Model Estimation Results

Table 4.5 reports ordinary least squares regression results based on the estimated empirical model. Since some of the project variables (cost, maturity, tranche number, and foreign currency) are simultaneously determined with the gearing ratio, specifications with and without these variables are considered. All regressions include sector fixed effects. Standard errors are clustered at the syndicate level to account for correlation among projects financed by the same syndicate of banks.9

Table 4.5: Drivers of Project Finance Deals in Asian Markets

FE = fixed effects, GDP = gross domestic product, LN = natural logarithm, MDB = multilateral development bank, NPL = nonperforming loan, PPP = public-private partnership, ROAA = return on average assets.

Notes:

1.  The table presents ordinary least squares regression results to examine the drivers of project finance deals. The sample includes 244 projects financed from 2011 to 2016 in India, Indonesia, Malaysia, the Philippines, the Republic of Korea, Thailand, and Viet Nam.

2.  The dependent variable is the gearing ratio.

3.  Standard errors (in parentheses) are clustered at the syndicate level to account for correlation among projects financed by the same syndicate of banks.

4.  The lower of part of the table reports the R2 decomposition for groups of variables (Shapley values, %).

*** p < 0.01 ** p < 0.05 * p < 0.10

Source: Author's estimates.

Column (1) in Table 4.5 shows regression results using only syndicate characteristics; column (2) augments the explanatory variables by including project characteristics. The results show a positive and significant association with the tier 1 ratio, the ratio of liquid to total assets, and the ratio of cost to income, while other variables are broadly insignificant. Banks with solid capital or liquid asset bases, as well as less-efficient banks, are more willing to lend to project finance infrastructure PPP transactions.

One explanation of why banks with higher ratios of cost to income lend more is that they may be looking to increase their performance by lending larger amounts, especially as administrative costs tend to be fixed regardless of deal size. Indeed, infrastructure projects provide an opportunity to increase the average size of single loan transactions. The coefficient estimates in columns (1) and (2) imply that a one-standard deviation increase in the (i) tier 1 ratio increases the gearing ratio by 2.9%-3.5%, (ii) ratio of liquid to total assets increases the gearing ratio by 3.8%-4.4%, and (iii) ratio of cost to income increases the gearing ratio by 3.3%-4.4%.

Column (2) shows some evidence that larger projects are less leveraged. Sector fixed effects-estimated from model specifications without the constant term in column (1)-show cross-sector heterogeneity in average gearing ratios of 29.96% for mining, and oil and gas; 33.03% for industry; 36.02% for transport; and 38.39% for energy. Projects outside these sectors are more leveraged, and have an average gearing ratio of 45.91%. All sector fixed effects are significant at the 5% level. Sector fixed effects in column (2) line up consistently with those in column (1), whereby projects in mining, and oil and gas, are the least leveraged (55.70%), and those in other sectors are the most leveraged (71.77%).

Table 4.5 also shows the R2 decomposition for the groups of variables (Shapley values). There is a substantial portion of gearing ratios that country, project, and syndicate characteristics are unable to explain, since quarter fixed effects account for about one-third of the variability in gearing ratios for the most complete specification. Sector fixed effects, however, do not appear to be important. Overall, the financing structure of the sample projects depends substantially on both observable and unobservable country characteristics. Bank balance sheets constitute the third most important source of variation in project finance gearing ratios and about 17% to the overall R2.

The analysis then assesses the robustness of these results to the inclusion of time trends common to project financing and country variables. In columns (3) and (4), quarter fixed effects are incorporated to the specifications in columns (1) and (2). Point estimates and their significance are consistent with those in columns (1) and (2). In columns (5) and (6), time-varying proxies for economic and institutional conditions are included, together with country fixed effects. For the country variables, results show that the gearing ratio depends positively on GDP per capita, which shows that project finance tends to be higher in larger markets where demand and purchasing power are higher; this is in line with Hammami, Ruhashyankiko, and Yehoue (2006).

The regression results also show a negative dependence between gearing ratio and government indebtedness, which can proxy for sovereign default risk. Banks may be unwilling to provide long-term funding under these conditions. Moreover, gearing ratios are positively associated with inflation risk (proxied by the volatility of inflation rates), though these are unaffected by economic and currency risks. Initially, this evidence may be counterintuitive. This is because of the long-term nature of project finance investments, which make them highly exposed to inflation risk. As such, one would expect gearing ratios and inflation risk to be negatively-rather than positively-related. Two explanations are offered for this result. First, it may well be that interest rates on project finance debt are floating or inflation indexed in which case the inflation risk gets mitigated. Second, anecdotal evidence shows that many project finance transactions are exposed to inflation, both for revenue and operational costs, which implies a natural hedge against inflation. But this reasoning is only suggestive, because information is lacking on loan rates and the profit and loss structure of projects.

The results show no evidence that political stability and institutional quality affect debt investment in project finance deals. One explanation is that the governance indicators used show limited time variation at the country level, and therefore country fixed effects absorb much of the cross-country variation.10

The introduction of country variables gives a different picture on the relevance of bank balance sheets. The tier 1 ratio continues to be positively associated with project gearing ratios and with a relatively stable coefficient. Liquid assets and ratios of cost to income, however, are no longer statistically significant. Since most projects are funded by local banks, this suggests that asset liquidity and cost efficiency tend to move together with local economic conditions. Columns (5) and (6) of Table 4.5 also suggest that banks with larger loan portfolios provide less funds to project finance deals. On the economic magnitude of this effect, a one standard deviation increase in the ratio of loans to assets decreases the gearing ratio by 3.1%-3.2%. Overall, no project variable appears to be a driver of the gearing ratio at standard significance levels.