,138 [0.079]	-0,123 [0.079]
m2/gnp	0,024 [0.095]	0,015 [0.10]	0,23 [0.10]	0,197 [0.103]	0,391 [0.081]**	0,369 [0.079]**
r-1	-0,07 [0.071]	-0,086 [0.074]	0,038 [0.085]	0,034 [0.085]	0,169 [0.076]*	0,166 [0.077]*
D1990	0,019 [0.011]		0,017 [0.013]		0,014 [0.011]
D1991		0,008 [0.012]		0,019 [0.015]		0,013 [0.011]
R2	0,99	0,99	0,99	0,99	0,99	0,99
F[7,22]=	581.3**	520.77**	417.69**	414.57**	580.4**	569.85**
DW	1,17	1,12	1,4	1,39	1,87	1,86
RSS	0,004	0,0045	0,006	0,006	0,004	0,004

^1/All variables are in logs and in real terms except m2/gnp.

The variables are defined as follows: y1p is permanent labor income; ypp is permanent private income; ypap is permanent private income adjusted by capital gains; gp is permanent public consumption; r is the real interest rate; D are step dummies that are 1 in the 1990s.The standard deviations are in parenthesis.

^**The coefficient is significant at the 1 percent level.

^*The coefficient is significant at the 5 percent level.

Nondurable consumption in Colombia is mainly explained by permanent income. The coefficient on permanent income varies from 0.588 to 0.736, depending on the regression analyzed (Table 3). In the 1990s, permanent income seems to be the only variable capturing the “euphoria” factor attributed to the structural reforms undertaken during the decade. In all the regressions, the dummy variable is insignificant. Consequently, until 1995, concerns regarding the sustainability of the balance of payments could not be attributed to private consumption. Rather, the balance of payments deficit that could be attributed to private consumption was mostly reflecting private sector’s borrowing against future output and/or the decumulation of previously accumulated assets (See Herrera, 1996, for a similar result).

As expected, wealth is important in explaining consumption behavior. And, when significant, the coefficient on wealth is positive and varies between 0.245 and 0.266. With regard to financial liberalization, some regressions show that relaxing liquidity constraints has a positive and significant effect on consumption. Still, this effect is not present only in the 1990s. Considerable evidence suggests that financial liberalization has been a slow and halting process in Colombia (see, among others, Ortega, 1982; Herrera, 1988).

The results also suggest that during the 1990s fiscal policy has had a negative impact on private saving in two ways.⁶ First, when significant, real interest rates have a positive effect on private consumption. And the high real interest rates of the 1990s (Table 1) could be associated with public consumption, which increased from 10 percent of GNP in 1990 to 15 percent of GNP in 1995. Second, evidence shows that private and public saving rise or fall together. In four out of six regressions, the coefficient of permanent public consumption is positive and significant, varying between 0.10 and 0.185. In this regard, the fall in the private saving rate could be explained by the rise in public consumption if it could be shown that public consumption complements private consumption.

A. Corporate Behavior: Stylized Facts and Policy Implications

Sánchez et. al (1996) conclude that in the 1990s Colombia’s corporate saving declined because of a fall in corporate profits associated with fiercer external competition brought about by trade reform. This hypothesis is appealing because it is consistent with the stylized facts presented by the National Accounts. In the early 1990s the stability of the private saving rate was disrupted when corporate saving declined sharply—from 50 percent of private saving during 1985–89, to 24 percent of private saving in 1990–95 (Figure 3). The decline was explained mainly by a deterioration of the gross operating surplus (GOS) generated by corporations. Consequently, and also as a result of tax increases and declines in the GOS generated by households,⁷ private disposable income fell 5.4 percent of GNP, explaining 126 percent of the decline in the private saving rate in the 1990s (Table 2).

Components of Private Saving

Percentage of GNP

Citation: IMF Working Papers 1998, 171; 10.5089/9781451858471.001.A001

Source: National Accounts

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Components of Private Saving

Percentage of GNP

Citation: IMF Working Papers 1998, 171; 10.5089/9781451858471.001.A001

Source: National Accounts

• Download Figure
• Download figure as PowerPoint slide

Components of Private Saving

Percentage of GNP

Citation: IMF Working Papers 1998, 171; 10.5089/9781451858471.001.A001

Source: National Accounts

• Download Figure
• Download figure as PowerPoint slide

Still, with the exception of the period 1993–95, declines (increases) in corporate saving have been usually compensated by increases (declines) in household saving, reflecting the fact that firms are ultimately owned by households (Figure 3). This suggests that the distinction between household saving and firm saving in Colombia is, to some extent, arbitrary: if firms retain more earnings, households will save less by a corresponding amount. In other words, households “pierce” the corporate veil, so that total private saving (the sum of household and corporate saving) is basically determined by household behavior.

There are several implications for fiscal policy when households pierce the corporate veil. First, a system of corporate income taxation, which induces corporations to retain or distribute earnings, would have little effect on aggregate level of capital formation (Feldstein, 1973; Poterba, 1991). Second, tax incentives, such as lower corporate tax rates, would reduce household saving and thus would not stimulate total private saving and investment. Third, in place of the first two, another option could eventually be advocated: a “fully integrated” system of corporate and personal income taxation. In this arrangement, the incomes of all entities would be attributed to the individuals who are their ultimate owners, and income tax would be collected from owners at the relevant rates, depending on their total income.⁸ It is thus important to test the absence of a corporate veil.

B. Testing the Absence of a Corporate Veil

A possible way to test the absence of a corporate veil is to study household behavior when firm’s perspective earnings are unchanged but policy on dividends is modified. If firm’s expected profits are unchanged but higher dividends are paid, corporate saving fall. In this case, the corporate veil is pierced if there is an increase in household saving equal to the decline in corporate saving. This behavior occurs if dividends do not increase consumption—agents realize that their higher income is offset by a decline of the same amount in the firm’s value (i.e., their wealth).

To construct a test that indicates consumer behavior after a change in dividend policy, it is useful to start from Hall (1978). Assuming a representative agent seeking to maximize the expected discounted utility function of consumption, a CES utility function with an intertemporal elasticity of consumption σ, and a discount rate of time preferences δ, the optimal consumption path must follow the following Euler equation:

where E is the expectations operator, C_t is real per capita consumption, and r is the after tax rate of return to savings.

Rewriting the expectation in stochastic form, applying logarithms to both sides of the equation, and renaming the log (C_t/C_t−1) by Δ c_t, it is possible to obtain:

where µ = −σlog[l+δ/l+r] and ε_t can be interpreted as the innovation or surprise in permanent income. This equation implies that no variable other than current consumption could be significant in predicting future consumption.

The permanent income hypothesis under rational expectations (PIREH) has been rejected several times in the literature (e.g, Flavin, 1981; Deaton, 1987). The existence of imperfect capital markets is perhaps the most common explanation for the failure of the PIREH. If some agents are liquidity constrained, consumption will be excessively sensitive to current income—anticipated changes in income have some explanatory value when modeling consumption. Using this evidence, dividing consumers between liquidity and non—liquidity constrained, and assuming that liquidity constrained consumers spend their entire income, equation 2 takes the following form (Campbell and Mankiw, 1989):

where Δ y^e_t is the logarithm of expected income growth and λ is related to the percentage of private income accrued by liquidity constrained households. Consequently, if λ>0, the PIREH is rejected—and λ should be higher in countries where capital markets are more imperfect (Jappelli and Pagano, 1989).

To test the absence of a corporate veil, expected income growth of liquidity constrained consumers should be divided between expected change in dividends and expected change in income net of dividends (Auerbach and Hasset, 1990). In other words, the test is:

where Δ d^e_t and Δynd^e_t are expected changes in dividends and income net of dividends of liquidity constrained consumers, respectively. The coefficient η is related to the percentage of dividends accrued by liquidity constrained consumers. If the corporate veil is pierced, η = 0. Changes in expected dividends should not affect consumption—agents realize that their higher (lower) income is offset by a decline (increase) of the same amount in the firm’s value (i.e., their wealth).

Two points should be highlighted about this test. First, the test is biased toward rejecting η = 0. Indeed, absent liquidity constraints, η would be zero unless the corporate veil exists. However, if liquidity constraints are binding, expected dividends would affect the consumption of liquidity constrained households: households could relax their constraints by selling their stock. Consequently, the coefficient η would be different from zero even if agents pierce the corporate veil. Second, the test indicates the degree of capital concentration. If there is high capital concentration, most stocks should be held by non—liquidity constrained agents, implying that η should be small when capital concentration is high.

C. Is the Corporate Veil Pierced in Colombia?

It is likely that Colombian households have pierced the corporate veil. The evidence is shown in Table 4. Each column in the table presents the results of the test in equation 4 using four definitions of income. The coefficient λ, which measures the response of consumption to changes in income, varies between 0.51 and 0.79, depending on the definition of income used, and is always significantly different from zero. Moreover, capital concentration seems to be very high in Colombia. Indeed, η varies between 0.03 and 0.05 when positive, although the coefficient is not significantly different from zero. This implies that changes in dividends have not affected consumption, indicating that households have pierced the corporate veil in Colombia.

Table 4.

Testing the Corporate Veil^1/

^1/The test was carried out using instrumental variables for the period 1974–95. The instruments used were second and third lags of adjusted private income (YPA), and third lags of dividends and real interest rates. T statistics are shown in parenthesis.

^2/YPA is adjusted private disposable income; YP is private disposable income as measured in the National Accounts; YD is households disposable income; YL is labor income net of taxes. In this test, these four variables are net of dividends. Appendix II has a detailed definition of the variables.

^3/λ is the coefficient measuring the response of consumption to changes in income; η is a coefficient related to the percentage of dividends accrued by liquidity constrained consumers; σ is the standard error of the regression; RSS is the residual sum of squares.

Table 4.

Testing the Corporate Veil^1/

	Income2/
Coefficients and Statistics3/	YPA	YP	YD	YL
λ	0.507 (1.604)	0.647 (2.278)	0.794 (2.053)	0.576 (3.245)
η	0.050	0.036	-0.091	0.030
Constant	0.007 (1.425)	0.009 (1.417)	0.005 (0.534)	0.005 (0.832)
σ	0.018	0.016	0.016	0.015
Residual Sum of Squares	0.005	0.004	0.004	0.004
Sargan Specification Test χ2 (2)	6.443	5.291	6.340	1.049
Testing B= 0 χ2 (2)	4.218	7.171	6.179	12.863
ARE(1) χ1	0.114	0.114	0.547	0.033
Heteroscedasticity F (4.12)	0.609	1.0621	2.123	0.487
Normality χ2 (2)	0.520	0.468	0.196	0.846

^1/The test was carried out using instrumental variables for the period 1974–95. The instruments used were second and third lags of adjusted private income (YPA), and third lags of dividends and real interest rates. T statistics are shown in parenthesis.

^2/YPA is adjusted private disposable income; YP is private disposable income as measured in the National Accounts; YD is households disposable income; YL is labor income net of taxes. In this test, these four variables are net of dividends. Appendix II has a detailed definition of the variables.

^3/λ is the coefficient measuring the response of consumption to changes in income; η is a coefficient related to the percentage of dividends accrued by liquidity constrained consumers; σ is the standard error of the regression; RSS is the residual sum of squares.

View Table

Table 4.

Testing the Corporate Veil^1/

	Income2/
Coefficients and Statistics3/	YPA	YP	YD	YL
λ	0.507 (1.604)	0.647 (2.278)	0.794 (2.053)	0.576 (3.245)
η	0.050	0.036	-0.091	0.030
Constant	0.007 (1.425)	0.009 (1.417)	0.005 (0.534)	0.005 (0.832)
σ	0.018	0.016	0.016	0.015
Residual Sum of Squares	0.005	0.004	0.004	0.004
Sargan Specification Test χ2 (2)	6.443	5.291	6.340	1.049
Testing B= 0 χ2 (2)	4.218	7.171	6.179	12.863
ARE(1) χ1	0.114	0.114	0.547	0.033
Heteroscedasticity F (4.12)	0.609	1.0621	2.123	0.487
Normality χ2 (2)	0.520	0.468	0.196	0.846

^1/The test was carried out using instrumental variables for the period 1974–95. The instruments used were second and third lags of adjusted private income (YPA), and third lags of dividends and real interest rates. T statistics are shown in parenthesis.

^2/YPA is adjusted private disposable income; YP is private disposable income as measured in the National Accounts; YD is households disposable income; YL is labor income net of taxes. In this test, these four variables are net of dividends. Appendix II has a detailed definition of the variables.

^3/λ is the coefficient measuring the response of consumption to changes in income; η is a coefficient related to the percentage of dividends accrued by liquidity constrained consumers; σ is the standard error of the regression; RSS is the residual sum of squares.

Why is η not different from zero if there are liquidity constraints? There are two explanations. First, η could still be different from zero, but the test may not have been able to capture it because the regression was carried out using a small sample size.⁹ This possibility would be increased if there were signs of misspecification, but this is not the case. As the last three rows of Table 4 suggest, the errors in the regression are not serially correlated, are homoscedastic, and appear to be normal.¹⁰ Second, if η is zero and there are liquidity constraints, it could be an indication that capital concentration in Colombia is so high that almost all stocks are held by non—liquidity constrained agents.

The second explanation suggests that the corporate veil can be pierced not only when agents have perfect information and rational expectations, but also when saving is carried out by the same agents who own and manage most of the capital stock (i.e., when capital concentration is high). This is likely to be the case in Colombia, where capital concentration is among the highest in the world, stock transactions represent a small percentage of GDP, and a concentration index measuring the main ten stocks as percentage of total transactions varies between 65 percent and 80 percent (Misión del Mercado de Capitales, 1996). Moreover, it is likely that this capital is owned by the very few households that save. Indeed, two thirds of total private saving is carried out by 5 percent of household heads, while the 77 percent less wealthy individuals save nothing at all (Ramirez, 1992).¹¹

Besides the implications for fiscal policy mentioned in section III.A, the finding that households pierce the corporate veil provides support for the tax measures undertaken during the last decade in Colombia. The finding indicates that total capital accumulation was not reduced when several distortions affecting firms financing decisions were eliminated. In particular, when the following measures were taken to reduce disincentives to issuing stocks: removal of the double taxation of dividends, gradual elimination of the deductible inflationary component of interest, removal of property tax on stocks, and a significant reduction of capital gain taxes on stocks. Still, in Colombia only large firms have access to the stock market. It is thus possible to conclude that reducing disincentives to the issuance of stocks favored mostly large corporations and might have increased capital concentration.¹²

A. Tax Policy and Private Saving: Interrelations and Stylized Facts

The recent fall in the private saving rate in Colombia has also been attributed to the increase in taxes in the 1990s. At a theoretical level, this hypothesis would be consistent with the Ricardian equivalence proposition, whereby fiscal policy would not be a matter of concern. According to this view, successive generations are closely linked through gifts and bequests, and individuals are assumed to be rational, forward looking, and not subject to liquidity constraints. If there is a fiscal deficit (surplus), Ricardian consumers will react today to future taxes (transfers) paid (received) by their descendants. Thus, changes in public saving are offset by opposite movements in private saving.

At an empirical level, most evidence for developing countries reject the pure form of Ricardian equivalence (Haque and Montiel, 1989; Corbo and Schmidt-Hebbel, 1991; Easterly, Rodríguez and Schmidt-Hebbel, 1994; Edwards, 1994, 1995). Although the evidence does not show conclusively by how much public saving is offset by private saving, the magnitude seems to be between 0.4 and 0.6 (see Schmidt-Hebbel, Servén, and Solimano, 1996, for a review of this literature). In case of Colombia, recent evidence suggests that private saving offsets approximately 0.5 of the change in public saving (Cárdenas and Escobar, 1997), but other results could not reject the validity of the Ricardian equivalence proposition (Carrasquilla and Rincón, 1990).

The stylized facts suggest that increases in public saving have been compensated to some extent by a fall in private saving since the late 1980s (Figure 4). Moreover, Table 5 shows that the increase in the public saving rate has been the result of a substantial rise in disposable income of the public sector caused, in part, by tax increments.¹³ It is thus relevant to examine the importance of tax increases in explaining the fall in private saving by using an econometric test.

National, Private and Public Saving

Percentage of GNP

Citation: IMF Working Papers 1998, 171; 10.5089/9781451858471.001.A001

Source: National Accounts

• Download Figure
• Download figure as PowerPoint slide

View Full Size

National, Private and Public Saving

Percentage of GNP

Citation: IMF Working Papers 1998, 171; 10.5089/9781451858471.001.A001

Source: National Accounts

• Download Figure
• Download figure as PowerPoint slide

National, Private and Public Saving

Percentage of GNP

Citation: IMF Working Papers 1998, 171; 10.5089/9781451858471.001.A001

Source: National Accounts

• Download Figure
• Download figure as PowerPoint slide

Table 5.

Accounting for the Change in Public Saving

Source: National Accounts and authors calculations. In the National Accounts saving is a current account measure. To have a capital account measure of saving consistent with consumption theory, this table adjusts the current account measure by net capital gains from revaluation of assets and liabilities due to inflation and changes in the real exchange rate. For a detailed derivation of the adjusted measures, see Arrau and Oks (1992) and Schmidt-Hebbel and Servén (1996). For the case of Colombia, see López (1997).

Table 5.

Accounting for the Change in Public Saving

Period	Adjusted Saving Rate (Percentage of GNP)	Movements in Public Saving (Percentage of GNP)
		Change	Explanations
			Consumption	Capital		Disposable Income
				Gains	Total	GOS	TAX	Other
1956-60	4.8	0.7	-0.5	0.7	-0.5	0.1	-0.1	-0.5
1961-65	2.1	-2.7	0.8	-0.5	-1.4	-0.1	-0.6	-0.7
1966-70	5.9	3.8	0.6	1.0	3.4	0.1	3.1	0.2
1971-75	4.8	-1.1	2.2	0.2	0.9	-0.5	0.6	0.8
1976-80	7.9	3.1	-0.8	0.4	1.9	-0.8	1.7	1.0
1981-85	1.0	-6.9	2.2	-3.0	-1.7	1.7	-2.0	-1.4
1986-90	3.8	2.8	-0.7	-0.1	2.2	2.5	2.5	-2.8
1991-95	11.6	7.8	2.1	3.7	6.2	1.0	1.9	3.3

Source: National Accounts and authors calculations. In the National Accounts saving is a current account measure. To have a capital account measure of saving consistent with consumption theory, this table adjusts the current account measure by net capital gains from revaluation of assets and liabilities due to inflation and changes in the real exchange rate. For a detailed derivation of the adjusted measures, see Arrau and Oks (1992) and Schmidt-Hebbel and Servén (1996). For the case of Colombia, see López (1997).

View Table

Table 5.

Accounting for the Change in Public Saving

Period	Adjusted Saving Rate (Percentage of GNP)	Movements in Public Saving (Percentage of GNP)
		Change	Explanations
			Consumption	Capital		Disposable Income
				Gains	Total	GOS	TAX	Other
1956-60	4.8	0.7	-0.5	0.7	-0.5	0.1	-0.1	-0.5
1961-65	2.1	-2.7	0.8	-0.5	-1.4	-0.1	-0.6	-0.7
1966-70	5.9	3.8	0.6	1.0	3.4	0.1	3.1	0.2
1971-75	4.8	-1.1	2.2	0.2	0.9	-0.5	0.6	0.8
1976-80	7.9	3.1	-0.8	0.4	1.9	-0.8	1.7	1.0
1981-85	1.0	-6.9	2.2	-3.0	-1.7	1.7	-2.0	-1.4
1986-90	3.8	2.8	-0.7	-0.1	2.2	2.5	2.5	-2.8
1991-95	11.6	7.8	2.1	3.7	6.2	1.0	1.9	3.3

Source: National Accounts and authors calculations. In the National Accounts saving is a current account measure. To have a capital account measure of saving consistent with consumption theory, this table adjusts the current account measure by net capital gains from revaluation of assets and liabilities due to inflation and changes in the real exchange rate. For a detailed derivation of the adjusted measures, see Arrau and Oks (1992) and Schmidt-Hebbel and Servén (1996). For the case of Colombia, see López (1997).

B. Testing Private Saving Response to Fiscal Policy

A possible way to test private saving response to fiscal policy is to study household behavior when tax changes occur but the stream of government spending is not modified. Under these conditions, tax increases will raise public saving. In this case, Ricardian equivalence is not rejected if private saving decline by a corresponding amount. This behavior occurs if taxes do not affect consumption—agents realize that their expected permanent income gross of taxes has not changed.¹⁴ Still, the timing of taxes will affect consumption if individuals have lower discount rates than Ricardian consumers, and/or a proportion of private saving is carried out by liquidity constrained agents.

A test to study consumer behavior after a tax change can be derived assuming a representative agent with rational behavior, a CES utility function, and households that incorporate the government budget constraint when selecting their optimal consumption path. The intertemporal maximization problem is the following:

where c_t is private real consumption and E is the expectations operator. If agents are faced with perfect capital markets and the government uses taxes to finance deficits, the consumer intertemporal budget constraint is:

for every state of the world and every t. According to the budget constraint, discounted future consumption is equal to households permanent labor income net of taxes plus their current wealth in assets, Wpt,¹⁶ and government bonds, Bt. Taxes, tx, include the inflation tax (Barro, 1979) and are net of government transfers, TR.

To test the Ricardian equivalence proposition, households must incorporate the government budget constraint:

indicating that present discounted total taxes net of transfers are equal to present discounted government consumption plus government current debt and wealth. Notice that the introduction of government’s wealth, Wg_t, brings the government budget constraint closer to the definition of fiscal deficit (i.e., public savings minus public investment, where public investment is measured as the change in public wealth).

Solving for consumption under Hall’s (1978) permanent income hypothesis model¹⁸ it is possible to get:

Multiplying by (l+r)^−s, substituting in and solving for c_t+s for every s larger than one, gives the transformed equation:

Substituting in the government budget constraint the equation changes as follows:

Adding private and public wealth, and assuming that government consumption and household labor income gross of taxes follow a random walk with drift, the model could be simplified to:²¹

This equation nests several hypothesis regarding households behavior after a tax change under a very simplified process for expected government expenditures and permanent income gross of taxes. The timing of taxes affects consumption when there are liquidity constraints (β4 < 0), and/or forward-looking consumers do not expect to bear the burden of deferred taxes (β1 < β2). In other words, Ricardian equivalence will not be rejected when liquidity constraints are absent (β4 = 0) and strong intergenerational links are present (β1 < β2).

C. What is the Response of Private Saving to Fiscal Policy in Colombia?

The collapse of private savings during 1989–95 cannot be entirely explained by tax increases. The timing of taxes matter; consequently, private savings do not completely offset public savings. To reach this conclusion several steps were taken.

First, equation 11 was log-linearized. Second, the order of integration for the variables were determined. Table 6 lists Dickey-Fuller statistics for private consumption, household disposable income gross of taxes,²² wealth, taxes, and government consumption. Unit root tests were given for the original variables (all in logs, in real terms, and per capita), and for their changes. According to the Dickey-Fuller statistics, all variables appear to be integrated of order 1, and the estimated roots for the levels are close to unity.

Table 6.

Statistics for Testing Unit Roots^1/

^1/For a variable x, the augmented Dickey-Fuller (1981) statistic ADF(k) is the ratio from the regression:

${\Delta }{x}_t={\pi }{x}_{t-1}+{\displaystyle \sum _{=}^k{{\Theta }}_t{\Delta }{x}_{t-i}+c+{\gamma }t+{\epsilon }_t}$

The critical values are those calculated by Mackinnon (1991). If there is no trend and the constant is included, the critical value at 5 percent is: −2.94, 1%= −3.612. The estimation period is 1958–95.

Table 6.

Statistics for Testing Unit Roots^1/

Null	Variable		Log C	Log Y	Log T	Log G	Log YD	Log W
I(1)		β	0.9813	0.9760	0.9542	0.9690	0.9603	0.9883
	t-adf(k)		(−1.0497)	(−1.1078)	(−1.1098)	(−1.1757)	(−1.2707)	(−1.0975)
	k		0	2	0	1	1	1
I(2)		β	-0.0369	-0.0187	-0.2265	0.2587	-0.4596	0.5041
	t-adf(k)		(−4.0956)**	(−5.8133)**	(−7.0759)**	(−4.6813)**	(−9.574)**	(−3.9046)**
	k		3	0	0	0	0	0

^1/For a variable x, the augmented Dickey-Fuller (1981) statistic ADF(k) is the ratio from the regression:

${\Delta }{x}_t={\pi }{x}_{t-1}+{\displaystyle \sum _{=}^k{{\Theta }}_t{\Delta }{x}_{t-i}+c+{\gamma }t+{\epsilon }_t}$

The critical values are those calculated by Mackinnon (1991). If there is no trend and the constant is included, the critical value at 5 percent is: −2.94, 1%= −3.612. The estimation period is 1958–95.

View Table

Table 6.

Statistics for Testing Unit Roots^1/

Null	Variable		Log C	Log Y	Log T	Log G	Log YD	Log W
I(1)		β	0.9813	0.9760	0.9542	0.9690	0.9603	0.9883
	t-adf(k)		(−1.0497)	(−1.1078)	(−1.1098)	(−1.1757)	(−1.2707)	(−1.0975)
	k		0	2	0	1	1	1
I(2)		β	-0.0369	-0.0187	-0.2265	0.2587	-0.4596	0.5041
	t-adf(k)		(−4.0956)**	(−5.8133)**	(−7.0759)**	(−4.6813)**	(−9.574)**	(−3.9046)**
	k		3	0	0	0	0	0

^1/For a variable x, the augmented Dickey-Fuller (1981) statistic ADF(k) is the ratio from the regression:

${\Delta }{x}_t={\pi }{x}_{t-1}+{\displaystyle \sum _{=}^k{{\Theta }}_t{\Delta }{x}_{t-i}+c+{\gamma }t+{\epsilon }_t}$

The critical values are those calculated by Mackinnon (1991). If there is no trend and the constant is included, the critical value at 5 percent is: −2.94, 1%= −3.612. The estimation period is 1958–95.

The third step was to test if the log-linear form of equation 11 was cointegrated. Since all the variables have a unit root, the existence of a cointegrating vector is a necessary condition for any sensible interpretation of the results (Engle and Granger, 1987). Table 7 reports Johansen’s maximum likelihood statistics for a fourth-order VAR,²³ and shows that the maximal eigenvalue and trace eigenvalue statistics reject the null of no cointegration (even with a degree of freedom adjustment). The null hypothesis in favor of more than one cointegrating vector is also rejected when these statistics are adjusted by degrees of freedom. This table also reports the standardized eigenvectors (β′) and adjustment coefficients (α). The first row of β′ is the estimated cointegrated vector:

Table 7.

A Cointegration Analysis of Consumption 1957-95 ^1/

^1/Vector Pormanteau 5 lags =143.96; ARE 1-2 F(2,23) = 1.2879 (0.2725); Normality χ²(10) = 42.148 (0.00) The vector autoregression includes four lags on each variable (C, Y, W, T, G), an unrestricted constant term, and a trend restricted to lay in the cointegration space.

The statistics λ_max and λ_trace are the Johansen’s maximal eigenvalue and trace eigenvalue statistics for testing cointegration. The null hypothesis is in terms of the cointegration rank. The critical values are taken from Osterwald- Lenum (1992, Table 1).

^**The null is rejected at a 1% significance level.

^*The null is rejected at a 5% significance level.

Table 7.

A Cointegration Analysis of Consumption 1957-95 ^1/

Eigenvalue	0.890785	0.580879	0.5548007	0.3714325	0.180129
Null hypothesis	r = 0	r ≤ 1	r ≤ 2	r ≤ 3	r ≤ 4
λmax	81.93**	32.18*	29.94*	17.18	7.349
λamax	37.65*	14.78	13.76	7.893	3.376
95% critical value	37.5	31.5	25.5	19	12.2
λtrace	168.6**	86.64**	54.57**	24.53	7.349
λatrace	77.46	39.81	25.03	11.2776	3.376
95% critical value	87.3	63.0	42.4	25.3	12.2
	Standardized Eigenvectors β
Variables	C	G	Y	Τ	W
	1.000	0.15875	-0.61019	0.03868	-0.28612
	-6.803	1.000	5.3	-0.86422	-2.345
	3.360	-0.48530	1.000	-0.56362	-3.634
	-0.87352	0.028169	-1.563	1.000	1.182
	-4.11	-1.116	4.155	-0.86087	1.000
	Standardized Eigenvectors α
C	-0.96871	0.0021633	-0.0556	0.0040491	-0.061165
G	-0.89258	-0.27754	0.27565	0.54508	-0.017318
Y	0.31003	-0.016834	-0.23921	0.095526	-0.10699
T	1.469	-0.36051	-0.72964	-0.42195	-0.20427
W	0.38947	0.018927	0.06591	0.0037514	-0.020298
	Weak Exogeneity Test Statistic					{G,T.Y,W}
χ2(1)	7.0878	0.497398	0.18435	0.46345	4.8038	9.0317
p-value	(0.0078)**	(0.4806)	(0.6677)	(0.4901)	(0.0284)*	(0.06073)
	Multivariate Statistic for Testing Stationarity
χ2(4)	62.205	65.558	68.934	68.109	68.734
p-value	(0.00)**	(0.00)**	(0.00)**	(0.00)**	(0.00)**
	Null: Ricardian Equivalence {βT = 0 & βY = −βG}
χ2(1)(p-value)	33.395/(0.00)**
Variable	C	G	Y	T	W
β restricted by Null	1	-0.63973	-0.63973	0	0.011399
	Statistics for Testing Significance of a Given Variable
Variable	Trend	G	Y	T	W
χ2(1)	40.233	40.874	37.294	4.0449	18.397
p-value	(0.00)**	(0.00)**	(0.00)**	(0.0447)	(0.00)**

^1/Vector Pormanteau 5 lags =143.96; ARE 1-2 F(2,23) = 1.2879 (0.2725); Normality χ²(10) = 42.148 (0.00) The vector autoregression includes four lags on each variable (C, Y, W, T, G), an unrestricted constant term, and a trend restricted to lay in the cointegration space.

The statistics λ_max and λ_trace are the Johansen’s maximal eigenvalue and trace eigenvalue statistics for testing cointegration. The null hypothesis is in terms of the cointegration rank. The critical values are taken from Osterwald- Lenum (1992, Table 1).

^**The null is rejected at a 1% significance level.

^*The null is rejected at a 5% significance level.

View Table

Table 7.

A Cointegration Analysis of Consumption 1957-95 ^1/

Eigenvalue	0.890785	0.580879	0.5548007	0.3714325	0.180129
Null hypothesis	r = 0	r ≤ 1	r ≤ 2	r ≤ 3	r ≤ 4
λmax	81.93**	32.18*	29.94*	17.18	7.349
λamax	37.65*	14.78	13.76	7.893	3.376
95% critical value	37.5	31.5	25.5	19	12.2
λtrace	168.6**	86.64**	54.57**	24.53	7.349
λatrace	77.46	39.81	25.03	11.2776	3.376
95% critical value	87.3	63.0	42.4	25.3	12.2
	Standardized Eigenvectors β
Variables	C	G	Y	Τ	W
	1.000	0.15875	-0.61019	0.03868	-0.28612
	-6.803	1.000	5.3	-0.86422	-2.345
	3.360	-0.48530	1.000	-0.56362	-3.634
	-0.87352	0.028169	-1.563	1.000	1.182
	-4.11	-1.116	4.155	-0.86087	1.000
	Standardized Eigenvectors α
C	-0.96871	0.0021633	-0.0556	0.0040491	-0.061165
G	-0.89258	-0.27754	0.27565	0.54508	-0.017318
Y	0.31003	-0.016834	-0.23921	0.095526	-0.10699
T	1.469	-0.36051	-0.72964	-0.42195	-0.20427
W	0.38947	0.018927	0.06591	0.0037514	-0.020298
	Weak Exogeneity Test Statistic					{G,T.Y,W}
χ2(1)	7.0878	0.497398	0.18435	0.46345	4.8038	9.0317
p-value	(0.0078)**	(0.4806)	(0.6677)	(0.4901)	(0.0284)*	(0.06073)
	Multivariate Statistic for Testing Stationarity
χ2(4)	62.205	65.558	68.934	68.109	68.734
p-value	(0.00)**	(0.00)**	(0.00)**	(0.00)**	(0.00)**
	Null: Ricardian Equivalence {βT = 0 & βY = −βG}
χ2(1)(p-value)	33.395/(0.00)**
Variable	C	G	Y	T	W
β restricted by Null	1	-0.63973	-0.63973	0	0.011399
	Statistics for Testing Significance of a Given Variable
Variable	Trend	G	Y	T	W
χ2(1)	40.233	40.874	37.294	4.0449	18.397
p-value	(0.00)**	(0.00)**	(0.00)**	(0.0447)	(0.00)**

^1/Vector Pormanteau 5 lags =143.96; ARE 1-2 F(2,23) = 1.2879 (0.2725); Normality χ²(10) = 42.148 (0.00) The vector autoregression includes four lags on each variable (C, Y, W, T, G), an unrestricted constant term, and a trend restricted to lay in the cointegration space.

The statistics λ_max and λ_trace are the Johansen’s maximal eigenvalue and trace eigenvalue statistics for testing cointegration. The null hypothesis is in terms of the cointegration rank. The critical values are taken from Osterwald- Lenum (1992, Table 1).

^**The null is rejected at a 1% significance level.

^*The null is rejected at a 5% significance level.

All the coefficients in the cointegrated vector have their anticipated signs. The long-run elasticity of household disposable income is similar to those found in the literature, but the elasticity of wealth is higher than traditional estimates for developed countries (for example, see Currie, Holly and Scott, 1990).²⁴ As can be seen, the timing of taxes matter in economic terms. A 10 percent increase in the growth rate of government consumption (i.e., in expected growth rate of taxes), diminishes the growth rate of private consumption by 1.5 percent. In addition, a 25 percent rise in the growth rate of current taxes lowers the growth rate of private consumption by 0.4 percent.

The rejection of Ricardian equivalence is confirmed statistically. In fact, the joint test restricting β₄=0 and β₁=-β₂ is strongly rejected (penultimate row of Table 7). Moreover, the last row of Table 7 reports chi-squared statistics for testing the significance of individual variables in the cointegrating vector. Each variable is significant, although taxes are very close to a 95% critical value.