HOUSING FINANCE: IMPACT OF HEAVY INTEREST INCOME TAX

Heavy taxation of interest income becomes a structural driver of property prices in a low-interest-rate environment. Inflation-adjusted price appreciation in 1996-2017 is approximately 200 basis points higher in 14 countries allowing no exemptions on interest income than in 37 countries that tax interest income at favorable rates or provide exemptions. Results for average returns over long-term periods are confirmed in models with annual frequencies, city-level data, and in a sample of 39 OECD countries for which price/rent ratios are available. It appears that investors view direct real estate, a heavily tax-favored asset, as an inflation hedge and/or alternative to fixed income asset. Higher interest income taxation may be fueling demand for direct real estate investments by retail investors. Separately, my empirical findings suggest that easy monetary policy effects can be magnified through the housing channel in countries that do not allow exemptions on interest income. Consequently, we should expect larger investment misallocations due to asset prices departure from fundamentals in some geographies.


INTRODUCTION
The global financial crisis has clearly demonstrated the macroeconomic imbalances the housing market can cause. It is telling that, in the aftermath of the crisis, the IMF established the Global Housing Watch, and the Federal Reserve Bank of Dallas created the International House Price Database. At the same time, rapid price run-up and collapse of house prices in the United States raised questions regarding the rationality of investor behavior in the real estate markets (Fu & Qian, 2014).
In parallel, a growing real estate finance literature attempted to explain variation in price dynamics prior to and after the implosion of the subprime mortgage market in the United States in 2008. Previously, several papers have documented changes in housing demand during the business cycle (Demary, 2010;Inglesi-Lotz & Gupta, 2013). This study examines the differential impact of short list of predictors. Further, expansion of predictor variables dataset would reduce sample size due to data unavailability outside the most developed markets.
One of the advantages of this study is that it puts to test a variable that is clearly exogenous in the context of employed econometric models. Whereas many macro and real estate sector-specific variables are tied together through feedback mechanisms, tax regimes change only infrequently, allowing to quantify the impact of tax policies on the housing market using relatively simple econometric techniques.
Methodologically, this paper is related to Liu and Mei (1992), who use common stock and bond returns as proxies for two risk factors to explain the performance of publicly traded real estate (REITs) and conclude that the real estate market is integrated with both capital market segments.

Model Specifications
We employ two model specifications in this study to differentiate between permanent and transitory effects -the short-run reaction of house prices to fundamental variables can be different from the longrun response (Adams & Füss, 2010;Kishor & Marfatia, 2017). First, we run an OLS regression with mean changes in inflation-adjusted real estate returns as the dependent variable, heavy tax categorical variable, and a group of control variables: Where, r is a mean inflation-adjusted return on real estate, HT is a heavy tax categorical variable and it X is a vector of country characteristics. Second, we re-visit the results using a panel data set and, annual frequency data: i i HT t it t-1,t-5 i,t-1,t-5 it r = α + β × HT + β × X + β × X +AR(1) + e , Where, i r represents an annual inflation-adjusted return on real estate, it X is a vector of static control variables, i, t-1, t-5 X is a vector of five-year averages for dynamic control variables, and AR(1) is a first-order auto-regressive term. My models with annual frequencies closely resemble the forecasting equation employed by Poterba et al. (1991). We put a heavy tax dummy to test in seven models. All OLS models with meanssee tables 5, 7, and 9use the same specifications as reported in panel A in table 5, but due to space considerations, I report only heavy tax betas and goodness-of-fit statistics. Likewise, all annual regressions reported in tables 6, 8, and 10 replicate models reported in panel A of Table 6. In each table, heavy tax betas are reported for five datasetsthe whole sample and high-and low-interest-rate environment in the United States and domestic markets.
Our research set-up is comparable with Poterba et al. (1991), who examine the interaction of high inflation rates and the U.S. income tax code changes that allowed households to deduct nominal interest payments on the housing market. The use of interaction low-interest-rate/heavy-interesttaxation variable does not alter conclusions reported in this paper, but we prefer to investigate taxation effects separately in high-and low-interest rate periods for clarity of exposition.
In annual regression models, we follow Lowry (2003), who introduced the autoregressive term of order one to account for non-stationarity in annual and quarterly IPO time series. In real estate finance literature, a similar approach was implemented by Gan (2010). The standard errors are Newey-West heteroskedastic autocorrelation consistent.

High-and Low-interest Rate Regimes Classification
Three areas represented by grey shading in figure 1 correspond to the high U.S. interest rate environment; the remaining two identify low-interest-rate periods. Choice of the U.S. interest rate regime for classification purposes is dictated by the growing integration of global markets and increasing correlation in housing market returns, a development documented in several studies on real estate (Pavlidis et al., 2016;Gomez-Gonzalez et al., 2018;de Bandt et al., 2010) and credit markets (Taylor, 2013). Yet, the weakness of this approach is its subjectivity, so I re-run my tests using a classification based on domestic interest rates. In the domestic markets low-interest-rate periods are identified as calendar years for which after-tax interest rate is below the domestic rate of inflation: Where, the rate of return on a 10-year government bond or another available interest rate that proxies for domestic risk-free rate (see Appendix A for sourced Datastream series) and T is marginal tax rate from the KPMG list (see data description). In addition to the national level data series, we put my hypothesis to test in a sample of capital cities. This allows for a more nuanced view of the heavy tax impact on housing prices.

DATA
There are several sources of data. National real estate prices are from the Organisation for Economic Co-operation and Development (OECD) housing prices database, and city-level data are from the residential property price series of the Bank of International Settlements (BIS). In addition, nationallevel data for the Philippines was added from the BIS database, and Teranet and National Bank of Canada data were used to track residential house dynamics in Toronto, Canada. My sample runs from 1996 to 2017. I obtain the real house price index by adjusting the nominal index changes by annual inflation rates.
Classification of income tax regimes is from Horan and Robinson (2008) and is based on information provided in International Business Guides from Deloitte Touche Tohmatsu and 0.00 1.00 2.00 3.00 4.00

5.00
PricewaterhouseCoopers online database of worldwide taxation. Fourteen out of fifty-one nations in my sample tax interest income in full. Table 1 reports geometric returns for fifty-one national markets and twenty-seven capital cities, the number of years for which data are available, and classification by type of interest income taxation.
Data on GDP growth, population changes, and foreign exchange rates were sourced from the World Bank and the International Monetary Fund for Taiwan. Net migration is reported by the World Bank in five-year intervals, so each figure was spread over a five-year period and scaled by starting population level to measure annual migration impact. Appendix A reports sources for interest rates and building permit series from the DataStream database system. The use of a 10-year Treasury bond yield as a measure of nominal long-term discount rate is consistent with Lai and Van Order (2017) and Campbell et al. (2009). Finally, total credit to households for forty-two national markets was obtained from the Bank for International Settlements credit to the non-financial sector (CRE) dataset, and tax rates were extracted from the KPMG individual income tax rates table (KPMG, 2021).

EMPIRICAL ANALYSIS Real Estate Capital Gains: Preliminary Analysis
Mean inflation-adjusted returns for heavy and light interest tax regimes are reported in table 2. Average real price appreciation equaled 3.45 percent versus 1.79 percent, and most of the gap is due to large price increases during low-interest rate periods in heavy tax geographies (see panels B and C in table 2).
Pairwise comparisons could reflect differences in cross-country characteristics, including institutional and legal frameworks. To isolate these effects, I examine housing price dynamics in five OECD-member countries with a common law legal system -Australia, Canada, New Zealand, the United Kingdom, and the United States, for which data are available for the whole 22-year period (see table 3).
Two patterns emerge. First, prices in heavy tax geographies exhibit stronger downward stickinessmassive price declines in 2008-2009 and 2011 in the United Kingdom and the United States are strikingly different from price dynamics in the other three common law countries. Consequently, over the 1996-2017 period, countries offering tax exemptions recorded a lower return, but much higher volatility. Second, the better performance of heavy tax countries is time dependent; heavy tax countries underperformed the U.K./U.S.A. pair by 1.4 percent in 1996-2006 but outperformed it by 4.1 percent in 2007-2017 with the largest differences accumulated during the bear markets (see table 3).
We proceed to examine pairwise correlations of inflation-adjusted returns, tax variables, and control variables. Pairwise correlations are reported for the whole sample and in high-and low-interest rate subsamples (see table 4). Panel A reports correlations for means and panel B includes estimates for annual data frequencies.  Preliminary analysis using pairwise correlations confirms that heavy tax impact could be regime-dependent (see table 4). In correlations for means (panel A of table 4), heavy tax dummy attains larger value when the U.S. interest rates are low, but not when domestic classification is used. Results in panel B for panel data with annual frequencies are more in line with expectations both for the heavy tax dummy and control variables -correlations are stronger and more significant when interest rates are low in both the U.S. and domestic interest rate classifications. GDP growth is significant in all samples and change in building permits is positive and significant in all but one. Other control variables attain larger significance in higher frequency data (panel B). Higher interest rates and domestic currency depreciation dampen demand for housing, change in credit to households fuels price increases, whereas immigration and inflation take on expected signs and are significant in most subsamples.
Further, marginal tax rates are positively related to price appreciation over longer periods in the whole sample and two subsamples (panel A in table 4). A positive correlation of marginal tax rates with price increases could be due to the tax-exempt status of the main residence. The effect can be similar for high-tax geographies in the same way interest payment deductions in computing taxable income had a larger impact on high-tax rate households in the United States (Poterba et al., 1991). The marginal tax variable is included in regression models 1 and 7 in tables 4 -10. Table 5 reports baseline estimates of regression models with means. Heavy interest income taxation is associated with an additional 1.3-2.4 percent increase in inflation-adjusted house prices (see panel A). In a low-rate environment, the magnitude is 1.6-3.5 percent higher depending on the classification used (see results in panels B and D). Interestingly, the marginal tax variable also attains a positive value in panel A, suggesting larger price appreciation in high-tax geographies. However, it does not render heavy tax variables insignificant.

MULTIVARIATE ANALYSIS
In annual frequencies, heavy tax variable attains a positive sign in models for the whole sample, but it is statistically significant in only some of the regressions in low-interest-rate subsamples (see table 6). Results suggest that heavy tax effects may best be measured over longer intervals.
Further, the momentum effect may be capturing some of the tax impacts in higher frequency data. My estimate of 0.49-0.51 for AR (1) term almost exactly matches Hwang and Quigley (2006), and my goodness-of-fit statistics of 33 percent to 41 percent are slightly higher than their R-squared of 23 percent reported for an autoregressive model with one term and no control variables.
Results in annual regressions lend partial support to the main hypothesis tested in this study. Heavy tax impact on housing prices is regime-dependent; heavy tax betas in panels B and D in table 6 are clearly different from estimates in panels C and E.
Next, we examine whether results can be replicated using data for twenty-seven capital cities, of which seven represent heavy tax geographies. Statistical significance for city-level OLS regressions with means are slightly weaker than for national markets, but heavy tax impact is slightly larger in magnitude than in the national market's sample, and its influence is larger in the low-interest environment in the United States (table 7).
Results in table 7 are weaker for classification based on domestic rates; in fact, heavy tax variable attains marginal significance in two models when domestic rates are high (see panel E). This could be due to momentum behaviorthe bull market can be triggered in a low-interest rate regime, but spillover effects may be felt in its aftermath. In city-level models with annual frequencies, heavy tax variable attains significance in all models when the U.S. rates are low except model 5 (panel B table  8). However, lower significance in model 5 is a consequence of a smaller sample sizewhen building permits and household credit are tested interchangeably, more observations are used, leaving heavy tax variables significant at a ten percent level. Overall, regression models for capital cities have slightly lower explanatory power, but this result could be due to the use of control variables specified at the national level and/or smaller sample size.
Finally, we examine heavy tax impact on changes in price-rent ratio, which corresponds to a price-dividend ratio for stocks and is used as an over-valuation metric in real estate finance literature.
Our sample in tables 9 and 10 are based on data for thirty-nine national markets from the OECD housing prices database. Heavy tax impact varies depending on interest rate regime; the strongest results in OLS regressions with means are observed in the low domestic rates subsample (panel D of table 9). Heavy tax attains significance in five models out of seven and is only marginally insignificant in models with demographic variables, in which it attains a p-value of 0.11 and 0.12, respectively (models 3-4 in panel D of table 9).    Results are much stronger for the Newey-West regressions with annual frequencies reported in table 10. The heavy tax dummy is significant in all models in panel D and is significant in all but one model in panel A. Even then, in model 3 of panel A, it's only marginally insignificant with a p-value of 0.114.
A certain pattern emerges in annual frequency regressions. Heavy tax impact on national house prices is felt when domestic interest rates are low (see tables 6 and 10). In city-level annual data, influence is observed when the U.S. embarks on easy monetary policy (table 8). The difference could be due to the arrival of immigrant cohorts, which may have larger exposure to international rather than domestic rates. This is indirectly confirmed by the weaker statistical of significance of heavy tax variables in regression models with demographic controls. Table 9. Heavy tax betas in regressions with means for price/rent ratio. Control variables are suppressed in all tables, except panel A in tables 5 and 6 due to space considerations (they are available upon request). One can indirectly gauge the impact of dependent variables by comparing goodness-of-fit statistics. For example, in panel A of table 5, R-squared is larger in models 5-7, which include building permits and GDP growth. GDP growth and building permits are significant in all specifications, foreign exchange variable attains significance in models with annual frequencies, and immigration is significant in some models. The presence of a statistically significant relationship between inflation-adjusted house prices and economic activity chimes with Kishor and Marfatia (2017) and validates my model specifications.

Panel A. Whole sample
Further, my estimates of economic growth influence, which range from 0.4 in Newey-West models with annual frequencies to 0.7 in OLS regressions with means, are in line with Adams and Füss (2010), who report that 1 percent increase in economic activity raises demand and house prices over 0.6 percent in the long run in a sample of 15 OECD countries in 197515 OECD countries in -2007 An increase in building permits is positively related to housing price increases in all regression models. Economic theory suggests that long-term relationships should be of opposite nature - Case (2008) observes that, in the U.S. market, the housing cycle peaked four times in 1972, 1978, 1984, and 2006 every time housing starts rose above 2 million on an annualized basis. However, my estimates are in line with previously reported results - Hwang and Quigley (2006) report a positive coefficient on housing supply for single homes in 74 U.S. metropolitan areas in 1987-1999. Separately, Case and Shiller (2003 argue that housing starts may proxy for supply restrictions. In a similar vein, population growth should have a positive association with demand for real estate, but I do not document such a relationship in my tests. However, Myers and Pitkin (2009) and Mankiw and Weil (1989) suggest that it is not population growth per se that impacts real estate prices, but changes in age structure and arrival of age cohorts in the household formation stage.
It is less surprising that real interest rate appears unrelated to housing prices -interest rates are negatively related to housing prices in pairwise correlations for annual data series (panel B of table 4), but the relationship is not strong enough to survive in multivariate models. My results do not contradict existing literature - Kishor and Marfatia (2017) find a positive relationship between real interest rates and house price dynamics in five out of fifteen OECD markets, and Kuttner (2012) shows that the effect of interest rates on house prices is relatively modest.
Many of the control variables in this study exhibit the same pattern of behavior as in Arrazola et al. (2015), who documented high demand sensitivity to the labor market situation and, to a lesser extent, to demographic changes, but the much smaller impact of real interest rate in the long run in the Spanish housing market in 1975-2009. Robustness checks in this study included the use of fixed-year effects and one-period lags in models with annual frequencies. Their impact on my results was limited. Outliers diagnosis using DFFITS statistic failed to identify influent observations, and size-adjusted cutoff suggested by Belsley et al. (1980) picked out at most two observations in my regression models. Their elimination lowered the p-value on heavy tax variables, but only marginally (results available upon request).

CONCLUDING REMARKS
Several studies have previously documented international transmission of real estate market bubbles prior to the housing crash in 2008(in't Veld et al., 2014Gomez-Gonzalez et al., 2018). Yet, in its aftermath, international real estate markets diverged.
This paper tested the impact of a heavy tax on interest income in both long-term series and higher frequency data. It showed that housing price increases were larger in countries with heavy interest income taxation. I do not suggest a central role of interest income taxation for predictable changes in housing prices. Several variableseconomic growth, building permits, and autoregressive term that captures momentum effecthave a larger explanatory power in tested models than my heavy interest income tax variable.
It is possible that heavy tax acts as a trigger in a low-interest-rate environment, but its ripple effects manifest over longer periods due to inertia. This would explain continuing outperformance of real estate markets in several heavy tax common law countries against the backdrop of rising interest rates at the tail end of my sample.
It is an established fact in the economics literature that easy monetary policies impact aggregate demand differently depending on the fiscal policy stance (Freedman et al., 2009). Given the positive relationship between housing wealth and consumption growth (Gan, 2010), easy monetary policy effects can be magnified through the housing channel in countries that do not allow exemptions on interest income. Consequently, we should expect larger investment misallocations due to asset prices departure from fundamentals in some geographies.
Understanding whether the macroeconomy or specific housing market conditions drive prices is crucially important for public policy initiatives. A case-in-point is a fiscal stance in Canada, which historically adjusted its monetary policies in lockstep with the United States, its largest trading partner, and competitor. However, if different interest income taxation regime induces higher price inflation in the residential sector in Canada, easy monetary policies may stimulate corporate sector and consumption directly and in addition promote excessive house price inflation, creating wealth effect and a further increase in consumption.
In addition to policymakers, my results could be of interest to the investment community. Real estate could be viewed by investors as an inflation hedge and/or alternative to the fixed income asset class in a low-interest environment due to expected appreciation on par with inflation. When interest rates drop, households in heavy tax geographies may be replacing fixed income allocation with additional exposure to the real estate asset class. This effect can be mitigated by an increased allowance for tax-free savings accounts (TFSAs), but these have been introduced quite recently -Canada mandated TFSAs in 2009, and the first contribution was capped at a meager $5,000, hardly enough to absorb large allocations to the fixed income asset class.
Interestingly, in the wake of the real estate boom, Canadian public debate shifted to foreign ownership of the real estate in gateway cities -Vancouver and Toronto, but no concerns have been raised about national tax policies which may be partially responsible for the domestic real estate boom (Deschamps, 2018).
Finally, my results are relevant for discussions within the academic community. My findings support the market efficiency hypothesis in the housing markets by identifying an additional factor that explains higher direct real estate returns in geographies without tax exemptions on interest income.
One of the shortcomings of this study is that it does not test several predictor variables that have been shown to impact housing prices in earlier literature, including credit spreads, survey measures, personal savings rates, and consumer confidence (Bork & Müller, 2018). The reason is data scarcity for international markets. However, this gap will eventually be filled in with the growing disclosure of economic data and its dissemination through online channels. I invite further research in this area. Interest rates are measuredin descending orderusing yields on 10-year government bonds, instruments with shorter maturities or bank regulators' re-financing rate.

Copyrights
Copyright for this article is retained by the author(s), with first publication rights granted to the journal. This is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/4.0/)