VIEWS: 25 PAGES: 25 POSTED ON: 3/28/2011
TAX REFORMS AND THE DEMAND FOR DEBT Paper to be presented at the 7th Nordic Seminar on Microsimulation Models in Helsinki, June 12. -13. 20031 This version: May 28, 2003 By Erik Fjærli Statistics Norway Abstract: In 1992 Norway carried out a major tax reform that replaced the progressive taxation of capital income and interest expenses by a so-called dual income tax, which implied a flat tax on capital income and deductible interest expenses and progressive taxes on labor income. The flat tax rate is 28 percent. Interest deductions are unlimited under the current tax system, similar to the tax system before the reform. However, the average marginal tax rates are lower now than before, and equal for most taxpayers. Consequently, the tax-subsidy to borrowers due to the difference between pre- and after-tax interest rates has dropped, in particular among the rich. The purpose of this paper is to examine more closely the relationship between borrowing and income, before and after the tax reform. Based on repeated cross-sectional data and Tobit regressions, we find strong evidence that the propensity of debt to increase with income is much less pronounced after the tax reform. Using a difference-in-differences estimator on panel data of identical individuals from 1991 to 1993, we also find that increased borrowing followed an increase in the marginal tax rate, while those whose marginal tax rate fell on average reduced their debt. 1 Work in progress - preliminary results. 1. Introduction In 1992, Norway went through a major tax reform that replaced the progressive taxation of capital income and interest expenses by a flat 28 percent tax on capital income and interest expenses. Interest deductions are unlimited under the current tax system, similar to the tax system before the reform. The average marginal tax rates are lower now than before, and equal for all taxpayers. Consequently, the tax-subsidy to borrowers due to differences between pre- and after-tax interest rates has dropped, in particular among the rich. Side by side with the reduction in marginal tax rates, the rate of inflation and the nominal interest rate fell too, which caused a further reduction in the tax-subsidy. Similar tax reforms were also carried out in Denmark, Finland and Sweden (Sørensen, 1994). In the US, the Tax Reform Act of 1986 partly addressed the same problems as the later tax reforms in the Nordic countries: By eliminating interest deductions on consumer credit and limiting the mortgage interest deduction, the goal was to increase private saving and government tax revenue. Gordon and Slemrod (1988) point at the role of the tax arbitrage opportunities in the US tax system before the TRA of 1986, caused by the combination of differences in the tax treatment of different types of capital and differences in personal tax rates. They find that in 1983, a shift to an R-base tax, which exclude all financial incomes and expenses from the personal tax base and taxes only the cash flow at the corporate level, would have practically no revenue costs. By contrast, Gordon, Kalambokidis and Slemrod (2002) concludes that in 1995, a switch to the R-base would reduce corporate tax revenue by 18 billions. current US $ and reduce revenue from personal taxes by 90.1 $, a total revenue loss of 108.1 bill. $. According to Gordon, Kalambokidis and Slemrod, the differences is explained by 1) a change in the ratio of capital allowances to new investments and 2) a change in the financial flows due to a drop in the level of nominal interest rates. Both in 1983 and 1995, the exemption of financial incomes from the US personal tax base would cause a significant revenue loss. In 1995, the reduction in personal taxable income would be approximately 8.2 percent. In Norway, one would expect quite different results than in the US, as far as the individual tax base is concerned: Due to the combined effect of unlimited interest deductions and very lenient taxation of the corresponding imputed rate of return on owner occupied housing, interest deductions exceed taxable capital income on average. Table 1 below indicates what the consequences would be like, if interest expenses and financial incomes were exempt from the individual tax base2. Quite apart from the fact that exemption of capital incomes from the tax base probably would cause severe income shifting problems, 2 The direct effect of a shift to a pure labor income individual tax base is calculated by omitting capital incomes- and expenses from the individual tax base in the tax-simulation model LOTTE, based on large samples of individual tax returns (secondary effects through changes in labor supply and real wages are not included). 2 this gives an idea of the magnitude of the revenue raised (lost) by the taxation of net financial income in the household sector. Table 1. Individual tax revenue and change in revenue from shifting to a pure labor income tax base (wealth tax retained).Current bill. NOK and percent. Year Individual tax Increase in revenue Increase in revenue, revenue, current from shift to percent system1) R-base1) 1987 95 24 25.3 % 1991 109 27 24.8 % 1992 106 26 24.5 % 1993 114 23 20.2 % 1994 121 24 19.8 % 1995 131 24 18.3 % 1996 142 24 16.9 % 1997 155 28 18.1 % 1998 169 31 18.3 % 1999 178 32 18.0 % 1) Source: Calculations on the micro simulation model LOTTE based on detailed income statements (tax-forms). As indicated by the results in table 1, the taxation of individual capital income leads to an enormous loss in tax revenue. Before the tax reform, the revenue loss was approximately 25 percent of the total tax revenue from the household sector. After the reform, the revenue loss has dropped to about 18 percent, in part due to lower average tax rates on interest deductions, partly to lower nominal interest rates. The change in the marginal tax rates that followed the tax reform can possibly have affected the demand for debt among households, in several ways: First, the households’ demand for debt is largely driven by the demand for housing and other consumer durables. As the tax subsidy to borrowing declines, one would expect to find a tendency among younger households to postpone home purchases and instead increase their capital before buying their first home. Second, older households would possibly increase their down payments. Third, borrowing can be motivated by tax arbitrage opportunities, typically when rich tax payers in high marginal tax brackets borrow (from low-tax clienteles), deduct interest expenses against high marginal tax rates and invest in relatively low-taxed stocks. After the reform, households that used to be located in high tax brackets would possibly sell some of their shares and pay off their debt. 3 All of the above effects are related to progressive tax rates and the households’ income level, and indicate that the tax reform possibly has made household debt less dependent of income. The purpose of this paper is to examine empirically the relationship between borrowing and income before and after the tax reform. Previous studies of the impact of the interest deductions on households' debt include Follain and Dunsky (1997) and Maki (2001), who look at portfolio shuffling by substituting mortgage debt for consumer debt, due to the differences in the after-tax cost of different types of credit. Using Tobit-regressions, Follain and Dunsky find that the demand for mortgage debt is highly responsive to changes in the deductibility of mortgage interest. Maki finds similar results, using difference-in-differences methodology on pooled time-series and cross-secton data. Skinner and Feenberg (1990) also find shuffling effects without being able to identify tax- or income effects directly, while Scholz (1994) finds that high-income households appeared to reduce consumer debt and increase mortgage debt to a greater extent than other households. Hendershott, Pryce and White (2002) study the impact of ceilings of mortgage interest deductions on loan-to-value ratios on UK data, and find "a major sensitivity of leverage to the debt tax penalty created by the partial deductibility of mortgage interest". Apart from Agell and Edin (1990), there are few related Nordic studies. Their analysis of the impact of the marginal tax rates on portfolio composition on Swedish data produced mixed results, some assets being more sensitive to differences in marginal tax rates than others. However, mortgage debt was clearly among the assets that showed the highest degree of sensitivity to variations in the marginal tax rates. The rest of the paper is organized as follows: In section 2 we model the demand for debt and provide empirical specifications. Section 3 gives a description of the data and details of the tax-system. The results are presented in section 4, and section 5 concludes the paper. 2. A model and empirical specifications Modeling tax incentives: The financing of consumer durables as a choice between debt and equity The literature about debt clienteles goes back to the model for optimal capital structure under corporate taxes of Miller (1977), further developed to include the impact of uncertainty and personal taxes by DeAngelo and Masulis (1980). In these models, debt clienteles arise as a result of differences in the tax treatment of bonds and stocks combined with differences in the personal tax rates. While Miller and the 4 subsequent work of DeAngelo and Masulis primarily focused on corporate debt policy, our focus is at the other side of the coin; savings decisions in the household sector. Our theoretical model follows along the lines of Follain and Dunsky (1997) and Alm and Follain (1987). This approach builds on a class of housing finance models where households choose among (owner-occupied) housing services, h, non-housing consumption, c, and expected future wealth, E(W), to maximize expected utility given by E U U (c, h) F E W , 2 W , (1) subject to W0 I Pc c PH H M Be , (2) E W 1 E rp p d PH H 1 rm M 1 E re Be 1 ErI I t y I Ere Be t m rm M p PH H , (3) 2 W E W E W 2 . (4) Following the notation of Follain and Dunsky (1997), H is the housing stock and Be is a risky financial asset. W0 denotes initial wealth in the first period, I is income in the first period, Pc is the first period price of non-housing consumption, PH is the price of one unit of housing stock at the beginning of the first period, E(rp) is the expected rate of appreciation of houses, d is the depreciation rate, rm is the (fixed) mortgage rate, E(re) is the expected rate of return on the risky financial asset and E(rI) is the expected increase in salary income. tp denotes the property tax rate, ty is the marginal income tax rate and tm is the tax rate at which mortgage interest rate and property tax can be deducted. E(W) is the expected value at the end of the period of the random variable wealth, W, with variance s2(W) . Mortgage demand models like the one above focus on the tax treatment of mortgage debt (in some countries, and/or to different times, tax deductions are restricted to mortgage debt only, not consumer credit). As pointed out by Maki (2001), taxpayers who own homes can borrow against their home equity to pay for purchases instead of using consumer credit. Consequently, other durables than housing should enter the utility function too. Another difference is that in Norway, all forms of interest expenses are 5 deductible, not only mortgage interest. As a consequence, our version of the model includes total debt, D, and all sorts of consumer durables, cd, Cd. The assumption of only one (and risky) financial asset is not a realistic one. We will assume that there exists a "risk-less" financial asset too; denoted F. To reduce the number of variables and keep the model easy, we assume that F can be sold short, as opposed to the risky asset B. Negative amounts of F means that the household issues debt, i.e. F=-D. Due to the difference between interest rates on deposits etc. and debt of the same risk class, the households will never hold debt and deposits simultaneously, at least not out of investment motives. However, the households may hold debt and stocks simultaneously, in particular if the return to stock investments is taxed leniently, compared to interest income and -expenses. To limit the arbitrage opportunities, we add a short sales restriction that debt cannot exceed the value of real assets at the beginning of the period3. Our version of the model is described by equations (1') to (5') below. E U U (c, c h ) F E W , 2 W , (1’) W0 I Pc c Pd Cd D Be , (2’) E W 1 E rp p d Pd Cd 1 rD D 1 E re Be , (3’) 1 ErI I t y I te Ere Be t D rD D 2 W E W E W 2 (4') D Pd Cd (5') p now summarizes property taxes and duties on consumer durables and we let d represent depreciation on the entire stock of durables. Compared to the starting point, we also modify the model by assuming that the effective personal tax on the return on the risky financial is te and that re is the rate of return net of corporate tax. As seen from the budget constraint (2’), one additional unit of durables can be financed by reducing the current period’s consumption of non-durables and/or by reduced financial saving (i.e., reduced future consumption), either by reducing the amounts of the risk-less asset, eventually borrowing if necessary, or reducing the stock of the risky financial asset. The tax rate tD not only affects the choice of financing, but also the composition of current consumption and the accumulation of wealth. However, our focus is 3 This is a simplification. A slightly more realistic assumption would be to allow for some unsecured consumer credit. 6 on demand for debt and we will limit the analysis to the impact of changes in the tax treatment of debt and interest-bearing securities vs. equity on the household’s demand for debt. Of course, the same tax incentives that apply to borrowing also apply to the use of interest-bearing securities, bank deposits etc. However, most households do not have the means to fully finance home purchases, cars etc. without at least some borrowing. While "household equity" generally can mean stocks as well as deposits, bonds, real assets etc., we will use the term equity when the household finance purchases out of the stock of risky assets, instead of borrowing. The tax regime We define the after-tax rate of return in the corporate sector as re=(1-tc)rc, where tc is the corporate tax rate and rc is the (expected) pre-tax corporate rate of return. We look at a small, open economy with free flow of financial capital and where rD is exogenous, determined by the international capital market and the national economic policy. In a Miller equilibrium, rc is affected by the tax rate asymmetries through capitalization effects in the financial markets. This is neither a necessary assumption for our purposes, nor realistic in a small open economy, so we will assume that rc is exogenously determined by "CAPM-mechanisms", and that r E D , where is determined by the risk premium (differences in risk between stock investments rc and bonds/debt), which is assumed to be constant over time. Under the progressive tax regime, the taxpayer is assumed to belong to one of the following three tax brackets: B1: (1-tD)>(1-te)(1-tc) B2: (1-tD)=(1-te)(1-tc) B3: (1-tD)<(1-te)(1-tc) The effective personal tax rate on equity investments (te) is small, due to various loopholes, so for all the taxpayers we assume (1-tD)<(1-te). Further, te is assumed to be the same for all taxpayers4. For simplicity, we assume that the return to owner-occupied housing is fully tax-exempt. In the actual tax system, housing services from own dwellings are subject to income taxation before as well as after 4 This is not an entirely correct description of the tax system in Norway before the 1992 tax reform, since dividends actually were taxed at a progressive rate. However, the assumption is justified by the fact that the dominating source of corporate equity financing before the tax reform was retained earnings and that distribution of profits largely came as capital gains which were taxed at a flat rate, if taxed at all (capital gains taxes could easily be circumvented and raised very little revenue). 7 the tax reform, but at a very low effective tax rate compared to the taxation of financial incomes and interest expenses. The B2 taxpayers will be indifferent to the choice between using debt or equity to finance their purchases, while the low-taxed people in B1 will prefer equity. High-taxed taxpayers in bracket B3 are likely to borrow more than they would do otherwise in the absence of taxes or under a flat tax system. After the reform, te=0 and (1-tD)=(1-tc) for all taxpayers, i.e., (1-tD)/(1-tc)=1. Accordingly, we can summarize the impact of the tax reform as follows: B1 : 1 t D 1 0 1 t e 1 t c B2 : 1 t D 1 0 1 t e 1 t c B3 : 1 t D 1 0 1 t e 1 t c Consequently, taxpayers in bracket B2 will still be indifferent to the use of equity vs. debt to finance their consumer durables after the tax reform. Previously low-taxed tax-payers in bracket B1 will experience a decrease in the cost of debt financing, while the tax-payers in bracket B3 will find that the cost of debt has increased compared to equity. Depending on transaction costs, risk considerations and other non-tax factors not modeled here, different taxpayers are expected to adjust their optimal mix of debt and equity according to which tax bracket they belonged to under the progressive tax-regime. Predictions An increase in tD will increase the cost of debt relative to the cost of financing durables by reducing the amount of risky financial assets. Households with positive amounts of debt and stocks will possibly reduce their holdings of stocks and use the proceeds to repay debt. Similar, a reduction in tD will have the opposite effect. Even though we have not incorporated the choice of tenure in our formal model, it is worth mentioning that reduced tD also affect the choice between rental and owner-occupied housing. Due the tax-advantage of home ownership, most households in Norway acquire their own dwelling sooner or later and rental housing is most common among young households. One explanation could be that young households balance the (non-tax) advantages of rental against the tax advantage of owner-occupied housing, and that the former decreases with age. If this is the case, reducing tD and the 8 tax advantage of home ownership could stimulate some households to postpone home purchases and instead accumulate capital5. Finally, reduced tD could reduce the amount of pure tax arbitrage that arises when high-taxed clienteles borrow (possibly against their home equity) and invest in low-taxed stocks. All the effects mentioned above lead to the general prediction that a tax reform where flat tax rates on interest deductions replace progressive rates will lead to increased borrowing among the previously low-taxed, low-income clientele. Similar, previously high-taxed, high-income clientele is expected to reduce their amount of debt. As a consequence, one should expect debt to become less dependent on income after the tax reform. Specification of empirical models (1) The amount of debt under different tax-regimes: Regressions on repeated cross-sectional data Since the demand for debt is largely driven by home purchases and the demand for other consumer durables, a structural model for borrowing behavior should at least contain two equations; one for the demand for debt and one for the demand for consumer durables. For our purposes, reduced form-regressions will do. A reduced-form equation for the demand for debt under progressive tax rates on interest deductions is Dit 0 1 I it 2 1 t D / 1 t e 1 t c it rD 3 Z it u it , rc or, since we assume that te and tc are equal for all tax-payers and rD=rc, simply Dit 0 1 I it 2 t D it 3 Z it uit , (5) where Dit=Household i's amount of debt at time t, Iit=Gross salary income tit=marginal tax rate on interest deductions Zit=vector of other exogenous variables like age, number of children, etc. uit= Error term with mean zero and constant variance The amount of debt is likely to depend positively on the amount of consumer durables owned by the household. Assuming that durables are normal goods, this will cause debt to depend positively on 5 King (1980) models a household's choice of tenure and demand for housing services as a joint decision that is affected by tax prices and other variables. 9 income (1). Alternatively, the stock of durables could be included in Z. This would probably lead to a drop in the estimated income parameter 1. A realistic model of consumer behavior should also take into account that purchases of durables (in particular home purchases) and borrowing are simultaneously determined, and that durables should be represented by instrumental variables. Endogenous marginal tax rates cause another, more serious problem. The parameter 2 (with positive sign) measures the impact of the tax rate from its effect on the user cost of debt. Under the progressive tax regime, the marginal tax rate on interest deductions depend on the income level as well as the amount of debt (amount of interest expenses). Consequently, we have a problem of endogenous disturbance terms (tD depends on D) and of multicollinearity (tD depends on I), as shown by the stylized tax function (6), where the tax rate is assumed to increase continuously with taxable income in a linear fashion: t D it 0 1 ( I it rD DD it ) it t D it 0 1 I it 2 DD it it (6) With endogenous marginal tax rates, the cross-sectional variation in observed marginal tax rates will typically be less than the variation in hypothetical, no-debt marginal tax rates and one may well observe high levels of debt together with low marginal tax rates. Also, even minor changes in the marginal tax rates of a progressive tax system is likely to offer limited possibilities with respect to measurement of the impact of the tax system on the demand for debt. For example, if the nominal marginal tax rates are reduced among the rich, it will increase the cost of debt financing and the taxpayers might respond by reducing their amount of debt until the marginal tax rate is restored on the original level. While it often can be difficult to track effects of the tax system on the demand for debt as long as the taxation of interest deductions remains progressive, a major tax reform with a shift from progressive and endogenous tax rates on interest deductions to flat, exogenous tax rates provides a great opportunity to compare the amount of debt held by households under quite different incentives. As shown below, it is possible to identify changes in the combined effect of income heterogeneity and taxation, by comparisons of the impact on debt from income before and after the tax reform. Substituting (6) for the tax variable tD in (5) gives the joint reduced form equation Dit 0 1 I it 2 Z it it , (7) where 10 0 0 2 0 , 1 2 2 1 1 2 1 , 1 2 2 2 3 , 1 2 2 and 2 u One can think of the demand for debt as consisting of a fixed component (independent of income, but possibly dependent of the households location, stage in life etc) and an income-dependent component. The relationship between income and debt is due to the fact that 1) owner-occupied housing and other consumer durables are normal goods and 2) the tax subsidy to borrowers is a function of gross income (in the case of progressive taxes), which reduces the cost of debt-financing for all borrowers, but more than average for those with the highest incomes and less than average for low-income groups6. So, as the tax system changes from progressive to flat taxes, one would expect debt to become less income-dependent as the rich repay their debt and the poor borrow more. In the simplified model (7), the income parameter 1 will contain the multiplicative term 21, which is zero after the tax reform and positive before. The term 2 2 is negative under progressive taxation but with interest deductions equal to rDD, then |2 | | 1 rD |< | 1 |, so 1 2 1 . 1 2 2 1 By pooling the cross sectional data for t=n years before and t=k years after the tax reform, and estimating the parameters of the equation Dit 0t 1t I it 2t Z it it , i I t , (8) we can test the hypothesis that 6 With a flat tax rate on interest deductions, the tax system can still cause debt to depend on income, because interest deductions works as a subsidy that reduces the user cost of consumer durables, which, in turn, increase with income. However, the percentage reduction of the user cost will now be the same for all taxpayers. With a flat tax rate, there will be no cross-sectional variation in marginal tax rates and the impact of interest deductions on the demand for debt will be hard to track. 11 H(1)1: 1,T-n = … = 1,T-2 = 1,T-1 > 1,T 1,T+1 = 1,T+2 = ... = 1,T+k, where T= year of implementation of the tax reform7, against H(0)1: There is no systematic differences in the 's before and after the tax reform (and no systematic similarities before the reform and no systematic similarities after). (2) Individual responses to changes in the tax incentives: Difference-in-differences using panel data Using panel data, we can identify one group of taxpayers that experienced a reduction of the marginal tax rate on interest deductions, another group that experienced an increase in marginal tax rates and one control group that was (almost) unaffected by the tax reform. We modify equation (5) above, by allowing for fixed (time-invariant) individual-specific effects, time-specific (individual-invariant) effects, and group- and time-specific effects: DiK ,t 1K 2t 3it 1 I it 2 t D it 3 Z it uit , (9) where K=H for high-taxed taxpayers in bracket B3, K=L for low-taxed taxpayers in bracket B1 and K=0 for the control group in bracket B2. By sweeping out the individual-specific effects by measuring all observations as the deviation from the previous period we get DiL ( 2t 2t 1 ) ( 3t 3t 1 ) iL 1I iL 2 t D iL 3 Z iL (ut ut 1 ) iL , (10) for the low-taxed group in bracket B1, and DiH ( 2t 2t 1 ) ( 3t 3t 1 ) iH 1I iH 2 t D iH 3 Z iH (ut ut 1 ) iH (11) for the high-taxed group in B3 (similar, the control group can be denoted by subscript 0). 12 Since E(2)=0, and by assuming that the change in income, I, and changes in other possible relevant variables, Z, on average is the same for the two groups, the difference-in differences estimator is: (DiL DiH ) ( 3t 3t 1 ) iL ( 3t 3t 1 ) iH 2 (t D iL t iH ) ( ) iK , (12) where E[( 3t 3t 1 ) iL ( 3t 3t 1 ) iH ] 0 if the two groups H and L are identical with respect to the determinants of the time- and individual-specific parameter 3, non-zero otherwise. The changes in the time- and individual-specific parameter 3 will typically be different if the two groups L and H respond differently to the impact of time. Figure 1 displays the cross-sectional average age patterns in the tax liability on net capital income before and after the tax reform, corresponding to the simulations in table 1 in the introduction. The age-tax patterns show a clear tendency among the taxpayers to borrow in the early stages of life, and repay their debt when they grow older. At the age of retirement, tax liability on net capital income is about zero, on average. Average tax on household capital income (expenses) by age of household head. Before the tax reform and after (current NOK) 5000 0 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 -5000 -10000 Before -15000 After -20000 -25000 -30000 Figure 1: Cross-sectional age patterns in the tax liability on capital income and interest expences. Wealth tax not included Source: Calculations with the micro simulation model LOTTE based on individual tax returns, 1986-2000. 7 The possibility that 1,T1,T+1=1,T+2 allows for slow (i.e., not immediate) adjustment in the transition year. The actual speed of adjustment will depend on many factors related to the financial flexibility of the households. 13 Due to the age-profiles in the amount of debt held by households indicated by figure 1 above, it is a possibility that one could observe a simultaneous increase in average debt and increase in marginal tax rates among the young as they get a full time job and settles. Similar, one could possibly observe a normal decrease in debt among older taxpayers, who also are most likely to have got a lower marginal tax rate on interest deductions after the tax reform. Certainly, the displayed age profiles calls for a careful interpretation of observed differences-in-differences. In section 5, we test the hypothesis H(1)2: that =H- L < 0 against H(0)2: =0. Further, we check whether the 's have the expected signs in the groups H, L and 0, while controlling for possible differences in the effects of ageing. 3. Description of the data and the tax system Our data source is the Survey of Income and Wealth (Statistics Norway), which is based on individual tax returns. The sample size varies from more than 9000 individuals in the 1980’s, increasing to an average of almost 28.000 in the 1990’s. The table2 below shows the sample size each year. Table 2. Sample size by year Year: 1986 1987 1988 1989 1990 1991 1992 1993 N: 14271 9582 9366 9319 22349 24451 24010 18404 Year: 1994 1995 1996 1997 1998 1999 2000 N: 41112 26305 37980 39504 38938 26825 34851 The annual cross-sections also contain a panel sub-sample, which is used in the difference-in-differences analysis. For the period 1991-1993, the net panel data set used in our calculations consists of 7056 individuals, after omitting certain observations. 14 Apart from taxable income from different sources (wages, self-employment, interest receipts, interest deductions, etc), the data also contains information on gross and net wealth, debt and to some extent portfolio composition, meant for calculation of the base for wealth taxes. In principle, the existence of a panel sub-sample could cause problems with biased estimates on the repeated cross-sectional data, as one might argue that the cross-sections are not (entirely) independent. However, it is hard to think of any reasonable mechanisms that can cause such bias and we have carried out the regressions as if the cross-sectional samples are drawn independently. Table 3 displays the maximum marginal tax rates on interest deductions from 1986 and on. Table 3, maximum statutory marginal tax rates Year: 1986 1987 1988 1989 1990 1991 1992-> tD, maximum 62 56 48 45.6 43 40.5 28 (55) While most tax-payers face a progressive (flat) tax rate before (after) the tax reform, a special arrangement applies to those who receive social security or public pensions. In short, these people pay min[T,s(Y-G)], where s= 50 (55) before (after) the tax reform, T = ordinary tax, Y=taxable income and G=ceiling on income. The probability of having the tax limited in this way depends on the amount of debt (interest expenses), which makes the marginal tax rate endogenous after the tax reform too. The difference-in-differences calculations are carried out with and without tax-limited observations included. [More details to follow...] 15 4. Estimation procedure and results The regressions on the repeated cross-sectional data are carried out at the household level, where all variables are aggregated within the household. Only single-family households are included. The difference-in-differences estimator is calculated on individual data, omitting observations with zero marginal tax rates. Tobit regressions of the amount of household debt Because many households have zero debt, the dependent variable is left-censored. To obtain non-biased parameter estimates in the regressions, we use the Tobit-estimator. The estimation procedure used is maximum likelihood, under two alternative assumptions about the distribution of the disturbance term: normal distribution and an extreme value distribution (on a log transformation). The estimations are carried out using the SAS Lifereg procedure. Only the results from the estimations under first assumption of normal distribution are reported in this version of the paper. The results from the Tobit regressions are reported in tables 4 - 7. 16 Table 4. Tobit estimates of household debt. Including income and time dummy variables only. By year # observations=102172, # left censored values=19749. All variables measured at 2000-prices. Normal distribution, Log likelihood=-1226858.9 Variable Parameter Chi-Square Significant Significant estimate at at 1 % level 5 % level Intercept -365276.9 329.41 x Dummy, year=1987 -47176.0 2.33 - - Dummy, year=1988 -67768.3 5.01 x Dummy, year=1989 34715.4 1.33 - - Dummy, year=1990 37335.9 2.24 - - Dummy, year=1991 -9185.7 0.14 - - Dummy, year=1992 245469.0 92.07 x Dummy, year=1993 483548.6 404.76 x Dummy, year=1994 335473.9 254.87 x Dummy, year=1995 301047.2 170.18 x Dummy, year=1996 563889.9 727.75 x Dummy, year=1997 689404.7 1081.1 x Dummy, year=1998 505977.0 575.49 x Dummy, year=1999 473745.2 424.67 x Dummy, year=2000 547418.9 1341.06 x Income in 1000 NOK 1691.3 1341.06 x Income x Dummy, year=1987 49.01 0.51 - - Income x Dummy, year=1988 190.13 7.84 x Income x Dummy, year=1989 -8.23 0.014 - - Income x Dummy, year=1990 54.7 0.89 - - Income x Dummy, year=1991 94.93 3.31 x Income x Dummy, year=1992 -506.47 80.16 x Income x Dummy, year=1993 -1125.7 477.05 x Income x Dummy, year=1994 -859.16 342.17 x Income x Dummy, year=1995 -884.12 291.70 x Income x Dummy, year=1996 -1358.9 866.18 x Income x Dummy, year=1997 -1622.7 1223.91 x Income x Dummy, year=1998 -1192.4 657.68 x Income x Dummy, year=1999 -1071.1 464.87 x Income x Dummy, year=2000 -1136.4 592.15 x (Income measured in 1000 NOK) 17 Table 5. Tobit estimates of household debt. Including income and time dummy variables only. By period. # observations=102172, # left censored values=19749. All variables measured at 2000-prices. Normal distribution, Log likelihood=-1231042.0 Variable Estimate Chi-Square Significant Significant at at 1 % level 5 % level Intercept 236679.4 9430.47 x Dummy, year < 1992 -274184.5 3328.82 x Income in 1000 NOK 230.94 48505.31 x Income x Dummy, year < 1992 684.63 23711.31 x (Income measured in 1000 NOK ) 18 Table 6. Tobit estimates of household debt. Including value of real business assets and durables. By year.# observations=102172, # left censored values=19749. All variables measured at 2000-prices. Normal distribution, Log likelihood=-1222110.2 Variable Parameter Chi-Square Significant Significant estimate at at 1 % level 5 % level Intercept -379202.2 389.68 x Dummy, year=1987 -12886.3 0.19 - - Dummy, year=1988 -53306.3 3.41 - - Dummy, year=1989 94025.6 10.77 x Dummy, year=1990 94185.4 15.67 x Dummy, year=1991 45193.0 2.92 - - Dummy, year=1992 192096.0 61.92 x Dummy, year=1993 378194.2 271.14 x Dummy, year=1994 286804.3 202.58 x Dummy, year=1995 239287.5 117.79 x Dummy, year=1996 424085.3 443.99 x Dummy, year=1997 510936.8 639.03 x Dummy, year=1998 409752.7 410.41 x Dummy, year=1999 378992.0 297.61 x Dummy, year=2000 443374.1 468.81 x Income in 1000 NOK 949.08 434.16 x Income x Dummy, year=1987 285.05 18.91 x Income x Dummy, year=1988 471.73 52.99 x Income x Dummy, year=1989 147.2 4.86 x Income x Dummy, year=1990 210.06 14.47 x Income x Dummy, year=1991 153.25 6.88 x Income x Dummy, year=1992 -96.19 3.03 - - Income x Dummy, year=1993 -542.80 115.49 x Income x Dummy, year=1994 -359.32 61.3 x Income x Dummy, year=1995 -435.67 73.46 x Income x Dummy, year=1996 -731.06 257.62 x Income x Dummy, year=1997 -910.81 396.61 x Income x Dummy, year=1998 -655.72 203.81 x Income x Dummy, year=1999 -539.18 121.87 x Income x Dummy, year=2000 -556.56 150.52 x Business assets 1.42 2365.39 x Business assets year < 1992 -0.56 261.43 x House and other durables 0.75 4794.48 x House and other durables year < -0.03 2.3 - - 1992 (Income measured in 1000 NOK) 19 Table 7. Tobit estimates of household debt. Including value of real business assets and durables. By period. # observations=102172, # left censored values=19749. All variables measured at 2000-prices. Normal distribution, Log likelihood=-1224341.0 Variable Estimate Chi-Squar Significant Significant e at at 1 % level 5 % level Intercept 46466.1 196.74 x Dummy, year < 1992 -181130.6 1032.73 x Income in 1000 NOK 133.44 12392.77 x Income x Dummy, year < 1992 451.19 6062.07 x Business assets 1.52 2818.80 x Business assets year < 1992 -0.61 313.42 x House and other durables 0.91 7170.70 x House and other durables year < 1992 -0.05 6.64 x (Income measured in 1000. Business assets and durables measured by their tax values) 20 Difference-in-differences in the change in individual debt The difference-in-differences estimator is calculated using the average change in tax rates and debt amounts for identical individuals from 1991 to 1993, omitting the transition year 1992. The sample is divided into three groups; one group where marginal tax rates increased after the reform, one group where marginal tax rates fell and one control group where marginal tax rates increased by only 1.5 percentage points (from 26.5 to 28 percent). Individuals with a zero marginal tax rate in 1991 or in 1993 were omitted. If there are effects related to typical age-profiles, it is possible that the time- and individual-specific parameter 3 in equation (12) differ between the groups H and L, i.e., E[( 3t 3t 1 ) iL ( 3t 3t 1 ) iH ] 0 . In figure 1 in section 2, the amount of debt seems to increase until the age of 40, then stabilize and decrease until retirement. To control for the effect of ageing, we have divided the sample into two sub-samples, below 40 and 40 years and older. The estimated difference-in-differences are reported in tables 8 and 9. 21 Table 8. Difference-in-differences, entire sample. Current NOK All Average change Average change Implied elasticity Average average Average age, in tax rate in debt, NOK (non-adjusted/adju income,1991. change in 1991 (percent points) sted) NOK income, percent Reduced (H) -10.81 -8224 0.12/0.20 236310 6.4 % 45 Control group (0) 1.50 5929 115503 17.7 % 42 Increased (L) 16.45 17153 0.55/0.36 95081 5.9 % 58 t value (H ne L) 96.37 4.10 Associated level <0.0001 <0.0001 of significance Below 40 Average change Average change Implied elasticity Average average Average in tax rate in debt, NOK (non-adjusted/adju income,1991. change in age, 1991 years (percent points) sted) NOK income, percent Reduced (H) -10.19 3429.71 -0.04/0.12 226223.52 9.0 % 32 Control group (0) 1.5 13413.85 118182.12 20.4 % 29 Increased (L) 21.29 37289.54 0.39/0.25 108891.61 13.8 % 29 t value (H ne L) 62.2 2.6 Associated level <0.0001 0.009 of significance 40 years and Average change Average cange Implied elasticity Average average Average in tax rate in debt, NOK (non-adjusted/adju income,1991. change in age, 1991 older (percent points) sted) NOK income, percent Reduced (H) -11.21 -15832.88 0.26/0.23 242895.88 4.8 % 53 Control group (0) 1.5 -1877.98 112707.54 14.8 % 57 Increased (L) 15.48 13118.07 0.57/0.65 92314 4.0 % 64 t value (H ne L) 75.23 2.86 Associated level <0.0001 0.004 of significance 22 Table 9. Difference-in-differences, excluding observations with possibly endogenous shift in marginal tax rates. Current NOK All Average change Average change Implied elasticity Average average Average in tax rate in debt, NOK (non-adjusted/adju income, 1991. change in age, 1991 (percent points) sted) NOK income, percent Reduced (H) -10.58 -8512 0.12/0.21 238039 6.3 % 45 Control group (0) 1.50 5929 115503 17.7 % 42 Increased (L) 20.93 21176 0.39/0.28 99284 5.0 % 57 t value (H ne L) 95.07 4.09 Associated level of significance <0.0001 <0.0001 Below 40 Average change Average change Implied elasticity Average average Average in tax rate in debt, NOK (non-adjusted/adju income, 1991. change in age, 1991 years (percent points) sted) NOK income, percent Reduced (H) -9.99 2941.28 -0.04/0.13 227581.39 8.9 % 31 Control group (0) 1.5 13413.85 118182.12 20.4 % 29 Increased (L) 24.83 33888.37 0.23/0.14 106110.64 14.9 % 28 t value (H ne L) 62.03 2.61 Associated level of significance <0.0001 0.009 40 years and Average change Average change Implied elasticity Average average Average in tax rate in debt, NOK (non-adjusted/adju income, 1991. change in age, 1991 older (percent points) sted) NOK income, percent Reduced (H) -10.97 -16019.27 0.26/0.23 244893.84 4.7 % 53 Control group (0) 1.5 -1877.98 112707.54 14.8 % 57 Increased (L) 20.02 18200.08 0.46/0.51 97685.64 2.5 % 64 t value (H ne L) 73.73 2.82 Associated level of significance <0.0001 0.005 5. Concluding remarks Summary of preliminary results The Tobit estimates of the impact on household debt from income show a remarkable shift from 1992 on. While the response of debt to cross-sectional income variations seems similar within the years before the tax reform and within after the years after, there are large and clearly significant differences between the pre- and post-reform period. As expected, adding the tax-valuation of consumer durables and real business assets among the explanatory variables reduces the estimated impact of income. However, the striking similarities within, and differences between the years prior to and after the tax reforms remains. 23 The Tobit estimations and difference-in-differences provide rather strong and significant evidence to safely reject the null hypothesis. It is hard to imagine other explanations for the observed pattern than the tax reform, but it cannot be ruled out completely. However, the estimated difference-in-differences will to a large extent control for such unknown factors, since it is carried out on identical individuals. In combination, the two results point at the change in tax incentives to be the major cause of the observed change in borrowing behavior. Further work with the paper The work is at an early stage and we hope to be able to improve the econometrics. While the present estimates of the relation between debt and income are based on a linear functional form, we will now concentrate more on what's going on in the top 5 or 10 percent of the income distribution. With respect to the estimated difference-in-differences, we have not yet been able to fully control for the impact of ageing, as demonstrated by the bottom panels of tables 8 and 9: In the group "40 years and older", the average age between the groups H and L differ slightly more than what we are comfortable with. We will now work more closely with this group to sharpen the results further. 24 References Agell, Jonas and Per-Ander Edin (1990): "Marginal Taxes and the Asset Portfolios of Swedish Households", Scandinavian Journal of Economics, 92 (1), 47-64. Alm, James and James R. Follain (1987): Consumer Demand for Adjustable Rate Mortgages. Housing Finance Review 6, 11-17. DeAngelo, Harry and Ronald W. Masulis (1980): "Optimal capital structure under corporate and personal taxation", Journal of Financial Economics, 8, 3-30. Follain, James R. and Robert M. Dunsky (1997): "The Demand for Mortgage Debt and the Income Tax", Journal of Housing Research, 8,155-199. Gordon, Roger and Joel Slemrod, (1988): "Do We Collect Any Revenue From Taxing Capital Income?", in Summers, Lawrence (ed.): Tax Policy and the Economy, Vol. 2, MIT Press, Cambridge, MA, 89-130. Gordon, Roger, Kalambokidis, Laura and Joel Slemrod (2002): Do We Now Collect Any Revenue From Taxing Capital Income? Paper presented at The Research Forum on Taxation ("Skatteforum 2002"), Losby Gods, April 29 and 30, 2002. Hendershott, Patrick H., Pryce, Gwilym and Michael White (2002): Household Leverage and the Deductibility of Home Mortgage Interest: Evidence from UK House Purchases. NBER Working Paper 9207, National Bureau of Economic Research, Cambridge MA. (http://www.nber.org papers/w9207). King, Mervyn A. (1980): An econometric model of tenure choice and demand for housing as a joint decision. Journal of Public Economics 14, pp 137-159 Maki, Dean M. (2001): "Household Debt and the Tax Reform Act of 1986", The American Economic Review, vol. 91 no. 1, 305-319. Miller, Merton(1977): "Debt and Taxes", Journal of Finance, 2, 261-275. Skinner, Jonathan and Daniel Feenberg (1990): "Impact of the 1986 Tax reform on Personal Saving", in Slemrod, Joel (ed.): Do taxes matter? Cambridge, MA: MIT, 1990, 50-79. Scholz, John Karl (1994): "Tax Progressivity and Household Portfolios: Descriptive Evidence from the Surveys of Consumer Finances" in Slemrod, Joel (ed.): Tax progressivity and income inequality. New York: Cambridge University Pres, 219-267. Sørensen, Peter B. (1994): "From the global income tax to the dual income tax: Recent tax reforms in the Nordic countries". International Tax and Public Finance, 1 57-79. 25