Determinants of the Price-to-Net Asset Value
in Brazilian REIT’s: The Role of Stocks and
Bonds
Real Estate Finance Master Thesis
João Mendes Succar July 2015
Statement of Originality
This document is written by João Mendes Succar who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
The study intends to investigate the price-to-NAV puzzle. This puzzle is characterized by the difference in value between a real estate fund shares and the fund’s property assets. Several authors argue that a small difference between share price and Net Asset Value would be normal, due to funds’ administration costs or due to the value added by funds’ management of several properties. Large differences, in contrast, would be avoided by arbitrage between public and private markets. However, price-to-NAV deviations can be large and persistent over time. So far, studies were restricted to the real estate funds market and to real estate funds’ characteristics. The present study contributes by adding the influence of stocks and bonds markets into the explanation of the price-to-NAV puzzle. The results show that stocks and bonds returns have a strong influence on the price-to-NAV difference. In terms of pricing, this implies a stronger disconnection between direct and indirect real estate investments. To put it differently, public real estate investments are influenced by other public markets in a way that is not replicated in the direct private real estate market.
Table of Contents
1. Introduction ... 1
2. Institutional setting ... 4
3. Literature review ... 5
3.1. Share price to net asset value ... 5
3.2. Asset “bubbles” and the price-to-NAV ratio ... 8
3.3. The relationship between real estate funds, stocks and bonds ... 9
3.4. Fund fundamental attributes ... 10
3.4.1. Fund/REIT size ... 10
3.4.2. Fund liquidity ... 11
3.4.3. Fund volatility ... 11
3.4.4. Dividend yield ... 12
4. Data and descriptive statistics ... 12
4.1. Dependent variable: price-to-NAV ratio ... 13
4.2. Independent variables ... 15
4.3. Control variables ... 15
5. Methodology ... 17
5.1. Model specification ... 19
6. Results ... 20
6.1. Main regression estimates ... 20
6.2. Robustness checks ... 25
7. Limitations ... 26
1
1. Introduction
Brazilian Real Estate Investment Funds (BREIF’s) became, in recent years, one of the main vehicles for investing in real estate in the Brazilian economy. The market capitalization of only the listed BREIF’s jumped from BRL 2.5 billion in December 2007 to BRL 22.5 billion (USD 8.5 billion) in December 2014, after reaching almost 28 billion of market capitalization in December 2013 (Uqbar, 2015). Listed BREIF’s market capitalization is still quite small compared to the USD 907 billion1 of the US REIT’s industry. However, Brazilian economy is the largest in Latin America and the country’s GDP is the seventh biggest in the world2, making BREIF’s an interesting study subject for a nascent public real estate market in an economically important developing country. BREIF’s are closed-end funds that buy properties or property rights and have their shares traded at the Brazilian Stock Exchange. These funds are very similar to the American and European Real Estate Investment Trusts (REIT’s) and, like them, experience the same “price-to-NAV puzzle”. This puzzle consists of a persistent and large difference between the appraisal value of the properties owned by a REIT – its Net Asset Value (NAV) – and the market value of the REIT as a whole, represented by its market capitalization in stock exchanges (Brounen et al., 2013; Lee et al., 2013; Patel et al., 2009). Figure 1 shows this difference for BREIF’s.
Figure 1 – Average price-to-NAV premium on BREIF’s’ shares
Note: Premium (discount) is calculated as: ((Fund Market Capitalization / NAV) – 1). Data in percentages.
1 Source: NAREIT - National Association of Real Estate Investment Trusts, 2014. Retrieved from: https://www.reit.com/data-research/data/us-reit-industry-equity-market-cap
2 Source: The World Bank, 2014. Retrieved from:
http://data.worldbank.org/indicator/NY.GDP.MKTP.CD?order=wbapi_data_value_2014+wbapi_data_val ue+wbapi_data_value-last&sort=desc -25% -20% -15% -10% -5% 0% 5% 10% 15% 20% 25% 30% D ec -0 7 Ma r-0 8 Ju n -0 8 Sep -0 8 D ec -0 8 Ma r-0 9 Ju n -0 9 Sep -0 9 D ec -0 9 Ma r-1 0 Ju n -1 0 S ep -1 0 Dec-1 0 Ma r-1 1 Ju n -1 1 Sep -1 1 D ec -1 1 Ma r-1 2 Ju n -1 2 Sep -1 2 D ec -1 2 Ma r-1 3 Ju n -1 3 Sep -1 3 Dec-1 3 Mar -1 4 Ju n -1 4 Sep -1 4 D ec -1 4
2 The price-to-NAV difference is peculiar, in part, due to this unique characteristic of real estate investments: the presence of a dual asset market, in which properties can be negotiated both privately and in public markets (Geltner et al., 2006). According to Geltner et al. (2006), this characteristic should prevent large and persistent differences between REIT’s market capitalization and NAV, because REIT managers could arbitrage between public and private markets whenever the difference between market capitalization and NAV was large enough to cover for transaction costs.
Several studies addressed this puzzle but, so far, there is no conclusive evidence (Brounen et al, 2013; Lee et al., 2013; Patel et al., 2009). Some authors consider the inaccuracy of NAV measurement as a possible explanation (Liow, 2003; Patel et al., 2009), as well as the relative lack of information efficiency in the private property markets compared to the stock markets (Chiang, 2009). Other explanations could be the presence of noise traders, that would drive prices away from its fundamental value (Lee et al., 2013) and short sale constraints (Brounen et al., 2013).
Despite explaining partially the Price-to-NAV puzzle, these studies do not take into account the relationship that exists between REIT’s and other investments in the financial markets. It is well documented, for example, that REIT returns can be influenced by the returns of both the stock market and the bonds market (Clayton and MacKinnnon, 2003). Peterson and Hsieh (1997) propose that, since REIT’s shares are traded in stock exchanges and have cash flows similar to an investment in bonds, REIT’s pricing and returns might be explained by common risk factors between them and stocks and bonds returns (e.g. trading noise and the term structure of interest rates). Given the importance of stocks and bonds returns on predicting REIT’s performance, the main idea of this research is to investigate the influence of stocks and bonds on price-to-NAV deviations, controlling for fund fundamental attributes.
The dataset consists of monthly information for 81 funds ranging from December 2007 to February 2015. Specifically, the data covers, for each fund and month: the NAV, market capitalization, dividends paid, traded volume, and fund volatility. Additionally, a stock index (Ibovespa) and a government bonds index (IMA-S) are used. Sample size for each fund varies and, for this reason, the data is an unbalanced panel dataset.
Ordinary Least Squares method is used to estimate the impact of stocks and bonds returns on the price-to-NAV difference. A dynamic model is also proposed and, in this case,
3 Arellano-Bover/Blundell-Bond linear dynamic panel-data estimation is used. A bottom-up approach is used for model construction, with regressor being added from one specification to the next. In this case, it is possible to check for parameter consistency across different specifications. Apart from fund fundamental variables, the study also controls for entity and time fixed effects.
The results of the study show that stocks and bonds have indeed a strong and significant influence on the price-to-NAV deviations. The stock market coefficient of 0.1 was statistically significant at the 1% level. The size doesn’t seem to be large at first sight, but taking into account the high volatility of the stock market, it’s not negligible. The bonds market coefficient, in its turn, showed to be very large (9.5) and also statistically significant at the 1% level. However, taking into account that bonds return volatility is usually small, the impact on the price-to-NAV ratio tends to be restricted. Additionally, the results showed that the price-to-NAV deviations are autocorrelated and that some fund fundamental attributes – fund size and dividend yield – are important predictors of the price-to-NAV difference.
These results have several implications. First, they put emphasis on the distinction between public and private real estate markets, in which assets of a different nature are traded (fund shares opposed to real estate properties), and in which different factors might be relevant to determine returns. This implies that using NAV’s as a parameter for fundamental value of REIT’s, BREIF’s, or property based companies, can be misleading. From an academic perspective, the importance of stocks and bonds on the explanation of price-to-NAV deviations stablishes a possible connection between large price-to-NAV deviations and the formation of asset bubbles. The results also add up to previous evidence of dividend yields being interpreted as a signal for future performance, in the case of passive property companies. In terms of novelty, apart from considering new factors to the understanding of the price-to-NAV puzzle, evidence is gathered from a new country, with refined dataset.
The study is particularly relevant to investors that look for a better valuation process of BREIF’s shares, as well as for investors that might consider BREIF’s shares as an investment vehicle to diversify into real estate assets.
The rest of the study is organized as follows. Section 2 briefly presents the institutional setting of BREIF’s. Section 3 addresses some of the main points present in the literature
4 that are useful to contextualize the price-to-NAV debate as well as to support the study. Section 4 describes and explain the data. Section 5 is dedicated to the methodology of the study. Section 6 shows the main results. Section 7 presents some of the research limitations and Section 8, the conclusion.
2. Institutional setting
In Brazil, one of the main vehicles for investing in real estate are the Brazilian Real Estate Investment Funds (BREIF’s), similar to the American and European Real Estate Investment Trusts (REIT’s). These real estate funds are closed-end funds that buy properties or property rights and have their shares traded at the Brazilian Stock Exchange, Bovespa.
According to their investment focus, BREIF’s can be roughly divided into three different categories. They can be equity funds and invest directly in real estate property, mortgage funds, investing in real estate debt (mortgages or mortgage-backed securities), and, finally, be a fund of funds, when they buy shares of other BREIF’s. It is also possible for a BREIF to be a hybrid real estate fund, when they combine the previously mentioned investments.
The operation of a BREIF is regulated by the Brazilian Securities Commission3 (a government agency similar to the American SEC), which imposes several restrictions to these investment vehicles, like:
1) Obligation to have at least 75% of its assets invested in real estate properties or property rights;
2) Distribution of monthly dividends derived from the rents paid by the tenants of the buildings. A minimum of 95% of all profits must be distributed to fund shareholders, as recognized on a cash basis; and
3) Forbiddance of debt financing.
Another important aspect concerning BREIF’s is that, since August 2009, BREIF’s’ monthly dividends are exempt of personal income taxation, if the following conditions are met:
5 1) The shareholder do not have more than 10% of the fund’s shares;
2) The fund has at least 50 shareholders; and
3) Fund’s shares must be negotiated exclusively on the Brazilian Stock Exchange on the open market or on the over-the-counter market.
In general, BREIF’s can be viewed as the Brazilian version of the American REIT’s. However, two main differences should be taken into account. First, as mentioned before, the regulation prohibits the use of debt by BREIF’s, which means that the use of leverage as a way of increasing equity return is not possible. American REIT’s, conversely, frequently use debt financing. Second, BREIF’s are obliged to distribute 95% of profits according to a cash basis accounting system. In American REIT’s, the use of accrual-accounting-based earnings measurement allows them to use freely the cash of depreciation expenses that is not actually a cash outflow. These two differences make BREIF’s’ managers less capable of adding value via active management and make BREIF’s even more passive investment vehicles than American REIT’s. As a consequence, it would be expected an even smaller difference between fund asset values (NAV’s) and fund share prices.
Finally, another regulation change that is particularly important to the study of price-to-NAV difference took place in December 2011. With this new regulation, BREIF’s were obliged to start to value their assets at fair-value, meaning that other valuation methods had to be abandoned and a professional real estate appraisal firms had to be hired to mark funds’ properties at fair-value. Under the old regulation, NAV’s were stablished as the historical purchase prices or investment costs, less accumulated depreciation. In this case, it would make no sense to compare market value and NAV’s, especially for properties bought many years ago. From one hand, this specific regulation made the dataset useful for studying the price-to-NAV difference but, from the other hand, it limits the dataset to more recent years.
3. Literature review
3.1. Share price to net asset value
Property investment companies, characterized as pools of professionally managed income producing properties, usually present a close correlation between the value of the property portfolio and the value of companies’ shares (Liow, 2003). Considering that the value of
6 a company is the sum of all its expected cash flows, discounted at rates that reflect investors’ risk premiums, and that NAV reflect REIT’s cash flow prospects, NAV should be a reasonable guiding parameter to a REIT share price (Patel et al., 2009). In general, the argument is that, at a fundamental level, property company share prices must reflect their underlying real estate investment value (Liow, 2003).
However, several studies mention that there is significant deviations between the price at which shares of REIT’s trade and their NAV’s (Brounen et al., 2013; Lee et al., 2013; Liow, 2003; Gentry et al., 2004). Numerous studies have addressed this puzzle and many possible explanations have been raised. Nevertheless, to the present time, there is no conclusive evidence or consensus regarding the reasons for this phenomenon (Brounen et al., 2013; Lee et al., 2013; Patel et al., 2009).
Another interesting aspect that is worth to mention is related to the results of the study of Gentry et al. (2004) that found large positive excess returns to a strategy of buying REIT stocks that trade at a discount to NAV, and shorting REIT stocks trading at a premium to NAV. The authors believe that some variation on the Price-to-NAV ratio makes sense, as premiums are positively related to recent and future NAV growth, but this variation appears to be too high, which opens the possibility for potential profits on a strategy of trading on mean reversion.
Liow (2003) analyzes the relationship between stock prices and NAV’s of Singaporean property investment companies and concludes that stock prices are nonlinearly linked to NAV’s, but the extent of mean reversion between the two is slow, making deviations prolonged across time. Some of the possible explanations Liow gives for his findings are related to problems on the measurement of the NAV’s: valuation methodology, valuation variance, valuation accuracy and valuation smoothing. The same belief is shared by Patel et al. (2009) to whom the risk premium on the private real estate market does not fully aggregate all the components of property risk (tenant credit risk, vacancy risk, liquidity, etc.), which gives rise to wrong property valuations and consequently a discount on REIT share prices relative to NAV.
In a similar manner, Chiang (2009) found that pricing on the private market (NAV) follows the pricing of the public market (REIT share price), that is, lagged share price returns are useful to predict returns on an NAV basis, but not the other way around.
7 According to this author, public markets, that are more efficient in processing information, tend to lead private markets.
Apart from inaccuracy of NAV valuation, possible explanations, according to the literature, are usually fitted in different backgrounds: a behavioral background, where sentiment-based investors drive prices away from their fundamental value (also known as noise trading); a rational background, largely based on firm’s characteristics (like size); and, finally, a structural background, with short sale constraints being an example. On the behavioral side, Lee et al. (2013) claim that retail investors, often influenced by fads in the market, can trade based on sentiments, which could drive asset prices away from their fundamentals. According to the authors, volatility and volume are two variables that measure the market sentiment and literature points out that high trading volume in the REIT market, not underpinned by positive fundamentals, is a sign of the presence of “noise-traders”, in a sort of herd effect. In their study, they empirically test the information and “noise” effects on Price-to-NAV. Their results show that trading volume has lagged and significant positive effects on Price-to-NAV premiums. Additionally, they found that Price-to-NAV premiums are positively and significantly correlated with the market capitalization of REIT’s, which implies that investors put a premium on firm size effects. Finally, they found that diversified REIT’s (multi property type) are traded at significant discounts to NAVs relative to focus-REIT’s.
Brounen et al. (2013) examine the role of short sale activity and short sale constraints in explaining the cross-sectional variance of REIT premiums to NAV. The rationale behind their study is that, if short selling is constrained (not enough supply of REIT stocks to be borrowed and then shorted), it is more difficult for negative information and opinions to reach the market. For this reason, stock prices may be set by the most optimistic investors. According to these authors, REIT short sale activity and premiums to NAV appear to be highly correlated (-0.89, on average). The authors found that the variation in short sale activity across individual REIT’s accounts for approximately one-third of the variation in NAV premiums. Moreover, the authors claim that the effect of short sale constraints on NAV premiums appears to dominate the influence of investor sentiment.
8
3.2. Asset “bubbles” and the price-to-NAV ratio
The interest on asset bubbles is justified by its importance to the economy. Specifically, price bubble bursts are followed by loan defaults, bankruptcies, and stress on the banking system, arising from loss of capital due to a mispricing of collaterals (Grover and Grover, 2014). Another reason why bubbles attract a great deal of interest is because they challenge the idea of rational market participants (Grover and Grover, 2014) and the efficient market theory (Zingales, 2010). But what exactly is a price bubble?
It is generally agreed that a price bubble exists when asset prices deviate from its fundamental value (Grover and Grover, 2014; Brooks et al., 2001; Stiglitz, 1990). A bubbling behaviour would be then characterized by asset prices being high today only because investors believe that the selling price will be even higher tomorrow, without any justification on fundamental factors (Stiglitz, 1990). However, this simple definition implicitly brings the challenge of previously defining what would be the fundamental value of an asset, without which, it becomes too vague (Lind, 2008).
An alternative explanation is proposed by Lind (2008), who states that “there is a bubble if the price of an asset first increase dramatically and then almost immediately falls dramatically”. Critics of this vision, on their turn, point out that defining a bubble based only on the development of prices is also problematic, as “dramatically” is a subjective measure and fundamental value might truly evolve over time.
Despite the controversy around the definition of a bubble, in a practical matter, technics used to detect asset bubbles in real estate markets usually involve some comparison between what can be considered as a fundamental value of the assets and their actual prices. These technics try to stablish the fundamental value via some income related measurement – dividend ratios and the Gordon Growth Model, for public real estate markets (Brooks et al., 2001), and rental income and terminal value, for the direct real estate market (Stiglitz, 1990).
Regarding some specific investment vehicles, like mutual funds and property funds, some authors consider that the NAV is a good proxy for fundamental value (Rosser, 1997; De Long and Shleifer, 1991) and that high price-to-NAV ratios are a sign of overly optimistic investors (De Long and Shleifer, 1991). These authors point out that, frequently, large price-to-NAV deviations precede a bubble burst (Rosser, 1997; De Long and Shleifer, 1991).
9 However, using NAV as a proxy for fundamental value and, consequently, price-to-NAV deviations as an indication of bubbling behaviour has an important deficiency: If bubbles persist for a certain period of time, market prices with an embedded bubble component start to be incorporated into property appraisals and, hence, to NAV’s (Penman, 2007). The debate around what defines an asset bubble, as well as how to properly detect the presence of one, remains controversial. Large price-to-NAV discrepancies may be an indication that market participants might be including in their share valuation expectations not related to the fund fundamentals. Nevertheless, since the bubble component can be incorporated into both the share price and the NAV, it seems that a bubbling behaviour is not enough to explain these deviations and that other causes might play a role.
3.3. The relationship between real estate funds, stocks and bonds
Since REIT’s (like the Brazilian Real Estate Investment Funds), have their shares negotiated in stock exchanges and also pay regular dividends, many studies have addressed the relationship between this real estate investment vehicle and stocks and bonds. Clayton and MacKinnon (2003), for example, point out that two confronting views about the role of REIT’s are presented in the literature. The authors argue that, in terms of risk exposure, REIT’s can be seen as a hybrid of stocks and bonds while, in terms of REIT pricing, there seems to exist a connection between the public market and the private, unsecuritized, market returns.
Several authors have pointed out the similarities between REIT’s and stocks (Goebel et al., 2013; Geltner et al., 2006, etc.). Geltner et al. (2006), for example, argue that most REIT’s are comparable to income stocks in the long run (instead of growth stocks). According to them, the main reason is because, by their very nature, REIT’s invest mostly in stabilized operational commercial properties, which are income-oriented, not growth-oriented assets.
Fuerst and Matysiak (2013), using panel data and researching on the factors driving total returns on non-listed real estate funds, found that the performance of the overall economy (lagged GDP), competing asset classes (lagged stock market returns) and contemporaneous government bond rates are significant and positive predictors of annual fund performance. Peterson and Hsieh (1997) found that Equity REIT returns are
10 significantly related to stock market factors, while Mortgage REIT returns are related to both stock market factors and bond market factors. Goebels et al. (2013) found that different conditions of monetary policy (restrictive versus expansive), may affect REIT returns. The authors mention, however, that several studies found different conclusions on the influence of government interest rates and the performance of REIT returns.
3.4. Fund fundamental attributes
On the literature related to investment funds, some fund attributes (or dimensions) like size, liquidity, dividend yield, and volatility are said to have an influence on overall fund performance. From an investor perspective, by affecting performance, these attributes would ultimately have an impact on the valuation of the shares of the fund and, by extension, on eventual premiums and discounts on the NAV. These four attributes will be briefly discussed below.
3.4.1. Fund/REIT size
REIT size is pointed out as an important dimension of REIT’s in several studies (Chatrath et al., 2012; Goebel et al., 2013; Clayton and MacKinnnon, 2003; etc.). Chatrath et al. (2012) suggest that size is related to greater price discovery and increase in trading activity. Larger REIT’s have less uncertainty related to their valuation due to a higher degree of transparency. The authors found, for instance, that size matters when it comes to how public information is incorporated into prices, with large REIT’s being more sensitive than small REIT’s.
The effect of size on REIT’s value was also investigated by Capozza and Lee (1995). They found that small REIT’s trade at much larger discounts than large REIT’s and the differences are highly statistically significant. One of the reasons that can explain this difference, according to these authors, is that small REIT’s are proportionally twice as costly to administer than large REIT’s (Capozza and Lee, 1995).
Goebels et al. (2013) didn’t find differences in returns between large and small REIT’s. However, in restrictive times, when interest rates are increasing, they found that large REIT’s outperform small REIT’s.
11
3.4.2. Fund liquidity
Liquidity refers to the ability to sell an asset quickly, and inexpensively, for a price close to market value (Clayton and MacKinnon, 2000). Measuring liquidity, however, can be a complex task, given its multi-dimensions. Some studies measured liquidity using bid-ask price spreads (Bhasin, Cole and Kiely, 1997). This way of measuring it was criticized by Clayton and MacKinnon (2000), who used the impact of trades on price movements as a better proxy for liquidity. Glascock and Lu-Andrews (2014) use the Amihud Illiquidity measurement and Turnover Ratio measurement to estimate REIT market trading liquidity. Huang et al. (2011), studying liquidity changes after REIT stock splits, used five different measures of liquidity: Amihud’s illiquidity ratio, Zeros, Dollar spread, Relative spread and Turnover Ratio, calculated as the average of the daily ratio of trading volume (in shares) to shares outstanding. According to these authors, the turnover ratio can be used as a proxy for liquidity because it is negatively correlated with the bid-ask spread and because it is a good measure for trading activity.
Liquidity is always considered an important attribute of any asset. In the case of real estate investments this attribute plays an even more important role. One of the explanations for the growth in real estate public markets, for example, is due to its higher level of liquidity compared to direct investments in real estate assets (Clayton and MacKinnon, 2000). Haβ et al. (2012), for example, report that discounts on open-ended fund value relative to its net assets can go up to 20% if investors fear that fund managers will have to “fire-sell” properties because of the lack of liquidity.
3.4.3. Fund volatility
“Risk can be defined as the likelihood that an asset's realized returns will differ from that which is expected” (Delisle, 2013) and “a cornerstone of the traditional asset-pricing theory is that higher risk should be rewarded with higher return. This relationship also holds true for real estate.” (Bardos and Zaiats, 2011). With this in mind, investors would require a risk premium on BREIF’s that are more volatile. In a discounted cash flow framework, the expected dividends of a riskier BREIF would be discounted with a higher rate and, with everything else equal, have smaller values. Simply put, it is expected a negative relationship between fund volatility and price-to-NAV premiums. A regular way of measuring riskiness or volatility of an asset is using the standard deviation (Bardos and Zaiats, 2011; Ross and Zisler, 1991).
12
3.4.4. Dividend yield
In their 1996 seminal paper, Miller and Modigliani proved that in an ideal world, with no market imperfections, dividend policy is irrelevant to share value. This proposition was later confirmed by Black and Scholes (1974), to whom a share price change would be expected after a change in the dividend yield only if the market believes that this change indicates something about the probable future course of earnings. If that is not the case, this temporary effect should disappear.
However, as pointed out by Kallberg et al. (2003), REIT’s (and BREIF’s) are not exactly similar to regular corporations, since they are obliged to pay out almost all of their funds from operations as dividends. This regulation constraint creates, then, a stronger link between current dividends and future cash flows.
In line with these proposition, Bradley et al. (1998), using a sample of REIT’s, found strong evidence of a link between future cash flow volatility and the current payment of dividends. This finding reinforces dividend policy as signal to investors, in the sense that managers that expect future cash flows to be more uncertain, tend to pay lower levels of dividends. In their research, Kallberg et al. (2003) also found evidence that current dividend payout is a credible signal of the future prospects of the firm and, therefore, that present value models that discount dividends should work better for REIT’s than for other equities.
4. Data and descriptive statistics
This study uses data on Real Estate Equity Funds in Brazil negotiated at the Brazilian Stock Exchange - Bovespa. All data utilized in the study consists of public information. However, two different sources were used for data collection. The first source was a private dataset provider called Economatica4. From Economatica, monthly data for funds’ share prices, NAV’s, funds’ market capitalization, monthly traded volume and monthly dividends were extracted. Additionally, two indices were extracted from this dataset provider: IMA-S and Ibovespa.
13 The IMA-S is an index that tracks the market prices of LFT’s5, Brazilian government bonds with a variable interest rate that is tied to the government basic interest rate (a sort of floating-rate note). Due to data restriction, it was not possible to obtain an index that covered corporate bond returns or an index that covered the Brazilian bond market as a whole.
Ibovespa is the main index of the Brazilian Stock Exchange and was created to be the average performance indicator of the prices of the most traded and representative assets of the Brazilian stock market. It is an index that accounts for the total return of stocks. To calculate the index, assets are weighted by their free-float market value.
The second source was BREIFs’ webpages, from which data was hand-collected from official documents that the funds are obliged by regulation to publish (balance sheets, management reports, etc.). These data consists of type of fund (equity, mortgage, etc.), property sector and number of buildings owned by the fund. These documents were also important to track the precise date when assets started to be appraised and marked by their fair-value, which made usable the data extracted from Economatica. The final dataset goes from December 2007 to February 2015. As mentioned in the institutional setting section, few funds used to mark their assets at (appraisal-based) fair-value. Most of them started to report appraisal-based NAV’s only after the regulation change of December 2011. For this reason, no data points before December 2007 could be used. Table 1 contains the descriptive statistics of the whole dataset. The dataset characteristics and descriptive measures will be presented and discussed next.
4.1. Dependent variable: price-to-NAV ratio
The dependent variable in this study is the percentage premium (or discount) on the NAV’s of BREIF’s. The premium is calculated as follows:
𝐹𝑢𝑛𝑑 𝑀𝑎𝑟𝑘𝑒𝑡 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛−𝑁𝐴𝑉
𝑁𝐴𝑉
– 1
(1)Premiums are positive discrepancies between market capitalization and NAV and mean that market participants value a BREIF shares more than the appraisal value of the
5 LFT stands for “Letras Financeiras do Tesouro”, a type of Treasury bond linked to the basic interest rate of the Brazilian economy.
14 BREIF’s assets. Contrarily, discounts are negative discrepancies and imply that public markets value fund shares less than the appraised fund’s assets.
The data for the NAV’s presented on this study are entirely based on appraised fair value of the properties, which means that a real estate appraisal firm was hired to do a valuation of the fund’s assets. To the knowledge of the author, in other contexts (other countries or older data), the data for the NAV’s are based on historical cost less accumulated depreciation, which are measures of NAV that do not take into account any market evolution and, for this reason, prevents a more detailed study of price-to-NAV deviations. As can be seen on Figure 1, the average premium was positive most of the time, for more than five years. This trend reverted only in June 2013. The average premium peaked on July 2008 (24.8%), right before the beginning of the financial crisis. During more recent years, it is possible to see that the discounts are close to 20%.
When it comes to the distribution of Premiums (see Figure 2), it is possible to perceive an approximately normal curve. For the whole sample, the price-to-NAV deviations ended up being slightly negative, -3.67 percent, on average. One of the reasons for this result is the fact that the more recent years, which were marked by negative deviations, are backed up by more data points. The standard deviation was of 21.25% and the spread of the deviations is quite big, going from -60 percent to 112 percent.
Figure 2 – The distribution of the Premium values
Notes: Premium is in percentages. Based on the whole sample of 2,791 observations, from December 2007 to February 2015.
15
4.2. Independent variables
In the study, the independent variables of interest are the indices that track the stock market and the bonds market returns. Regarding these variables, represented respectively by Ibovespa and IMA-S, it is possible to visualize their volatility on Figure 3. As expected, the stocks monthly returns are much more volatile than the bonds’ returns. For the 87 months covered in the study, the standard deviation for Ibovespa was 6.6% and for IMA-S, 0.14%. The average monthly return for the former was close to zero and, for the later, 0.81%.
The performance of the bonds index (grey solid line) is relatively steady across the whole period. The stock market performance (black solid line) reveals a stationary characteristic but with varying volatility across time. It is possible to observe a certain increase in volatility in the second half of the time period, with the exception of the 2008 financial crisis in the first half, when volatility also increased substantially.
Figure 3 – Stocks and bonds returns
Note: Percentage returns based on data from December 2007 to February 2015
4.3. Control variables
The study control variables are the four fund attributes mentioned in the literature review section: fund size, fund volatility, fund liquidity and dividend yield. Regarding fund size, it is usually measured by market value, and so it will be in this study. This means that BREIF’s’ sizes will be the month-end product of shares outstanding and price per share. The sample, by this dimension, is very heterogeneous, with the market cap ranging from
-30% -25% -20% -15% -10% -5% 0% 5% 10% 15% 20% Dec -0 7 Ma r-0 8 Ju n -0 8 Sep -0 8 Dec -0 8 Ma r-0 9 Ju n -0 9 Sep -0 9 Dec -0 9 Ma r-1 0 Ju n -1 0 Sep -1 0 Dec -1 0 Ma r-1 1 Ju n -1 1 Sep -1 1 Dec -1 1 Ma r-1 2 Ju n -1 2 Sep -1 2 Dec -1 2 Ma r-1 3 Ju n -1 3 Sep -1 3 Dec -1 3 Ma r-1 4 Ju n -1 4 S ep -1 4 Dec -1 4
Ibovespa and IMA-S Returns - Dec 2007 to Feb 2015
16 only BRL 5.03 million6 to BRL 2.97 billion. The average fund size is BRL 258 million and the standard deviation is BRL358 million. Table 1 presents descriptive statistics for the dataset.
Table 1 – Descriptive Statistics
Note: based on data from December 2007 to February 2015 for Brazilian Real Estate Investment Funds.
Concerning the dividend yields, the mean value was 0.92% per month, a value fairly in line with the mean value for the IMA-S, the government bonds’ index (0.81%). The standard deviation of the dividend yields (2.2%), however, is much higher than that of IMA-S (0.14%). It is also noticeable the high value of the maximum dividend yield. Amongst the whole sample, 21 observations had dividend yields higher than 5% per month. Most of them occurred in the event of a divestment, when capital was returned to investors in the form of dividends. By eliminating these 21 observation, the mean value for the dividend yield falls to 0.77% per month.
In the present study, a turnover ratio will be used as a proxy for liquidity. However, because of data constraints it will be measured as the ratio of trading volume (in Brazilian Reals) to the market value of the company, that is, what percentage of the value of the company was traded in a given month. The average liquidity in the sample was equal to 1.67% with a standard deviation of 2.76%.
Formula (2) indicates the calculation of funds’ liquidity: 𝐹𝑢𝑛𝑑′𝑠 𝐿𝑖𝑞𝑢𝑖𝑑𝑖𝑡𝑦 = 𝑀𝑜𝑛𝑡ℎ𝑙𝑦 𝑇𝑟𝑎𝑑𝑒 𝑉𝑜𝑙𝑢𝑚𝑒
𝑀𝑜𝑛𝑡ℎ−𝑒𝑛𝑑 𝑀𝑎𝑟𝑘𝑒𝑡 𝐶𝑎𝑝 (2)
6 All currencies of the study are in Brazilian Reals (BRL)
Variable Obs Mean Std. Dev. Min Max
Premium (%) 2,791 -3.67 21.25 -60.11 112.13
MarketCap (BRL Million) 2,791 280 378 5 2,970
Liquidity (% Traded Volume) 2,791 1.67 2.76 0 71.34
DY (%) 2,791 0.92 2.20 0 66.89
Risk - SD (%) 2,170 4.50 2.18 0.87 34.37
Ibovespa - Stocks (%) 87 months -0,007 6.61 -24.80 15.55
17 Finally, fund volatility is a measure of the risk investors are exposed to, when investing in shares of BREIF’s. It is measured by the standard deviation of the previous year month-end share price. The mean standard deviation for BREIF’s shares is 4.5%. It is, therefore, way higher than that of the bonds index (0.14%), but fairly lower than that of the stock market (6.6%). This is in line with the idea that real estate is an investment less risky than stocks, but more risky than government bonds.
Regarding the correlations between the variables under study, Table 2 depicts the Pearson correlation coefficients. It is possible to notice the general low values for them, with the highest positive value, a mere 20%, occurring between Premium and Fund Size. It is also noticeable the strong negative correlation between Premium and IMA-S, the bonds index, with a coefficient of -0.36. The signs of the coefficients are very mixed, with the majority being negative. By inspecting the coefficients of Premium with both the stocks and the bonds indices, it is possible to see a negative correlation that is close to zero for the stocks and approximately -37% for the bonds index.
Table 2 – Correlation table between study variables
Note: based on data from December 2007 to February 2015 for Brazilian Real Estate Investment Funds.
5. Methodology
Since BREIF’s were launched in different dates and started to mark their assets using appraised fair-value also in different dates, the sample size for each fund varies across the dataset. As such, the data is an unbalanced panel dataset.
Premium Fund Size Fund Liquidity Dividend Yield Fund Volatility Ibovespa (Stocks) IMA-S (Gov. Bonds) Premium 1 Fund Size 0.2017 1 Fund Liquidity -0.1023 -0.0159 1 Dividend Yield 0.0579 -0.2099 0.0177 1 Fund Volatility -0.0078 -0.1850 0.0025 0.0081 1 Ibovespa (Stocks) -0.0053 0.0051 -0.0499 -0.0084 -0.0225 1
IMA-S (Gov. Bonds) -0.3678 -0.0932 -0.0356 0.0797 -0.0693 -0.0357 1
18 The regressions focus on the role of the stocks and bonds returns to explain the price-to-NAV deviations. In order to prevent omitted variable bias, the four fundamental attributes mentioned in the literature review section will be included as control variables.
Additionally, since appraisals (that are the base for NAV’s) are usually lagged measures of asset value, it is reasonable to imagine that the Price-to-NAV ratio is also influenced by its lag. For this reason, a way to predict Price-to-NAV deviations is to use previous months’ ratios, that is, to apply an autoregressive model. Another important indication of the need of an autoregressive model is by the observation of the correlations between Premium and its lagged values. Table 3 depicts the correlations between this variable and its first four lags.
Table 3 – The correlation between premium and its first four lags
Note: based on data from December 2007 to February 2015 for Brazilian Real Estate Investment Funds.
Table 3 shows that Premium is strongly positively autocorrelated. The first autocorrelation is 0.96. The sample autocorrelation decreases as the lag increases, but even at a lag of 4 months it remains above 0.85. The final model, therefore, contains one lagged value of the dependent variable, Premium.
The study will also control for entity (BREIF’s) fixed effects and time effects. As the study uses panel data, this procedure is a way of reducing the chances of omitted variable bias. It is also expected to increase model fit.
The fixed-effects variable will capture the influence of omitted variables that vary across entities but do not change over time. Examples of these variables could be management style and the long-term strategy of the fund. The time-effects variable, on its turn, will capture the influence of omitted variables that change over time but affect all entities in
Premium 1st Lag 2nd Lag 3rd Lag 4th Lag
Premium 1 1st Lag 0.9547 1 2nd Lag 0.9215 0.9540 1 3rd Lag 0.8868 0.9204 0.9536 1 4th Lag 0.8511 0.8859 0.9206 0.9536 1 Observations: 2,467
19 the same fashion. Examples of these omitted variables could be the GDP growth and macroeconomic policy changes.
5.1. Model specification
On the four first model specifications, a multiple regression with ordinary least squares estimators is used to analyze the influence of stocks and bonds on the price-to-NAV deviations. Arellano–Bover/Blundell–Bond linear dynamic panel-data estimation is used in Specification 5, the dynamic model. This specification contains one lag of the Premium variable and controls for unobserved entity fixed effects, as well as time fixed effects. The estimation method, which uses additional moment conditions, is built on the works of Arellano and Bover (1995) and of Blundell and Bond (1998).
A bottom-up approach was used for model construction with regressor being added from one specification to the next. One of the reasons for adopting this procedure was to check for parameter consistency, that is, to check if the coefficients would remain stable after the addition of new regressors. The objective is, then, to show that the estimates of the model parameters are not very sensitive to the exact specification that is being used and that the true, but not observable, value of the coefficient is close to those reported on the regression.
The first specification considers only the influence of stocks and bonds returns. From specification one to four, time fixed effects, fund fundamental variables and fund fixed effects are added as control variables. Specification 5, with all control variables and the first lag of the Premium variable – the most elaborated specification and final model – is as follows:
20 In equation (3), Premiumit is the price-to-NAV deviation, that can be, in fact, positive or negative; β0 is the constant parameter; DYit is the monthly dividend yield paid by each fund; Riskit is the fund volatility, measured by the standard deviation of the previous-year share price; Sizeit is the logarithmic transformation of the market capitalization of each fund in each month; Liqit is the liquidity, measured by the ratio between monthly traded volume to the market capitalization; Stockit represents the monthly return of Ibovespa;
GovBondit is the monthly return of IMA-S; αit and λit represent, respectively, the entity fixed effects and time fixed effects. Except from Sizeit, all the variables are in percentage terms.
6. Results
6.1. Main regression estimates
Concerning the regression results, several model specifications are presented in Table 4. As can be seen in this table, the first model specification includes only the bonds and the stock indices. The two variables of interest for the study show statistically significant coefficients at the 1% level. The effect is negative for both variables, but the simplicity of the first specification suggests the presence of omitted variable bias. The specification also does not explain much of the total variance of the data (9.86%).
In Specification 2, by adding just the time fixed effects, a significant increase in the R² is observed. Both coefficients of the variables of interest become positive on this specification. However, the total variance explained by the model is still small, of approximately 34%.
In Specification 3, the four fund fundamental attributes – fund size, fund volatility, fund liquidity and the monthly dividend yield – are included in the regression specification as control variables. Except from the dividend yield coefficient, none of the other control variable coefficients is statistically significant.
As expected, in Specification 4, it is possible to verify a huge increase in the model fit, going from only 37.5% to almost 78%. This shows the importance of fund fixed effects on the behavior of the Price-to-NAV deviations, computing for the importance of the individual characteristics of the funds. In the forth specification, controlling for entity fixed effects does not change the previous specification main conclusions. Stocks and bonds indices coefficients remain significant at the 1% level, as well as the dividend yield
21 coefficient. According to this specification, an increase of one percentage point in stock returns leads to an increase of approximately 0.13 percent in premiums and an increase of one percentage point in bonds returns leads to an increase of approximately 7.5 percent in premiums.
Table 4 – Regression estimates with price-to-NAV as dependent variable
Notes: 1) Fund size is a logarithmic transformation and all the other variables are in percentages. 2) The t-statistics for the coefficients of the first four model specifications are presented in parentheses. The z-score is presented in parentheses, for Specification 5.
*** Significant at the 1% level ** Significant at the 5% level
Finally, in Specification 5, the first lag of the dependent variable is added, together with all the control variables and fixed effects. To confirm the need of adding lags to the model, a Wooldridge test for autocorrelation in panel data (derived from Wooldridge, 2002) was performed. The test is applied in Specification 4 (without lags) and the null hypothesis of no first order autocorrelation is strongly rejected.
The first important result is that there is a strong evidence that the price-to-NAV deviations are influenced by their first lag. This is consistent with the presence of
Specification 1 (Stocks & Bonds)
Specification 2 (With Time Effects)
Specification 3 (With Controls)
Specification 4 (With Fixed Effects)
Specification 5 (Dynamic Model) Premium 1st Lag 0.8400389 (25.90)*** Ibovespa (Stocks) -0.0758154 0.1307006 0.1078587 0.1252366 0.0946533 (-2.61)*** (5.50)*** (4.26)*** (5.24)*** (4.26)*** IMA-S (Gov. Bonds) -48.87141 5.455144 6.961455 7.472928 9.416763 (-6.75)*** (2.42)** (2.50)** (3.08)*** (2.85)*** Fund Size 3.665571 3.35458 2.815776 (1.71) (0.39) (1.01) Fund Liquidity -0.910639 -0.2383797 -0.1159096 (-1.67) (-1.31) (-1.63) Dividend Yield 1.347247 0.918252 0.5022039 (5.42)*** (4.82)*** (11.07)*** Fund Volatility 0.2884691 -0.2781281 0.0915344 (0.50) (-0.39) (0.59) Entity Fixed Effects No No No Yes Yes Time Fixed Effects No Yes Yes Yes Yes Clustered Standard Errors Yes Yes Yes Yes Yes Estimation Method OLS OLS OLS OLS Arellano–Bover/
Blundell–Bond # Obs. 2,791 2,791 2,170 2,170 2,143 R² 9.86% 33.93% 37.45% 77.93%
22 momentum effects on the valuation of BREIF’s shares, which end up being reflected on the price-to-NAV difference. The first lag coefficient is equal to 0.84 and is statistically significant at the 1% level. This means that an increase of one percentage point in the past-month premium leads to a 0.81 percentage point increase in the present-month premium.
Focusing on the influence of the stock market and of the bonds market on the price-to-NAV deviations, it can be seen a positive and statistically significant relationship, in the regression. Starting by the stock market’s influence, it is possible to observe a coefficient of approximately 0.1, which means that an increase of 1 percentage point in the stock market return leads to a 0.1 percentage point increase in the premiums.
A significant relationship is an evidence that, in some sense, BREIF’s market is tied to the whole stock market, like was already found in previous studies for REIT’s. The economic sense is that stock markets affect fund share price and fund NAV in different ways, which causes a change in the price-to-NAV ratio. BREIF’s shares might be affected quickly by the stock market movements, while NAV’s reflect slower changes in the space markets, that might occur across months or years. Economic expectations revealed by the stock market performance might be incorporated quickly to the valuation of BREIF’s shares, while appraisal-based NAV’s tend to rely more on hard evidence, like transactions and actual rent levels. Furthermore, NAV’s, like any appraisal-based measurement, can suffer from appraisal smoothing, which also might explain part of the price-to-NAV difference. The contradiction, however, is that price-to-NAV deviations are persistent over long periods (several years, sometimes), which is not compatible with a simple lagging on appraisals. Other considerations regarding appraisal smoothing are presented in Section 7.
The magnitude of the stock index coefficient is not so large but, taking into account the standard deviation of stock returns (6.6%), it is not negligible. Additionally, a z-score of 4.26 shows it is a significant relationship at the 1% level. The coefficient is also quite stable throughout the different model specifications – always around 0.11 – which brings some confidence to the true magnitude of the impact of the stock market on the BREIF’s’ price-to-NAV deviations.
With regards to the influence of the bonds market in the price-to-NAV deviations, it is possible to observe a huge impact with a coefficient of approximately 9.5. The coefficient
23 is positive and statistically significant at the 1% level. According to this estimate, an increase of 1 percentage point in the return of the government bonds index increases premiums by 9.5 percentage points. However, despite presenting a large impact, bonds returns tend to change little from one month to the next (standard deviation of 0.14%), which limits major changes on the price-to-NAV ratio.
A possible explanation for the phenomenon lies on the dynamics of the monetary policy, with government interest rates being used as the primary instrument to control inflation in Brazil (Banco Central do Brasil, 2015). In principle, changes in the basic interest rates should be incorporated into the valuations performed by both BREIF’s investors and NAV appraisers, resulting in no price-to-NAV deviation. However, since most rental contracts are inflation-indexed, commercial properties become inflation hedging assets and, therefore, more valuable to investor seeking protection. To put it another way, small investors that cannot buy properties, or investors that do not want to do so, might be willing to pay a premium on BREIF’s shares for its inflation-hedging characteristics, increasing the difference between price-to-NAV. In this case, the price-to-NAV deviation could be attributed to the securitization of real estate assets.
Concerning the control variables, fund size presents a non-significant coefficient of 2.82. The positive sign is consistent with the idea that bigger funds are viewed by investors as better than smaller funds. However, this result is different from the results found by Lee et al. (2013) and by Capozza and Lee (1995), who found a significant impact of size on price-to-NAV deviations. The coefficient is fairly stable but non-significant in all model specifications presented in Table 4.
Regarding fund liquidity, an unexpected negative coefficient was found. The coefficient was -0.12, meaning that an increase in one percentage point in liquidity would lead to a decrease in premiums of -0.12 percentage points. Despite the relatively small value, this result was definitely not expected, since liquidity is usually a valuable attribute to investors. The negative sign is present in all models, but in none of them the liquidity coefficient is statistically significant.
Considering the influence of the dividend yield, it is possible to observe a coefficient of 0.5, meaning that a 1 percentage point increase in dividends leads to a 0.5 percentage point increase in the premiums. This result is statistically significant at the 1% level and reflects the idea of BREIF’s being viewed as a fixed-income securities that pay a monthly
24 “coupon”. In this sense, a BREIF that is capable of, through its operations, generate and distribute a good amount of cash to investors should be more valuable than a BREIF that cannot do the same thing. This result is in line with the work of Bradley et al. (1998) and Kallberg et al. (2003), who found dividend policy to be a good signal of the future prospects of the firm. Bearing in mind that, according to BREIF’s regulation, a minimum of 95% of all profits (in a cash base) must be distributed to fund shareholders in the form of dividends, the dividend yield can be seen as a sign of the operational health of the fund, in terms of its capacity to generate a constant stream of cash. Throughout the different model specifications, the coefficient for the dividend yield remains stable and statistically significant at the 1% level.
When it comes to the fund volatility coefficient, it is possible to observe a small and non-significant influence on the premiums. A positive coefficient was not expected, since the more volatile share price returns are, the higher the risk for investors and, consequently, the higher the expected discounts on the price-to-NAV ratio. However, the coefficient of 0.09 is not statistically different from zero and is also not stable throughout other specifications.
The study brings an important implication to investors of BREIF’s, REIT’s and property companies traded on public markets. From a practical perspective, in terms of pricing, the results emphasize the distinction between real estate direct and public, indirect, investments. That is, while appraisals tend to be more restricted to the space market, investors should take into account the role of stocks and bonds markets when valuing BREIF’s.
Furthermore, the link between stocks and bonds markets to the premiums show that viewing NAV – a valuation of real estate properties – as an indication of the fundamental value of a fund’s shares, might be inaccurate and risky. This means that investors should be careful in adopting strategies that consider that share price will converge to NAV, like the one proposed by Gentry et al. (2004), at least in the short- and medium-terms. In terms of academic implication, the results might help to explain why asset bubbles are also characterized by large price-to-NAV deviations. As proposed by Penman (2007), the extent of the deviations themselves are not caused by bubble behaviour. However, the price-to-NAV deviations might grow when an asset bubble is being formed because of the influence of the stock and the bonds markets on it.
25 Finally, since there is an evidence that the influence of the stocks and bonds markets are relevant to explain the price-to-NAV puzzle, previous studies on the topic that do not account for it might have biased results.
6.2. Robustness checks
Starting from the literature inputs and observing the t-statistics and R-squares of the proposed regression give the researcher some confidence on the reliability of the chosen model specification. Nevertheless, it is possible to use some additional checks to assess its robustness and stability. The presence of outliers, for example, might contaminate the whole sample and affect the main conclusions of the study. Another check, is to test new specifications, without the variables that didn’t have significant coefficients. Table 5 shows the results for some changed specifications.
In order to control for the possible presence of outliers, standardized residuals are generated and observations whose residuals are more than three standard deviations away from the mean are deleted from the sample, as potential outliers. This procedure resulted in the elimination of 38 observations.
Specification 6 is the same as Specification 5 without the presence of potential outliers. The conclusions for the main variables of interest remains absolutely the same. The coefficients for both stocks and bonds indices do not change significantly and the z-scores remain statistically significant at the 1% level. However, regarding the four fund attributes, an important change is noticed on the results of Specification 6. Fund size coefficient jumps from 2.8 to 6.2. Moreover, it becomes statistically significant at the 5% level.
Specification 7 presents another check: Since fund volatility had non-significant coefficients in all previous models, it is removed from this specification. This procedure has the additional benefit of increasing the sample size. This occurs because the calculation of fund volatility is based on the data of the previous 12 months, which renders missing values to the eleven first months’ observations.
As can be seen in Specification 7, a bigger sample is considered and the fund size coefficient of 3.65 becomes significant at the 1% level. The other parameters remain consistent with previous models both in terms of magnitude and significance levels.
26
Table 5 - Regression estimates without potential outliers
Notes: 1) Fund size is a logarithmic transformation and all the other variables are in percentages. 2) The t-statistics for the coefficients of the first four models are presented in parentheses. The z-score is presented in parentheses, for Specification 5.
*** Significant at the 1% level ** Significant at the 5% level
Finally, in Specification 8, fund liquidity, that was not significant in any model, is removed, leaving only statistically significant parameters. In Specification 8, the influence of stock returns is a bit higher than 0.09 and is statistically significant at the 1% level. The coefficient of the bonds returns is 6.7 and also statistically significant at the 1% level. Fund size and dividend yield have significant coefficients of 3.6 and 0.4.
7. Limitations
Despite the attempt to develop a rigorous study, some limitations are present in the research. In order to provide a better assessment of the study implications, it is important to acknowledge them. Specification 6 (Without Outliers) Specification 7 (Without Volatility) Specification 8 (Without Liquidity) Premium 1st Lag 0.7737593 0.7727626 0.7736154 (27.12)*** (35.09)*** (35.04)*** Ibovespa (Stocks) 0.0996333 0.0900241 0.0910809 (4.54)*** (4.83)*** (4.86)***
IMA-S (Gov. Bonds) 9.666926 6.664411 6.696366
(3.94)*** (2.72)*** (2.73)*** Fund Size 6.153706 3.652246 3.630914 (2.57)** (2.83)*** (2.80)*** Fund Liquidity -0.0961153 -0.1005672 (-1.57) (-1.60) Dividend Yield 0.3770469 0.3961913 0.3939919 (2.37)** (2.30)** (2.28)** Fund Volatility -0.0843865 (-0.40)
Entity Fixed Effects Yes Yes Yes
Time Fixed Effects Yes Yes Yes
Clustered Standard Errors Yes Yes Yes
Estimation Method Arellano–Bover/
Blundell–Bond Arellano–Bover/ Blundell–Bond Arellano–Bover/ Blundell–Bond # Obs. 2,097 2,629 2,629
27 The first limitation is related to the time span of the dataset, which have data points concentrated in only 5 years. According to Clayton and MacKinnnon (2003), some studies reveal that the relation between “wall street” and “main street”, that is, public and private real estate markets, is more clear in the long term. According to Liow’s findings (2003), share prices and NAV’s are linked, but the extent of the mean reversion between them is slow, making deviations prolonged across time. This indicates that a better understanding of the puzzle would require longer time series of data. This is reinforced by the relatively young nature of BREIF’s market, which might imply a lack of familiarity with these funds and their modus operandi. For these reasons, the research would benefit from a longer time horizon.
Another limitation is related to the lack of data on short sales. Brounen et al. (2013) used data related to shares available for lending as well as the value and quantity of shares currently on loan to short sellers to measure short sales activity and constraints. They found that the variation in short sale activity accounts for approximately one-third of the variation in NAV premiums. Due to the lack of data, this study do not control for short sales constraints.
A third limitation is related to the occurrence of lagging and smoothing in appraisal-based measurements. Fund NAV’s might suffer from lagging and smoothing, which are not considered once the price-to-NAV ratio is investigated directly. In order to adjust for lagging and smoothing, it would be necessary to treat share price and NAV separately, and to apply a different methodology. The presence of lagging could explain part of the price-to-NAV puzzle, but, since these differences are persistent across long periods of time, it is unlikely that it is a single determinant factor.
Finally, it is important to bear in mind that the study was conducted with Brazilian data, in a very young and relatively small public real estate market. This means that the generalizability of its conclusions for more mature and bigger public real estate markets is limited.
8. Conclusion
The present study started from a puzzle that is not completely understood: the price-to-NAV deviations. The puzzle is born in large part due to the presence of a dual asset market in which real estate properties can be negotiated: a private market and a public market.
28 The possibility of negotiating assets in these two different markets allows for an arbitrage opportunity that should prevent price deviations bigger than those of transaction costs. However, what is observed in several countries is that share price, a public market price, is different from Net Asset Value per share, a direct private market price. Moreover, this happens over long periods of time and with deviations that would more than enough cover for transaction costs.
Several studies addressed the puzzle, and successfully helped to explain part of it. Nevertheless, no conclusive evidence was reached yet. The present study brought new indications to this matter and evidence from a new country, that haven’t been studied yet, with a new dataset.
The study conception parts from the idea that public markets’ real estate investments behave often more like stocks and bonds than like direct private real estate. In this sense, if investors see BREIF’s more like stocks, it will also be subjected to the market “mood”, bringing in its valuation methods and its characteristics, like noise and herding. In a similar manner, if investors see BREIF’s more like a bond, that provides monthly “coupons”, it will tend to be aligned with the bonds market.
After controlling for important fund attributes and entity and time fixed effects, the regression results showed that the stocks and bonds markets have, indeed, a strong influence on the price-to-NAV deviations. Controlling for entity and time fixed effects proved to be important in order to avoid omitted variable bias and to improve model fit. After controlling for them, the model coefficients changed significantly and the R² jumped from 9.86% to 77.93%.
A first finding is that the price-to-NAV deviations seem to be autocorrelated. Since NAV measurements tend to be lagged, it would be expected that deviations would be lagged too. The results showed that the first lag have strong explanation power for the discrepancies, being significant at the 1% level.
Regarding the stock market influence, the coefficient reported in Model 5 was of 0.1, meaning that an increase in 1 percentage point in the stock market return leads to an increase of 0.1% on price-to-NAV deviations. The impact doesn’t seem to be enourmous at first sight, but considering the high volatility of the stock market, it is not negligible. The coefficient is also statistically significant at the 1% level, with a z-score of 4.26.
29 This result reveals that the stock market has a statistically different influence on BREIF’s share price and BREIF’s NAV, explaining partly the price-to-NAV deviation. This might be due to the quicker incorporation of expectations brought by the stock market into the share price.
Regarding the IMA-S coefficient, that is, the government bonds index coefficient, it showed to have considerable explanatory power for price-to-NAV deviations, with a coefficient of 9.4 (in Model 5), significant at the 1% level. High returns on the government bonds market are associated then with even higher premiums. However, since the government bonds’ interests vary little from one month to the next, major changes on the price-to-NAV ratio due to the bonds market are not expected.
The positive influence found in the study for the IMA-S might be a consequence of the basic interest rate being used as the main tool to control inflation and to the fact that commercial real estate is often an inflation-hedging asset. In this sense, investors could be willing to pay a premium on BREIF’s shares, resulting in a price-to-NAV difference. Concerning the control variables, dividend yield showed to have an impact on premiums, with a coefficient significant at the 5% level. Fund size significance level was inconsistent. Without removing potential outliers, fund size did not show any significant impact on the price-to-NAV deviations. However, after removing potential outliers, fund size coefficient became a significant parameter. Fund volatility and fund liquidity didn’t show to have significant coefficients in any specification.
The implication of the study findings are several. First, investors of BREIF’s should take into account that NAV and fund share price are valuations of different assets, negotiated in different markets, that are influenced by different factors. Therefore, to simply treat NAV’s as a fundamental value of BREIF’s shares is misleading. For this reason, to adopt strategies that consider that share price will converge to NAV might be very risky. Second, the study stablishes a possible connection between large price-to-NAV deviations and the formation of asset bubbles. Third, among the control variables, the consistency of the dividend yield coefficient adds up to the evidence found by Bradley et al. (1998) and by Kallberg et al. (2003) that, in the case of passive property companies, dividend yields work as a signal for future performance.
Finally, even bringing a strong evidence of the role of stocks and bonds in explaining NAV deviations, the mechanisms by which these markets influence the
