Brazilian REIT: Alternative Investment to Real Estate, Stock and Bonds

Given the recent expansion of real estate securities in the Brazilian market, the present study examines Brazilian REITs (FIIs) returns’ exposure to underlying market returns (real estate, stock and bond) in order to assess evidence of diversification power provided by this investment type in light of Modern Portfolio Theory. The research considers a sample of FIIs listed on the São Paulo Stock Exchange, during the period of 2008-2014, applying Clayton and Mackinnon´s (2003) methodology, where the econometric model explaining REIT returns is decomposed into three market factors. Results indicate that although FII returns reflect their hybrid nature, the proposed model is not sufficient to explain their total returns, suggesting that FII performance is not primarily driven by any of these underlying markets. In fact, FII would consist of a unique asset class and as such may provide diversification benefits in a mixed portfolio, depending on the selected FII type.


Introduction
Brazilian REITs -Real Estate Investment Trust-known in Brazil as FII (or "Fundos de Investimento Imobiliário"), were created in Brazil in 1993 and have experienced rapid expansion over the last five years, boosted by regulation changes initiated in 2008.Total primary emissions increased from BRL 1 billion in 2007 to BRL 11.5 billion in 2012, and the secondary emissions reached BRL 3.5 million in the same year (UQBAR, 2009).
REITs combine features of different markets: the real estate marketmain source of REIT income -and the stock market, environment in which they are traded.REITs benefits from taxes incentives, however, investment is subject to specific regulatory obligations that limit investment management, when compared to stock investments.Moreover, REITs tend to be compared to bonds, as part of gains comes from regular rent distributions (Bertin, Kofman, Michayluk, & Prather, 2005).
The impact of each of these underlying markets in behavior, performance and risk of this investment type, however, has not been clearly defined, being one of the major themes under analysis, both in the academic literature, as in the international funds industry.
Understanding the nature of this financial asset has implications that go beyond the analysis of this investment alone.To support this, we can resort to Modern Portfolio Theory (Markowitz, 1952).According to this approach, the characteristics of the investments, mainly its correlation/covariance with other asset classes, can maximize returns and/or minimize the total risk of a multi-asset portfolio, through diversification.Thus, by increasing the total return of a portfolio, with no increase in risk, or by keeping the return of this portfolio, with reduced volatility, efficiency frontier can be increased (Case, Yang, & Yildirim, 2012).
The main objective of this paper is to analyze whether listed real estate securities distinctive features can characterize Brazilian REITs as an alternative investment despite exposure to real estate, stock or bonds markets, considering the Modern Portfolio Theory approach.
Therefore, we developed an empirical study, from a sample of Brazilian REITs listed on the São Paulo Stock Exchange in the period of 2008-2014.Additionally, the present study aims to contribute to the expansion of national and international research on Brazilian REITs, taking into account that scientific analysis of Brazilian REITs is still scarce in Finance literature.
Finally, our findings have practical implications for REIT industry and investors as well.
The remainder of this paper is organized as follows.Section 2 presents theoretical background and brief overview of related literature.Section 3 introduces the selected data during the examination period, the model framework and the variables description; Section 4 provides empirical evidence and Section 5 presents concluding remarks.

Brazilian REIT overview
The Securities and Exchange Commission of Brazil (CVM) defines FII-Fundo de Investimento Imobiliário (or Brazilian REIT) as a capital communion raised through the securities distribution system, intended for application in real estate projects.Therefore, Brazilian REIT may invest in real estate assets, fixed income and other real estate securities, provided that at least 75% of fund assets are applied to real estate properties or rights (Securato, Amato, Takaoka, & Lima Jr, 2005).
Brazilian REIT is a closed-end investment fund, which means it is publicly traded, raising capital through an initial public offering and investors are not allowed to redeem their investment from the fund.Thus, REIT shares shall be traded at secondary market.
Dividend yield comes from income distribution related to lease contracts or asset sale, both exempt from taxes, which attracts mainly individual investors.At least 95% of income must be distributed every semester.In addition, investor may profit from shares traded at stock exchange.
Regulation changes in 2008 and 2009 significantly expanded the types of real estate investment allowed for REITs and the entry of pension funds, driving a new and stronger expansion of this market.

Modern Portfolio Theory
According to Modern Portfolio Theory -MPT, systematized by Harry Markowitz (1952), an optimal allocation of resources consists of selection of financial assets that promote the maximization of returns, given a risk limitation (Geltner, & Mcgrath, 2007).Markowitz (1952) proposed a methodology for portfolio selection through quantitative analysis of mean-variance characteristics of assets, where expected values (mean) is a measure of return and the variance of the overall rate of return (or standard deviation) is a measure of risk (Muller,1988).
More importantly, Markowitz mathematically demonstrated that the mere selection of different assets do not necessarily incur in diversity benefits, but the optimal combination of assets with low or negative correlation would improve risk-return characteristics of overall portfolio, since the total variance is impacted by covariance between these assets (Markowitz, 1990).Thus, combinations of low or non-correlated assets would maximize return and minimize the risk of a portfolio comprising the so-called efficient frontier.A point on the efficient frontier corresponds to the lowest volatility for a given return or the maximum return for a given level of risk.
The benefit of diversification of a portfolio assumes that different asset classes react differently to economic changes, in other words, different asset classes do not substantially share common characteristics of risk-return.
Alternative investments are asset classes that have significantly different risk and return characteristic when compared to traditional investments (eg.bonds and stocks), and consequently are expected to provide risk diversifying benefits in mixed portfolios by broadening the efficient frontier set (Donald, 2013, Ibbotson, 2006).
Academic studies indicate real estate as an investment alternative for diversification (Bekkers, Doeswijk, & Lam, 2009).Real estate securities, due to its hybrid characteristics of real estate, stock and fixed income, on the other hand, is subject of further investigation.

International literature
The study of influence of real estate, stock and bonds market on REIT returns and real estate securities as alternative investment have been the focus of major scientific research efforts (Bouldry, Coulson, Kallberg, & Liu, 2012;He, Webb, & Myer, 2003), however, there is no general consensus on findings.
According to Glascock, Lu, So (2000), considering that a relevant percentage of the REIT´s capital resources have to be invested in real estate assets, it is intuitive that REIT prices follow property market, however, as there are more market participants, especially non individual investors, it is possible that the stock market factors have a greater influence on return behavior, partially due to increase in the volume of transactions and the proper monitoring of the market.Westerheid (2006) found evidence that international REITs (samples from 1990-2004 of 8 countries) are a different asset class from bonds and stocks, and that REITs tend to function as substitute to direct real estate investments in the long term.Other empirical researches on international REITs, and especially US REITs concluded that REITs behavior differ in time, tending to behave as stocks in the short term and as direct real estate investments in the long term (Bouldry et al., 2012;Clayton;&Mackinnon, 2003;Hoesli, Oikarinen, 2012;Ling, & Naranjo, 1999;).On the other hand, cointegration with bonds was found by Giliberto (1990) and He, Webb and Myer (2003).
Lee e Stevenson (2005) and Sebastian and Zhu (2012) focused on the diversification role of REITs in a mixed portfolio and their benefits concluding that despite major influence of real estate market in return behavior in the long term, their characteristics of the risk-return make them an unique alternative asset class, different from stocks, bonds or real estate itself.Muller and Muller (2003) reached the same conclusion, through sensitivity analysis on hypothetical mixed portfolio comprised of bonds and stocks and bonds, stocks and direct real estate.Applying Markowitz approach, they demonstrated that the inclusion of REIT in a mixed portfolio substantially improves efficient frontier design in both scenarios.

Brazilian literature
While international research on REITs is quite vast, the subject was still little explored in Brazilian literature, especially after 2008, when the Brazilian REITs had its greatest expansion.Securato et al. (2005) concluded that the price of Brazilian REIT´s shares were impacted by dividends payment; secondly, that FIIs were perceived as non-liquid assets with high risk when compared to other investments available in the market and that the profitability was in general heavily influenced by real estate industry.Consentino and Alencar (2011) and De Castro (2012) found that the increase in prices of FII shares, samples from 2007-2011/2012, also followed property market.
On the other hand, FII shares appreciation indicated gains not only by dividends distribution, but probably from greater exposure to the capital markets.No cointegration with stock market was found also suggesting that REITs would function as hedge for stock investments (De Castro, 2012).
Considering sample from 2011-2013, Milani and Ceretta (2013) have not found significant correlation between FII returns and construction companies' performance, but have concluded that REIT relative high average returns and low standard deviation suggest that these funds are an interesting alternative to the investor, since further increase in return of the portfolio did not incur in additional risk.

Sample and Data
The overall sample includes 37 real estate funds listed on the Stock Exchange of São Paulo (BM&F Bovespa), extracted from ComDinheiro data base, a new high-end financial data platform developed by the Business Faculty of São Paulo University.This data base is web-based, easily accessible and focused on providing financial analysis tools for students and researchers in Economics, Finance and Accounting.
Real estate funds included in the sample represent approximately 32% of total listed funds.Only the actively traded funds with monthly data available for at least 3 consecutive years were selected, covering the period of 2008 to 2014.
Brazilian REIT market experienced significant regulatory changes over the last eight years and for this reason data before 2008 was not included in the sample despite sample size limitations.

Research Design
The econometric model applied to the empirical study is based on the methodology developed by Clayton and Mackinnon (2003), which assumes that the market returns can act as proxies for unobservable state variables that are common to REITs, Stocks, real estate and bonds.The linear regression and decomposition of the total variance of REIT is analyzed against these three components: Real Estate Market, Stock Market, Bond Market, as described below: where the RFII corresponds to the Brazilian REIT return, calculated by the monthly variation in the share's closing price adjusted for dividends paid; RRE the return of the direct real estate market, calculated by the monthly variation of a rate of return of the property market; RS is the return of the stock market, calculated by the monthly variation in the closing price of a stock index; RB the return on bonds, calculated by the monthly variation of a fixed income security.Clayton and Mackinnon (2003) assumed that regression residual or error term (ε) represents the idiosyncratic risk.Nevertheless, as they highlighted, this portion of REIT returns not explained by the market factors may comprise other variables not stated in the model, not necessarily idiosyncratic risk.Thus, we will not address the error term as idiosyncratic risk at this point.

Pure factors
According to Clayton and Mackinnon (2013), the pure form of variables was extracted through orthogonalization method, which consists in regressing each return variable, in this case RRE (real estate) and RB (bonds) against the other return factors.The final variables included in the model will be, then, the residuals of those regressions.Thus, residuals represent the pure influence of each market.
where: ν is the constant term; ω1 and ω2 : regression coefficients (or beta); RS: Stock market returns; е: regression residual that corresponds to bonds market pure factor (RB*).
This way, final econometric model correspondents to the equation bellow:

Variance decomposition
Thereafter, we proceeded to the decomposition of the variance of the Brazilian REIT returns, calculating the portion attributable to the variance of each independent variable as applied by Mackinnon and Clayton (2003, p. 43): The individual contribution of each component is calculated by dividing each component (Βi 2 * σ 2 ) by the total variance of Brazilian REIT returns (σ 2 RFII).

Brazilian REIT Returns (RFII)
The proxy selection for Brazilian REIT market (dependent variable) is a sensitive step of the study, due to data limitation.Although created in early 90s, the Brazilian REIT market was only recently expanded.
The IFIX -Brazilian REIT return index published by Sao Paulo Stock Exchange was adopted by Milani and Ceretta (2013) in their study, however, this index began only in 2011, making its period of analysis very limited to the purposes of our study.Also the index construction does not allow further analysis by property type.
As such, a return index of Brazilian REIT from a hypothetical portfolio was built, composed of the individual REITs available in each period, selected according to the criteria described in Sample and Data.
The monthly return of each REIT was calculated from the natural logarithm of REIT closing price as of the last day of each month in relation to previous month.Price corresponds to share price adjusted by dividends paid, as to capture gains distribution in the total REIT return index.The final index of the hypothetic portfolio was then calculated from the average of the returns weighted by the size of the individual measure by equity value.
It is noteworthy that this return calculation differs from the dividend yield, commonly considered a return measure in the Brazilian REIT market.The dividend yield considers only the rent paid per share divided by the share´s price.This metric shows the relative return on investment, and does not capture the appreciation/ depreciation of the REIT shares.

Real Estate Market Returns (RRE)
Structured information and statistics on the Brazilian real estate market are quite scarce.Information about real estate transactions, valuations and market studies are considered strategic for companies operating in the sector and therefore rarely available to external users.Therefore, a major challenge of this kind of study is the selection of a suitable proxy for the direct property market returns.
De Castro (2012) tested two indexes in his cointegration model: the IMOB and the INCC.Milani and Ceretta (2013) also used the IMOB index as a proxy for the real estate market.
The IMOB published by São Paulo Stock Exchange measures the performance of shares of leading real estate companies traded on the stock exchange.The INCC (National Index of Construction Cost) is an inflationary price index published by the Getúlio Vargas Foundation, which measures the evolution of housing construction costs, equipment, service and hand labor.
Although IMOB and INCC may provide some sense of the real estate market performance, these rates do not directly address the not securitized real estate market returns.IMOB is highly correlated with stock market, so its inclusion in the model as a proxy of the real estate market may lead to a biased conclusion of influence of this market on Brazilian REIT returns.In the other hand, the INCC is also highly correlated with the economy interest rate (Selic), and as such, probably would present multicollinearity with NTN-B.
Therefore, this study broke new ground by adopting a proper index return of the property market, the IGMI-C index from Fundação Getúlio Vargas (FGV).
According to the definition provided by Fundação Getúlio Vargas (FGV): "IGMI-C is a profitability index of the Brazilian market of commercial real estate, which aims to portray the most comprehensive way possible the evolution of price appreciation and income of the commercial property segment throughout Brazil ".
The composition of this index has resemblance to the composition of Brazilian REIT type's distribution in the hypothetical portfolio in the full sample, covering commercial offices, malls, hotels, industrial and logistic facilities and others.It is noteworthy that the highest concentrations are found in commercial offices (about 50% of the total) and malls (about 25% of total) (FGV, 2011).
The IGMI-C presents data available since 2000, but on a quarterly basis.Due to the limitation of REIT observations in the sample, it was necessary to adopt a monthly frequency.The IGMI-C index aggregates both income returns (IGMIC income) as capital gains (IGMIC capital).

Stock Market Returns (RS)
The IBRX-100 index measures the return on a theoretical portfolio composed by 100 stocks selected among São Paulo Stock Exchange's most actively traded stocks in terms of number of trades and financial value.This index was selected as proxy for overall Brazilian Stock Market.

Bond Market Returns (RB)
International studies (Giliberto, 1990;He, Webb, & Myer, 2003) have concluded that investment in real estate funds have similar behavior to investment in fixed income securities, since a significant part of their income comes from space lease.
The NTN-B is a leading Brazilian direct treasury bill and was chosen as a proxy variable for Brazilian bond market.In order to capture investor's future expectations a long maturation bond (2025 bond) was selected.

Descriptive Statistics
Means and standard deviations of monthly returns variables RFII, RRE, RS and RB are presented in Table 1.Brazilian REIT Portfolio´s return (RFII) was on average superior to stock and bonds average returns in the sample.The standard deviation as a measure of return´s volatility, or risk, of RFII was lower than stock market measured by IBRX100 index; however, higher than direct real estate market (IGMI-C) and bonds (NTN-B).

Correlation
Pearson correlation matrix is presented in table 2. As observed, despite general belief, correlation between RFII and IGMI-C (Real estate variable) and NTN-B (bonds variable) was not statistically different from zero.RFII was positively correlated with stock market, but less than 50%.

Regression Analysis and Variance Decomposition
Preliminary analysis through dispersion scatterplots and Stata´s tests indicated that model framework meets the econometric requirements for linear regression analysis, such as minimum sample size, normality of residuals and linearity between dependent variable and predictors.
Model framework was also evaluated for heteroskedasticity through Breusch Pagan/Cook Weisberg test which tests the null hypothesis that the error variances are constant.Variance inflation factor test (VIF) was applied in order to assess multicollinearity issues, considering collinearity tolerance rate of 0.1 (1/VIF).Both tests did not indicate heteroskedasticity or multicollinearity problems.
Samples were tested for outliers by applying 3 standard deviation analysis, excluding observations above this rate.Only one outlier in REIT return was identified in the sample.Additionally robust standard errors regression was applied in order to control for high influential observations in overall sample.
Table 3 presents regressions main results: regression coefficients, pvalue, sample size, F-statistics and coefficient of determination (R 2 ).The regression coefficient indicates the impact of each variable on the dependent variable and the p-value shows the significance of the variable for the model.Significance levels of 1% and 5% were adopted.The Fstatistic refers to the model predictive capacity as a whole and resulted significant p-value of 0.0042.Coefficient of determination (R 2 ) refers to the explanatory power of the model, resulting 16%.Regression results indicated that real estate market (RRE) as measured by IGMI-C, is significant in the model, at significance level of 1%, as well as the stock market variable (Rs), both with positive coefficient's signs: 0.64 and 0.14, respectively.Bond market factor (RB), in the other hand, does not appear to have influence on Brazilian REIT returns in the sample.
Variance decomposition reveals that although real estate market and stock market can explain part of Brazilian REIT returns, their influence on total return does not surpass 16%.Stock market responds for 9% and real estate market responds for 7% of variance in the full sample.
These results support the notion that REIT returns reflect its hybrid characteristics, however, Brazilian REIT returns are not primarily driven by real estate, stock or bonds market return performances.

Results for subsamples by REIT property type
Brazilian REIT formatting structures can have different configurations, depending on property focus (i.e.office, industrial, shopping mall, hotel, retail and others) and the project development stage (i.e.constructed, built-to-suit, sale-leaseback and development projects).REIT funds can also be composed of shares of other REITs and investments on fixed income financial instruments, such as debentures, certificates of deposit, real estate receivables and mortgage notes, known as Fixed Income REITs.Multiclass REITs in the other hand is a mix of several REIT types, according to UQBar classification (UQBAR, 2013).In order to verify the consistency of results for the overall sample, subsamples were extracted by prevailing Brazilian REIT types: Offices, Malls, Fixed Income, and Multiclass.
The subsample Offices consisted of REITs focused on lease or sale transaction of corporate space.The subsample Malls is composed of REITs focused on retail space.The subsample Fixed Income refers to funds backed by real estate receivables, mortgage notes, and shares of other REITs as well.The subsample Multiclass comprises REIT funds that invest in different types of REITs, such as residential, office buildings, industrial / logistics and receivable credits.REITs based on housing, industrial/logistics, hospitals and education facilities did not totaled enough observations to linear regression purposes, and therefore they were not analyzed.Table 5 presents the results for subsamples by REIT property type.Linear regression of Office subsample indicated that the returns of these funds are at 5% significance level, explained by variables real estate market (IGMI-C) and stock market (IBRX100 index) with positive beta signs.The results of Office subsample are the closest of the full sample, which was expected, since such funds are the most frequent in the full sample.
In the subsample formed of retail developments (Malls subsample), only stock market return was a significant explanatory variable, at 5% level of significance.Explanatory power of model remained at 15%-16%.
On the other hand, F-statistics significance results on Fixed income and Multiclass subsamples indicated that model or its components are not statistical significant at 5% significance level, as informed by F (p-value) at table 6. F-statistics p-value ranged 0.11 and 0.32, respectively, to fixed income and multiclass subsamples.Robust regressions were used in order to control for outliers and heteroskedasticity.Variables were orthogonalized as described in the methodology as to apply pure factors in the model; as such model was also controlled for multicollinearity issues through VIF test.
In order to allow further examination of these subsamples, control variables were included in the model: inflation index, measured by IGP-M (General Price index), published by Fundação Getúlio Vargas, and two real estate market parameters: property vacancy and lease price.
The IGP-M inflation index is the most common inflation index used in lease agreements in Brazil.
Vacancy corresponds to the vacant stock percentage, considering corporate lease offer in the cities of São Paulo and Rio de Janeiro -vacant space divided by total lease space offer.Lease price refers to natural logarithm variation of average asked price for AA and A corporate space rental.Data was extracted from real estate consulting firm, Jones Lang LaSalle´s quarterly market reports.
Table 6 presents regression results with inclusion of control variables: inflation, vacancy and lease price.Results for Office and Malls subsamples suggest that substantial part of explanatory power of real estate market return (IGMI-C) covers other real estate parameters, such as vacancy and lease price information.Model specification for fixed income subsample becomes significant with inclusion of control variables inflation and lease price.Both variables were also significant at 5% level.This result is consistent with subsample´s characteristics, since fixed income REIT returns are mainly comprised of lease receivables and mortgage letters.
The inclusion of lease price variable did not improve the model statistical significance for Multiclass subsample.Inclusion of inflation and vacancy variable, on the other hand, resulted in F(p-value) statistics of 0.07 against 0.32 of original model, although these variables were not individually significant, according to t-statistics.The stock market return variable emerged as the only significant explanatory variable for Multiclass returns.
In the overall, we can observe that REIT return sensitivity to explanatory variables differ from a subsample to another, grouped by property type.This finding can be more easily observed through variance decomposition at table 7: Relatively, real estate market return seems to have a higher impact on Office subsample return's variance (8%), followed by stock market return (5%), however, as similarly observed at full sample results, their total influence is not higher than 13% of total variance.Results also reinforces that other real market parameters can affect REIT return.Stock market responds for around 3% of variance of Malls hypothetic portfolio return and vacancy although not statistical significant responds for 74% of variance.Stock market influence on variance was 4% for Fixed income and 5% for Multiclass subsamples.Lease price variation responds for nearly 28% of fixed income REITs variance, followed by inflation with 12%.Property vacancy was responsible for 4% of variance in Multiclass returns.Real estate market returns and bonds presented no significant influence on variance of this subsample.

Consistency and Robustness checks
Results of linear regressions described at section 4.3.1 suggest that investment markets' performance -measured by Real Estate, Stock and Bonds returns -is not primary driver of Brazilian REIT returns, although these investments may share common explanatory variables as indicated by inclusion of control variables.
In order to evaluate robustness of results given samples limitations a closer examination of findings´ consistency was undertaken.
Notwithstanding bond returns (RB variable) measured by NTN-B treasure bill, was not significant in full sample regression, nor in property type subsamples´ regressions; the negative relationship between RB and RFII variables was examined for consistency.One possible interpretation for results is that bonds and REITs are in fact alternative investments.Other potential explanation is the effect of not stated variables captured by NTN-B returns.
NTN-B return is comprised of two components: the tax rate measured by Selic (Government tax rate) plus inflation variation measured by IPC-A -National Wide Consumer Price Index, which is the main general inflation index used in Brazil for Government policies.
As expected, NTN-B is strongly positively correlated with Selic tax rate (approx.71%) and negatively correlated to IPC-A (approx.35%).In contrast, RFII variable presented poor correlation both with Selic (approx.4%) and IPC-A (approx.-1%).For robustness check, model was respecified (results not reported) by replacing the RB variable by Selic and IPC-A variables in the model.Although market general belief is that tax rate has strong effect on Brazilian REIT investments, in the sense that the higher the tax rates the lower the attractiveness of real estate fund to investors, selic rate was not significant in the regression.
The IPC-A inflation index was also insignificant, while IGP-M inflation rate was significant for Fixed income subsample,.As previously punctuated the IGP-M inflation index is more commonly applied in lease agreements reviews in Brazilian real estate market and as so a stronger relationship between IGP-M and REITs returns was expected.
In the overall; variance decomposition analysis indicated that with the inclusion of control variables, the influence of the market return variables (RE, RS and RB) tend to decrease.As highlighted in previous section, key real estate market operating metrics such as vacancy rate and lease price presented higher influence on RFII´s variance, especially for Office and Malls subsamples, while Fixed income funds are mainly explained by lease price.
These results seem consistent with how these underlying real estate markets operate.For instance, shopping malls revenues mainly depends on its leasable gross space rate, although there are also revenues from parking and commercial rights (key money) whereas fixed income real estate funds are mainly based on mortgage and receivables, as such would not be as highly exposed to future vacancy rate fluctuations.This way, although REIT returns may be impacted by real estate market parameters, its performance is not subjected to direct real estate performance, which makes sense considering that the securitization structure intend to maximize strengths of real estate investment and minimize its weaknesses by providing higher liquidity through stock market.
Regarding the stock market variable it is interesting to note that stock market return (RS) remained significant after inclusion of control variables for full sample (table 8), malls and multiclass subsamples (table 6).One possible interpretation regarding multiclass subsample, for instance, it that the inner diversification of multiclass funds approximates them to common stocks.Stock returns explanatory power of total variance, however; was only of 5% (table 7).
According to some international researches (Ling, & Naranjo, 1999;Clayton, &Mackinnon, 2003;Lee, &Stevenson, 2005) US REIT behaved as small cap stocks, this way it is possible that the IBR-X index adopted in our study, which comprises the most traded stocks in the market, could be inadequate to capture relationship between RFII and RS.For robustness purpose, SMLL index (small cap index of São Paulo Bovespa Exchange) was computed as proxy for stock returns; however no substantial change in results was detected.
Lastly, it is possible to observe that even with the inclusion of control variables total variance not explained by the model is still considerably high, except for malls subsample, indicating that there are other not-stated variables influencing Brazilian REIT returns not captured by market returns variables nor selected control variables.Nonetheless; this evidence does not contradict, but may reinforce, findings that suggest REIT as unique class of investment.

Final Remarks
This paper examined listed Brazilian REIT as an alternative investment in a multi-asset portfolio by questioning whether Brazilian REIT returns are significantly and predominantly driven by real estate market, general stock market or bond market returns.
Descriptive statistics and correlation matrix indicated that Brazilian REITs have the potential for diversification in a mixed portfolio since sample data presented slightly superior performance of Brazilian REITs both in average return and standard deviation in comparison with the other types of investments, except from not securitized real estate, for the observed period and there was no evidence of strong positive correlation among the variables.
Regression analysis indicated that real estate market and stock market have significant explanatory power towards REIT returns; however, variance decomposition shows that overall impact is not higher than 15%.These results lead us to believe that the hybrid nature of the REIT is reflected somehow in returns, but performance does not predominantly depend on underlying market returns, reinforcing the conclusion that Brazilian REITs consist of a unique asset class.More importantly, the subsample analysis indicates that the diversification power of this asset class is significantly influenced by REIT property/asset type, since the explanatory factors and their impact on returns differed from one type to another.This result has also an important practical implication since we can understand that the choice of the investor is not limited to whether to include REITs in portfolio, but what kind of real estate fund as well.
In light of the Modern Portfolio Theory, these results indicate that the inclusion of REITs may provide potential diversification benefit in a multiasset portfolio, by increasing the total return of a portfolio consisting of stocks and bonds, without an increase in risk; or keeping the return of this portfolio, with reduced volatility, thereby broadening the efficient frontier of the portfolio.This result put in question the traditional equilibrium fund of portfolios composed only by stocks and bonds, as REITs emerges as an alternative investment, and mainly as a unique asset class.
It is worth noting that our results do not rule out the hypothesis that some other common factors acting in these markets could have some influence on Brazilian REIT return/risk structure, such as vacancy rates and rental rates of the direct real estate market, as well as macroeconomic variables, such as inflation and interest rates not captured by return variables.

Limitations and Developments
According to Clayton and Mackinnon (2003), it is possible that the increased influence of idiosyncratic risk is related to the higher capitalization and institutionalization of real estate funds, with consequent greater dissemination of information on each fund and increased influence of their capital structure and information quality on share pricing.Therefore, in order to minimize effect of idiosyncratic risk in the sample, a hypothetical portfolio index was adopted as proxy for Brazilian REIT returns.
The adoption of portfolio index instead of the actual funds returns was also intended to prevent effects of using pooled data against independent variables in index form, such as real estate return (IGMI-C index) and stock return (IBRX index).
The differences in results due to use of monthly and quarterly samples as well as the use of indexes were reported in several studies, including Case, Yang and Yildirim (2009).However; this assumption also incurs in limitations to the present study regarding the effects of REIT share´s liquidity and size, and their impact on results herein.
Theoretically, the lower the liquidity of an investment, the higher expected returns by the investor or, in the other direction, the higher liquidity provided by an investment, the lower required rate of return.One of REIT´s main theoretical advantages is to provide liquidity to a traditionally lower liquidity underlying asset such as real estate through securitization.
Assessment on the relationship between liquidity risk (or illiquidity risk) and REIT returns has being examined in international literature and could provide additional evidence on linkages among investment classesdirect real estate, stocks and bonds-and on diversification benefits as suggested by empirical results from our study.Subrahmanyam (2007), for example, found evidence that US REITs liquidity was forcastable by non-REIT and that order flows and return in the stock market negatively forecast REIT order flows, indicating that real estate market as a substitute investment for the stock market.Other research question related to liquidity that could affect present findings is the impact of underlying property type on REIT liquidity as investigated by Danielsen and Harrison (2007) and hence on returns and diversification.
Nevertheless, by adopting dependent variable in the aggregate level, samples could not be controlled for liquidity, what should be addressed in future research, when Brazilian REIT market is expected to be more mature and more observable data shall be available.
Finally, among other possible developments for empirical of the present findings, it is possible to build a multi-asset portfolio, applying sensitivity analysis with the inclusion of REIT shares in the portfolio, analyzing the issue of asset allocation.Further investigation on other variables that may explain REIT returns, such as interest rates and specific real estate market operating parameters is also an intended development of this work.

Table 1
Descriptive Statistics: mean, standard deviation, minimum and maximum returns.
Note: RFII: Brazilian REIT return; RRE: real estate market return; Rs: stock market return; RB: bond market return.

Table 4
presents the variance decomposition computed according toClayton and Mackinnon (2003)methodology.Variance decomposition provides each independent variable's contribution to total variance of REIT returns in the sample.

Table 6
Results of linear regression with inclusion of control variables -Subsamples by property type.

Table 7
Variance decomposition -Subsamples by property type.

Table 8
Results of linear regression with inclusion of control variables -Full sample.