ENSAYO DE POTENCIAL A CIRCUITO ABIERTO (OCP)

I evaluate the relative performance of the models by comparing their implication for the marginal likelihood of the common dataset. According to the first row in Table 4.5, the

Table 4.5: Log Marginal Likelihood

Model Variants Marginal Likelihood

Baseline −4573.15

EFP −4884.10

ICC with financial frictions on all types of loan −4771.97

EFP with lending−borrowing spread series −4861.75

ICC with financial frictions on entrepreneurial loan alone −4802.08

Note: The model variants with the same dataset are compared by the marginal likelihood, which is computed using Geweke’s (1999) modified harmonic mean estimator. By contrast, for different models estimated with different sets of financial series I evaluate the marginal data density using a Laplace approximation at the posterior mode. The computations are based on a Monte Carlo Markov chain of length 500,000 for each model.

log marginal likelihood of my Baseline model is −4573.85. The second and third rows show that with financial frictions`a la BGG or GK, the fit of the satellite models decreases significantly. In particular, the marginal likelihood drops roughly 310 and 200 log points for the EFP and ICC models, respectively. The larger deterioration in fit from dropping either of the financial frictions, comes from the model with financial frictions between banker and entrepreneur. Financial frictions on this relationship reduce 110 additional log points to the fit below what is achieved by remaining financial frictions on the household−banker relationship.

Using different proxies for the financial frictions may affect the goodness of fit. Thus I give the two satellite friction models a fair competition chance in terms of fit. Therefore I use observables for the lending−deposit spread instead of the borrowing−riskless wedge in the re-estimation of the EFP model. That attempt adds a little to model fit using entrepreneurial spread. In particular, the marginal likelihood increases roughly by 22 log points. On the other hand, the literature on demand-side financial frictions focuses solely on funding for entrepreneurs, so I force the ICC model to take a fair competition chance in terms of fit by considering the case where only entrepreneurial loans are subject to the moral hazard problem, in the spirit of Gertler and Karadi (2011). That reduces 30 log points compared to the case of all types of loans subject to the moral hazard problem in the ICC model, while it still adds 60 log points to fit beyond the second EFP model. However, all variants of the two satellite friction models achieve lower fit than the Baseline model.

To sum up, three results can be inferred from the analysis of findings in Table4.5. First, it appears that a combination of two types of financial frictions improves the ability of the model from the overall measure of goodness of fit. Second, the incorporation of financial frictions arising in the banking sector results in better performance, if one chooses one of two types of financial frictions. Third, financial frictions on all types of loans show the better performance between the two alternative GK models.

4.3.5 Steady state

I now assess which of the estimated models is the most reliable representation of the two economies by comparing their steady-state properties to the data. Table 4.6 reports se-

lected model variables and ratios evaluated at each model’s posterior mode, along with their empirical counterparts. Overall, the models and the data match well. Two discrep- ancies lie in the inflation rate and the short-term riskless rate, which are lower in the data. This can be explained by the zero lower bound monetary policy for an extended period of time since the latest financial crisis. It is therefore not surprising that the models don’t perform well on these dimensions. Another exception to the goodness of fit is the some- what low ratio of capital stock to GDP in the friction models, which results partly from the effects of financial frictions on capital accumulation, even more so than in the Baseline model due to the enhanced accelerator effect. I deliberately do not include the data’s relevant ratios in computing the posterior distribution of the model parameters because I want to make a comparison between the pure open economy model and the friction models on a level playing field.

The leverage elasticity of external risk premium implied by the mode of my estimated EFP and Baseline models is within the range of 0.04–0.08 in other studies that work with developed countries. Interestingly, the Australian non-financial business sector has a lower leverage multiple and, simultaneously, smaller elasticity of the external finance premium to the leverage multiple compared to the U.S. counterpart, suggesting that U.S. firms rely more on external finance associated with higher costs. This is because with higher leverage the U.S. entrepreneurial sector imposes a greater cost on its banks in the event of default. Since there’s no empirical evidence of the feasible value range of divertible banking asset fraction, it suffices to say that relatively low values are broadly in line with my interpretation of this parameter as a fractional asset diverted for discretionary spending.26