Citation
Real Estate in the Sunshine State: An Analysis of the Impacts of Local Economic Conditions on the Performance of REITs Investing in Florida

Material Information

Title:
Real Estate in the Sunshine State: An Analysis of the Impacts of Local Economic Conditions on the Performance of REITs Investing in Florida
Creator:
Alkhatib, Dima
Publisher:
University of Florida
Publication Date:
Language:
English

Subjects

Subjects / Keywords:
Undergraduate Honors Thesis/Project

Notes

Abstract:
This paper studies the effect of local macroeconomic conditions on the annual returns of REITs. Specifically, I look at Florida’s economic performance using four fundamental indicators: state GDP, employment, personal consumption expenditure, and consumer sentiment, and assess whether there is a difference in the relationship between the local economy and REIT returns across the firms that are highly concentrated in Florida relative to those that are not. The empirical results initially confirm a higher equity return for firms with a high exposure to Florida. However, the differential in geographic exposure generally does not affect the sensitivity of the returns to the economic variables, with the exception of consumer sentiment, which displays a negative impact on the returns of high exposure REITs. When the economic variables are scaled by the time varying concentrations of each REIT, the coefficients of each variable are still significant, with short term mean reversion detected from the change of the regression coefficients from negative in a lagged regression to positive in a contemporaneous regression. The growth rate of Florida’s GDP is the most significant economic indicator. Further analysis shows that its effect varies across property sectors, with significant effects observed in retail, residential, and industrial REITs but not in healthcare REITs. ( en )
General Note:
Awarded Bachelor of Science in Business Administration, magna cum laude, on May 6, 2023. Major: Finance
General Note:
College or School: Business
General Note:
Advisor: Andy Naranjo. Advisor Department or School: Finance, Insurance and Real Estate.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Dima Alkhatib. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

! " ! ! ! ! "#$%!&'($(#!)*!(+#!,-*'+)*#!,($(#. ! /*! /* $%0')' ! 12 ! (+#! 3 45$6( ' ! 1 2 ! 7 16$%! & 61*14)6 ! 8 1*9)()1*'!1*!(+#! : #;21;4$*6#!12! "&3<' ! 3 *=#'()*>!)*!?%1;)9$ ! ! ! @0 .!A)4$! / % ! B +$()C ! ! ! ! ! ! ! D*9#;>;$9-$(#!E1*1;'!<+#')' ! @,@/ F ?3G ! D*)=#;')(0!12!?%1;)9$ ! E#$=#*#;!,6+11%!1 2!@-')*#'' ! /5;)%!HIHJ! ! ,-5#;=)'1;.!/*90!G$;$*K1 ! ! ! ! ! ! ! ! ! ! ! ! ! !

PAGE 2

! # ! " #$ %&'(% ! This paper studies the effect of local macroeconomic conditions on the an nual returns of REITs. Specifically, I look at Florid a ' s economic performance using four fundamental indicators: state GDP, employment, personal consumption e xpenditure, and consumer sentiment, and assess whether there is a difference in the relationship between the local economy and REIT returns across the firms that are highly concentrated in Florida relative to those that are not. The empirical results initially confirm a higher equity return for firms with a high exposure to Florida. Howe ver, the differential in geographic exposure generally does not affect the sensit ivity of the returns to the economic variables, with the exception of consumer sentiment , which displays a negative impact on the returns of high exposure REITs. When the econo mic variables are scaled by the time varying concentration s of each REIT, the coef ficients of each variable are still significant, with short term mean reversion detected from the change of the regression coefficients from negative in a lagged regression to positive in a contemporaneous regression. The growth rate of Florida ' s GDP is the most significant economic indicator. Further analysis shows that its effect varies across property sectors, with significant effects observed in retail, residential, and indus trial REITs but not in healthcare REITs. ! 1. ! I ntroduction In finance, a simple question leads to vast outcomes. When a conjecture is made, it opens a door to growing curiosity that contributes to it continually over time, either with supporting empirical evidence or counterarguments. The development of real estat e investment trusts in the 1960s has piqued the attention of investors and economists alike and led to the emergence of simple questions that later translated to an extensive body of literature. Although that body is not as established as in other areas of finance, it is advancing quickly in tandem with academics' desire to understand the unique structure and characteristics of REITs. REITs have become a popular scope of interest for several reasons. Because they are a relatively modern investment vehicle, they offer a valuable opportunity to learn about their behavior and performance. Secondly, because they are neither exclusively stocks nor real estate, their classification as a hybrid asset class provides insights regarding how the two distinct asset clas ses may be re lated. 1 Moreover, being publicly traded securities, expansive data is available from these securities to analyze, which reveal more accurate and timely information than what would be collected from the less responsive private real estate market . In this p aper, I am attempting to add to the increasing body of literature that explores the sensitivity of REITs to macroeconomic factors. Such contributions became prevalent after researchers deduced that REITs share similar behavior patterns to stock s, specifica lly small capitalization ! 1 Brounen, D., & De Koning, S. (2012). 50 years of real estate investment trusts: An international examination of the rise and performance of REITs. Journal of Real Estate Literature , 20 (2), 197 223.

PAGE 3

! $ ! stocks 2 . The implications of this finding are pivotal when we consider that stocks have demonstrated sensitivity to economic variables and business cycles, as per the seminal paper by Chen, Roll, and Ross (1986). The aut hors found that changes in inflation, industrial production, and market risk premium significantly affect stock returns, with the sensitivities varying across time. Researchers were able to extend this framework to other asset classes, including REITs, and found that they can further explore their risk and return behavior once macroeconomic conditions are accounted for. Particularly, Chaudhry, Bhargava, and Weeks (2022) demonstrated that the default risk premium and unanticipated inflation adversely affect REIT return s, while GDP and the federal funds rate have a positive effect on REIT returns. Glackcock and Lu Andrews (2013) present strong evidence for the relationship between funding liquidity in the REIT market and macroeconomic factors, which in turn in fluences th e market trading liquidity of REITs and consequently their stock prices and returns. A s funding liquidity increases, REIT market liquidity increases, with the effect changing across business cycles. The recency of these papers suggest s that the conclusions remain relevant today , which is an important assumption because the structure and characteristics of REITs before 1990 differed from those of REITs that went public after 1990 3 . However, many papers that examined REITs prior to 1990 present robu st conclus ions that still hold. Ling and Naranjo (1997) identify the growth rates in per capita consumption, the treasury bill rate, the term structure, and unexpected inflation as macroeconomic variables that affect commercial real estate returns, which a re the und erlying assets of REIT. Rather than looking at macroeconomic variables that proxy for national economic conditions, I have narrowed my scope to focus on local macroeconomic factors related to the state of Florida. My main interest is to investig ate wheth er there is a relationship between local economic conditions in Florida and REITs that have a portion of their real estate portfolio located in Florida. Some papers emerged in the REIT literature that examine geographic concentration and local eco nomic var iables, which motivate my interest in this topic and my desire to examine an individual market. Feng and Wu (2022) assert the relationship between local economic growth and REIT growth by discovering that REITs with more assets in higher economic growth ar eas, proxied by the lagged GDP growth rates, provide higher stock returns to shareholders. Florida is particularly of interest because it is home to 3 of the top 25 MSAs that REITs typically hold commercial real estate properties in. The three MSA s are Mia mi, Orlando, and Tampa and are characterized by a large population, strong employment growth, and high real estate demand 4 . Florida also has the fourth highest GDP in the nation, with a GDP growth rate exceeding the ! 2 Liow , K.H., &Li, X. (2006). Are REITs unique? A comparative analysis of major asset classes. Journal of Real Estate Finance and Economics, 22(2), 299 318. ! ! Brounen, D., & De Koning, S. (2012). 50 years of real estate investment trusts: An international examination of the rise and performance of REITs. Journal of Real Estate Literature , 20 (2), 197 223. " ! Ling, D. C., Naranjo, A., & Scheick, B. (2019). Asset location, timing ability and the cross ! section of commercial real estate returns. Real Estate Economics , 47 (1), 263 313. !

PAGE 4

! % ! national average from 2010 to 20 19 5 . Moreo ver, Florida is a gateway to international trade and investment due to its primal geographic proximity to Latin America. These attributes collectively make Florida an attractive hub for commercial real estate investment, and a region worth examini ng indepe ndently. Throughout the paper, I wish to understand whether there is a relationship between a REIT's degree of property portfolio exposure to Florida as a geographic region of interest and its annual returns. If a relationship exists, I will proc eed by lo oking at the effect of changes in or shocks to local macroeconomic conditions on annual REIT returns across different levels of geographic concentration in Florida to analyze the marginal effect of local economic conditions on REITs that are geogr aphically concentrated in Florida relative to REITs that are not. My dataset contains REITs with a wide spectrum of concentration levels in Florida, ranging from 0 to 68 percent . Eviden tly, the exposure is time varying for each REIT, and understanding the time vary ing nature of allocation and disposal of commercial properties in Florida will be helpful in the analysis. I want to detect the extent to which real estate exposure in Florida affects REIT returns in the cross section of REITs and th roughout my sample period, which ranges from 199 9 to 2021. I will accomplish this by using a panel regression to regress annual returns against a contemporaneous explanatory variable that represents the annual percentage of Florida exposure for REIT i in y ear t in a categorical format. If the time series and cross sectional variations in annual REIT returns are significantly explained by changes in the degree of exposure to Florida, I will introduce local economic variables such as the growth rates of GDP, employmen t, consumption, and consumer sentiment as controls in the panel regression to check for an asymmetric response of annual REIT returns to local economic factors depending on the whether the level of exposure to Florida is low or high. I later defin e low and high exposure levels depending on the average cross sectional concentration in Florida in each year. ! The remainder of the paper will proceed in the following format. The next section details relevant papers in the literature that review the hist ory of RE ITs and their performance summary before versus after the IPO boom, the risk and return performance of REITs with respect to the broader equity market, the pricing of macroeconomic factors into general stock returns and REIT returns, and the corre lations b etween REIT returns and local or regional economies. Section s 3 and 4 detail the methodology, data, and descriptive statistics. I present my dataset, introduce my local variables and explain why I selected them to proxy for Florida's economic perf ormance, demonstrate how I construct my Florida E xposure explanatory variable by converting the percentage of exposure for each REIT in each year from a continuous to a categorical variable. I also display my summary ! ! ! "#$%!&'()*!+!,-.--/+! !"#$%&'()*#+#,%*(-$.+&/0(/1'1/(2($'+3%+4/(5(67689#$"& +!0)'($*1! 2 ! 34145!67'8'9$7!:(';$)5+!<54($5=5*!>?($)! @A!-.-BA!;('9!C44?%DEEFFF+$#$%F'()*+7'9EG8$45* 2 %4145%E57'8'9$7 2 ?(';$)5%E;)'($*1E! ! !

PAGE 5

! & ! statistics and correlation matrices. In section 5 , I present my main results and analysis. Lastly, in section 6 , I summarize my results and point out further implications in the conclusion. 2. ! Literature Review 2.1 ! History of REITs and performance summary after IPO Boom Commercial Real Estate properties hav e been known for their profitability and ability to generate stable income streams for investors. However, they are illiquid, expensive, and inaccessible to retail investor s . The pillars of investing, such as diversification and portfolio opt imization, are difficult to apply to physical properties and real assets because they require substantial access to capital that perhaps only institutional investors can obtain. Even so, trading of commercial real estate by institutional investors will lik ely sway marke t prices significantly, especially since trades cannot be done frequently, quickly, and in large volumes as in the stock market. The informational inefficiency and heterogeneity of real estate also increase search costs and induce limits to a rbitrage oppor tunities, causing mispricing to persist. The creation of Real Estate Investment Trusts (REITs) by the US C ongress addressed the liquidity problem by designing a structure that enables investors with an ever increasing appetite for innovative financial inst ruments to invest in large and diversified portfolios of commercial properties without directly purchasing those properties; through buying shares of stocks in publicly traded real estate companies . REITs were created as a result of Congress adopting the Real Estate Investment Trust Act in 1960, a period in which one of the biggest US bull markets was witnessed 6 . The legislation allowed retail investors to gain indirect access to income generating real estate without purchasing such properties . REITs are pu blic companies that have at least 75 percent of their assets and income tied to revenue producing real estate properties spanning residential, retail, office, hotel, industrial, and healthcare, among others, and are required to distribute at least 90 perce nt of their taxable rental income as dividends to shareholder s . As a result of persistent and high dividend payout ratios , REITs enjoy a tax exempt status at the cost of foregoing cash reserves that could be utilized to expand organically. REITs were preve nted from managing their own properties and selecting their tenants as they were only allowed to operate as holding compa nies until the Tax Reform Act of 1986 relaxed many REIT related restrictions , thereby expanding their profitability and growth opportun ities . Nevertheless, there were only 58 equity REITs on the market with a combined market capitalization of $5.6 billion ; o wners and developers did not see the advantage in capital ! " ! HGII18A!&+!,-.-@A!05#(G1(J!K/+! -:%/(;'<(%+(,'$3.1(:%/1#$<=(>?@A/(BC" "(,'$3.1(.+&/ +!L1C''M!<54($5=5*!>?($)!-A!-.-BA!;('9! C44?%DEEFFF+J1C''+7'9E=$*5'E*1J 2 91(N54 2 C$%4'(J 2 @KO.% 2 #G)) 2 @P@-..OK@+C49)! ! !

PAGE 6

! ' ! market access at the cost of abiding by operating restrictions because they had sufficient access to debt at low rates and hardly any taxable income. REITs witnessed an ÔIPO Boom' between 1993 19 94 in which 95 private U.S. real estate companies went public, significantly expanding the size of the REIT industry and allowing capit al to flo w from mutual funds, insurance companies, and pension funds. During a period when overbuilding was prominent, pr operties were highly leveraged , refinancing no longer easy to obtain , and companies ambivalent to liquidate their assets at a loss , pri vate real estate companies resorted to the public equity market. Many followed the footsteps of major companies such as K IMCO ($128M IPO) in 1991 and Taubman ($330M IPO) in 1992, as REIT valuations were optimistically high despite falling real estate price s 7 . By accessing equity markets, REITs were able to reduce their mounting debt, purchase more properties, and execute rapi d growth strategies with improved interest coverage ratios. Between 1992 and 1996, the number of equity REITs grew from 82 to 166 with the market capitalization growing from $12.9 billion to $78.3 billion. In 1992, the 20 largest REITs had a collective mar ket capitalization of $3.6 billion, only two of which being valued at more $500 million. Five years into the IPO boom, more than 23 REI Ts accumulated a market capitalization in excess of $1 billion each 8 . Eventually, more private companies were able to exec ute successful debt to equity conversions, trading volume increased, and conflicts of interests were resolved with the transformation o f such companies from Ômutual fund like' to active and self managed . As of 2022, 167 REITs trade in the NYSE and have a combined value exceeding $1 trillion. 9 2.2 ! REIT risk and return performance in relation to stock market The rise of REITs generated endurin g interest among academics, as the current literature addresses many questions concerning their structure, r eturn predictability , and sensitivity to market conditions. There is much to be researched about REITs, understandably so seeing that they span a gr ey area with pure real estate on one end of the spectrum and pure equity common stock on the other end. Vari ous papers in the finance literature suggest that REITs have comparable return predictability to stock portfolios and are integrated with the genera l stock market. Seck (1996) found that REIT returns are driven more by equity market effects than direct com mercial property. Direct commercial property valuation is appraisal based and infrequent valuations prevent the timely absorption of market informa tion into property prices. Liow and Li (2006) demonstrate that REITs behave similarly to small capitalizatio n stocks and share similar characteristics . Ling and Naranjo (1999) found that the market for real estate securities is integrated with the market f or ! # ! Brounen, D., & De Koning, S. (2012). 50 years of real estate investment trusts: An international examination of the rise and performance of REITs. Journal of Real Estate Literature , 20 (2), 197 223. 8 Ambrose, B. W., & Linneman, P. D. (1998). Old REITs and New REITs . Real Estate Center, Wharton School of the University of Pennsylvania. 9 REIT Industry Financial Snapshot . Nareit. (n .d.). Retrieved April 3, 2023, from https://www.reit.com/data research/reit market data/reit industry financial snapshot

PAGE 7

! ( ! non real estate securities, with the degree of integration increasing significantly during the IPO boom. Glascock et al. (2000) contend that REITs post 1992 are now more cointegrated with stocks, although they previously behaved similarly to bonds. Li ( 2016), by covarying expected REIT returns with Fama French factors, shows that time varying REIT returns price risks related with the stock market premium and small stock premium, and Gyourko and Keim (1992) find that market returns are significant in expl aining real estate stock returns , and in a second paper show that REIT stock returns readily abso rb information about market fundamentals in a similar fashion to general stocks ; such that the performance of an appraisal based portfolio can be implied from the lagged performance of real estate stocks. 2.3 ! Equity returns price macroeconomic factors As r esearch and interest in equity markets surged, it became an indispensable notion that stock returns are sensitive to macroeconomic factors and local economic c onditions. In their seminal paper, Chen , Roll, and Ross (1986) model equity returns as a function of macro variables and demonstrate that they depend on their exposures to economic state variables such as industrial production, changes in the risk premium and yield curve, and inflation . Smajlbegovic (2019) adds to the literature by showing that stock returns and profitability positively correlate with predicted economic conditions. Not only that, but there exists a strong relationship between equity prices and local macroeconomic variables; as Pirins ky and Wang (2006) prove that companies' stock returns are affected by the local macroeconomic variab les, specifically GDP and the unemployment rate, of their headquarters' regions. Korniotis and Kumar (2012) est ablished that local stock returns fluctuate with respect to local business cycles , particularly recessionary periods , which stimulate shifts in local risk aversion and produce predictable return patterns. When examining state constructed portfolios, they e arned higher future returns when state level unemployment rates were high and loan to value ratios were low. Since some similarities can be found between REITs and non REIT stocks, and stock markets are exposed to economic forces, one may naturally hypot hesize that REITs are also exposed to such e conomic forces. In Chan et al (1990), unexpected inflation and changes in the risk structure and term spread of interest rates impact equity REIT returns 60% as much as their impact on common stock returns , and F ei et al. (2010) adds that the unemployment rate and inflation rate affect returns. Thus, although equity REITs are less risky than corporate stocks, they do not hedge against systematic conditions. Lin g and Naranjo (1997) support the discernable co moveme nt between macroeconomic events and real est ate markets by identifying the growth rate in real per capita consumption as an additional fundamental state variable that bears a risk premium which is consistently priced ex ante i n real estate returns across R EIT and appraisal based portfolio s . Prior to their work, risk adjusted performances of REITs endured an omitted variables problem, as multifactor models that overlooked the role of consumption as a state variable were possibly biased in their predictabilit y of stock returns. The conclusion reaffirms Geltner's (1989) finding about the

PAGE 8

! ) ! systematic risk of real estate indices being significantly positively correlated with national consumption. Thus, I include the growth in per capit a consumption expenditures as one of my explanatory variables. In more general terms, real estate portfolio betas display procyclical behavior with respect to recessionary periods, according to Glascock (1991), although the effect is temporary and period s pecific. Lastly , by estimatin g a two factor regression model of a sample of REIT returns after 1992 with changes in interest rates and the stock market as covariates, Allen, Madura, and Springer (2000) demonstrate a statistically significant result that su pports the sensitivity of REI T returns. Their analysis also suggests that REITs may not completely isolate their stock performance from economic and broader market forces, although they can control their degree of exposure to them. 2.4 ! Equity returns price ma croeconomic factors Many pa pers proved that commercial real estate is sensitive to local economic conditions. Given that such properties constitute the underlying assets of REITs, these findings offer valuable implications that we can extend later to test on REIT returns . Pla zzi, To rous, and Valkanov (2018) highlight how the expected returns of commercial properties , which rely on expected rent growth and rent price ratios , are not only time varying depending on the state of the national economy, but are subject to idiosyncratic fluc tuations, such as location differences in the cross section of properties . It is important to capture the between property aspect because individual effects stemming from differences in geographic, demographic, and urban factors convey re gion specific econ omic fluctuations and heterogeneous degrees of propagation of national economic shocks across regions. Additionally, Cotter, Gabriel, and Roll (2014) identify MSA allocations as a significant factor in commercial real estate performance , as between MSA per formance differs according to the MSA's exposure to macroeconomic shocks. Feng (2021), relying on the conjecture that changes in local economic conditions should impact the income return and capital appreciation of commercial real estate in such local econ omies, uses GDP level and GDP growth of different geographic locations to identify a relationship between the local economy and CRE performance. Using a Fama Macbeth regression, he finds a positive correlation between CRE returns and GDP level and g rowth. Specifically, the size of the economy, proxied by GDP level, significantly affects CRE income and capital return, and the growth of the economy, proxied by GDP growth rate, significantly affects the capital return. The paper supports the migration of CRE i nvestment from areas with low GDP level but high GDP growth to areas with high GDP level but low GDP growth, reaffirming the conclusion by Ling , Naranjo, and Scheik (2018) that commercial property portfolios have recently been moving to gateway markets, w hich are regions that enjoy good economic health. It also backs their uncovered evidence for the ability of REIT geographic exposures to explain the cross section of REIT returns, thus influencing portfolio allocations across time toward and away from geog raphic markets depending on their market conditions and performance.

PAGE 9

! * ! 2.5 ! Correl ation between REIT returns and the local economy Given the discussed literature above, the natural next step is to hypothesize that , theoretically, REIT returns must also be res ponsive to the local economic conditions that impact their underlying commercial real estate propert ies . In fact, this sensitivity should be more easily detectable with REIT stocks, which behave like general stoc ks , than with commercial proper ties due to the fact that commercial properties' returns and market values are appraisal base d 10 . Appraisals are not done instantaneously and frequently, whereas REIT stock prices appreciate and depreciate continua lly , so REIT returns are more likely to refl ect the general market factors governing the underlying properties than the properties themselves. Prevailing arguments exist for the presence of a Ôlocal beta', which means that the market risks associated with local real estate markets are non diversifia ble and priced into REIT equity returns. Many academics agree with this notion, among them are Zhu and Lizieri (2022). They construct an explanatory variable known as local beta by finding the weighted average s um of the betas of each local property marke t (§ m ) for each firm's property portfolio, with the weights representing the proportion of properties of firm i in MSA m to their total number of properties. § m , derived for each MSA, reflects the loading of syst ematic market factors on local commercial re al estate. Their results show that as RE I Ts' exposure to the most volatile property markets (high local beta) increases, their returns increase , as highly concentrated REITs' returns increase by 4.7% per a one st andard deviation in the local beta measure. Feng and Wu (2021) suggest that local GDP growth affects REIT firm growth through the growth of equity. For each REIT, a value weighted aggregated measure of local GDP growth is constructed in which the GDP growth rate for each MSA is scaled by the net boo k values of the properties located in that MSA. REIT firm growth at year is regressed against lagged firm level GDP growth . The results indicate that REITs with more assets concentrated in high econ omic growth areas experience faster growth in the ir book value and market value of assets. Lastly, Hartzell et al. (2014), by using Herfindahl indices to measure geographic concentration, identify that REITs that employ a geographic diversification strateg y are valued lower than REITs with a stronger geo graphical focus. Their findings relate to my scope, as I will also attempt to observe whether a tighter geographical focus on Florida is a driver of annual returns. 3. ! Methodology In Chen et. Al's influential paper, the authors found that changes in industr ial production index, the risk free rate, the inflation rate, and the term structure were statistically significant in predicting stock prices. Based on this framework, many researchers have adopted these ec onomic indicators in subsequent studies as a basi s to measure the sensitivity of REIT returns to macroeconomic conditions and shocks. For example, Chan, Hendershott, and Sanders use the factors presented by ! 10 Gyourko, J., & Keim, D. B. (1993). Risk and return in real estate: evidence from a real estate stock index. Financial Analysts Journa l , 4 9 (5), 39 46 . !

PAGE 10

! "+ ! Chen et. A l to understand how various ma croeconomic factors affect real estate returns. T he constraint with these widely used factors is that they are broad national indicators, with few regional economies having state specific indicators that mimic them. Since my aim is to analyze the effect of local economic conditions in the state of Florida on the annual returns of REITs that own or operate properties in Florida, my explanatory variables should be as localized as possible. FRED and other databases do not prov ide a measure of industrial production that is sp ecific to Florida. Other possible proxies such as the manufacturing production index and capacity utilization rate were also not available. As a proxy for the inflation rate, the consumer price index for Flo rida urban consumers was found, but is only speci fic to Miami, which is one out of three MSAs of interest. Lastly, data on the term structure of municipal bonds in Florida may be used, but municipal bonds do not accurately represent economic conditions in a state. Economic Variables I collect all my eco nomic variables, excluding the consumer sentiment index, from the St. Louis Fed database (FRED). A. ! GDP growth I examine a time series of the percentage change from the prior year in the all industry total gross domestic product from 1998 Ð 2021 . The data i s annual and not seasonally adjusted . FRED constructs the percentage change year over year as follows: !" # ! $ % & ' "# $ ! "# $ ! " #$%&'()*(+) ( ) $ * + ),, Where the subscript -./01 23425 3 denotes the number of observations per year, which differs by frequency. 11 In our case, since the data 's frequency is annual, it is set to 1. ! " # ! is the most recent estimate of GDP for the given year. Although the estimate for ! " # ! is not exactly r eleased at the end of year t to match annual REIT returns conveying the change in the stock price between the end of years t and t Ð 1, it still reflects total economic activity (as measured by the total production of goods and services) in Florida during year t, so it is concurrent with fluctuations in the stock price. Since GDP is a coincident indicator, it reflects current information rather than information that has already been priced, which affects investors' sentiment in tandem and subsequently stock prices throughout the year B. ! Employment growth ! $$ ! 9:'1(D#$,C"'/('$.(C/.&(1#(*'"*C"'1.(4$#E1:($'1./F ! Q544$8I!R'!S8'F!0<6H+!,8+*+/+!<54($5=5*!>?($)!BA!-.-BA!;('9! C44?%DEE;(5*C5)?+%4)'G$%;5*+'(IE;(5*E*141EG8*5(%418*$8I 2 4C5 2 *141E;'(9G)1% 2 71)7G)145 2 I('F4C 2 (145%E! ! !

PAGE 11

! "" ! A time series of the seasonally adjusted percentage change from the prior year in the total nonfarm employment in Florida from 1998 Ð 2021 is collected. The percentage change in employment ye ar over year is ca lculated similarly to the GDP growth variable. Employment reports are released more frequently than GDP reports as the state of Florida releases monthly employment reports compared to annual GDP reports. FRED aggregates higher frequency d ata series, such a s the monthly time series of total nonfarm employment to a lower frequency annual time series by taking the average of the twelve monthly employment values 12 . Employment growth is an important barometer of the health of the economy because it is closely corr elated with consumer spending and economic growth. Employment growth is used in the asset pricing literature, specifically when looking at macroeconomic risk factors and asset returns, as changes in employment patterns affect market volat ility, monetary po licy, and corporate earnings, which consequently affect stock returns. REITs can use employment growth to identify markets and MSAs with strong job growth, which helps to assess future demand for office, retail, and industrial spaces . C. ! Consumption growth A time series of annual personal consumption expenditures for Florida was used to obtain the percentage change from the prior year from 1998 Ð 2021. The collected data measures the total spending on goods and service s purchased by households residing in Florida. FRED uses the similar aggregation mechanism as in finding GDP and employment growth to calculate consumption growth and convert the data from quarterly to annual. Like GDP, personal consumpti on is a coincident economic indicator. I employ consumption as a variable following Geltner's (1989) finding that appraisal based real estate returns are sensitive to changes in national consumption, and Ling and Naranjo's (1997) conclusion that excluding the change in per capita consumption as a source of systemic risk in multifactor models that price the sensitivity of real estate returns elicits an omitted variables problem. D. ! CSI growth The University of Florida's Bureau of Economic Research surveys 563 individuals from F lorida, to represent an unbiased demographic cross section of Florida, on a monthly basis and asks each respondent a set of questions related to their current personal financial situation, their expected financial situation one year from now, the expected national economic outlook over the next year and the next five years, and whether it is a good time to buy a major household item 13 . UF' CSI ! $% ! 9:'1(%/(D$.GC.+*<('44$.4'1%#+F ! Q544$8I!R'!S8'F!0<6H+!,8+*+/+!<54($5=5*!>?($)!BA!-.-BA!;('9! C44?%DEE;(5*C5)?+%4)'G$%;5*+'(IE;(5*E*141EG8*5(%418*$8I 2 4C5 2 *141EFC14 2 $% 2 ;(5TG587J 2 1II(5I14$'8EUDVD45W4X0(5TG587JY-.1II(5I14$'8Y-.7'8=5(4%Y-.C$IC5(Y-H;(5TG587JA4C5Y.)'F5%4Y-.*141Y-.$%Y-. 188G1)+ ! $& Z+6+Z+<+! [ ! ZG(51G!';!67'8'9$7!18*!ZG%$85%%!<5%51(7C+!,8+*+/+!<54($5=5*!>?($)!BA!-.-BA!;('9!C44?%DEE#5#(+G;)+5*GE;)'($*1 2 7'8%G95( 2 %584$9584E! ! !

PAGE 12

! "# ! calculations follow the University of Michigan's consumer sentiment index, with the same base year used, which is 196 6. I aggregate the end of month index figures to an annual basis by taking a 12 month average. I find the year over year percentage change in consumer price index as follows: 67 8 ! $ % & 9 %& ' ! %& ' ! " , ( ) $ : + ),, REIT Data I obtain REIT stock data from the Bergstrom Real Estate Center at the University of Florida for 171 equity REITs. For each REIT, the dataset contains a description of the property sector it opera tes in, its annual returns as of the en d of each year, and its total concentration in Florida at the end of each year. Annual returns in each year are measured by finding the percentage change in the REIT's closing stock price at the end of the year from t he closing price at the beginning of th e year. The REIT dataset contains firms that IPO after 1998, firms that halt operations before 2021, and firms that have never owned or operated properties in Florida from their year of inception until either their ye ar of dissolution or the end of the sam ple period in 2021 (whichever comes earlier). Since my aim is to observe annual returns between 1998 and 2021, I only include REITs that are in operation as of the beginning of the sample period but keep any firms tha t halt operations or have no available data in any year during the sample period to avoid survivorship bias. I exclude REITs that have no exposure to Florida for the entirety of the sample period to avoid negatively skewing my Florida concentration measure , as my aim is to study REITs that have invested in Florida properties whether as of the beginning or eventually during the sample period Florida Concentration Total Florida concentration predictor variable is calculated for each REIT in each year as: 6 ; < (! & = >?.3@A5 B #3.1 2 3C40DE C () * + ) , = = 7C5C2 B #3.123C40DE C (). * + ) , / . , Where >?.3@A5 B #3.123C40DE C () * is the value in square footage of property j located in Florida and the sum of square feet is found for all commercial real estate properties going from j = 1 to N for REIT i. 7C5C2 B #3.123C40DE C (). * is the value in square footage of property j located in one state and summed over all properties go ing from j = 1 to N for REIT i in that state. The result is summed o ver all states going from k = 1 to M for REIT i. The denominator assumes that each REIT's portfolio is geographically diversified across several MSAs or at least in one other geographic l ocation outside of Florida. The average cross sectional concentrati on is calculated as: >F B 6; < ! & = 6; < (! + ( , <

PAGE 13

! "$ ! Where the cross section in each year contains all the REITs in the sample. Annual Florida concentration is a continuous variable. It does reveal whether a REIT is geographically concentr ated in Florida due to the cross sectional and time series dispersion in annual concentration. To convert 6; < (! * to a more relative measure, I compare it to >F B 6; < ! each year. if 6 ; < (! $ is greater than >F B 6; < ! , REIT i will be classified as high concentration and low concentration otherwise, since it is above the average cross sectional concentration, but only in the year at which >F B 6; < ! is evaluated. Since >F B 6; < ! is time varying, 6; < (! is com pared to it each year to account for the fact that REITs vary their allocation decisions in Florida over time depending on market demand, economic conditions, the regulatory environment, supply and demand dynamics, and the competitive landscape. After cl assif ying each REIT into high or low concentration. I create the following indicator variable: G8!GF; H (! & I $ ) $ @E $ 6 ; < (! * J >F B 6; < ! , $ .C K 23L@02 $ $ Since our data includes both cross sectional and time series components, I will utiliz e panel regressions to first identify a relationship between the level of geograp hic exposure to Florida and annual REIT returns. Since the REITs in my sample will display individual level variation that may not have been captured by the data, I will utili ze a fixed effects panel regression. Fixed effects is useful to control for unobs ervable characteristics associated with each individual firm that may correlate with the independent variables such that the covariance between the independent variable and the error term is not zero. However, since the economic data is time varying and identical in the cross section of REITs, I will also utilize a random effects panel regression. To assess the suitability of each model, I will rely on the Lagran ge Multiplier test, which tests for the presence of individual varying effects. I f the null hypothesis can be rejected, it will be necessary to control for cross sectional heterogeneity by using an individual fixed effects regression. If the null hypothesi s fails to be rejected, I will use a time fixed effects regression. 4. ! Data and De scriptive Statistics I begin my analysis with the data set obtained from the Bergstrom Real Estate Center. The dataset compiles the REITs' names, main operating sectors, annual stock returns, and ann ual Florida concentration measured by square foota ge from 1998 Ð 2021 . Although some papers in the literature use book or market value to measure concentration, using square footage is a more relevant proxy for the size of the REITs' portfolio allocation toward a speci fic location. Firstly, book values could distort exposure due to depreciation exp ense. Second, market values are appraisal based and appraisals are not updated contemporaneously with evolving information about market fundamentals. The market value availabl e in year t may not be as relevant to that year as it was to previous years. Henc e, under or overpricing may arise in that year, leading to

PAGE 14

! "% ! under or over estimation of Florida exposure. Changes in square footage can reflect the acquisition and disposal of properties by portfolio managers depending on current market conditions. Addition ally, square footage provides a more homogenous measure of exposure in the cross section of REITs than book and market values, enabling the comparability of exposure across RE ITs. This is especially important when we take into account the fact that the sam ple REITs operate in different sectors that may not apply uniform valuation mechanisms. I exclude REITs that IPO after 1998 to avoid lookahead bias and REITs that have never invested in Florida from 1998 Ð 2021. I keep REITs that invest in Florida in any given year but cease operations during the sample period and have missing data to avoid survivorship bias. I also keep REITs that have zero exposure in Florida as of 1998 but increase or decrease exposure subsequently, in addition to REITs that have some e xposure in 1998 but cut exposure to zero subsequently in the sample period. Consequently, I have 43 REITs which I analyze over 24 years, yielding 1032 observations. Two firms Ð Post Properties Inc and Equity One Inc contain missing data. Upon first examini ng the data, I observe cross sectional and time series variation in REIT level exposures (by % square feet) to the Florida market. As exposure varies across time within each R EIT , the average FL exposure varies in the cross section of all 43 REITs in each year. I create three Florida exposure groups to allocate the sample REITs; high (above average), low (below average), and 75 th quartile. For each year, I compute an equally w eighted cross sectional average REIT concentration and the 75 th quartile concentr ation . I then compare each REIT's annual concentration to those values; where if it is above the average concentration in that year, I allocate it to the high exposure group, if it is above the 75 th quartile concentration, I allocate it to the 75 th quartil e group, and if it is below the average concentration, I allocate it to the low exposure group. I repeat the process for the entire sample period and constantly rebalan ce the portfolios' components by regrouping the REITs at the end of every year in the sa mple period according to the average cross sectional Fl exposure in that year. For any given exposure group, the combination of REITs in year t may not be the same as t he combination in year t+1, because we construct a time varying measure of geographic co ncentration in Florida. I compute the time varying measure of average Fl concentration as stated in the methodology. The concentration is calculated at the end of each year t. If REIT i's concentration (%) exceeds 6. ! , it is added to the high exposure portfolio. N is fixed so that if a REIT's concentration is zero at any given year, this observation is still factored into the average cross sectional exposure of that year. Ignoring zero exposure observation will overstat e the cross sectional average. The reallocation of REITs into and ou t of exposure portfolios is necessary to account for the time variation and individual variation in Fl exposure. A REIT in 1998 would be considered above average if its exposure to Florid a exceeded the average REIT concentration in that year of 9.6%. Later on, its exposure in a given year may be below the average REIT concentration in that year, so it should not remain in the above average exposure group in that year. Analyzing the raw dat a shows that although a short term persistence exists in exposure, it dissipates in the long term, in which a REIT that begins the sample period with a certain degree of Fl exposure does not necessarily maintain it for the enti re time horizon .

PAGE 15

! "& ! The methodology above is consistent with what Ling, Naranjo, and Scheick observe when analyzing the cross sectional and time series variation of MSA exposures of REITs over time and REIT returns. They find that MSA exposures that are significant in explaining REIT returns in a given year differ substantially over their sample period. In a given year, MSAs that do not have significant explanatory power do have significant coefficients, whether positive or negative, in another year. For this reason, the authors construct time varying measures of geographic concentrations across their MSAs of interest that adjust at the beginning of each year for the duration of their sample period. For each REIT, they allocate each of its properties into its corresponding MSA and calculate the % of the REIT's geographic exposure to a given MSA by summing the market value of its properties in that MSA and dividing the result by total sum of properties across all MSAs, repeating the proce ss each year. Similarly, Fl exposure values that do not have significant explanatory power do have significant ex planatory power in another period, depending on whether they are above or below the time varying average exposure. For this reason, I repeat th e classification each year. Figure 1 shows that any REIT with an annual concentration exceeding 10% is deemed abo ve average before 2000, but 10% will be slightly below average in 2009. By constructing exposure groups to categorize the REITs in the sample , I create equally weighted REIT portfolios that follow an investment criterion related to the extent of Florida exposure. The portfolio is monitored at the end of each year to ensure it meets the investment criterion. If a REIT no longer satisfies the cri terion, adjustments must be made. This logic is similar to selling stocks that do not meet investment criteria an d buying stocks that do. The rebalancing at the end of each year does not necessitate that the number of REITs in each portfolio be fixed in ea ch year. Thus, the number of REITs in the high exposure portfolio in year t is not necessarily equal to the numbe r in the subsequent year. Figure 1: Average Fl concentration in the cross section of REITs in the sample from 1998 2021 Figure 1 shows the a verage annual Fl exposure in the REIT cross section over time. The average is equally weighted. On average, in any given year, a REIT will invest 9.361% of its property portfolio in Florida. The minimum average exposure in the REIT c ros s section is 7.564% and the maximum is 11.448%. In 2018, the average REIT invested 7.564% of its assets in Florida properties. In

PAGE 16

! "' ! 2008, the average REIT invested 11.448% of its assets in Florida properties. The highest figures recorded for average Fl expo s ure across all REITs occurred between 2007 and 2009. Average Fl exposure declined by 1.5% after 2009 and never returned to the levels recorded between 2007 and 2009 as of 2021. Between 2003 and 2009, National Health Investors had the highest exposure in bo th the cross section and throughout the sample period (67.67%). Since the financial crisis, no REIT in the sample had comparable Fl exposure to National Health Investors; maximum Fl Exposure in the cross section ranged from 43.91% to 48.03% after 2010. Figure 2: Average an nual Fl concentration by exposure group from 1998 Ð 2021 We can observe from figure 2 that REITs belonging in the high and 75 th percentile categories exhibit more volatile behavior than those in the low category in terms of the proportion of their rea l estate asset portf olio concentrated in Florida, especially in the years preceding the financial crisis and after 2009. Between 2002 and 2007, a persistent increase in the cross sectional average exposure is noted for the high and 75 th percentile categori es. Despite the vary ing composition of REITs across the years in each portfolio, the average in the cross section consistently increases and is not skewed by any removals or additions of REITs fr om the portfolios. This shows that the exposure pattern or be havior throughout th e years is not an individual specific phenomenon, rather it is observed regardless of which REITs are in which exposure category. For comparability purposes, I also create a zero exposure group to classify any REITs that had no propert ies in Florida in a given year. Many REITs in the sample invest in and divest from Florida through time, but some divest completely and reacquire some assets subsequently. In figures 3 to 6, I compute the average annual returns in the cross section of REI Ts in each exposure group from 1998 Ð 2021 and plot them in a time series.

PAGE 17

! "( ! Figure 3: Comparing the average annual returns between the zero exposure REIT subgroup and the high (above average) exposure REIT subgroup For zero exposure and high exposure, w e observe similar vo latility patterns across time. In the 2000 2002 bear market, the annual returns of a portfolio simulating the high exposure group is higher, and the decline in annual returns during the financial c risis is less steep. After the Covid cr isis, average annual returns for the high exposure groups rise more sharply. Figure 4: Comparing the average annual returns between the low (below average) exposure REIT subgroup and the 75 th percentile exposure REIT subgroup

PAGE 18

! ") ! Figure 5: Comparing the average annual return s between the low (below average) exposure REIT subgroup and the high (above average) exposure REIT subgroup Similar patterns to Figures 3 and 4 are observed in the 2000 2002 bear market, financial crisis, and Covid cri sis. There is a more persistent out performance in average annual returns from 2016 Ð 2020 for the high exposure group when compared to the low exposure group versus when compared to a portfolio of REITs that do not invest in Florida at all. Figure 6: Comparing the average annual returns between the low (below average) exposure REIT subgroup and the zero exposure REIT subgroup A low exposure portfolio is more severely impacted by the financial crisis than a zero exposure portfolio and underp erforms from 2016 2020. We can imply from this figure that exposure to Florida in broader terms may not drive outperformance, rather a high degree of exposure is what drives outperformance relative to portfolios that are not as geographically concentrated in Florida. The following table presents the summary statistics for the high, low, and 75 th quantile exposure portfolio groups. Throughout the sample period, the portfolio groups' REIT composition does not

PAGE 19

! "* ! remain constant. Each REIT either remains in its preceding year's portfolio or is regrouped dep ending on the current year's average Florida concentration in the REITs' cross section. Regrouping accounts for the fact that the exposure to Florida within each REIT is time varying, which distorts average exp osure every year. The low Fl exposure group ha s the lowest standard deviation, suggesting that there is low volatility in the average annual concentration throughout the sample period. The smooth line in figure 2 displays no evidence of volatility: an aver age annual Fl exposure of 3% is considered low (below average) in 2001 and in 2013. The high Fl exposure group displays a higher standard deviation with slightly higher volatility in average annual Fl exposure. The 75 th percentile group contains that highe st volatility, with average annual concentrati on ranging from 21% to 65%. The low group has the lowest mean annual returns compared to the high and 75 th percentile group. The mean annual returns for those two groups are almost similar, but the latter group 's returns are more volatile, offering higher return but greater risk. Table 1: Summary Statistics for Sample REITs Grouped by Florida Exposure: 1998 Ð 2021 Group 1: High (Above Average) FL Exposure Concentration Ann ual Returns Minimum 0.1761 Minimum 0.1670 Maximum 0.2730 Maximum 0.5697 Mean 0.2079 Mean 0.1469 Med ian 0.1933 Median 0.1084 Standard Deviation 0.0307 Standard Deviation 0.1970 Group 2: Low (Below Average) FL Exposure Concentra tion Annual Returns Minimum 0.0304 Minimum 0.4182 Maximum 0.0569 Maximum 0.4687 Mean 0 .0402 Mean 0.1190 Media n 0.0378 Median 0.1964 Standard Deviation 0.0077 Standard Deviation 0.2331 Group 3: 75 t h Quartile FL Exposure Concentration Annual Returns Minimum 0.2132 Minimum 0.1793 Maximum 0.6535 Maximum 0.6535 Mean 0.1514 Mean 0.1514 Median 0.1097 Median 0.1097 Standard Deviation 0.2093 Standard Deviation 0.2093 Ta ble 1: After categorizing the REITs into high, low, and 75 th percentile for the duration of the sample period, I find the cross sectional average annual Fl concentration and return in each year t. Then, I compute the average and standard deviations for the se values in addition to the minimum, maximum, and median. Besides exposure groups, I break down the sample REITs by property sector and form property type portfolios to classify each REIT. I compute the summary statistics for annual Florida concentratio n and annual returns in each sector in table 2. We observe t hat diversified REITs have

PAGE 20

! #+ ! the highest mean annual returns and the highest standard deviation in both annual returns and annual exposure, while office REITs have the lowest standard deviation in a nnual returns. The high variability in returns for diversifi ed REITs confirms Gyourko and Nelling's (1996) conclusion that REIT diversification across property types may be a na•ve diversification strategy. The trend for office REITs may have changed after the Covid 19 pandemic as office property investments look l ess lucrative due to the changed work environment. These impacts were only captured in the last year of the sample period, so they may not be noticeable enough to skew the mean. Hotel REITs exhibi t the lowest mean in annual returns and annual exposure acro ss all groups, and specialized REITs have the highest mean Fl exposure. Industrial and residential sector types have comparable average annual returns and volatility. Table 2: Summary Statistics for Sample REITs grouped by sector (1998 Ð 2021): After grouping the firms into their respective property sector groups, I subdivide each sector group into a high and low exposure subgroup relying on my previous methodology of computing the average annua l concentration in the REITs' cross section at each year. For each sector group, I compare each REIT in the cross section to its corresponding year's average concentration and categorize it into an above or below average subgroup within its property sector , accordingly, repeating the same steps for all 24 y ears. Although each sector group contains the same combination of REITS across time (since each REIT's property type focus is unchanged), the exposure subgroups contain a different combination of REITs ev ery year due to their annual reclassification betwee n the high and low exposure subgroups. The results show that for all property sectors, with the exception of hotel and diversified, we note a marginal effect of the degree of geographic concentration in F lorida on annual REIT returns. I demonstrate from th ese results that the return pattern is not a phenomenon specific to a single property sector as the effects are observed between sectors .

PAGE 21

! "# ! I sub divided each sector group into high and low exposure sub g roups to classify REIT i in year t, repeating the pr ocess for all i = 1, 2, É, 43 and t = 1, 2, É, 24, taking the average across time for each sub group within each sector afterwards. For REITs falling under the hotel sector group, all the REITs had below average exposure . I set the summary statistics for the high exposure sub group within the hotel sector to 0.000 to account for the absence of data. The summary statistics show that returns are higher in high exposure groups than low exposur e groups, with the exception of diversified REITs. The most notable differences in annual returns between high and low exposu re groups are in the industrial, office, and specialized REIT property sectors.

PAGE 22

! "" ! In addition to annual Fl concentrat ion, I consider a set of macroeconomic variables as my explanatory variables: All industry total Florida Gross State Product growth rate, the total nonfarm employment growth rate, the personal consumption expenditure growth rate in Florida, and the Florida consumer sentiment index growth rate. I collected data from 1998 Ð 2021 to create lagged explanatory variables that can estimate the effects of annual REIT returns in the subsequent year. This is based on Feng and Z hu's finding that REIT firm growth is po sitively correlated with the lagged firm scaled measure of economic growth, which follows from Ling, Naranjo, and Scheick's methodology of lagging the economic variables in the panel regression. However, for regressi on models testing the effects of contemp oraneous variables, I use the data from 1999 Ð 2021. Table 4 presents my descriptive statistics. Table 4 : Summary Statistics for explanatory variables: 199 8 Ð 2021 GDP: All Industry Total in Florida (% Change from Year Ago) Min 3.6577 Max 12.4 320 Mean 4.840 St. Deviation 3.652 Total Nonfarm Employment (% Change from Year Ago) Min 6.2736 Max 4.5955 M ean 1.3489 St. Deviation 2.8411 Personal Consumption Expenditure (% Change from Year Ago) Min 2.8636 Max 15.6238 Mean 5.073 St. Deviation 3.5680 Florida Consumer Sentiment Index (% Change from Year Ago) Min 16.696 Max 56.969 Mean 15.76 St. Deviation 7.7088 Annual data for GDP, Employment, and Consumption are taken from the FRED database . 1 Monthly data for Florida CSI are taken from UF's Bureau of Business and Economic Research and are averaged to obtain annual data. Growth rates are determined by finding the percentage change from the prior year. GDP, Employm ent, Consumption, and CSI gro wth are lagged by 1 year in the panel regression, so REIT returns between 1999 Ð 2021 are regressed against the economic variables from 1998 Ð 2020. Since I am also looking at contemporaneous relationships, I collect data on th e economic variables for 2021 .

PAGE 23

! "# ! Table 5 displays the correlation matrix for the state variables. I initially compute the correlations between the explanatory variables while excluding the y variable of annual returns. For comparative purposes, I org anize the sample REITs into two portfolio grou ps: high (above average) and zero exposure. This helps me detect whether REITs are responsive at all to any regional shocks to Florida's economy even if they are fully divested from properties in that region. I compute the correlation matrices between the economic variables and the average annual returns in each of the portfolios. To broadly observe whether contemporaneous and lagged variables interact differently with REIT returns, I find correlations between l agged variables and returns and contemporaneou s variables and returns respectively for each exposure portfolio. Table 5 : Correlation Matrix for Economic Variables 1998 Ð 2021 GDP % Emp % Cons % Florida CSI GDP % E mp % 0.8507 Cons % 0.8819 0.8122 Florida CSI % 0.1831 0.3875 0.0486 GDP % is the percent change from the end of year t Ð 1 to year t in the all industry total gross state product in Florida observed from 1998 to 2021. Emp % is the percent change from the end of year t Ð 1 to year t in the total nonfarm employment in Florida ob served from 1998 to 2021. Florida CSI % is the annual percent change in the consumer sentiment index which surveys Florida residents. Monthly collected data is averaged over twelve months to yield annual observations. The annual data is taken from 1998 to 2021. Cons % is the percent change from the e nd of year t Ð 1 to year t in personal consumption expenditure recorded from 1998 to 2021. Table 6 : Correlation Matrix for Economic Variables (no lags) and Average Annual Returns of the High Exposure REIT Gro up 1 between 1999 Ð 2021 Return GDP % Emp % Cons % Florida CSI Return GDP % 0.4011 Emp % 0.2860 0.8486 Cons % 0.4468 0.8807 0.8105 Florida CSI 0.2149 0.1710 0.3760 0.0367 Correlation Matrix for Economic Variables (no lags) and Average Annual Returns of the Zero Exposure REIT Group betwe en 1999 Ð 2021 Return GDP % 0.3313 Emp % 0.2534 0.8486 Cons % 0.2121 0.8807 0.8105 Florida CSI 0.3712 0.1710 0.3760 0.0367 1 In the upper half of the table, the variables GDP %, Emp %, Consumption %, and Florida CSI % are contemporaneous. Average annual portfolio retur ns in year t are predicted from state variabl es in year t. When there are no lags, all the economic variables are positively correlated with the average annual returns of the high exposure group and the low exposure group. Correlations between the predicto r variables and annual returns decline in a z ero exposure portfolio, except for CSI growth. CSI growth and per capita consumption expenditures are very weakly correlated. Predicted consumption patterns for the next 12 months do not reveal valuable informat ion about current consumption patterns.

PAGE 24

! "$ ! Table 7 : Correlation Matrix for lagged Economic Variables* and Average Annual Returns of the High Exposure REIT Group 2 Return GDP % Emp % Cons % Florida CSI Return GDP % 0.2114 Emp % 0.3800 0.8557 Cons % 0.3137 0.8669 0.8809 Florida CSI 0.3054 0.3123 0.4592 0.2364 Correlation Matrix for lagged Economic Variables* and Average Annual Returns of the Zero Exposure REIT Group Return GDP % 0.2222 Emp % 0.039 9 0.8557 Cons % 0.0421 0.8669 0.8809 Florida CSI 0.0694 0.3123 0.4592 0.2364 *GDP %, Emp %, Consumption %, and Florida CSI % are lagged by 1 year and correlation is measured between these variables at the end of year t Ð 1 and average annual return at the end of year t. Observations for those variables are taken in 1998. Average annual returns for the high exposure REIT groups in year t are n egatively correlated with GDP %, Emp %, Consumption %, and Florida CSI % in year t 1. GDP % is highly correlated with Emp % when lagged by 1 year and when there are no lags. GDP % is highly correlated with Cons %. Emp % in year t is highly correlated with Cons %. Florida CSI is negatively correlated with high exposure group REIT returns, but the correlation weakens to almost zero for the zero exposure REIT group returns. A por tfolio of REITs that is fully divested from Florida during 1998 to 2021 is insign ificantly correlated with lagged macroeconomic variables, with increased correlations in the contemporaneous case. 5. ! Empirical Results For my sample period (1999 to 2021), I conduct regression analysis on annual REIT returns and their concurrent annual concentrations in Florida, which is measured by a REIT's square footage in Florida real estate properties as a percentage of its total squ are footage of owned properties across all geographic locations in its portfolio. The aim is to run a simple OLS regression to broadly measure how much real estate exposure in Florida affects annual REIT returns. Since I classify the REITs in my sample int o high concentration or low concentrati on categories each year depending on its corresponding year's average cross sectional exposure, I use an indicator variable in my regression instead of a continuous variable denoting concentration. The indicator (HIGH LOW) is set to one if REIT's exposure t o Florida properties is above the average cross sectional concentration in that year and 0 otherwise. The use of a categorical variable standardizes and simplifies my desired definition of concentration, as my goal is to measure a difference in returns acr oss levels of exposure and not a difference in returns per a one unit increase in the percentage of concentration. Additionally, it also allows me to account for the time varying dispers i on in average exposure since t he definition of high exposure is relat ive and should not be assumed constant as of the beginning of the sample period.

PAGE 25

! "% ! Table 8 displays the results for the following ordinary least squares regression: The OLS regression coefficient for the forecasting variable is reported. The indicator variable is statistically significant at the 5% alpha level; when HILOW = 1, on average, t he mean response function is 3.999% higher for the high concentration group than the low concentration group. It can be implied that in a given year, there is a posit ive relationship between a REIT holding commercial properties in Florida in an amount (mea sured by square feet) exceeding the equally weighted average Fl concentration in that year and its annual returns. Table 8 : One variable OLS regression summary stati stics (1998 Ð 2021) Variable Coefficient t statistic P value Intercept 0.12185 10.426 0.000*** HILOW 0.03999 2.087 0.0372* Residual Standard error 0.287 Adjusted R squared 0.0035 F statistic 4.354 However, the OLS model is na•ve and merely gives a descriptive insight. It ignores the underlying cross sectional and time varying effects. Th is model also assumes that we are selecting random and i ndependent samples at different points in time and treats every REIT year combination as an individual observation, ignoring heterogeneity and correlations between the combined error term and the pred ictor. It is important to recognize the inherent unobser ved heterogeneity, and that the trend in annual returns could stem from a time dependent or entity dependent component that we do not define. Thus, the regression is biased and inconsistent. The low a djusted R squared implies that exposure alone caries lit tle explanatory power, which might disappear if we introduce controls. To solve the omitted variables problem, I utilize panel regressions, which hold time specific unobservable factors constant over the observation span and introduce a temporal dimension . I create an unbalanced panel, since I have missing data for two entities Since a pooled OLS does not solve the assumption violations that we encounter in the na•ve regression, I run a fixed effects regression and begin with testing for individual fixed e ffects as follows: Where ! ! accounts for the differences between individual firms that are constant over time. The estimated coefficient represents a common effect of exposure on returns across all entities controlling for individual heterogeneity. I also run a fixed effects p anel regression to control for unobserved effects that are constant across entities but vary over time: Where " " accounts for variables that change over time. A panel regression with time fixed effects captures whatever a set of t Ð 1 dummy variab le would capture, including any variables that are

PAGE 26

! "& ! present for all REITs in a given time period and that may influence annual returns. Table 9 represents summary statis tics for individual and time fixed effects panel regressions. Within each REIT, the leve l of exposure does not significantly impact annual returns. Between REITs, the level of exposure is significant in explaining concurrent REIT returns. However, it is im plied by the Lagrange multiplier test that it is not as crucial to control for REIT firm characteristics as it is to control for macroeconomic variables when estimating the predictability of REIT returns. We conclude that the estimated relationship between annual Fl exposure and annual REIT returns in an individual fixed effects model is insi gnificant and not affected by omitted variable bias due to factors that are constant over time, but significant and responsive to omitted variable bias arising from fac tors that are constant across entities in a time fixed effects model. Of course, this co nclusion may vary by looking at different sample periods. Table 9 : Summary statistics of panel regression with indicator variable (199 9 2021) Individual effect s w ithin m odel Coefficient t statistic P value Variable HI GH LOW 0.0 4981 1.3012 0.1 935 R squared 0.0018 F statistic 1.6930 Lagrange Multiplier test: (H 0 = individual effects are not significant) P value 0.9992 Time effects within model Variable HIGHLOW 0.02985 2.2232 0.02644* Adjusted R squared 0.00525 F statistic 4.9425 Lagrange Multiplier test: (H 0 = ti me effects are not significant) P value 0.0000*** This table shows the effect of contemporaneous Fl exposure on REIT returns, using a sample period from 1999 2021. After controlling for individual specific unobservable factors, the effect of exposure on returns is no longer sig nificant. However, a LaGrange multiplier test indicates that we should fail to reject the null hypothesis tha t individual effects are not significant. Although using a fixed effect estimator helps obtain a consistent estimate of beta, there is no need to e liminate the time invariant unobserved component because it does not significantly correlate with the regress or, thus not affecting the zero covariance assumption between the predictor variable (HIGHLOW) and the correlation coefficient. On the other hand, a time fixed effects regression shows that a change in Fl exposure at year t from 0 to 1 significantly affect s returns at year t after excluding unobserved variables that vary over time but are constant across entities. Since time effects are significant , I introduce our specified macroeconomic variables, GDP %, EMP %, CSI %, CONS% as controls to our panel regr ession. I run the two following regressions, in which the first tests for the combined effects of Fl exposure at year t and lagged economic variabl es at year t Ð 1 on annual returns at year t, and the second introduces interaction terms. Specifically, I in teract each of the lagged economic variables with the HIGHLOW indicator variable to test for differences in the lagged local economic effects acros s the levels of Florida exposure on annual returns. By using interaction terms, I can answer the question: By how much

PAGE 27

! "' ! will a REIT's sensitivity to economic conditions change depending on how geographically concentrated their portfolio current ly is in Florida? In a five factor model with the aforementioned economic variables, whether or not you are highly exposed to Florida relative to the average concentration of the sample REITs in Florida in the same year does not significantly affect your returns. As illustr ated in table 10 , lagged GDP and e mployment growth are statistically significant, while la gged p er capita consumption and CSI growth are only significant at the 10% level. The explanatory power of the level of geographic exposure dissipates, as most of the variation in returns is explained by time varying controls. In a multi factor model with highly correlated economic variables, issues with collinearity arise. To investigate whether and how the economic variables separately affect REIT returns across diff erent levels of Fl exposure, I run the following regression for each individual variable. I find that there is no significant evidence to support that lagged local macroeconomic effects on annual REIT returns vary across geographic exposure levels. Tab le 10 : Summary statistics of panel regression with lagged economic variables (1999 2021) A ) Five factor model Variable Coefficient t statistic P value HIGHLOW 0.0544 1.4799 0.13924 GDP % 2.7001 4.4467 0.000*** EMP % 3.6606 4.1516 0.000*** CSI % 0.2782 1.8755 0.0610 CONS % 1.565 4.4467 0.0669 Adjusted R 2 0.0437 B) Interaction terms Model 1 HIGHLOW 0.0503 1 .0624 0.2883 GDP % 1.2634 3.4766 0. 000*** HIGHLOW_GDP 0.0533 0.0897 0.9299 Model 2 HI GHLOW 0.0534 1 .3992 0.1621 EMP % 2.5662 6.0787 0.000 *** HIGHLOW_EMP 0.1966 0.2838 0.7766 Model 3 HIGHLOW 0.0581 1.5361 0.1248 CSI % 0.5597 3.5767 0.000*** HIGHLOW_CSI 0.2787 1.0694 0.2852 Model 4 HIGHLOW 0.0436 0.8687 0.3852 CONS % 2.1959 5.0923 0.000*** HIGHLOW_CONS 0.0227 0.0319 0.9745

PAGE 28

! "( ! Th e table above displays the combined and individual effects of the lagged macroeconomic variables on annual REIT returns. We can see that the effect of the lagged economic factors in year t Ð 1 are not significantly different across low and high levels of geographic expo sure to Florida in year t. The predictive power of exposure disappears as the effec t of a REIT's exposure in year t on its corresponding year's equity returns changes from positive and significant in a one factor panel regression with time fixed effects to positive but not significant when the effects of the economic variables are combin ed and isolated. To check whether the results above change for contemporaneous economic variables, I run the same fixed effects panel regressions. In the following five fa ctor panel regression, I estimate the combined effects of geographic exposure and e conomic concentration in year t on REIT returns in year t: The same results hold for the indicator variable. In a multi factor model with contemporaneous economic variables, most of the returns' sensitivities are explained by changes in the local macro economic conditions in that year than by changes in exposure levels. However, this model differs in that consumption growth and CSI growth are significant an d have positive coefficients, implying that REIT returns are more responsive to changes in consump tion expenditure patterns that occur between the end of year t and the end of year t Ð 1 than to changes occurring between the end of years t Ð 1 and t Ð 2. T he explanatory power of GDP weakens as the p value increases, but a positive coefficient is report ed in the lagged and contemporaneous model. The adjusted R squared increases from 0.0437 to 0.1419, which is expected due to the addition of model parameters that are common across all firms. As we introduce time varying economic factors, the panel data set becomes more time dominant and the effects of heterogeneity in the cross sections are reduced, increasing the R squared. Employment growth is significant as in the lagged model but maintains a negative coefficient. Table 11 displays the results of running a separate panel regression for each contemporaneous economic variable with an interaction term between the HIGHLOW indicator variable corresponding with the following regression:

PAGE 29

! ") ! Table 11 : Summary statistics of panel regression with contempo raneous economic variables (1999 2021 ) Variable Coefficient t statistic P value Model 1 HIGHLOW 0. 0461 1.0235 0. 3063 GDP % 2.3551 7.5612 0. 000*** HIGHLOW_GDP 0. 2375 0.4510 0.6521 Model 2 HIGHLOW 0.0 434 1.1292 0. 2591 EMP % 2. 5457 6.2907 0.000 *** HIGHLOW_EMP 0.6754 0.2838 0.3275 Model 3 HIGHLOW 0.0290 0.7853 0.4325 CSI % 1.2062 8.0540 0.000*** HIGHLOW_CSI 0.7969 3.2292 0.0012** Model 4 HIGHLOW 0.0304 0.6610 0.5088 CONS % 2.3182 7.2547 0.000*** HIGHLOW_CONS 0.1812 0.3368 0.7363 In a separate regression for each variable, the coefficients for GDP, Employment, CSI, and Consumption growth are all negative when lagged and positive wh en contemporaneous. In terestingly, the interaction term between CSI growth and HIGHLOW is negative and statistically significant. Although a positive change in consumer sentiment between the end of year t and year t Ð 1 positively effects the level of REIT returns, the effect i s not uniform across all levels of geographic exposure to Florida. REITs whose portfolio exposure to Florida exceeds the average cross sectional exposure in a given year are negatively impacted in their annual returns by a positive ch ange to consumer senti ment from the prior year. This result is unusual because behavioral finance theory demonstrates that asset returns increase when consumers and investors are optimistic and decrease when they are pessimistic about the future direction of the economy. Consum er sentiment is a leading economic indicator which predicts changes in economic activity based on consumers' predicted buying patterns in the next 12 months. Because the CSI at the end of year t reflects predicted economic activity fo r year t + 1, it could be possible that the following year's returns for high concentration REITs will be positive or less negative as a way of correcting the over reaction in the prior year that led to the decrease in returns. For the remaining economic v ariables, the interact ion terms are not significant and do not suggest a difference in return sensitivity to local economic conditions between above average versus below average concentration portfolios. Possible reasons for why the predictive power of g eographic exposure is only significant at the 5% alpha level in a one factor model and no longer significant with the inclusion of time varying controls are ample. Firstly, the heterogeneity of Florida concentration in the cross section is not as

PAGE 30

! #* ! high as o ne would expect. Altho ugh the differences in concentration as measured by square footage are not fully uniform, they are also not fully dispersed. A simple scatterplot of geographic concentration for any given year reveals that REITs with above 40 percent concentration by squar e feet are outliers, with the majority of REITs in the sample having a concentration between 0 to 20 percent. For the entirety of the sample period, the proportion of REITs with an annual concentration between 0 to 20 percent is at le ast 80 percent in any year, and the proportion of REITs with an annual concentration exceeding 30% ranges between 5 and 12 percent. In only 7 out of the 24 years in our sample period, annual concentration surpassed 50%. However, two REITs at most in any ye ar in that subperiod h ad such exposure density. Such REITs match the desired specification of what constitutes a highly geographically concentrated firm, but they are scarce in the data set and their presence does not skew the average concentration conside rably. The equally wei ghted average de emphasizes those highly concentrated REITs and emphasizes those with 0 percent concentration. As a result, the definition of high concentration for this sample is relaxed to place equal importance on a firm with 20% e xposure and another fi rm with more than double that exposure despite the stark differences in their investment strategy and risk profile. This leads to the firms that are truly highly concentrated to be considered outliers in theory when they are not in pr actice . If we had 100% exposure cases in the cross section for each year, the spread in annual concentration between firms would be wider, and the increased heterogeneity will be more insightful in the panel regressions such that the time varying effects w ill not be as signific ant, and the cross sectional variation will be observed after introducing economic variables as controls. Another reason is that REITs, even those headquartered in Florida, do not have a high exposure to Florida to begin with. Altho ugh Florida contains 3 cities falling under the top MSA category (Miami, Orlando, and Tampa), the total number of major cities in that category are 25, according to Ling, Naranjo, and Scheik, with the top gateway markets being Boston, Chicago, Los Angeles, New York, San Francis co and Washington DC. These gateway markets have the greatest allocations of investments, with a lower allocation towards the remaining cities. This implies that REIT returns are more likely to be responsive to changes in geographic e xposure to gateway ci ties than to non gateway cities, especially since investments in gateway markets are drastically higher. In fact, it is observed in the authors' panel regression results examining the effect of geographic exposure on the cross section of REIT returns: the rel ationship between annual excess returns and the percentage of a firm's total RE portfolio located in gateway markets is positive and statistically significant. However, a breakdown by MSA shows that the relationship is not statistical ly significant, as Tam pa is the only Florida MSA with significant p values. Nevertheless, a panel regression with only local economic variables produces significant results that cannot be ignored merely because geographic exposure to Florida in the time series and cross secti on is not high enough. In table 10, a fixed effects panel regression of lagged and contemporaneous economic variables produces significant results for all the variables. Whether lagged or contemporaneous, the GDP growth rate is positi vely related to REIT r eturns. The

PAGE 31

! #+ ! continuation of positive returns suggests that GDP carries a positive but short term momentum effect on REIT stocks. Returns respond positively to information related to GDP that is at least one year old and respond to inf ormation arising as recently as year t similarly to how it responds to older information. The p value in the contemporaneous regression is still significant but increases slightly , which signifies that the returns may not instantly respond to newly emergin g inform ation about GDP growth and thus will not fully price them in the same year . We also note that returns respond negatively to increases in employment growth at the end of year t Ð 1 and at the end of year t, and returns are more responsiveness to con current changes to CSI and per capita consumption growth than prior year changes. It can be implied that investors still rely on past year information about consumer sentiment and consumption patterns, but once newer information emerges prior data becomes less val uable to them. Overall, the predictability of returns using contemporaneous variables is stronger, since the adjusted R 2 increases by almost 10 percent. Table 1 2 : Summary statistics of panel regression with economic variables (199 8 Ð 2021) Panel A : Lagged Coefficient t statistic P value GDP % 2.7035 4.4494 0.000*** EMP % 3.6156 4.1003 0.000*** CSI % 0.2721 1.8342 0.0670 CONS % 1.6270 1.9081 0.0567 Adjusted R 2 0.0443 Panel B: Contemporaneous GDP % 1.4954 4.4494 0.0084** EMP % 4.8478 6.9196 0.000*** CSI% 1.3907 10.1005 0.000*** CONS % 4.0725 7.0420 0.000*** Adjusted R 2 0.14 17 Although our aim was to inspect whether the effect of economic variables on REIT returns changes across high and low levels of geographic exposure to Florida, we deduced that the marginal effect of exposure on returns disappear when introducing such e conomic variables to our panel data. In general, REIT returns are explained more by changes to local economic conditions, in which the changes are not necessarily economic shocks, than by altering the firms' degree of exposure from low to high. However, it is important to cons ider that in any given year a REIT may have zero exposure to Florida , and because its annual concentration falls below the average concentration corresponding to that year, it will still be included in the low exposure portfolio. Thus, the indicator variab le is set to zero whether a REIT's exposure falls below the average in the cross section or has a concentration of zero. As a result, we have an identifiability problem that leads us to conclude that changes to local economic conditions affect REIT returns regardless of whether or not a REIT chooses to dispose of its Florida based properties from its portfolio to have a concentration of zero in Florida in any year. Using this conclusion , we make a generalization that REIT returns are always entirely expos ed to Florida's local economic conditions. A possible solution would be to

PAGE 32

! #" ! scale each REIT's exposure to the local economy to the proportion of its assets located in Florida, because a R EIT's sensitivity should be limited to the extent of its exposure to F lorida, especially if it is geographically diversified across different MSAs and accordingly across different regional economic variables. I scale each economic variable as follows, re peating the step for each REIT: #$%&'%()*(+$ % " , -& $ %. * ) " Where the concentration s for each REIT are calculated as follows: #$%&'%()*(+$ % " / 01$)+2* 3 456*)'7 ( " 8 9:; 3 456*)'7 ( #" $ # % & Where 9:; 3 456*)'7 ( #" measures the total propo rtion of properties in each MSA by square feet in a given year t and is summed over all the MSAs composing the property portfolio of a REIT in that year. Concentration displays cross sectional and time series variation, and each of the economic variables d isplay only time series variation. The product of concentration and any of the specified economic variables now displays cross sectional and time series variation. As a result, the REITs in our sample are no longer uniformly expose d to the economic variabl es as we construct firm specific economic growth measures, in line with Feng and Wu's methodology. Indirectly, we account for possible unobserved heterogeneity between firms, arising from different investment strategies and attitud es towards geographic con centration/diversification that lead portfolio managers to allocate investments towards Florida differently across space and time. Table 1 3 displays the results from running a fixed effects panel regression after scaling the econom ic variables. In the lagg ed regression, scaled GDP and employment growth are still significant. The results concerning the explanatory power of lagged GDP is consistent with what is prevalent in the literature. Feng and Wu found a positive correlation betw een local GDP growth and expected stock returns in the following year, suggesting that higher economic growth is associated with better future stock performance. CSI becomes significant at the 5% alpha level and per capita consumption growth becomes significant at the 1% alpha lev el in explaining REIT returns when they are scaled by geographic exposure. Changes in consumer sentiment and per capita consumption are negatively related to future REIT returns. In the contemporaneous regression, GDP growth is no longer significant when i t is scaled, and employment growth is significant at the 1% alpha level. REIT returns may not instantaneously price recent information about GDP growth, which proxies for economic growth, rather their responsiveness is detect ed one year ahead. CSI and cons umption growth are still significant. We can deduce that, whether or not CSI growth is scaled, its relevance to REIT returns increases once new information is collected and becomes available to investors. It may be that CSI a t the end of year t Ð 1 reveals information about year t that has already been incorporated in the stock price. Many economists agree that although CSI

PAGE 33

! ## ! has considerable predictive power, its marginal value decreases when used in conjunction with other econ omic variables, especially thos e that are highly correlated to CSI. Meanwhile, CSI at the end of year t reveals newer information concerned with the following year, with such information not being reflected in the stock price. Overall, lagged CSI contains information that is outdated en ough to no longer be as relevant to investors as the information contained in contemporaneous CSI. The future economic conditions it predicted have already manifested in the following year and became incorporated in the stock returns. Table 1 3 : Summary st atistics of panel regression with scaled economic variables (1999 Ð 2021) Panel A: Lagged Coefficient t statistic P value GDP % 17.0766 3.6829 0.000*** EMP % 16.1067 3.4655 0.000*** CSI % 2.1843 2.0761 0.0382* CONS % 13.3579 2.6174 0.009** Panel B: Contemporaneous GDP % 0.3207 0.9372 0.9372 EMP % 11.5445 2.7743 0.0056** CSI% 4.7830 5.1187 0.000*** CONS % 14.9452 3.6474 0.000*** It is necessary to run univariate regressions for each economic variable to isolate the effects of GDP, Employment, CSI, and per capita consumption expenditures on REIT returns. Multicollinearity problems are prevalent in a fo ur factor model of local economic variables, especially between GDP, Employment, and Consumption. Table 1 4 shows that the estimated coefficients for all the economic variables are negative in lagged regressio ns but positive in contemporaneous regressions, which shows that the signs of the coefficients in the multivariate panel regression are not totally accurate. For example, the persistence in negative performance that is explained by lagged and contemporaneo us employment growth is not present when we esti mate the effects separately. For each variable, we observe high and positive contemporaneous effects. The lagged effects are almost as high but steer in the opposite direction.

PAGE 34

! #$ ! Table 1 4 : Summary statistics of panel regression for each individual variable scaled for geographic exposure (1999 Ð 2021) Panel A) Lagged Coefficient P value R 2 GDP % 4.6977 0.0072** 0.0078 EMP % 11.9058 0.000*** 0.0281 CSI % 3.0894 0.000*** 0.0129 CONS % 7.0899 0.000*** 0. 0140 Panel B) Contemporaneous GDP % 7.4471 0.000*** 0.0253 EMP % 7.9973 0.000*** 0.0154 CSI % 3.2232 0.000*** 0.0177 CONS % 8.1478 0.000*** 0.0269 This table displays the results of a univariate regression d one for each economic variab le. Each lagged variable is scaled for geographic exposure in year t Ð 1. Each contemporaneous variable is scaled for geographic exposure in year t. Predictability increases slightly for GDP %, CSI %, CONS % when returns are reg ressed against contemporaneo us variables as the R squared increases. The differences in the estimated coefficients between lagged and contemporaneous regressions for each local variable suggest subsequent mean reversion as we do not observe evidence of persistence or momentum in a nnual REIT returns. Mean reversion is the presence of transitory components in equity prices, in which subsequent period returns are positively autocorrelated with past returns over short horizons (three to twelve months as demo nstrated by Jegadeesh and Titman) but are negatively autocorrelated over longer horizons. If past returns are high, they are likely to be low in later periods. However, this conclusion depends heavily on the duration of the holding period, the sample peri od under stu dy, and the time interval or frequency at which the returns are being analyzed. Although there is not enough or widespread mention about mean reversion in the REIT context in the finance literature, a study by Graff and Young (1997) tests for s erial persis tence in monthly sample intervals and observes performance reversals in REIT returns. In a more generalized context, many studies demonstrate that equity securities that initially outperform have weaker long term performance. The accepted reaso ning is that equities experience a temporary shock which causes their prices to increase or decrease by more than the expected value of the new information, but over time they inevitably return to a fundamental value, which is the Ôwould be' value of the s tock in the preceding period if it had behaved in accordance with the efficient market hypothesis. In the context of our study, the temporary shock constitutes the reaction to information carried from changes in local macroeconomic factors, which proxy for current and predicted economic conditions. Specifically, GDP growth, measured by the percent change of the most recent reading from a year ago, proxies for current changes in Florida's output during the year and thus its economic growth. A positive change to GDP caus es annual REIT returns to increase significantly in tandem, but to start decreasing as significantly in the following year. The estimated coefficient of lagged CSI growth being approximately equal in magnitude but opposite in direction to conte mporaneous C SI growth suggests that changes in consumer sentiment reported by Florida residents do not enhance predictability in the long term because the reversal that follows a year later almost completely erases the initial overreaction. The case is sim ilar for emp loyment growth, perhaps because the information derived from this variable can be derived

PAGE 35

! #% ! from other indicators that are stronger predictors of REIT return performance. However, the mean reversion observed from employment growth is strong er . Th is may be du e to employment growth being a lagging indicator that does not shift in tandem with the economy. At the end of year t, stock prices seem to have already reflected all available information related to employment growth with no new information to price. Empl oyment growth in year t Ð 1 is not only a lagging indicator by definition but contains old information at this point that may have been fully realized. We can also conclude that although mean reversion is noted for GDP growth and per capita con sumption gro wth, the overreaction from the positive shock still leaves a trace in the following year as the reversal is not as high in magnitude as the initial spike in returns. Also, in the context of our study, mean reversion is observed during a short er interval for REIT returns than what is commonly observed in the finance literature, in which price reversals occur several years after investing in past "winner" or "loser" stocks while momentum extends over a one year horizon. This supports the notion that REITs b ehave like general non REIT stocks but are not completely identical in their performance characteristics to stocks. GDP growth seems to be the most relevant predictor of annual returns for firms that allocate a proportion of their portfolio in Florida. Alt hough some degree of price reversal is observed over the following year, the coefficient in the lagged regression indicates that REIT stock price changes as predicted by GDP growth do not return to a fundamental level, at least not within the n ext year. In vestors do pay attention to changes in Florida's GDP, understandably so since the economy of the state of Florida is the fourth largest in the US. The effect of the size of the local economy is documented in the REIT literature, as Feng conclud ed that the income return and capital appreciation of commercial real estate are significantly positively impacted by the size of the economy, which is measured by local GDP. Accordingly, Florida's economy is not fully negligible, so an economic variable p roxying for Florida's economic growth is influential. Theoretically, if we construct a weighted average measure of GDP across all geographic areas in which a REIT is invested, Florida's GDP would still be emphasized due to its size. Overall, I find that th e scaled fir m level local macroeconomic measures are positively associated with the equity value of REITs that allocate assets in Florida in the short term. If an investor is forming a momentum based investment strategy of buying REIT stocks that respond p ositively to positive economic shocks related to Florida's GDP, they will not realize positive returns if their holding period exceeds one year, and cumulative returns will be low. To see whether the effect of GDP growth on concurrent REIT returns varies by the property type in which the REIT is specialized, I categorized my REITs into portfolios of different property sectors and run a fixed effects panel regression for each portfoli o. In the previous panel regressions that combined all REITs in the panel data, property type may have been an unobservable individual variant component. By accounting for property type, we can check for heterogeneity in the response to GDP growth between REITs belonging to different property sectors. In table 1 5 , the p values d emonstrate that the effects of contemporaneous GDP growth are not uniform across property types. Retail REITs' returns in year t are significantly positively impacted by a positive

PAGE 36

! #& ! c hange to GDP between the end of years t and t Ð 1. This sensitivity can be explained by the fact that the retail sector fluctuates with business cycles and economic conditions. Residential REITs are significant at the 1 percent alpha level, and industrial REITs are significant at the 5% level. Office REITs are not sensitive to c hanges in GDP, which is expected as the pandemic's impacts on the office industry and remote work landscape was only observed in two years in our sample period and should not skew th e results radically. Healthcare REITs are not at all sensitive to changes in GDP, as the underlying properties and the healthcare industry are recession proof and the demand for healthcare is inelastic. The differences in the estimated coefficients for GDP growth and their significance on contemporaneous REIT returns suggest tha t diversification across property types and sectors can provide a hedge against changes to GDP. However, persistent negative changes to GDP (reduction in GDP growth between the end o f years t Ð 1 and t) implies an economic downturn and a recession, to whic h commercial real estate is susceptible since rent growth, net operating income , and vacancy rates can be impacted regardless of the intended use of the commercial real estate proper ty. Realistically, perhaps only properties used for healthcare related pur poses and thus healthcare REITs are not significantly impacted by negative GDP growth. Table 1 5 : Summary statistics of panel regressions for annual REIT returns against scaled GDP growth by property type (1999 Ð 2021) Property type Coefficient p value R 2 Retail GDP % 11.1027 0.000*** 0.0378 Residential GDP % 8.8954 0.0029** 0.0585 Office GDP % 10.2437 0.1 894 0.0157 Industrial GDP % 11.7382 0.0311* 0.0378 Healthcare GDP % 1.364 0.6149 0.0019 Observing the lowest p value in the regression of retail REIT s' annual returns against contemporaneo us GDP growth also supports Gyourko and Nelling's empirical results. They conduct a property type analysis by running a regression of equity betas for REITs between 1988 Ð 1992 against the percentage of REIT i's investm ent in healthcare, industrial, office residential, and retail property sectors. Their results show that as the percentage of portfolio allocation towards retail properties increases, so does the REIT's equity beta, indicating that retail focused REITs have higher systematic risk. Retail REITs ' sensitivity to changes in Florida's GDP growth verify that the retail sector is procyclical, which affects retail tenants' operating income and cashflows.

PAGE 37

! #' ! 6. ! Conclusion The empirical evidence shows that REITs are se nsitive to changes in the degree of investment allocation towards real estate in Florida. A firm's annual returns increase by almost 3 percent on average when its geographic concentration is above the time varying average concentration across REITs as oppo sed to when it is below the average. My aim was to identify whether there was a significant differential in the effect of local economic conditions on annual REIT returns across high and low levels of geographic exposure to Florida. After accounting for th e growth rates in GDP, employment, p ersonal consumption expenditure, and consumer sentiment, I find that the significant result obtained disappears . Only the interaction term between the indicator variable HIGHLOW and CSI growth produce significant results in a contemporaneous panel regressi on, where a positive change in CSI is associated with lower returns for a highly concentrated REIT. I can attribute this phenomenon for the following reasons. Firstly, the dispersion in Florida exposure is not as stark as previously anticipated. Secondly, the exposure amounts are not high in the first place; as we do not observe firms that have extremely high concentration values, such as a concentration exceeding 50 percent, specifically after the financial crisis. REI T portfolio managers are not necessarily diversifying geographically across all regions. At the same time, they do not view geographic concentration in one state as a viable investment strategy. The summary statistics imply that REITs are more likely to be concentrated in a property sector than in a geographic region. If the exposure values were higher than what was observed in the dataset, or more frequently observed, so that a concentration above 50 percent is not an outlier, the indicator variable and th e interaction terms with the local economic variables may have been significant. To avoid an overgeneralization from claiming that the impacts of changes in local economic conditions propagate uniformly across all the REITs in our sample, the time varyin g geographic concentration is multiplied with each economic variable for each REIT to produce scaled economic variables. All four scaled economic factors are significant in a multifactor lagged regression, but GDP growth is not significant in a contemporan eous regression, suggesting that REITs take time to price information related to Florida's GDP. However, conducting univariate regressions shows all the economic variables to be statistically significant in explaining annual returns. The change in the sign s of the coefficient estimates for each variable between lagged and contemporaneous regression signals that REIT stocks may exhibit mean reverting behavior over short horizons. The degree of mean reversion varies across the economic variables, with the str ongest mean reversion observed for employment growth. For growth rates in CSI and consumption, subsequent year returns seem to fall back almost exactly to their fundamental value as the mean reversion occurs in the same magnitude as the initial spike in th e stock price . For GDP growth, the lagged regression suggests that REIT stock prices do not fully return to their fundamental value within a one year time horizon, suggesting that GDP carries the most influential and persistent effect on returns among the set of economic factors. However, the impact of GDP growth on annual returns varies

PAGE 38

! #( ! by property type, with retail REITs displaying higher sensitivity to scaled GDP growth and healthcare REITs not being sensitive. The results have implications for existing literature by verifying that REITs are indeed sensitive to macroeconomic factors similarly to other asset classes. Although REITs are an innovative financial instrument that provide retail and institutional investors a higher degree of stability a rising f rom dividend payouts and income producing assets, they are not a perfect hedge against broad as well as regional economic forces. My study focused on using ex post data to explain historical REIT returns , assuming that the sensitivity of returns to macroec onomic factors will be constant over time. However, many economists agree that sensitivities and loading s on system at ic risk s have a time varying nature. Fama and Macbeth pioneered a famous regression method that allows beta coefficients of the ind ependent variables to vary across time. It would be interesting to analyze the changes in the relationship between the scaled economic factors and annual REIT returns over time, especially once commercial real estate investment in Florida increases in the future a nd REIT portfolios become more exposed to regional macroeconomic conditions due to changing investment allocation strategies.

PAGE 39

! #) ! Ref erences Allen, M.T., Madura, J. & Springer, T.M. REIT Characteristics and the Sensitivity of REIT Returns. The Journal of Real Estate Finance and Economics 21 , 141 Ð 152 (2000). https://doi.org/10.1023/A:1007839809578 Ambrose, B. W., & Linneman, P. D. (1998). Old REITs and New REITs . Real Estate Center, Wharton School of the University of Pennsylvania. Brounen, D., & De Koning, S. (2012). 50 years of real estate investment trusts: An international examination of the rise and performance of REITs. Journal of Real Estate Literature , 20 (2), 197 223. Chan, K. C., Hendershott, P. H., & Sanders, A. B. (1990). Risk and Return on Real Estate: Evidence from Equity REITs. Real Estate Economics, 18 (4), 431 452. http://dx.doi.org/10.1111/1540 6229.00531 Chaudhry, M. K., Bhargava, V., & Weeks, H. S. (2022). Impact of economic forces and fundamental variables on REIT returns. Applied Economics , 54 ( 53), 6179 6201. Chen, N. F., Roll, R., & Ross, S. A. (1986). Economic forces and the stock market. Journal of business , 383 403. Cotter, J., Gabriel, S., & Roll, R. (2015). Can housing risk be diversified? A cautionary tale from the housing boom and bust. The Review of Financial Studies , 28 (3), 913 936. Fei, P., Ding, L., & Deng, Y. (2010). Correlation and volatility dynamics in REIT returns: performance and portfolio considerations. The Journal of Portfolio Management , 36 (2), 113 125. Feng, Z., & Wu, Z. (2022). Local economy, asset location and REIT firm growth. The Journal of Real Estate Finance and Economics , 65 (1), 75 102. Geltner, D. (1989). Estimating real estate's systematic risk from aggregate level appraisal < based returns. Real Estate Eco nomics , 17 (4), 463 481. Glascock, J., & Lu Andrews, R. (2014). An examination of macroeconomic effects on the liquidity of REITs. The Journal of Real Estate Finance and Economics , 49 , 23 46. Glascock, J. L. (1991). Market conditions, risk, and real estate portfolio returns: Some empirical evidence. The Journal of Real Estate Finance and Economics , 4 , 367 373. Glascock, J. L., Lu, C., & So, R. W. (2000). Further evidence on the integration of REIT, bond, and stock returns. The Journal of Real Estate Finance and Economics , 20 , 177 194. Graff, R., & Young, M. (1997). Serial persistence in equity REIT returns. Journal of Real Estate Research , 14 (3), 183 214. Gyourko, J., & Keim, D. B. (1992). What does the stock market tell us about real estate returns?. Real Estate Economics , 20 (3), 457 485.

PAGE 40

! $* ! Gyourko, J., & Nelling, E. (1996). Systematic risk and diversification in the equity REIT market. Real Estate Economics , 24 (4), 493 515. Hartzell, J. C., Sun, L., & Titman, S. (2014). Institutional investors as monitors of corporate diversification decisions: Evidence from real estate investment trusts. Journal of Corporate Finance , 25 , 61 72. Jegadeesh, N., & Titman, S. (2001). Profitability of momentum strategies: An evaluation of alternative explanations. The Journal of finance , 56 (2), 699 720. Korniotis, George M. and Kumar, Alok, State Level Business Cycles and Local Return Predictability (March 6, 2012). Journal of Finance, Forthcoming, Available at SSRN: https://ssrn.com/abstract=1094560 or http://dx.doi.org/10.2139/ssrn.109456 0 Li, Y. (2016). Time variation of expected returns on REITs: Implications for market integration and the financial crisis. Journal of Real Estate Research , 38 (3), 321 358. Naranjo, A., & Ling, D. C. (1997). Economic risk factors and commercial real estate returns. The Journal of Real Estate Finance and Economics , 14 , 283 307. Ling, D. C., & Naranjo, A. (1999). The integration of commercial real estate markets and stock markets. Real Estate Economics , 27 (3), 483 515. Ling, D. C., Naranjo, A., & Schei ck, B. (2019). Asset location, timing ability and the cross < section of commercial real estate returns. Real Estate Economics , 47 (1), 263 313. Liow , K.H., &Li, X. (2006). Are REITs unique? A comparative analysis of major asset classes. Journal of Real Estate Finance and Economics, 22(2), 299 318. Pirinsky, C., & Wang, Q. (2006). Does corporate headquarters location matter for stock returns?. The Jo urnal of Finance , 61 (4), 1991 2015. Alberto Plazzi, Walter Torous, Rossen Valkanov, Expected Returns and Expected Growth in Rents of Commercial Real Estate, The Review of Financial Studies , Volume 23, Issue 9, September 2010, Pages 3469 Ð 3519, https://doi.org/10.1093/rfs/hhq069 Seck, D. (1996). The substitutability of real estate assets. Real Estate Economics , 24 (1), 75 95. Smajlbegovic, E. (2019). Regional economic activity and stock returns. J ournal of Financial and Quantitative Analysis , 54 (3), 1051 1082. Zhu, B., & Lizieri, C. (2022). Local beta: Has local real estate market risk been priced in REIT returns?. The Journal of Real Estate Finance and Economics , 1 37 .


xml version 1.0 encoding UTF-8 standalone no
fcla fda yes
!-- Real Estate in the Sunshine State: An Analysis of Impacts Local Economic Conditions on Performance REITs Investing Florida ( Book ) --
METS:mets OBJID AA00092421_00001
xmlns:METS http:www.loc.govMETS
xmlns:xlink http:www.w3.org1999xlink
xmlns:xsi http:www.w3.org2001XMLSchema-instance
xmlns:daitss http:www.fcla.edudlsmddaitss
xmlns:mods http:www.loc.govmodsv3
xmlns:sobekcm http:digital.uflib.ufl.edumetadatasobekcm
xmlns:lom http:digital.uflib.ufl.edumetadatasobekcm_lom
xsi:schemaLocation
http:www.loc.govstandardsmetsmets.xsd
http:www.fcla.edudlsmddaitssdaitss.xsd
http:www.loc.govmodsv3mods-3-4.xsd
http:digital.uflib.ufl.edumetadatasobekcmsobekcm.xsd
METS:metsHdr CREATEDATE 2023-05-10T13:43:09Z ID LASTMODDATE 2023-05-10T13:28:30Z RECORDSTATUS COMPLETE
METS:agent ROLE CREATOR TYPE ORGANIZATION
METS:name UF,University of Florida
OTHERTYPE SOFTWARE OTHER
Go UFDC - FDA Preparation Tool
INDIVIDUAL
UFAD\renner
METS:dmdSec DMD1
METS:mdWrap MDTYPE MODS MIMETYPE textxml LABEL Metadata
METS:xmlData
mods:mods
mods:abstract displayLabel Abstract lang en This paper studies the effect of local macroeconomic conditions on the annual returns of REITs. Specifically, I look at Florida’s economic performance using four fundamental indicators: state GDP, employment, personal consumption expenditure, and consumer sentiment, and assess whether there is a difference in the relationship between the local economy and REIT returns across the firms that are highly concentrated in Florida relative to those that are not. The empirical results initially confirm a higher equity return for firms with a high exposure to Florida. However, the differential in geographic exposure generally does not affect the sensitivity of the returns to the economic variables, with the exception of consumer sentiment, which displays a negative impact on the returns of high exposure REITs. When the economic variables are scaled by the time varying concentrations of each REIT, the coefficients of each variable are still significant, with short term mean reversion detected from the change of the regression coefficients from negative in a lagged regression to positive in a contemporaneous regression. The growth rate of Florida’s GDP is the most significant economic indicator. Further analysis shows that its effect varies across property sectors, with significant effects observed in retail, residential, and industrial REITs but not in healthcare REITs.
mods:accessCondition Copyright Dima Alkhatib. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
mods:language
mods:languageTerm type text English
code authority iso639-2b eng
mods:location
mods:physicalLocation University of Florida
mods:url access object context https://original-ufdc.uflib.ufl.edu/AA00092421/00001
mods:name
mods:namePart Alkhatib, Dima
mods:note Awarded Bachelor of Science in Business Administration, magna cum laude, on April 29, 2022. Major: Finance
College or School: Business
Advisor: Andy Naranjo. Advisor Department or School: Finance, Insurance and Real Estate.
mods:originInfo
mods:publisher University of Florida
mods:dateIssued 2023
mods:recordInfo
mods:recordIdentifier source sobekcm AA00092421_00001
mods:recordContentSource University of Florida
mods:relatedItem series
mods:part
mods:detail Year
mods:caption 2023
mods:subject
mods:topic Undergraduate Honors Thesis/Project
mods:titleInfo
mods:title Real Estate in the Sunshine State: An Analysis of the Impacts of Local Economic Conditions on the Performance of REITs Investing in Florida
mods:typeOfResource text
DMD2
OTHERMDTYPE SOBEKCM SobekCM Custom
sobekcm:procParam
sobekcm:Aggregation ALL
UFIR
IUF
UFHONORS
UNDERGRADWORKS
sobekcm:MainThumbnail Alkhatib_Dima_Honors_Projectthm.jpg
sobekcm:Wordmark UFIR
sobekcm:Tickler 2231_SPRING_2023_HONORS
sobekcm:bibDesc
sobekcm:BibID AA00092421
sobekcm:VID 00001
sobekcm:Publisher
sobekcm:Name University of Florida
sobekcm:Source
sobekcm:statement UF University of Florida
sobekcm:SortDate 738520
METS:amdSec
METS:digiprovMD DIGIPROV1
DAITSS Archiving Information
daitss:daitss
daitss:AGREEMENT_INFO ACCOUNT PROJECT UFDC
METS:techMD TECH1
File Technical Details
sobekcm:FileInfo
METS:fileSec
METS:fileGrp USE reference
METS:file GROUPID G1 PDF1 applicationpdf CHECKSUM 7b57810ce7a17585dbf496376167a9d5 CHECKSUMTYPE MD5 SIZE 2180913
METS:FLocat LOCTYPE OTHERLOCTYPE SYSTEM xlink:href Alkhatib_Dima_Honors_Project.pdf
G2 TXT2 textplain
Alkhatib_Dima_Honors_Project_pdf.txt
G3 METS3 unknownx-mets 9c2cebbd4a793b62be1f73182c7c604f 7245
AA00092421_00001.mets
METS:structMap STRUCT2 other
METS:div DMDID ADMID ORDER 0 main
ODIV1 1 Main
FILES1 Page
METS:fptr FILEID
FILES2 2
FILES3 3