The influence of a mortgage interest deduction on house prices: evidence across tax systems in Europe

Abstract

In a large majority of European countries, a tax relief on mortgage interest payments is granted in order to enhance homeownership. Although there is a common belief stating that a mortgage interest deduction (MID) is capitalized into house prices, empirical evidence remains scarce. Therefore, our paper is about the influence of this fiscal relief on house prices. The scope of our analysis includes fourteen European countries, for which an unbalanced panel dataset over the period 1990–2015 is constructed. All our regression results support the hypothesis that a MID has had a significant price-increasing effect in the selected countries over the observed period. However, this result does not hold for a MID in the countries where a dual income tax (DIT) is applied, suggesting a significant difference between tax systems. These results are relevant for governments because there has been much debate about whether and to what extent countries should limit its MID over the last decades.

KEYWORDS:

• Fiscal Stimulus
• Housing Price
• Mortgage Interest Deduction
• Dual Income Tax

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 A DIT is a tax regime that combines a low proportional tax rate on capital income (i.e. interest, dividends, capital gains, etc.) with a progressive tax schedule for labour income (see e.g. Sørensen 1994; Cnossen 1999). In its pure version, the tax rate on capital income is aligned with both the corporate tax rate and the lowest marginal tax rate on labour income.

2 A single-country (US) study cannot test the influence of fiscal stimuli (such as a MID) across tax systems.

3 Caldera and Johansson (2013) indicate that supply elasticities vary widely across (and within) countries, but that they are usually low (<1) for European countries. According to this study of 21 OECD countries, Sweden and Denmark are the only two European ones with a supply elasticity above unity.

4 Our empirical research (with respect to Hypothesis 1) adds to Andrews (2010) by assessing the influence of a MID on (the level of) house prices directly and quantitatively (see also the robustness section) through a dummy variable that indicates whether a country grants such tax relief over time. Because we measure the effect of mortgage cost reliefs on house prices in a different way, we provide additional evidence about the link between the two. Our research differs even further from Andrews (2010) in terms of data and methodology (see the respective sections): (1) we investigate a smaller sample of (solely European) countries, with –on average– more observations per country; (2) we study a more recent time period i.e. 1990–2015; (3) we include a more compact set of (control) variables as well as use different proxies for similar control variables (such as (nominal) Gross Domestic Product (GDP) per capita instead of (real) household disposable income as a proxy for income); and (4) we employ random-effects (RE) models, as our baseline models, to analyze house price determinants.

5 A CIT is a tax regime in which, typically, the same progressive tax rate schedule is applied to all income, i.e. irrespective of its source (see e.g. Haffner 2002). A flat income tax is a tax regime with a single tax rate applied to income (see e.g. Paulus and Peichl 2009).

6 After all, multiple studies suggest that a MID does not boost homeownership, most probably because of capitalization effects (see e.g. Bourassa et al. 2013). The literature provides nevertheless several other arguments (i.e. aside from its distortionary effect on housing prices) against a MID (see e.g. Ventry 2014).

7 As can be seen in Table 1, the Danish tax system is defined as a hybrid between a DIT and a CIT in our study.

8 In a CIT, homeowners deduct interest payments at their marginal tax rate. In a DIT, capital income's marginal tax rate is usually lower than earned income's marginal tax rate. After having implemented a DIT, it is observed that the Nordic countries (i.e. Denmark, Finland, Norway, Sweden) classified mortgage interest as negative capital income. Mortgage interest payments are therefore deducted from income at a considerable lower tax rate than the (highest) marginal tax rate, which was applicable before (see e.g. Sørensen 1994; Haffner 2002).

9 The higher the MID, the lower the mortgage payments and thus the higher the homebuyer's ability to pay (ATP) (et vice versa). And a homebuyer's ATP ultimately translates into house prices (see e.g. Damen, Vastmans, and Buyst 2014).

10 Although we are aware that housing markets are typically local or regional, each country is treated as a single housing market. Our analyses at the country-level give an idea of a MID's influence on house prices, and differences in influence across tax systems.

11 Since 2000, interest was fully deductible (if not exceeding 120m²) in Greece. Ireland, Portugal, and Spain introduced a tax credit in 2004, 1998, and 1999, respectively. Undesirably, we do not have sufficient reliable information about these countries’ housing policies before these dates.

12 As can be seen in Table 1, we categorize three countries as applying a DIT over 1990–2015. The third condition (3) of Table 1, note a, indicates that the capital tax rate needs only to be at ‘approximately’ the same level as (and does not need to be perfectly equal to, like in the purest version of a DIT) the labour tax rate in the first income bracket, to qualify as a DIT in our study. After all, over the observed period, no country applied a tax system which contains all characteristic features of a DIT (see Cnossen 2016). In the three countries’ initial DIT reforms (i.e. Finland (1993), Norway (1992) and Sweden (1991)), only Norway perfectly aligned the capital tax rate with the labour tax rate in the lowest income bracket. In Finland, the capital tax rate was slightly higher (1.5 %) than the labour tax rate in the lowest income bracket (1.5%); whereas in Sweden, the capital tax rate was slightly lower (1.6%) than the labour tax rate in the lowest income bracket (see e.g. Genser and Reutter 2007; Bird and Zolt 2010).

13 To our regression models, we would also like to add some control variables that characterize the housing market (such as construction costs, homeownership rates, rental costs or rental stock). Unfortunately, data on these variables is not available to us for all investigated countries over our entire period covered. Though, as robustness checks, we also added (1) construction costs and (2) rental prices separately to our baseline (RE) models (i.e. model 1 and model 2, see the methodology section). The two variables were not put together in one robustness model, since they were found to be highly correlated (> 90%). Residential construction cost indices (2015 = base year) were obtained from Eurostat.; and actual rentals for housing indices (2015 = base year) (as a proxy for rental prices) were obtained from OECD Main Economic Indicators database. When construction costs or rental prices were included, the number of observations dropped from 323 to 268 or 286, respectively. A priori, we expected a positive coefficient on both variables: higher construction costs reduce residential construction (i.e. lower supply), and higher rental prices make buying a home more desirable (i.e. higher demand). Results are described in a footnote (25) in the robustness section.

14 Nevertheless, as a robustness check, we also ran our empirical baseline (RE) regression models (i.e. model 1 and model 2, see the methodology section) with construction costs added (see footnote (13)). And construction costs are a frequently used supply-side factor in housing price studies (see e.g. Andrews 2010; Sivitanides 2015). Results are desctribed in a footnote (25) in the robustness section.

15 A FE model only considers the within variation of data (see e.g. Šlander 2007). This means that the estimation of the coefficients of our key variables of interest (i.e. ‘Tax Benefit’ (model 1) and ‘Tax System’ (model 2)) only considers information from countries, which changed their tax policy over 1990-2015.

16 Nevertheless, we also fit fixed-effects (FE) models to the data, similar to our baseline (RE) ones. There are, after all, substantial cross-country differences in institutions that are fixed over time and which might affect house prices. The FE estimation results for model 1 (Hypothesis 1) and model 2 (Hypotheses 2, 2a and 2b) are reported in the second column (2) of Table 5 and Table 6, respectively; and briefly analysed in a footnote (22) under results.

17 Additionally, we expand our RE and FE (see footnote (16)) equations: year dummies are added. These dummies account for aggregate macroeconomic trends (such as the economic crisis of 2008-2009) which could have affected house prices. The results including year dummies for model 1 (Hypothesis 1) and model 2 (Hypotheses 2, 2a and 2b) are reported in the last two columns of Table 5 and Table 6, respectively; and briefly analysed in a footnote (22) under results.

18 There are two concerns here. First, the decision to eliminate the deduction may be related to economic and housing market conditions in the country. For example, a major economic downturn may force the government to raise additional tax revenue and one way of doing this is to eliminate the MID. Now it would seem that house prices go down because the country eliminated the deduction, even though the general economic downturn was responsible for both house price decline and the elimination of the interest deduction. Second, although this is very unlikely in practice, it could be the case that countries who experience price bubbles or large price swings are also more likely to abolish the deduction. As such, house prices drive the decision to eliminate the deduction, and not the other way around. However, we have looked at both in detail, but did not find such evidence.

19 Genser and Reutter (2007) define the Danish tax system as a hybrid between a DIT and a CIT. For taxpayers with positive net capital income, the tax system works like a CIT. However, for taxpayers with negative net capital income, the tax system works like a DIT. Therefore, as a robustness check, we also ran baseline model 2 (Hypotheses 2, 2a and 2b) when the Danish tax system was classified as a DIT. Similar results were produced. Moreover, Belgium, Italy and Portugal applied an indirect form of a DIT over the observed period. More specifically, all three countries introduced a final withholding tax on both interest income and dividend income, while applying a progressive tax scale on labour income (see e.g. Genser and Reutter 2007). However, we did not classify these countries’ tax systems as DIT, because the withholding tax rate is not aligned with the labour tax rate in the first income bracket (see Table 1, note d). Though, as an additional robustness check, we also ran baseline model 2 (Hypotheses 2, 2a and 2b) when both the Belgian tax system and Italian tax system were classified as DIT since 1993 and 1991, respectively (see e.g. Savjetovanje 2006). We did not include Portugal in this robustness regression, because we had no reliable information about the introduction year i.e. when Portugal introduced an indirect form of a DIT. Consequently, our number of observations dropped from 323 to 305. Again, similar results were yielded; except that the link between a MID in countries where a DIT is applied and housing prices, although with the expected sign, was no longer statistically significant.

20 As explained in the literature section, by switching to a DIT, mortgage interest became categorized as negative capital income and the marginal tax rate for capital income was lowered. Following the study of Berger et al. (2000), this should consequently lead to a fall in house prices.

21 From a large number of house price studies (see e.g. Meen 1999) we know that prices adjust relatively slowly. Hence, we also ran our baseline (RE) models with all explanatory variables lagged for one year (t-1). Results were very similar; although the positive influence of ‘Tax Benefit’ and ‘Inflation’ was now slightly less significant (5% level).

22 Neither a FE model, nor the inclusion of year dummies (in both a RE and FE model) changes the sign of any independent variable's coefficient (see columns (2), (3) and (4) of Table 5 and Table 6). It is however worth mentioning that the negative influence of a MID in countries where a DIT is applied, appears to be no longer significant. Nonetheless, both confirm that a MID generally has a significant positive effect on house prices (Hypothesis 1), and that there is a significant difference in a MID's influence between tax systems (Hypothesis 2).

23 ‘Dual’ is a dummy variable that takes the value ‘1’ when a country applies a DIT, and ‘0’ otherwise.

24 We do not include the main effect of ‘Dual’ in the interaction model (3), since it is perfect collinear with the interaction term.

25 As described in the data section (see footnote (13)), we also added (1) construction costs and (2) rental prices separately to our baseline (RE) regression models (i.e. model 1 and model 2) as robustness analyses. When construction costs were added, in both models, ‘Inflation’ turned out to be significant negative and ‘Income’ turned out to be no longer significant. Furthermore, in model 2, the negative influence of a MID in countries where a DIT is applied, appeared to be no longer significant. When rental prices were added, in both models, ‘Inflation’ turned out to be significant negative. Nonetheless, both robustness equations confirm that a MID generally has a significant positive effect on house prices (Hypothesis 1), and that there is a significant difference in a MID's influence between tax systems (Hypothesis 2).

26 The variable ‘Interest Rate’ is already provided in the form of percentage. The variables ‘Tax Benefit’ and ‘Tax System’ sometimes equal zero, for which the natural logarithm is not defined.

27 The percentage effect of the dummy ‘Tax Benefit’ (model 1) or categorical variable ‘Tax System’ (model 2), in which we are particularly interested, can be estimated by taking the coefficient's antilog (to base e) and subtracting 1 (see Halvorsen and Palmquist 1980).

28 Higher inflation might increase the longer-term attractiveness of home debt to the extent that it erodes the real value of mortgage debt over time, as mortgage debt is extended in nominal terms. Alternatively, lower inflation implies lower nominal interest rates, allowing financial institutions to increase the maximum amount that it can lend to households. It is, after all, not uncommon for financial institutions to impose lending limits based on repayment ratios (see Stevens 1997).

29 Our multi-country European study offers the possibility to test MID's effect on house prices across different tax regimes, which is not possible with a single-country (US) study.

30 However, as mentioned in the literature section (see footnote (6)), the literature provides several other arguments (i.e. aside from its price-increasing effect) against a MID (see e.g. Ventry 2014).

31 E.g. Sweden and Finland replaced its taxation of imputed rent by a property tax in 1991 and 1993, respectively.

Additional information

Notes on contributors

Wouter Vangeel

Wouter Vangeel received his degree in business engineering (2018) from KU Leuven. Currently, he is a PhD candidate in business economics at Vrije Universiteit Brussel (VUB). His research interest is in the area of fiscal policy and housing economics.

Laurens Defau

Laurens Defau holds a Ph.D. in business economics from the Vrije Universiteit Brussel. Currently, he works as a postdoctoral researcher at the Johannes Kepler University Linz. He is an experienced economist with an interdisciplinary mindset - having studied business economics, communication sciences, and history. His research interests include behavioral economics, digital marketing, fiscal policy, and housing economics.

Lieven De Moor

Lieven De Moor received his degree in business engineering (2000) and his PhD in applied economics (2005) from KU Leuven. He is professor of finance at Vrije Universiteit Brussel (VUB). His teaching activities cover corporate and international finance and quantitative methods. His research activities include empirical asset pricing, international finance, portfolio management, ESG investing, credit rating industry, financial literacy, SME finance, infrastructure investments, pension funds and housing economics.