Abstract
This Selected Issues paper addresses some of the key policy and economic challenges facing the Canadian economy. The paper presents a new approach to predicting the business cycle in the context of the Canadian economy. This approach uses a range of parametric and nonparametric tests to gauge the ability of various indicators to predict turning points in the business cycle. The paper also presents a model that links the inflation rate to the business cycle and the rates of change in the exchange rate and in unit labor costs.
V. Recent Trends in Personal Saving in Canada 1/
1. Introduction
The personal saving rate in Canada rose from about 13 percent in the late 1960s to 24 percent in 1982, before falling to about 14 percent in 1987. It remained at about this level until 1993, but dropped further to a low of 8 percent in 1995. These developments have raised concerns that an attempt by households to increase their saving rates to more normal levels would have an adverse effect on domestic demand in Canada. There also are concerns that if the decline in the household saving rate is permanent, it could contribute to a deficiency in domestic investment, reduce future living standards, and increase Canada’s reliance on foreign saving.
This paper briefly reviews recent trends in the saving rate and examines the factors that may have contributed to the recent decline, including the possible effects of the reduction in inflation and demographic shifts. It concludes that the decline in the saving rate appears to be largely related to the transition to a low inflation environment since the early 1980s. When the saving rate is adjusted to take into account the depreciation of nominal assets that is due to inflation, the decline appears less pronounced. Section 2 discusses issues related to the measurement of the effects of inflation, Section 3 examines the statistical properties of the saving rate and its relationship with a number of other economic variables, and Section 4 offers concluding remarks.
2. Recent trends in the saving rate
Chart V-1 illustrates the sharp rise in the personal saving rate that occurred between 1970 and 1982. During this period, the personal saving rate rose from about 10 percent to about 22 percent of disposable income, before trending downward to about 14 percent during the period 1992-1987, in part because of the cyclical recovery. The saving rate remained at about this level for five years, but declined further during 1994 and 1995 to 8 percent.
CANADA NATIONAL ACCOUNTS MEASURES OF PERSONAL SAVING RATES
(In percent of disposable income)
Citation: IMF Staff Country Reports 1996, 038; 10.5089/9781451806830.002.A005
Sources: Statistics Canada (supplied by DRI): and Fund staff estimates.1/ As a shore of disposable income less depreciation.2/ As a shore of disposable income fess inflation adjustment. The inflation adjustment is equal to a three-year moving overage of CPI inflation times net financial assets less stocks one foreign investments.A number of measurement issues are relevant for assessing these trends. 2/ In particular, the concept of household saving in the national accounts is intended to provide a measure of household accumulation of financial and nonfinancial assets. However, the national accounts data for the personal sector are estimated as a residual from the activities of the other sectors that are easier to measure. As a result, the data for the personal sector include measurement errors from elsewhere in the national accounts. Moreover, the personal sector includes the activities of the unincorporated business sector, since this sector is not accounted for elsewhere in the national accounts. The unincorporated business sector encompasses private non-profit institutions, trusteed pension plans, and the investment activities of life insurance companies. As a result, employers’ contributions to pension plans and the investment income of life insurance companies and trusteed pension plans are included in personal income. These flows increase the measured amount of saving, even though the household sector may not have discretion over these flows. 1/
An additional factor impinging on the usefulness of the conventional national accounts measure of saving is that it does not factor in the depreciation of physical assets owing to wear and tear. In addition, no account is taken of the erosion of the real value of financial assets as a result of inflation. Chart V-1 shows that depreciation of the personal sector’s physical assets has been a significant component of the saving rate, but does not explain the decline since 1982. The net saving rate-i.e., personal saving net of depreciation--shows roughly the same decline because depreciation has been a relatively constant share of disposable income averaging about 5 percent since the mid-1960s.
However, the effect of inflation on the personal sector’s financial assets has been considerably more volatile. In particular, Chart V-1 also shows an estimate of the inflationary bias of the conventional measure of the saving rate. 2/ As can be seen, the inflationary component of personal saving rose sharply up to the late 1970s and fell significantly thereafter, owing to the rise and subsequent decline in inflation. This suggests that a significant portion of the rise in the saving rate to 1982, and much of its subsequent decline, was related to a similar rise and fall in the inflation rate.
The inflation-adjusted measure of saving still has a number of drawbacks. For example, the national accounts do not include the capital gains on physical and financial assets as income and therefore may underestimate saving. 3/ A. particular example would be the case of financial assets whose price could be affected by changes in nominal interest rates and inflation. The measure in question also treats the full amount of purchases of consumer durables as consumption in the year that they were purchased, thereby understating saving.
An alternate measure of saving that addresses these problems (and also accounts for depreciation) can be derived from data on the personal sector’s balance sheet by defining saving as the change in households’ real net worth. Chart V-2 shows this measure of saving as a share of disposable income and does not indicate any downward trend in the saving rate over time. The factors underlying the shifts in saving behavior are examined in greater detail in the next section.
CANADA BALANCE SHEET MEASURES OF PERSONAL SAVING RATE
Citation: IMF Staff Country Reports 1996, 038; 10.5089/9781451806830.002.A005
Sources: Statistics Canada (supplied by DRI); and Fund staff estimates.1/ Real disposable income.3. Factors affecting the saving rate
A number of recent studies have focused on explaining movements in household consumption behavior in Canada rather than saving. 1/ Macklem (1994) finds a long-run or cointegrating relationship between real per capita consumer expenditures, disposable income, and total wealth, and concludes that changes in the unemployment rate and changes in the consumer price deflator help explain short-term movements in real consumption. Lamy (1995) supplements Macklem’s analysis by adding a long-term interest rate and the dependency ratio to the cointegrating vector and finds that the yield curve, the unemployment gap, and indirect taxes provide additional explanatory power for short-run movements in consumer expenditures (net of rent) in Canada.
Chart V-3 illustrates the correlation between a number of variables described above and the saving rate as measured by the national accounts. This chart suggests that the relationship between this measure of the saving rate and the real short-term interest rate (the nominal rate less the CPI inflation rate) is relatively weak. Similarly, the saving rate does not appear to be strongly correlated with a measure of demographic trends (the share of the population aged 45 to 65), real disposable income growth, and the net household worth as a share of disposable income. However, the saving rate, defined from the household balance sheet, appears to bear a negative contemporaneous relationship with real interest rates and to be positively correlated with the growth of disposable income (Chart V-4). This possibility is explored further below.
CANADA DETERMINANTS OF THE NATIONAL ACCOUNTS MEASURE OF THE SAVING RATE
Citation: IMF Staff Country Reports 1996, 038; 10.5089/9781451806830.002.A005
Sources: Statistics Canada (supplied by DM); and Fund staff estimates.CANADA DETERMINANTS OF THE BALANCE SHEET MEASURE OF THE SAVING RATE
Citation: IMF Staff Country Reports 1996, 038; 10.5089/9781451806830.002.A005
Sources: Statistics Canada (supplied by DRI); and Fund staff estimates.1/ Measured as the change in real net worth as a percent pf real disposable income.a. data issues
As Macklem (1994) notes, research into household saving/consumption behavior has tended in recent years to focus on searching for long-run relationships between saving (or consumption) and selected macroeconomic variables. In large part, this reflects the fact that the data in question have been found to be nonstationary--i.e., do not show signs of mean reversion. This means that conventional statistical methods cannot be used to test the relationship between short-run movements in the level of saving and the level of other (nonstationary) macroeconomic variables. It also admits the possibility that the level of saving may be related in the long-run to the level of variables such as income, interest rates, etc. while its short-run movements are driven by, among other things, deviations from this long-run relationship.
As a preliminary to a multivariate analysis of the saving rate, the time series properties of the data are examined. Table V-1 illustrates a series of tests on various measures of the saving rate, as well as on other variables often thought to influence saving behavior. The tests confirm that most of the variables considered are nonstationary. Notably, the national accounts measure of the saving rate was nonstationary, even when measured in real terms on a per capita basis. However, the various balance sheet measures of saving did not exhibit the same property. In particular, the change in net wealth expressed either as a share of disposable income or on a real per capita basis appeared to be stationary.
The stationarity of the saving rate as measured by the change in household net wealth suggest that this measure of the saving rate does not have to be examined in the context of a cointegrated or long-run relationship. Instead, the behavior of the saving rate and the possibility that it has experienced a structural break can be examined within the framework of a standard partial-adjustment model.
b. Multivariate analysis
The factors that have affected the short-run evolution of the saving rate--defined on the basis of the household balance sheet--are examined in this section. The discussion above suggests that household saving is likely to be related to the evolution of income, real interest rates, and demographic variables. In particular, the permanent income hypothesis suggests that households tend to save and consume constant fractions of their permanent income, and that fluctuations in the saving rate would therefore result mainly from transitory fluctuations in income.
Consumer theory suggests that the effect of changes in interest rates on the level of saving is ambiguous. An increase in interest rates raises the return on saving, thereby encouraging a substitution away from consumption, but also increases the return on current saving thereby reducing the need to forego current consumption to finance future consumption. As regards demographic variables, life cycle models of saving suggest that as the share of the population aged 25 to 65 increases, saving tends to increase because this age cohort saves in order to provide for retirement.
In order to test these hypotheses and to examine the possibility of a structural break in saving behavior, an equation was estimated relating the change in real household net worth as a share of real disposable income (ΔW/Y) to the real short-term interest rate (rtb) and the percentage change in real per capita disposable income (%ΔY). 1/ In addition, the possible effect of demographic variables on saving behavior was considered by including variables measuring the share of the population aged 25 to 45 and the share of the population aged 45 to 65. Lagged values of the dependent variable also were included on the assumption that households adjust their portfolios gradually to shocks, either because of uncertainties regarding their permanence or because of adjustment costs.
The results for the period 1966Q4-1995Q3 are summarized below. 2/
The fit of the estimated equation was relatively low, partly reflecting the definition of the dependent variable, which included the effects of changes in asset prices. Tests on the residuals of the equation rejected the hypothesis of autocorrelation and heteroskedasticity. The Ramsey RESET test also rejected the possibility of more general mis-specification errors. 1/ Demographic variables were not found to be significant determinants of the saving rate.
Saving Rate Regression
Citation: IMF Staff Country Reports 1996, 038; 10.5089/9781451806830.002.A005
The results confirm a strong and positive relationship between household asset accumulation and the change in real per capita disposable income. Moreover, the significance of the estimated coefficient seems consistent with the effects of income found in other studies of saving and consumption behavior. 2/ The results also suggest a relatively large negative effect of the real interest rate on the saving rate--a one percentage point increase in the real short-term rate would reduce the saving rate by 0.13 percentage point. The long-run coefficient of rtb is 0.27, and a Wald test rejected the hypothesis that it equaled zero at the 95 percent confidence level. 1/ Other specifications were examined but did not appreciably change this result. For example, the long-term interest rate was not found to be a significant determinant of the saving rate, and the estimated coefficient was not affected by including the interest rate in nominal terms.
The possibility that the saving rate as defined above was subject to the same structural break exhibited by the national accounts measure is examined by performing a Chow test, with a break at 1982Q3. The hypothesis of a structural break is not rejected at the 5 percent confidence level. However, when the equation was estimated separately over post- and pre-1982Q3 periods, the coefficients did not differ appreciably. Moreover, examination of the recursive residuals showed no significant evidence of a structural break. However, when the equation was re-estimated with a zero-one dummy variable with a break at 1982Q3, the coefficient estimate for the dummy was positive and significantly different zero at the 95 percent confidence level, suggesting an upward shift in the saving rate.
Cumulative Sum of Squared Residuals
Citation: IMF Staff Country Reports 1996, 038; 10.5089/9781451806830.002.A005
The fact that the estimated coefficient on the interest rate was negative was somewhat surprising--multi-country studies generally find a positive relationship between saving and the real rate of interest (for example, see Ogaki, Ostry, and Reinhart (1995) or Masson, Bayoumi, and Samiei (1995)). However, the evidence in favor of a positive interest rate effect tends not to be robust; for example, Bosworth finds a positive interest rate effect on saving in his time series regressions, but a negative coefficient in his cross-country estimates.
In order to examine this issue further, the saving rate equation was re-estimated using as the dependent variable the national accounts measure of saving less the inflation adjustment described above. 1/
The results in this case were similar to those for the saving rate defined from the perspective of the household balance sheet except that the coefficient on the real interest was positive. 2/ One possible explanation for the difference in the interest sensitivity could be that the balance-sheet measure of saving treats purchases of consumer durables as saving rather than consumption.
For example, Ogaki and Reinhart (1995) suggest that estimates of the intertemporal elasticity of substitution have been biased downward because they ignored the effect of changes in interest rates on the demand for durables. In particular, they argue that an increase in interest rates will tend to cause a substitution toward saving, but will also increase the demand for consumer durables. An implication of their work is that the interest elasticity of a measure of saving that treats the purchase of durables as saving will tend to have a lower (and possibly negative) interest elasticity than the conventional measure of saving.
4. Concluding remarks
The purpose of the analysis was principally to examine the factors that led to the decline in the national accounts measure of the saving rate, and the results suggest a number of observations. First, while the national accounts measure of the saving rate has declined sharply since 1982, a large portion of the decline can be related to the reduction in inflation. Adjusting the measures of household income and saving by the amount of capital losses resulting from inflation leads to a saving rate that is considerably more stable. A more comprehensive measure of the saving rate that accounts for the effect of inflation, depreciation, purchases of consumer durables, and changes in asset prices--the change in real household net worth as a share of disposable income--also shows little evidence of a decline after 1982.
Second, the fact that conventionally defined saving rates are nonstationary was confirmed. However, unlike the national accounts measure, the saving rate defined on the basis of the change in real household net worth was found to be stationary. This suggests that the latter measure of saving exhibits a long-run relationship with real disposable income.
The multivariate analysis indicated that the saving rate defined from the household balance sheet is positively related to the rate of growth of real disposable income and negatively related to the real rate of interest. Unlike the results of other studies, demographic variables did not appear to have had a significant effect on the saving rate. While there still was some evidence that the saving rate exhibited a structural break in 1982, the evidence would point toward an increase rather than a decrease in the saving rate since that year.
The negative sign on the interest rate coefficient also was unusual, and contrasted with a positive relationship between the interest rate and the saving rate that is defined using the national accounts (adjusted for the effect of inflation). One explanation for the difference was that the balance sheet measure of saving treated the purchases of consumer durables as saving, which would tend to be negatively related to changes in interest rates.
Canada: Stationarity Tests
An asterisk denotes that the hypothesis of stationarity was rejected at the a 95 percent confidence level. The ADF tests were conducted using a constant and a time trend. The lag length was determined sequentially by testing the significance of more restrictive specifications (see Campbell and Perron (1991) for a discussion).
Canada: Stationarity Tests
| ADF Statistic 1/ |
Lags | Period | ||
|---|---|---|---|---|
| Personal saving measures | ||||
| Saving as percent of disposable income | 1.29* | 1 | 65:3-95:3 | |
| Real saving per capita | 0.92* | 1 | 66.3-95:3 | |
| Inflation adjusted saving as percent of disposable income | 3.68 | 0 | 66:4-95:3 | |
| Disposable income | ||||
| Real disposable income | 0.64* | 0 | 66:3-95:3 | |
| Real disposable income per capita | 0.10* | 1 | 66:3-95:3 | |
| Population | ||||
| Share of population aged 25 to 45 | 1.19* | 8 | 67:2-95:3 | |
| Share of population aged 45 to 65 | 1.19* | 8 | 67:2-95:3 | |
| Share of population aged 25 to 65 | 1.43* | 8 | 67:2-95:3 | |
| Inflation | ||||
| 4-quarter CPI inflation rate | 2.46* | 5 | 68:4-95:3 | |
| Interest rates | ||||
| 3-month treasury bill | 2.41* | 1 | 65:3-95:3 | |
| 3-month treasury bill less inflation | 2.43* | 4 | 68:2-95:3 | |
| Net wealth | ||||
| Net wealth as share of disposable income | 0.06* | 8 | 67:2-95:3 | |
| Real net wealth per capita | 2.22* | 1 | 66:3-95:3 | |
| Change in net wealth | 1.58* | 0 | 66:4-95:3 | |
| Change in real net wealth | 4.27 | 3 | 62:2-95:3 | |
| Change in real net wealth per capita | 5.38 | 0 | 66:4-95:3 | |
| Change in real net wealth as a share of real disposable income | 5.36 | 0 | 66:4-95:3 | |
An asterisk denotes that the hypothesis of stationarity was rejected at the a 95 percent confidence level. The ADF tests were conducted using a constant and a time trend. The lag length was determined sequentially by testing the significance of more restrictive specifications (see Campbell and Perron (1991) for a discussion).
Canada: Stationarity Tests
| ADF Statistic 1/ |
Lags | Period | ||
|---|---|---|---|---|
| Personal saving measures | ||||
| Saving as percent of disposable income | 1.29* | 1 | 65:3-95:3 | |
| Real saving per capita | 0.92* | 1 | 66.3-95:3 | |
| Inflation adjusted saving as percent of disposable income | 3.68 | 0 | 66:4-95:3 | |
| Disposable income | ||||
| Real disposable income | 0.64* | 0 | 66:3-95:3 | |
| Real disposable income per capita | 0.10* | 1 | 66:3-95:3 | |
| Population | ||||
| Share of population aged 25 to 45 | 1.19* | 8 | 67:2-95:3 | |
| Share of population aged 45 to 65 | 1.19* | 8 | 67:2-95:3 | |
| Share of population aged 25 to 65 | 1.43* | 8 | 67:2-95:3 | |
| Inflation | ||||
| 4-quarter CPI inflation rate | 2.46* | 5 | 68:4-95:3 | |
| Interest rates | ||||
| 3-month treasury bill | 2.41* | 1 | 65:3-95:3 | |
| 3-month treasury bill less inflation | 2.43* | 4 | 68:2-95:3 | |
| Net wealth | ||||
| Net wealth as share of disposable income | 0.06* | 8 | 67:2-95:3 | |
| Real net wealth per capita | 2.22* | 1 | 66:3-95:3 | |
| Change in net wealth | 1.58* | 0 | 66:4-95:3 | |
| Change in real net wealth | 4.27 | 3 | 62:2-95:3 | |
| Change in real net wealth per capita | 5.38 | 0 | 66:4-95:3 | |
| Change in real net wealth as a share of real disposable income | 5.36 | 0 | 66:4-95:3 | |
An asterisk denotes that the hypothesis of stationarity was rejected at the a 95 percent confidence level. The ADF tests were conducted using a constant and a time trend. The lag length was determined sequentially by testing the significance of more restrictive specifications (see Campbell and Perron (1991) for a discussion).
References
Bosworth, B.P., Saving and Investment in a Global Economy (Washington, DC: Brookings Institution, 1993).
Bayoumi, Tamim, “Financial Deregulation and Household Saving,” Economic Journal, Vol. 103 (November 1993), pp. 1432–43.
Bovenberg, Lans and Owen Evans, “National and Personal Saving: Measurement and Analysis of Recent Trends” in Yusuke Horiguchi (ed.) The United States Economy: Performance and Issues (Washington: International Monetary Fund, 1992), pp. 19–51.
Bradford, David F., “What is National Saving? Alternative Measures in Historical and International Context”, in The U.S. Saving Challenge, ed. by Christopher E. Walker, Mark A. Bloomfield, and Margo Thorning (Boulder, CO: Westview, 1990), pp. 31–88.
Campbell, John Y. and Pierre Perron, “Pitfalls and Opportunities: What Macroeconomists Should Know about Unit Roots,” in Olivier Jean Blanchard and Stanley Fischer (eds.) NBER Macroeconomics Annual: 1991(Cambridge MA: MIT Press, 1991), pp. 141–201.
Clift, Barbara, “Components of Personal Saving,” Canadian Economic Observer (November 1988).
Dagenais, Marcel G., “Measuring Personal Savings, Consumption and Disposable Income in Canada since 1962,” University of Montreal, Centre de recherche et développement en économique Cahier 4789 (December 1989).
Lamy, Robert, “Income Measurement, Term of Interest Rate, Long-Term Canadian Consumption Function and Error Correction Model: An Application of Cointegration with the Engle and Granger Two-Step Procedure,” Department of Finance Canada mimeo (May 1995).
Lau, Hung-Hay, “Behaviour of the personal savings rate in Canada in recent years: A note,” Bank of Canada Review (Spring 1993), pp. 57–67.
Macklem, T.R. “Wealth, Disposable Income and Consumption: Some Evidence for Canada,” Bank of Canada Technical Report No. 71 (November 1994).
Masson, Paul R., Tamim Bayoumi, and Hossein Samiei, “International Evidence on the Determinants of Private Saving,” IMF Working Paper WP/95/51 (May 1995).
Ogaki, Masao and Carmen M. Reinhart, “Measuring Intertemporal Substitution: The Role of Durable Goods,” Rochester Center for Economic Research Working Paper No. 404 (May 1995).
Ostry, Jonathan D. and Joaquim Levy, “Household Saving in France,” IMF Staff papers, Vol 42 No. 2 (June 1995), pp. 375–397.
Prepared by Alun Thomas and Christopher Towe.
For a more detailed discussion, see Dagenais (1989). These issues are reviewed for the case of the United States by the Congressional Budget Office (1993) and Bradford (1990).
Clift (1988) suggests, however, that these flows are relatively small and do not affect overall personal saving trends.
The inflation adjustment is calculated by adjusting saving and disposable income by the product of a moving-average of inflation and net financial assets (excluding stocks and foreign assets). For a further discussion of this type of adjustment in the case of Canada see Lau (1993).
A number of authors have noted that national accounts measures of saving also are understated because personal consumption includes outlays for education, which could instead be considered an investment. This issue is beyond the scope of the present paper and is not addressed (for a detailed discussion see Macklem (1994)).
Carroll and Summers (1987) compared private saving rates in Canada and the United States and explain 80 percent of the variation as a result of differences in income growth, wealth-GDP ratios and government saving. They also concluded that tax incentives had an important effect on differences between saving rates. Bovenberg and Evans (1993) find that movements in real per capita consumer expenditures in the United States are well explained by movements in personal disposable income and net wealth. Ostry and Levy (1995) examine the behavior of saving in France and find that income growth, inflation, demographics, and interest rates have the expected effect. Bayoumi (1993) examines household saving and the effect of financial market deregulation in the United Kingdom.
Ostry and Levy also use the growth of income as a determinant of the saving rate. In this case, the %ΔY variable could be considered a proxy for transitory income, which avoids the need for a more complicated trend/cycle decomposition. Note that rtb is included as a regressor, despite showing signs of nonstationarity, on the assumption that over the long run its variance is bounded. Moreover, Hackles (1994) assumes that the real interest rate is stationary even though the ADF test does not reject the null hypothesis of a unit root.
Absolute values of t-statistics are shown in parentheses. The LM(4) statistic tests for fourth-order autocorrelation and is asymptotically distributed X2(3); the ARCH(4) statistic tests for auto regressive conditional heteroskedasticity and is asymptotically distributed X2(4); the White statistic tests for heteroskedasticity and is distributed X2(8); the Chow statistic tests for a structural break at 1982Q3 and is distributed F(4, 108); and the RESET statistic tests for model mis-specification and is distributed X2(3).
In particular, the result of this test relieves concern about the possibility of simultaneity between saving and disposable income.
For example, Hackles finds that the coefficient on income is 0.74 in Canada and Bovenberg and Evans (1992) find an income coefficient of 0.82 for the United States. The long-run coefficient on %ΔY was 0.82 and significantly different from zero at the 95 percent confidence level. However, the interpretation of this coefficient is problematic since %ΔY reverts to trend, and is assumed to proxy temporary deviations of Y from potential.
Lamy (1995) reports a positive coefficient, but this is with respect to the long-term interest rate.
A further difference was that in this case the hypothesis that the errors were autocorrelated could not be rejected, and the Chow test rejected the hypothesis of a structural break in 1982.
