Do Traditional Real Estate ETFs Increase the Volatility of REITs?

Abstract

We examine the impact of the introduction of traditional, non-U.S., real estate exchange traded funds (ETFs) on the realized volatility of real estate ETFs’ component stocks in a global setting. We first estimate the volatilities of the individual constituent stocks and then test for breakpoints and jumps. Analyses are conducted separately across each ETF and its underlying securities and jointly over the securities and their attributes. Under a comprehensive, robust approach that properly pulls the securities and their attributes together and includes new statistical tests that can be applied across the literature, present findings suggest that the initiation of traditional, non-U.S., real estate ETFs across global markets only modestly impacts the attributes of the ETFs’ underlying component securities. The overwhelming consensus derived from studies of non-Real Estate Investment Trust (REIT) stocks that ETF launches impact the volatilities of the underlying securities of the ETF portfolio is not supported for REIT ETFs and their constituent stocks.

Keywords:

• REITs
• ETFs
• breakpoints
• global volatility
• global impact analyses

JEL CODES:

• G15
• C58
• G14
• F36

Notes

Notes

1 Data from both NAREIT and EPRA provide evidence of the rapid expansion of REITs and similar investment structures over the last 20 years, especially starting in 1993 with regulatory changes in the United States. Hardin et al. (2017) highlight the role of institutional ownership in U.S. REITs and the increase in institutional ownership and its impact on governance and performance. The introduction ETFs would reduce the role of governance and performance attributes associated with institutional ownership as found in Mauck and Price (2018) who look at international diversification and REIT management.

2 Soyeh and Wiley (2019) distinguish between traded and non-traded REITs and indicate that the exit strategies for non-traded REITs in large part are conversions to traded REITs, which further expands the universe of investments.

3 There is substantial literature regarding the related area of real estate mutual funds, which also provide diversified returns. See studies by Kallberg et al. (2000), Lin and Yung (2004), Ling and Naranjo (2006), Bond and Mitchell (2010), and Chou and Hardin (2014) for background on real estate mutual funds.

4 We are deeply indebted to an anonymous reviewer for sharpening our thought on prior literature in this regard.

5 The international traditional real estate ETFs scrutinized come from four continents and are categorized into five distinct groups, with each ETF having a different country of origin and a distinctive inception date. These groups are: Australia (inception date February 18, 2002), Ireland (March 20, 2007), Switzerland (November 3, 2009), Brazil (February 22, 2010) and South Africa (May 30, 2011).

6 International traditional real estate ETFs have more of their constituent stocks in non-REIT real estate firms than do the U.S. ETFs, where real estate investment trusts dominate. Many countries did not have until recently the required legal framework for REITs. For instance, the U.K. REIT legislation was enacted in 2007 (rewritten in 2010), and REITs were introduced in Ireland in 2013 by the Finance Act. Similarly, South African REITs were only introduced in 2013.

7 For instance, Curcio et al. (2012) derive volatility from residuals of estimated equations that are based on securities and market (the S&P 500 index) returns. In addition, they consider both leveraged and traditional ETFs.

8 Studies that investigate the liquidity of the underlying stocks for general (all sector) ETFs include, among others, Van Ness et al. (2005), Yu (2005), Hamm (2014), and De Winne et al. (2011). The impact of general (all sector) ETFs on constituent stocks of the fundamental indices is addressed by Subrahmanyam (1991) and Gorton and Pennacchi (1991). Subrahmanyam (1991) and Gorton and Pennacchi (1991) models, which foresee a transfer of liquidity traders toward basket securities, accordingly cause a reduction in liquidity trading in individual securities. The individual stocks are traded predominantly by less-informed traders and therefore have increasing adverse selection costs (bigger bid-ask spreads). Later outcomes are mixed: Hegde and McDermott (2004) show that the introduction of two ETFs on the Dow Jones Industrial Average results in increased liquidity of the Dow 30 composite stocks through the initial 50 days of trading. Alternatively, the introduction of QQQ on the liquidity of Nasdaq 100 composite stocks is found to be less pronounced. Extending the work of Hegde and McDermott (2004), Richie and Madura (2007) assess the asymmetry in liquidity effects and the risk of component securities after the introduction of the Nasdaq 100 ETFs (QQQ). They find that the liquidity of the underlying stocks rises, especially those stocks that have inferior weights in the index, which display more conspicuous liquidity increases. The impact of the introduction of ETFs on constituent securities’ trading characteristics (bid-ask spread, trading volume, etc.) is also characterized by a modification in the proportion of informed versus noise traders, in other words a relocation of investors from one kind of instrument to another.

9 See, e.g., “SEC Reviewing Effects of ETFs on Volatility" by Andrew Ackerman, Wall Street Journal, October 19, 2011, “Volatility, Thy Name is E.T.F.?” by Andrew Ross Sorkin, New York Times, October 10, 2011, and “SEC To Look Into ETFs Exacerbating Volatility,” Money Management Letter, October 20, 2016.

10 See Curcio et al. (2012) for an investigation of U.S. real estate ETFs. It is also recognized that the new techniques introduced in the present study could also be applied to U.S. ETFs.

11 Our ETF selection process is exactly the same as the domestic ETF selection process extensively documented in the literature when using data sources like http://etfdb.com.

12 In Australia, the SLF AU; long name = SPDR S&P/ASX 200 LISTED PROP.

13 For Ireland: IUKP’s LN; full name = ISHARES UK PROPERTY.

14 Chiang (2010) uses market capitalization and age as proxies for a firm’s information openness. Book-to-market ratio and dividends per share are standard variables in most REIT’s capital structure studies (see, for instance, Hardin and Wu, 2010, Harrison et al., 2011, and Chiang, DeWitt, Folkman and Jiao, 2018, among others). These variables may also be argued to signal a firm’s wealth effect (Kyle & Xiong, 2001) and financial constraints (Yuan, 2005).

15 Empirically, a number of other macro variables are equally relevant for consideration. The inclusion of LIBOR is due to the importance of interest rates in the real estate market and the depth of this variable globally. The choice of foreign currency and MSCI index is based on similar grounds. LIBOR skyrocketed due to asset-backed commercial paper in October 2008. However, that period does not coincide with our window designation, being too late for Ireland's and too early for Switzerland's observations. LIBOR scandal in 2013 and subsequent changes in LIBOR are not close to any ETFs’ inception dates to influence our analysis.

16 In practice, and speaking in general terms, the financial industry uses six measures of historical volatility estimates: close-to-close, exponentially weighted, Garman-Klass, Parkinson, Rogers-Satchell, and Yang-Zhang. The Yang-Zhang method is recommended for small samples, whereas close-to-close is deemed equally appropriate for large samples. See Poon and Granger (2003).

17 It is shown that under specific computations, the outcome of daily data could mimic the attributes of minute-by-minute high frequency data. See Corwin and Schultz (2012) and Figueiredo and Parhizgari (2017).

18 Garman and Klass (1980) calculate their volatility measure under the assumption that stock prices are log normally distributed and follow a geometric Brownian motion without drift. This methodology includes not only high and low but also open and close historical prices and is one of the often known as high-low-open-close volatility estimates (HLOCs). However, it ignores overnight jumps (close-to-open) in prices. To this extent, this measure would underestimate volatility if overnight jumps in prices exist, thus generating a potential source of bias. It is computed using the following reduced form estimator: $$R{V}_{HLOC}=\sqrt{\frac{252}{n}{\displaystyle \sum _{i=1}^n\left[\frac{1}{2}\right(\ln {\left(\frac{H_i}{L_i}\right)}^2-(2\mathrm{\hspace{0.17em}ln\hspace{0.17em}}2-1){\left(\ln \frac{C_i}{O_i}\right)}^2]}}$$ where H, L, C, and O, are, respectively, high, low, close, and open daily values; n defines the interval used to measure volatility; and the 252/n annualizes the measures. Relatively speaking, RV_HLOC is shown to be about eight times more efficient than the classical close-to-close estimators.

19 All our markets are closed overnight and during holidays and have no lunch breaks.

20 Shu and Zhang (2006) determine that considerable differences exist among different estimators if asset returns distribution entails an opening jump or a large drift.

21 See Yang and Zhang (2000) for further details.

22 Bai and Perron extend the Quandt-Andrews framework by allowing for several unknown breakpoints.

23 Gallant et al. (1992) report evidence of a contemporaneous positive correlation between conditional volatility and volume. Chan and Fong (2006) corroborate the findings of Jones et al. (1994) that the number of trades is the predominant factor behind the volume-volatility relation. Giot et al. (2010), after decomposing realized volatility into a continuous varying component and a discontinuous jump component, conclude that the volume-volatility relation holds for the continuous component only.

24 Securitized property markets are cointegrated with key macroeconomic factors in the long-run and also influenced by respective economies in the short-run (Yunus, 2012).

25 ln (P_t) is assumed to follow a random walk (Parkinson, 1980).

26 It should be noted that simple Pearson correlations are inadequate to capture the dynamics of the changes that occur.

27 All corresponding results for the period of 121 days are available upon request.

28 Over the periods of 121 days and under the roll-over dates, Australia is not in possession of any breakpoints, Ireland has three out of 27 stocks, and Switzerland has three out of 13 stocks exhibiting breakpoints. Similarly, Brazil has one out of 16 stocks and South Africa has two out of eight stocks that are confirmed by Monte Carlo robustness

29 We report only the 181 days results. All corresponding results for the period of 121 days are available upon request.

30 Exact dates are available upon request.

31 As in the Chow tests, the statistical significance is at the 5% level or below.

32 Over the period of 121 days, the total number of breakpoints is 13 out of 27 for Ireland (date of inception 3/20/2007). However, only one security DJAN cannot be rejected based on the Monte Carlo results. Similarly, for Switzerland (date of inception 11/3/2009) the total number of breakpoints is six, yet only one security (SIAT) dominates the Monte Carlo results at the desired nominal 5% significance level. The disparity in these results could reflect the pre- and post-effects of global financial crisis. The Monte Carlo bootstrapping methodology is powerful enough to exclude noise or spurious effects, if any. This conclusion deserves further examination.

33 For example, none of the selected markets are closed during the day, though trading in them is still subject to time zone differences. Suffice to add at this point that in general, range-based estimators depend on the assumption of continuous geometric Brownian motion and may have unexpected biases, especially for assets with low liquidity and low transaction volumes.

34 The total number of ETFs’ constituencies considered is 78. Eight stocks are excluded due to lack of data.

35 As in the other tables, statistical significance is at the 5% level or below.

36 It should be pointed out that as the number of control variables increase, the potential for finding statistically significant breakpoints may decrease. This is expected on a priori grounds since there is often a possibility that “other” factors may be the cause of the shifts. We observe this phenomenon when we compare the results of our experiments with determining and macro variables (Table 7) against the experiments that include solely the market indices as the only explanatory variables. See Table 8 under Bai and Perron’s methodology with a single explanatory variable.

37 Some of these dates are the same as the inception or news dates, hence the headings in the subsequent tables.