RDP 2022-05: The Real Effects of Debt Covenants: Evidence from Australia 4. Indirect Effects of Covenants – Monetary Policy Transmission

I now turn to how debt covenants can amplify or mitigate the transmission of monetary policy to real business activity. This may differ based on the type of covenant, given interest rate changes may or may not change the associated constraint.

Changes in interest rates directly affect (variable-rate or floating) interest payments on outstanding debt and, in turn, the interest coverage ratio. That then moves firms closer to or further from the threshold associated with their ICC. Therefore, for firms subject to ICC, how much debt can be issued without violating these covenants and how much scope the firms have to expand and take on other expenses is highly sensitive to changes in monetary policy.

Conversely, changes in interest rates may have minimal effects on ratios used for other types of covenants: for OEC there may be some indirect effect via earnings, while for ABC there will be little to no effect since the ratios are measured with book values to avoid feedback from market prices of assets. If firms with such covenants are already constrained, changes in interest rates may have little effect on their behaviour. As such, OEC and ABC may dampen the effects of monetary policy.

To empirically test the indirect effects of debt covenants, I modify Greenwald's (2019) framework and split firms into three categories: subject to ICC, not subject to ICC but to either OEC or ABC or both (NICC), and the residual group of firms not subject to any covenants (NC).[1] To measure the heterogeneous transmission across different covenant configurations, I employ the panel local projections method of Jordà and Taylor (2016) and estimate the following regression for dependent variable y and horizon h:

(4) Δ h y i,t+h = α i h + Cov I i,t1,Cov ( β 0,Cov h + β 1,Cov h ε t )+( γ 0 h + γ 1 h ε t ) X i,t1 + θ h t + k δ k h ε tk + η i,t+h

where Δ h y i,t+h = y i,t+h y i,t1 is the response of the variable of interest (log investment and log staff expenses) in year t + h to a monetary policy shock in year t, notated as ε t . The categorical variable Ii,t–1,Cov indicates the firm's covenant configuration in year t–1, which is either (1) ICC, (2) NICC or (3) NC.

Similar to the empirical model employed to study the direct effects of covenants, I control for firm-level time-varying financial measures, firm fixed effects and time trends. As suggested by the varying financial statistics across covenant groups, selection into different covenants is potentially non-random. That is, firms whose business strategies expose them to changes in interest rates differently may select into different covenants. For instance, large firms in capital-intensive industries with more volatile streams of earnings are often subject to interest coverage limits. However, they are also more likely to invest in high-yielding projects that face fiercer credit rationing and, in turn, are more likely to be particularly sensitive to changes in interest rates (Barea Lugo 2006). To address this potential endogeneity issue, I include interactions between the firm-level controls (including firm size and firm sector) and monetary policy shocks. This allows for firms with different financial situations and in different industries to react differently to the same shocks and, to the extent that selection is correlated with my controls, alleviates the selection bias. Since business activity is likely to respond to past monetary policy, controlling for the lagged measures of monetary shocks ε tk can help improve the precision of my estimates.[2]

Another key departure from Greenwald (2019) is that instead of changes in the cash rate, I employ the newly available series of monetary policy shocks constructed by Beckers (2020). This is because changes in the cash rate are not exogenous, but instead reflect systematic responses to economic conditions, which may have confounding effects on business activity. This is concerning in my case as ICC and OEC are affected by earnings, which depend on aggregate economic conditions, while ABC are not. Therefore, the effects of monetary policy need to be separated from macroeconomic and financial conditions. The monetary policy shocks are constructed following the method of Romer and Romer (2004), but augmented to control for the systematic response of the central bank to financial conditions, as represented by credit spreads (as in Caldara and Herbst (2019)). They are then purged of financial market expectations for the cash rate to ensure that changes in monetary policy are plausibly exogenous to and unanticipated by private agents. I aggregate the shocks up to an annual frequency to match with the annual data on covenants and finances (Appendix E).

The primary statistics of interest are the differences between the coefficients attached to monetary policy shocks on firms of different covenants configurations ( β 1,Cov h ) , for each horizon h :

D ICCNC h = β 1,ICC h β 1,NC h D NICCNC h = β 1,NICC h β 1,NC h D ICCNICC h = β 1,ICC h β 1,NICC h

In other words, the statistics measure the responses of firms subject to a particular covenant configuration relative to otherwise-equivalent firms subject to a different covenant configuration.

The left panel of Figure 7 plots the differential coefficients for investment up to four years after an expansionary monetary policy shock. While imprecise, the estimates suggest a role for ICC in amplifying transmission of monetary policy to investment; firms subject to ICC respond more strongly to shocks relative to firms not subject to any covenants. In contrast, responses of NICC firms appear weaker than otherwise equivalent non-covenants firms, suggesting that other types of covenants dampen the effect of monetary policy on investment. Similarly, transmission of monetary policy to firms' staff expenses is amplified by ICC and mitigated by NICC (right panel).

Figure 7: Differential Responses by Covenant
To a 100 basis point expansionary monetary policy shock

Notes: Measured as the differences between the coefficients attached to each covenant configuration dummy interacting with monetary policy shock. Dashed lines depict 95 per cent confidence intervals.

Sources: Author's calculations; Beckers (2020); Connect4; Morningstar


I exclude firms not specifying the exact types of covenants. These firms share virtually similar financial and non-financial characteristics as firms reporting the types of covenants, suggesting little concern for bias in sample selection. [1]

I employ k = 2 but results are robust using fewer or more lags. [2]