Social Network Matters: Capital Structure Risk Control on REITs

Abstract

This paper uses biographical information of executives and directors of REITs in the US to show whether, and through what channels, top executives of REITs are influenced by their social peers when determining capital structure risk control strategies, especially in critical periods. Our focus is on the period of the 2007–2009 Financial Crisis. We find that peer influence through past employment and sharing activities significantly facilitate peer learning in making decisions on debt maturity extension, but does not affect leverage reduction. We find that being educated from the same school carries some effects on leverage reduction, possibly via its influence on managers’ personal traits. However, concurrent employment does not play a role in determining either of the strategies. We further verify the existence of influence of social network in decision making of REITs in 2015 in preparation for a boom at the beginning of the up-market. Hence, our study highlights the strength of peer connections in clarifying possible sources of herding in REITs decisions.

A. Appendix

A.1 The Bayesian Model

The models in (8) and (9) are also known as the Spatial Autoregressive Model (SAR) in the spatial literature (see LeSage & Pace, 2010). We first express (8) and (9) in the compact matrix form such that

$$Y={\alpha }_{0}+{\alpha }_{1}WY+X\beta +{D}_{s}\eta +U$$

(12)

where Y is the n × 1 vector of D.MLev_i or D.Mat23_i, W is the n × n network matrix such that the (i, j)th element is zero if firms i and j are not socially connected and equals to the centrality of j otherwise, X is the n × 4 matrix of regressors specified in (8) and (9), D_s is the dummy matrix for sector effects, $U = \left( {u_{1} , \ldots u_{n} } \right)$’ and ui ∼ N (0,${\sigma }_{u}^{2}$). Define S(α₁) = I_n − α₁W where I_n is the identity matrix of size n, and let θ = (α₀, β’, η’)’. The likelihood function of Y, conditional upon W, X, α₁, θ, is given as,

$$p\left(Y|W,X,{\alpha }_{1},\theta \right)={\left(2\pi \right)}^{-\frac{n}{2}}\bullet {\left({\sigma }_{u}^{2}\right)}^{-\frac{n}{2}}\bullet \left|S\left({\alpha }_{1}\right)\right|\bullet \mathrm{exp}\left(-\frac{{U}^{^{\prime}}U}{2{\sigma }_{u}^{2}}\right) .$$

(13)

We specify the priors p(α₁) = Uniform[− 1, 1] and p(θ) be conventional non-informative normal prior. The posterior of α₁ and θ is thus,

$$p\left({\alpha }_{1},\theta |W,X,Y\right)\propto p\left({\alpha }_{1}\right)p\left(\theta \right)p\left(Y|W,X,{\alpha }_{1},\theta \right) .$$

(14)

The posterior draws from (14) are obtained from the following MCMC iterations.

• 1. Sample α₁ from $p\left({\alpha }_{1}|\theta ,Y,W,X\right)$

• 2. Sample θ from $p\left(\theta |{\alpha }_{1},Y,W,X\right)$

The MCMC sampling iterates through the above two steps until convergence. For each step, the draw is conditioned on the rest of parameters with the most updated values at the current iteration.