How do housing cycles influence listed firms’ R&D investment: evidence from the collateral channel

ABSTRACT

Firm innovation is essential to long-run economic growth. Financially constrained R&D firms may use firm-owned properties as collateral to finance their R&D projects. Therefore, the housing price cycle can affect firms’ R&D investment through influencing their real estate value. By examining listed R&D firms during the housing boom period 2002–2006 in the U.S., we find that a $1 increase in real estate value leads a firm to increase its R&D investment by $0.38. We also find that this collateral effect is more pronounced among financially constrained R&D firms than that among unconstrained ones. Additionally, we examine the housing bust period 2008–2012, and find that real estate depreciation retarded R&D investment, especially among constrained R&D firms.

KEYWORDS:

• R&D investment
• real estate price
• financial constraints
• firm innovation

JEL CLASSIFICATIONS:

• D21
• O31
• G31
• R30

Acknowledgements

We would like to thank the Editor, Cristiano Antonelli, two anonymous referees, Ramirez Carlos, Raghavendra Rau, and Yang Yao for their helpful comments. We would like to acknowledge the support from Real Capital Analytics Inc. for generously providing the access to the Real Capital Analytics commercial property price index (RCA CPPI). We are also grateful to seminar participants at the 2014 CES Purdue Meeting, the internal workshop at the Research Institute of Economics and Management, Southwestern University of Finance and Economics, and University of Nebraska at Omaha.

Disclosure statement

No potential conflict of interest was reported by the authors.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1 Real estate appreciation is also associated with higher leverage ratios (Cvijanovic 2014), higher bank debt-to-total-debt ratios (Lin 2016), fewer cash holdings (Chen, Harford, and Lin 2014), and more entrepreneurship (Corradin and Popov 2015).

2 Common unobservables refer to some local policies that may influence both local commercial property prices and local investment opportunities. Reverse causality refers to the R&D investment that may boost the local economy and thus local commercial property prices.

3 Regarding firms’ initial real estate holdings, another concern on our identification strategy is that real estate ownership and R&D investment are correlated. If R&D firms that own more properties are also more sensitive to local shocks, our estimates would be upward biased. Following the literature (e.g. Chaney, Sraer, and Thesmar 2012; Cvijanovic 2014), we address this concern by controlling for determinants of real estate ownership by interacting a set of initial firm-level characteristics with commercial property prices. Doing so does not change our major results, thus it is unlikely that the above concern drives our major results.

4 There are numbers of other related studies (e.g. Chan, Martin, and Kensinger 1990; Chan, Lakonishok, and Sougiannis 2001; Chambers, Jennings, and Thompson 2002; Li and Liu 2010; Lin 2016).

5 By examining U.S. listed firms for 1993–2006, Mao (2015) shows that real estate appreciation enhances firms’ patenting through three different channels, with one being internal R&D. Using similar data, Cao et al. (2015) also find that real estate value is positively related to firms’ patenting.

6 Theoretically, the bubble effect on innovation can go the opposite direction. Using the same class of models, Saint-Paul (1992), Grossman and Yanagawa (1993), and King and Ferguson (1993) point out that the speculative bubble has a negative effect on R&D investment. Their argument is that the speculative bubble diverts savings away from R&D investment. Rong, Wang, and Gong (2016) examine manufacturing firms’ R&D propensities during China’s housing boom period 1999–2007. They show that firms invest less in R&D associated with a surge in housing prices. Their argument is that housing price surges created high-return investment opportunities and firms reacted accordingly by diversifying into the real estate industry, thereby investing less in R&D.

7 Though adjustment costs may make R&D firms more cautious to the increase in cyclical asset value, the benefit of using enhanced debt accessibility to fund R&D may be high enough to overcome the cost, resulting in firms investing more in R&D. Therefore, given the existence of the R&D-smoothing concern, the fact that we still observe a positive relationship further confirms our hypothesis of the existence of a collateral pledging effect on firm innovation.

8 See Kerr and Nanda (2015) or Hall and Lerner (2010) for a review.

9 Chava, Nanda, and Xiao (2017) show that recognizing the value of innovative firms’ patent stock, banks provide them with cheaper loans. Saidi and Zaldokas (2017) find that increased innovation disclosure helps patenting firms to switch lenders, leading to lower costs of debt. Hochberg, Serrano, and Ziedonis (2014) document that patents are used as collateral for venture debt.

10 A competing hypothesis regarding the extent of financial constraints is that R&D firms, relative to non-R&D firms, are so financially constrained that they solely use internal funds to finance their R&D investment. Consequently, regardless of real estate appreciation, their R&D expenditures remain unchanged. This hypothesis can be ruled out by the findings that debt financing is an important source of capital in innovative firms as we have mentioned.

11 Figure 1 shows that the growth rate of national housing prices in 2007 was already negative. It therefore seems reasonable to include 2007 in the housing bust period. However, cities’ housing cycles are not the same and in some cities’ housing prices peaked during 2007. To be conservative, we choose to start the housing bust period at 2008. The major results persist when 2007 is included.

12 We use the headquarter location as a proxy for the location of firm-owned real estate and trace the location’s commercial property prices to measure how the value of real estate evolves. Doing so relies on two approximations, the approximation for the location of a firm’s real estate assets and the quality of the real estate information. Chaney, Sraer, and Thesmar (2012) have shown strong evidence validating these two approximations.

13 We treat the case that a firm was listed before 1993 but has missing value of real estate assets at cost in 1993 as it being listed in a later year when the value of its real estate assets at cost is firstly available.

14 The founding year comes from Jay Ritter’s website on IPO data. Loughran and Ritter (2004) provide founding dates for 9,826 IPOs from January 1975 to December 2013 in the U.S.

15 When these two years are very close, this approximation is reliable. However, when there is a large gap between these two, this approximation becomes less reliable. To check whether the inclusion of those firms with a large gap drives our major results, we repeat the regressions by excluding those firms with a gap greater than 15 years and the results are similar (not reported).

16 For the case where a firm’s real estate assets at cost in 2002 are zero, the corresponding market value is also zero.

17 In the literature, properties, plant and equipment (PP&E: item #8) is also used to do the normalization (e.g. Chaney, Sraer, and Thesmar 2012). Our major results persist when PP&E is used (not reported). Different from using total assets, the coefficient on real estate value turns larger. It suggests that R&D investment is less sensitive to real estate value in those R&D firms with higher PP&E/TA ratios, which, by definition, have relative fewer intangible assets and thus tend to be less innovative. This difference is thus consistent with Hypothesis 1.

18 Our major results persist when we use the five percent winsorising method (not reported).

19 We only have commercial property prices for these 22 major MSAs. More details are discussed in section 3.2.

20 As a robustness check, following the specification of Cvijanovic (2014), we include R&D firms initially holding no property in our sample and rerun the regressions. Our major results barely change.

21 The NCREIF CPPI is available at the website: https://www.ncreif.org/property-index-returns.aspx. The NCREIF CPPI is appraisal-based while the RCA CPPI is a regression-based repeat-sales index. We use the Consumer Price Index (CPI) before 1978.

22 The HPI is available for 64 MSAs with a start year between 1977 and 1987.

23 As shown by Chaney, Sraer, and Thesmar (2012), older, larger, and more profitable firms are more likely to be real estate owners.

24 The S&P/Case-Shiller U.S. National Home Price Index tracks the market value of single-family houses in the U.S. The index measures changes in housing prices given a constant quality level.

25 To obtain the full sample, step 6 of the sample screening process (dropping observations with zero R&D expenditure) is skipped. The summary statistics of the full sample are not reported but available upon request.

26 Note that the coefficient is still larger than that found by Chaney, Sraer, and Thesmar (2012). This deviation may be because our measure of initial real estate assets is more up-to-date. We only include the period 2002–2006 while Chaney, Sraer, and Thesmar (2012) extend the examination period back to 1993.

27 So far, we have confirmed that real estate value influences firms’ R&D at the intensive margin (e.g. how much to invest in R&D). It is also interesting to check whether real estate value influences firms’ R&D at the extensive margin (e.g. whether to become an R&D firm). We thus rerun the regression of column 1 by using the R&D dummy as the dependent variable. The R&D dummy is equal to one if R&D expenditures are positive, and zero otherwise. The mean and standard deviation of the R&D dummy are 0.57 and 0.50, respectively. It is somewhat surprising to find that the coefficient on $REValu{e}_{i,t}$ is −0.047 but insignificant, which indicates that real estate value may adversely influence a firm’s propensity to become an R&D firm.

28 We regress capital investment on R&D investment with no constant.

29 As a robustness check, we also regress capital-unrelated R&D investment on real estate value, and find that the effect of real estate value on capital-unrelated R&D investment remains significantly positive (not reported).

30 The above IV approach strictly follows Cvijanovic (2014). Though not reported, as a robustness check, we also follow Chaney, Sraer, and Thesmar’s (2012) IV approach and the major results persist.

31 Its mean and standard deviation are 0.041 and 0.067, respectively.

32 Our major results persist when we use total investment instead of R&D investment as the dependent variable.

33 The regressions using either the MSA-level HPI or the state-level HPI deliver similar results as shown in the Online Appendix Tables.

34 Z-test results show that the coefficient for constrained firms is significantly larger (at the 1% level). Since our focus is on whether the coefficient on real estate value among constrained firms is larger than that among unconstrained firms, a one-tailed test is used.

35 The same pattern is found when either the MSA-level HPI or the state-level HPI is used (Online Appendix Tables 2 and 6).

36 See Rong, Wang, and Gong (2016) for an example from China, where numbers of listed firms diversified into the real estate industry during the housing boom period.