We measured individual REITs unused debt capacity as the difference between its regulatory debt limit and actual debt ratio. This unused debt capacity or debt buffer measures flexibility or financial slack because it indicates how much additional debt a REIT could issue before reaching its regulatory debt limit. The amount of buffer kept by individual REITs also reflects their debt policies. Conservative REITs tend to keep higher debt buffer and it is vice versa for aggressive REITs. We also considered a second debt boundary, i.e. optimal debt ratio that is widely used in the literature. Optimal debt ratio is the predicted value of debt ratios on a set of trade-off variables which includes firm size, firm age, profitability and growth opportunities (MB ratio). 3.2

Following Hovakimian et al. (2001) and de Jong et al. (2011), we estimated a binary logit model with marginal financing decisions (Marginal) as a function of unused debt capacity. This Marginal variable is equal to one for debt issues and zero for equity issues. We use quarterly changes in balance sheet information to identify debt and equity issues/repurchases. Equity issues (repurchases) are defined as increase (decrease) in total capital stock (exclude retained earnings) while debt issues (repurchases) correspond to increase (decrease) in total debt. The following filters were applied so that only material events were included in the sample of financing activities: The sum involved must, firstly, be larger than US $5million and the amount must also constitute more than 1% of the REIT’s total assets. Cases where REITs issued (repurchased) both debt and equity issues were omitted. The estimated logistic model is as follows: Marginalit =α1 + α2Debt bufferit-1 + θXit-1 + ϕMit + Pi + Yt + uit where Xit-1 and Mit, Pt, Yt are the vectors of firm characteristics, market timing variables, property and year dummies respectively, and uit is the residual assumed to be serially uncorrelated with mean zero. The vector X contains the controls for the traditional capital structure determinants such as cash holding, firm size, firm age, growth opportunities, profitability and asset tangibility. Vector M, on the other hand, controls for market timing behavior including past stock returns, interest rate and term structure. These variables capture financial managers’ market timing behavior by taking advantage of the relative costs of debt and equity capital in raising capital. We mitigated the reverse causality concern by lagging all firm-level explanatory variables. The construction of all of the variables used in this study is detailed in Appendix A. The correlation matrix in Appendix B indicates minimal problem with multicollinearity. We only elaborate here the predictions for issuing decisions since the predictions of repurchasing decisions follow the same logic albeit with the opposite sign. Traditional capital structure theories (pecking order and trade-off) and financial flexibility predict a positive α2 where REITs with higher unused debt buffer in the previous period are more likely to issue debt in the current period. This is consistent with the idea of debt convergence towards either an optimal debt ratio (trade-off) or debt capacity (pecking order). The financial flexibility hypothesis also predicts a positive coefficient α2 indicating REITs make use of the financial slack (debt buffer) they preserved in the previous period to issue more debt in the current period. A negative α2 in hand is consistent with the prediction of financial stability where firms that maintain a lower (higher) debt buffer in the previous period tend to remain as such in the next by issuing lesser debt (higher equity). In order to discern debt conservatism behavior, we allowed for the differential of α2 in Equation (1) to vary between high and low debt buffer subsamples. The specification is as below: Marginalit = α1 + α2Debt bufferit-1Dt-1 High buffer + α3Debt bufferit-1Dt-1 Low buffer + θXit-1 + ϕMit +Pt + Yt + uit where Dt-1 High buffer (Dt-1 Low buffer) is a dummy variable equal to one if debt buffer is above (below) the median value of debt buffer in the sample and zero otherwise for REIT i at time t-1. The signs of coefficients α2 and α3 indicate whether REITs are moving towards or away from their target debt ratio. The basic premise of debt conservatism behavior is that firms would opt for financing choices that make them deviate further (below) from their debt capacity, hence, making them carry more unused debt capacity. Empirically, we expect REITs with higher debt buffer to either deviate more from their debt capacity than low buffer REITs (α2<0 & α2< α3) or to converge at a slower rate than their low buffer counterparts towards debt capacity (α2>0 & α2<α3). Notice that we did not require REITs to adjust until they reached the debt capacity since this is maximum debt boundary of debt ratio. A move towards or away from debt capacity would suffice for us to infer debt conservatism behavior.

Our sample was a quarterly panel data set of Singapore and Malaysian listed REITs during the period 2003 to 2015. The financial data and stock prices used in this research were obtained from Bloomberg’s database. Individual REIT’s unused debt facilities were extracted from annual reports and converted to quarter using the linear interpolation method.7 We avoided survivorship bias by including both active and delisted REITs that appeared in Bloomberg’s database.8 The variables used in the regression analysis were winsorized at the 1st and 99th percentile to avoid the influence of extreme observations. After omitting observations with missing values REITs, the unbalanced panel data consisted of 1344 firm-quarter observations covering 54 REITs publicly traded between 2003:Q1 and 2015:Q4. This sample contained a total of 569 security issues and 475 security repurchases defined as in Hovakimian et al. (2001) and de Jong et al. (2011) mentioned above. 3.4

Table 1 presents the descriptive statistics for all variables used in this study. On average, REITs are less likely to issue (45.0%) and repurchase (40.6%) debt than equity. The prevalence of equity issues despite the fact that all REITs are operating below their regulatory capacity provides a crude test to reject the pecking order theory which hypothesizes firms to issue equity only when they have used up their debt capacity. Regulatory buffer is distributed symmetrically across the sample of REITs with the sample mean and the median equal to 22.2% and 23.1% respectively of the total assets on the back of the average regulatory debt limit of 52.3%. Our measure of debt buffer was higher than the estimated buffers of 16.2% and 17.2% reported by, respectively, de Jong, Verbeek and Verwijmeren (2012) and Hess and Immenkotter (2012) using credit rating regression models. The mean debt ratio of 30.1% was substantially lower than 52% documented by Giacomini et al. (2017) for US REITs for the period 1990-2012. The average optimal debt ratio of 29.6% indicated firstly, REITs were operating near their optimal debt limit and secondly, optimal limit was the first debt boundary for REITs before the regulatory limit which was the maximum boundary. Unused private and public debts constituted of 4.6% and 13.8% of the total assets respectively. The mean cash holdings were low at 2.7% while tangible assets were high at 92.4% of REITs’ total assets due to the regulatory structure of REIT. The average firm size was US$1.52 billion. REITs in our sample were relatively young with an average age of 5.3 years as REIT sector only began to flourish in Asia in the early 2000s. The average Tobin’s, a proxy to firm value was approximately 0.81 times with a range of 0.37 to 1.29. In terms of profitability, a typical REIT averaged 1.5% (6.0% p.a.) as measured by ROA. As for the market timing variables, the average stock performance of 6.1% represented the price appreciation of individual REIT stocks over the last six months. Term structure spread averaged 1.4% implying an upward sloping yield curve. The relative interest rate of 1 time over the average interest rate in the past one year implied a generally stable interest rate regime.

Note. The descriptive statistics for the explanatory variables was based on a full sample of 1344 firmquarter observations with 54 individual REITs. Total observations for unused private and public debts were 1,276 each due to missing value problem. Except for actual debt ratio, debt buffer and optimal debt ratio, all control variables are winsorized at the bottom and top 1% level. We used the logarithm of REIT size and REIT age for regression analysis.

Figure 1 presents the time series evolution of the average debt ratio and debt limit during Q1:2003 to Q4:2015. We observed a dramatic surge in debt buffer from 8.8% in 2004 to 21.0% in 2006 due to an increase in the regulatory debt limit for REITs in Singapore and Malaysia from 35% to 60% (for rated REITs) and 50%, respectively. REITs clearly chose to maintain debt ratio at a range which was comfortably well below the regulatory debt limits so that they could access external financing without any constraints. The relatively stable debt ratio masked REITs’ deliberate actions in enlarging their debt capacity. Four out of six Singapore REITs obtained credit rating within 1 year after the regulatory limit change in 2005 to tap into the additional debt capacity afforded by the existence of a credit rating. Moreover, 44% of rated Singapore REITs that had the option to leverage up to 60% chose to keep their leverage ratio below the first hurdle of 35% as in Dec 2015.

Figure 2 shows that high buffer REITs (above sample median) consistently maintain higher unused private debt capacity (mainly bank lines of credit) than low buffer REITs (below sample median) every year during the sample period. This reflects over-conservatism in debt policy since REITs need to pay annual fees (based on total lines size) and commitment fees (based on unused private debt) for maintaining this standby facility (Hardin and Hill, 2011). These results also suggest REITs that keep a higher debt buffer are not constrained in their access to external financing. In other words, low leverage is not an outcome of credit rationing by the lenders. We also observed similar pattern in unused public debts in Figure 3. Except for years 2006, 2011 and 2012, unused public debts were higher for high buffer REITs than low buffer REITs. Not reported here, the mean test of difference for unused private/public debts was significant at the 5% level between the high and low buffer subsamples. Note that we only showed in Figures 2 and 3 unused private and public debts from year 2005 onwards since there was no observation for subsample above the sample median debt buffer for years 2003 and 2004. In sum, these descriptive results suggest debt conservatism behavior where REITs prefer to operate significantly below their regulatory debt limit and at the same time, maintain a huge sum of unused debt facilities to back this unused debt capacity.

Table 2 presents the results of the logit regressions. Robust t-stats clustered by firms (Peterson, 2009) are reported in parentheses below the regression coefficients. Models 1-2 are debt-equity issue regressions while Models 3-4 are debt-equity repurchase regressions. Model 1 shows a strong significant negative (positive) relationship between debt (equity) issues and debt buffer. This finding provides strong evidence against the trade-off/pecking order/financial flexibility theories that predict a positive relationship between the probability of debt issue and previous period debt buffer. This finding is however, in line with Lemmon et al.’s (2008) financial stability hypothesis which posits that firms that carry low leverage (high buffer) tend to remain as such by issuing more equity (debt) in the future. In REITs, Harrison et al. (2011) also documented a negative relationship between lagged leverage and debt ratio and debt issuance events which the authors interpret as evidence of substantial level of stability within a REIT’s capital structure from period to period. Model 2 examines whether the impact of debt buffer on debt-equity choice is symmetric between high and low buffer REITs. We do this by decomposing debt buffer variables into those that are above sample median (Debt Buffer t-1*D high buffer ) and those that are below sample median (Debt Buffer t-1*D Low buffer). The coefficient of Debt Buffer t-1*D high buffer is negative and strongly significant while the coefficient of Debt Buffer t-1*D Low buffer is insignificant. These results suggest that the deviation from debt capacity is driven by high buffer REITs that chose to issue equity over debt, making them more underleveraged rather than by low buffer REITs that aggressively issue debt which push them nearer to their debt capacity. With regard to repurchasing decisions in Model 3, the negative coefficients of debt buffer variables suggest debt convergence towards the debt capacity. Hovakimian et al. (2001) also found firms moving towards their target debt ratio via repurchase instead of issuing decisions. These authors suggest that firms have more discretion when repurchasing existing capital than in raising new capital. This might be relevant to the REITs sector that is known to use bank lines of credit or equity to finance investment and subsequently refinance these debt/equity with public debt of longer maturity when the timing and pricing is right (Brown & Riddiough, 2003; Ooi et al., 2012). Repurchase decisions are therefore, likely used by REITs to reconfigure firm capital structure, while issuing decisions are driven by investment needs. This conjecture is supported by the positive and significant MB ratio, a proxy to growth opportunities that is positive and significant in issuing decisions but not in repurchasing decisions. Model 4 shows that convergence is larger in magnitude for low buffer REITs than high buffer REITs suggesting that high buffer REITs move much slowly towards their debt capacity compared to their low buffer counterparts. Three out of the four control variables are significant in the issuance equation and are in the direction predicted by the trade-off theory where financially constrained REITs (smaller, younger and lesser cash holdings) prefer to issue more equity over debt to avoid the cost of financial distress. The positive and significant coefficient of MB ratio is in line with the pecking order’s prediction that firms with the greatest degree of information (high growth opportunity) have incentives to issue debt over equity. Market timing variables (stock performance, term structure and interest rate) do not exert significant influence on issuance choices though the signs are consistent with the implication of this theory. Specifically, REITs appear reluctant to issue debt when market interest rates and/or risk premiums are high, as well as when past stock performances have been strong. These market conditions however, appear to play an important role in repurchasing decisions. REITs are significantly more likely to repurchase debt when risk premiums and interest rates are high and when their stocks are highly valued. The significance of this market timing variables adds credence to our claim that repurchases are initiated to reconfigure capital structure, taking advantage of capital market conditions. The only firm characteristic variable that is significant in the repurchasing equation is asset tangibility. The positive relationship is inconsistent with the collateral value of tangible real assets that should, all else being equal, lead to reduced probability of debt repurchase due to increasing REITs’ capacity to borrow more. 4.2

This section provides several robustness checks to the main results in Table 2. First, despite the short panel of our sample data, we re-estimated the logistic regression using GLS random effects and firm fixed effects. Second, we used quarterly change in debt ratio scaled by total assets as the dependent variable. Third, we used an alternative method to classify debt buffer variables, i.e. dummy variables equal to one for debt buffer at the top and bottom 10% of the sample. This classification allowed us to directly test our claim that the negative coefficient Debt Buffer t-1*D High buffer in Table 2 is driven by REITs’ high buffer REITs that chose to issue equity instead of debt.

Notes. We estimated a logit model for REITs’ debt-equity choices. The dependent variables for models 1 and 2 (3 and 4) were designed to equal one for debt issues (repurchases) and zero for equity issues (repurchases). Cases where firms issued (repurchase) both material debt and equity issues were omitted. The cut-offs for material debt/equity issues were that the issues need to be larger than US$5 million and constitute 1% of total assets. T-statistics which appear in parentheses. *, **, *** indicate significance at the 10%, 5%, and 1% confidence levels, respectively.

The results reported in Panels A and B demonstrate that the main findings of this paper are robust across different econometric and empirical specifications. Our main findings that high buffer REITs tend to deviate away from their debt capacity in issuing decisions and converge at a slower pace to their debt capacity than low buffer REITs remain intact. Panel C provides strong support to our claim that the deviation from debt capacity is driven by high buffer REITs (Top 10% buffer) that chose to issue equity instead of by low buffer REITs (Bottom 10% buffer) that chose to issue debt.

Notes. We re-estimated our results using the same set of control variables in Table 2. Models 1-2 were estimated with the logistic regression method (fixed/random/pooled) while Model 3 was estimated with panel regression (fixed/random/pooled). T-statistics that appear in parentheses. *, **, *** indicate significance at the 10%, 5%, and 1% confidence levels, respectively.

This section addresses the question of whether debt conservatism behavior is persistent. Similar in spirit with Lemmon et al. (2008), we extended the lag value of capacity buffer one year at a time up to five years prior to a debt/ equity issuance decision. We expect the impact of capacity buffers on future capital issues/repurchases to be short-lived and weaken as we increased the lag period of the debt buffer variables. Table 4 reports the regression results using the specification in Models 2 and 4 in Table 2 for issuing and repurchasing decisions, respectively. The results show that REITs’ debt-equity issue choices are influenced by their previous debt buffer up to a period of two years. The coefficient for low buffer REITs’ flip to positive in year 5 indicates that debt ratios are converging. The impact of debt buffer in debt-equity repurchases was short-lived where we did not document any significant relationship beyond 1 quarter period.

Notes. We re-estimated our results with lagged value of debt buffer up to five years using the same set of control variables in Table 2. T-statistics that appear in parentheses. *, **, *** indicate significance at the 10%, 5%, and 1% confidence levels, respectively.

Having provided evidence in favor of the debt conservatism behavior, we now present additional tests to rule out the alternative explanations, i.e. financial flexibility, trade-off and pecking order theories.

Notes. The regression models are exactly the same as in Table 2 except that we replaced debt buffer variables with the actual debt ratio. T-statistics that appear in parentheses. *, **, *** indicate significance at the 10%, 5%, and 1% confidence levels, respectively.