Finance, growth and (macro)prudential policy: European evidence

Abstract

This paper examines the interactions between financial development, economic growth and (macro)prudential policy on a sample of 12 euro area countries. Our main takeaway is that active (macro)prudential policy supports the positive finance-growth nexus instead of disrupting it. These benefits are found to be more likely to materialize during tightening of (macro)prudential policy measures and not during easing. This result is conditional on the ability of (macro)prudential policy to curb excess credit growth and mitigate systemic risk, which would otherwise disrupt the market. Moreover, we assert that when analysing the effects of (macro)prudential policy, it is important to account for the direction of (macro)prudential measures, not just for the frequency at which they are implemented.

Appendix

1.1 Appendix 1: Data

In order to provide more information, Table 8 reports all our variables, their summary statistics and their data sources.

Table 8 Summary Statistics

1.2 Appendix 2: Monetary conditions index estimation procedure

There are a number of approaches that allow a large number of time series to be combined into a single composite index. In the case of the indicator presented in this paper, we use a factor model estimate. Consider an n-dimensional vector of stationary observable variables $X={\left({X}_{1},\dots ,{X}_{n}\right)}^{^{\prime}}$ that are linearly dependent on an m-dimensional vector of originally unobservable factors $F=\left({F}_{1},\dots ,{F}_{m}\right)$. The baseline factor model then takes the following form:

$${X}_{t}=\Lambda {F}_{t}+{\varepsilon }_{t}, where{F}_{t}={\sum }_{i=1}^{p}{A}_{i}{F}_{t-i}+{u}_{t},$$

(1B)

where $\Lambda$ is a matrix of factor loadings, ${A}_{i}$ is a matrix of autoregression coefficients for $p$ lags and ${\varepsilon }_{t},{u}_{t}$ are i.i.d. Gaussian error terms. We use the maximum likelihood method to estimate the factor model. While more complicated to calculate, the maximum likelihood method, unlike the principal components method, makes it possible to test whether the number of common factors selected is sufficient. The optimal number of factors to estimate is primarily based on parallel analysis. The optimal number of lags is chosen based on the Schwarz information criterion. For our data, results of statistical tests prefer a factor model with 3 estimated factors and 1 lag. The robustness and sensitivity analysis of the selected model specification to calculate MCI was performed with respect to the number of lags used, number of factors estimated, the estimation period, and the variables included in the estimation.

Table 9 summarizes the set of 15 variables that reflect the monetary conditions in the euro area. Variables in respective blocks were treated as follows: (1) interest rates enter the estimation in levels; (2) monetary aggregates are expressed in year-on-year change and in reciprocal values (switched sign) so that that an increase would correspond to a monetary tightening, as for interest rates; (3) ECB balance sheet items are expressed in year-on year change with a negative sign for all these variables; and (4) exchange rate is transformed into a year-on-year change with the sign left unchanged.

Table 9 Dataset Used to Extract the Factor Loadings. Note: Data covers the period from January 1999 to December 2018 and was extracted from the Thomson database (time series codes are available in brackets), except for the nominal exchange rate, which is

To save space, we do not report all the robustness checks performed; they are available upon request. Figure 

Fig. 4

Monetary Conditions Index for the Euro Area. Note: In order to calculate the synthetic indicator, we weigh the sum of the three factors (with weights given by the percentage of the overall data variability explained by each factor, i.e., 44%, 34%, and 22%)

4 (left-hand graph) shows the relative contribution of each of the estimated factors to the final index. The figure also plots the MCI as normalized using the mean and standard deviation of the 3-month EURIBOR. The right-hand graph shows results of a simulation exercise in which the index was estimated multiple times, each time with one variable excluded from the input data set. This approach is very similar to a more formal bootstrapping proposed by Gospodinov and Ng (2013).

1.3 Appendix 3: Dating of financial crises in the Euro area

In identifying crisis-type events, we rely on a financial crises database for European countries maintained by the ESRB. The dates, presented in Table 10, were selected by the ESRB based on a quantitative identification approach, which had been cross-checked with expert judgement from national and European authorities (qualitative approach). This expert judgement was sought by the ESRB whenever the adopted quantitative approach did not identify event dates included in Laeven and Valencia (2013) and Babecky et al. (2014). In such cases, the dates were submitted to national authorities for revision in order to assess the most appropriate dates to be included in the dataset.

Table 10 Episodes of Systemic Crises According to the ESRB Database

1.4 Appendix 4: Additional estimates

Section 4 discussed several additional empirical specifications. Results are reported below in Tables 11, 12, 13. The first reports estimates of the interaction term from Eq. 2 using richer lag structure. The second contains estimates of the interaction term from Eq. 2 for various sample perturbations. The final table reports the estimates of the model as specified in Eq. 2 using alternative estimator (Table 14).

Table 11 Richer Lag Structure of the Interaction Term. Note: Robust standard errors are in parentheses. Statistical significance at the 1%, 5% and 10% level is indicated by ***, ** and * respectively

Table 12 Sample Sensitivity. Note: Robust standard errors are in parentheses. Statistical significance at the 1%, 5% and 10% level is indicated by ***, ** and * respectively

Table 13 Financial Development and Economic Growth Estimated via BBBC Estimator. Note: Bootstrapped standard errors are in parentheses. Statistical significance at the 1%, 5% and 10% level is indicated by ***, ** and * respectively

Table 14 Finance-Growth Nexus with (Macro)Prudential Policy Estimated via BBBC Estimator. Note: Bootstrapped standard errors are in parentheses. Statistical significance at the 1%, 5% and 10% level is indicated by ***, ** and * respectively

1.5 Appendix 5: The index of macroprudential authority strength

We use the ESRB (2014) assessment of the implementation of the ESRB's Recommendation on the macro-prudential mandate of national authorities to set up a macroprudential authority strength index. In 2014, the ESRB evaluated the degree to which EU member states are compliment with the Recommendation. We are interested in the assessment related to the Sub-recommendation B.3. which requires that the central bank plays a leading role in macro-prudential policy, given their institutional and functional strengths.

Table 15 summarizes the extent of the central bank’s role in the macro-prudential policy of individual countries. The ESRB grades countries according to their efforts to implement the ESRB Recommendation on a zero to one scale (0 = Non-compliant; Materially non-compliant = 0.25; 0.5 = Partially compliant; 0.75 = Largely compliant; 1 = Fully compliant). Table 16 show the standards as set by the ESRB regarding grading the Sub-recommendation B.3.

Table 15 Institutional Framework of the National Macroprudential Authority in EA-12 Countries.

Table 16 ESRB Implementation Standards.

Figure 5 shows the index values for our sample countries. Since the ESRB performed its assessment in 2014, we re-do the assessment for 2017 to check for any changes in the institutional setup.

Fig. 5

Source: ESRB (2014); assessment for the year 2017 was performed by the authors

Overview of $INS{T}_{i,t}$ Values for the Sample of Countries.