How Effective is Macroprudential Policy? Evidence from Lending Restriction Measures in EU Countries

Contributor Notes

Author’s E-Mail Address: TPoghosyan@imf.org

This paper assesses the effectiveness of lending restriction measures, such as loan-to-value and debt-service-to-income ratios, in affecting developments in house prices and credit. We use data on 99 lending standard restrictions implemented in 28 EU countries over 1990–2018. The results suggest that lending restriction measures are generally effective in curbing house prices and credit. However, the impact is delayed and reaches its peak only after three years. In addition, the impact is asymmetric, with tightening measures having weaker association with target variables compared to loosening measures. The association is stronger in countries outside of euro area and for legally-binding measures and measures involving sanctions. The results have practical implications for macroprudential authorities.

Abstract

This paper assesses the effectiveness of lending restriction measures, such as loan-to-value and debt-service-to-income ratios, in affecting developments in house prices and credit. We use data on 99 lending standard restrictions implemented in 28 EU countries over 1990–2018. The results suggest that lending restriction measures are generally effective in curbing house prices and credit. However, the impact is delayed and reaches its peak only after three years. In addition, the impact is asymmetric, with tightening measures having weaker association with target variables compared to loosening measures. The association is stronger in countries outside of euro area and for legally-binding measures and measures involving sanctions. The results have practical implications for macroprudential authorities.

I. Introduction

Macroprudential policies have gained importance in EU countries, especially following the global financial crisis. The authorities have used them to address externalities associated with two main dimensions of systemic risk: time-series and structural (Claessens 2014; Galati and Moessner 2018). In the time-series dimension, collateralized borrowing generates externalities and facilitates the emergence of boom-bust cycles. Fire sales represent a vivid example: simultaneous deleveraging in bad times by individual borrowers who do not take into account how their behavior collectively affects the entire system may lead to swings in asset prices and credit. In the structural dimension, externalities arise from the financial market structure, such as interconnectedness and size. Systemic risks may arise if financial institutions, especially systemically important ones, do not internalize the impact of their exposures on other financial institutions and the rest of the economy.

While containing the systemic financial risk is the ultimate objective of macroprudential policies, in practice policymakers pursue intermediate targets—such as house prices and credit (IMF 2014; FMF-FSB-BIS 2016; BIS 2018).2 Despite the widespread use of macroprudential instruments in recent years, understanding of their effectiveness is limited. First, these policies have become popular following the crisis and relatively few measures were implemented in individual countries so far. Expanding the analysis to a cross-country sample helps expanding the number of observations but requires exercising care when drawing inferences for individual countries. Second, macroprudential policies rely on multiple instruments to tackle multiple intermediate targets (IMF 2014; 2018). This differentiates macroprudential policies from monetary and fiscal policies, where the number of instruments and targets is smaller. Measuring effectiveness of macroprudential policies is thus more complicated since assessment should be made for a multiple combination of instruments and targets.

Several recent papers have provided cross-country empirical evidence on the effectiveness of macroprudential policies. The results are mixed. One reason is that the samples typically include a large number of heterogeneous countries (EU and non-EU) to expand the number of observations and this can dilute the results. In addition, most studies focus on the short-term—usually one period ahead—effects of macroprudential policies, while in many cases the full impact of the measures takes time to materialize. It is also notable that macroprudential stance is typically measured using indices of macroprudential measures, while policymakers are typically interested in the effects of discretionary tightening and loosening actions. Finally, none of the studies explores whether the effectiveness of macroprudential measures varies across types (e.g., legally-binding measures versus recommendations, measures with and without sanctions).

The purpose of this paper is to fill this gap and evaluate the effectiveness of macroprudential policies in 28 EU countries over the period 1990–2018. We focus on lending restriction measures, such as loan-to-value (LTV) and debt-service-to-income (DSTI) ratios, and assess their dynamic association with house prices and credit for up to 16 quarters. We also check whether the impact varies across different types of measures and country groups. The results suggest that lending restrictions have a significant association with house prices and credit, peaking at -1.5 percent after three years. However, there is asymmetry between tightening and loosening measures, with the former being weaker. There are also notable differences of effectiveness across country groups and types of measures. The results should be interpreted with caution given reverse causality between discretionary macroprudential actions and developments in target variables. Robustness check suggests that the association is stronger (-3 percent) when reverse causality is controlled for.

The remainder of the paper is structured as follows. Section II reviews cross-country empirical literature on the effectiveness of macroprudential policy. Section III discusses the dataset and describes the stylized facts. Section IV presents the estimation methodology and discusses results. The final section concludes.

II. Crosst-Country Empirical Studies on the Effectiveness of Macroprudential Policy

A growing body of cross-country empirical studies attempts to provide evidence on the effectiveness of macroprudential measures. The analysis draws on several databases of macroprudential measures that were put together by various authors for a large number of countries. The databases use official publications or surveys of regulators and central bank officials as sources of information. They cover both bank-based instruments (such as capital buffers, dynamic loan-loss provisioning, concentration limits) and borrower-based instruments (such as LTV, DSTI).

Table 1 lists cross-country empirical studies on the effectiveness of macroprudential policies, including dependent variables (intermediate targets), macroprudential tools, empirical methodology, sample period, and key results. As shown in the table, most commonly used target variables are house prices and various forms of credit (including total, bank, mortgage, and household). Most commonly used macroprudential instruments are various forms of lending restrictions (LTV, DSTI), but some studies also analyze the impact of capital buffers, reserve requirements, taxes on financial institutions, and dynamic loan-loss provisioning, among others.

Table 1.

Cross-Country Empirical Studies on the Effectiveness of Macroprudential Policies

article image
Source: IMF staff literature review.Note: Abbreviations used in the table: CCG – ceilings on credit growth, CEE – Central and Eastern Europe, CFM – capital flow measures, CG – limits on domestic currency loans, CR – capital ratio limits, DP – time varying/dynamic loan-loss provisioning, DTI – debt-to-income limits, DSTI – debt-service-to-income limits, FC – limits on foreign currency loans, GMM – Generalised Method of Moments, LEV – leverage limits, LIQ – liquidity requirements, LTV – loan-to-value limits, MP – macroprudential, OLS – Ordinary Least Squares, RR – reserve requirements, TAX – levy/tax on financial institutions, VIX – a measure of the implied volatility of S&P 500 index options.

The review of the literature suggests that the following issues complicate the empirical assessment of the effectiveness of macroprudential policies.

  • Insufficient number of measures. Some macroprudential measures have been implemented only recently and in a small number of countries, limiting the number of observations for the empirical analysis. Moreover, for measures implemented only recently is difficult to assess dynamic effects given the lack of sufficient observations on target variables following the implementation.

  • Intensity of measures. The intensity of macroprudential measures is difficult to quantify. For instance, a decrease in LTV by 5 percentage points and increase in the annual amortization requirement by 1 percentage point are both tightening measures, but which of these measures is more “biting” is controversial and depends on a number of factors. Most databases use categorical variables to denote tightening and loosening measures.3

  • Endogeneity. Macroprudential measures are typically implemented in reaction to developments in target variables, such as house prices and credit. This reverse causality biases the coefficient of the macroprudential variable upward.4 As a result, the estimated coefficients are typically interpreted as lower bounds.5 Most studies employ GMM methodology to alleviate the impact of endogeneity.

The evidence on the effectiveness of macroprudential measures is mixed. Some studies find that macroprudential policies are effective in curbing both house prices and credit, while others find that the effectiveness varies for different target variables (Jacome and Mitra 2015). There is also disagreement on the effectiveness of different types of macroprudential instruments: for instance, Fendoglu (2017) finds that borrower-based macroprudential measures are more effective in curbing credit compared to financial-institutions-based measures.

In addition to the challenges mentioned above, the mixed results could be the explained by the following reasons. First, most studies include a heterogenous sample of countries with different levels of development and financial deepening to expand the number of observations. Inclusion of various control variables may not be sufficient to address crosscountry heterogeneity and restricting the sample to a more homogenous group of countries may be warranted. Second, most studies evaluate the effectiveness of macroprudential policies one period ahead and do not assess the dynamic effects.6 Given that transmission from changes in the macroprudential stance to target variables can take time, medium- and long-term effects may differ substantially relative to impact effects. Third, most papers use an index of macroprudential measures that is a cumulative sum of tightening (+1) and loosening (-1) measures implemented from a certain period of time (see, e.g., Cerutti and others 2017; Fendoglu 2017; Akinci and Olmstead-Rumsey 2018). While it allows to proxy the cumulative stance starting from the initial period when the data became available, it does not represent the discretionary change in the policy stance.7 Finally, none of the studies makes a distinction between different types of measures (e.g., legally-binding versus recommended, measures with and without sanctions). Understanding how the effectiveness of macroprudential instruments varies across their types has practical importance for policymakers deciding on the appropriate mix of macroprudential measures.

Our objective is to fill these gaps and assess the effectiveness of lending restriction measures in EU-28 countries using the database of Budnik and Kleibl (2018). We use discretionary changes of lending restriction measures (tightening and loosening) and assess their dynamic effects on house prices and credit for up to 16 quarters. Finally, we assess how the effectiveness varies across different types of measures and groups of countries.

III. Lending Restriction Measures in the Eu: Stylized Facts

This section provides stylized facts on lending restriction measures in the EU using the Budnik and Kleibl (2018) database.8 We restrict the sample to 1990q1-2018q2 and exclude the microprudential measures. We code macroprudential measures as a categorical variable that takes the value of: (i) 1 if a country has implemented a tightening measure in that quarter, (ii) -1 if a country has implemented a loosening measure in that quarter, and (iii) 0 if a country has not implemented any macroprudential measures or implemented measures that had a neutral impact.9

Table 2 presents the list of lending restriction measures implemented in 28 EU countries during 1990q1-2018q2. There are 99 lending restriction measures in total. Most frequently used measures are loan-to-value (41 measures) and debt-service-to-income (20 measures).

Table 2.

List of Lending Restriction Measures Implemented in 28 EU Countries Over 1990–2018

article image
Source: Budnik and Kleibl (2018), IMF staff calculations.

Other important measures include maturity and amortization restrictions, other restrictions on lending standards, and other income requirements for loan eligibility.

Figure 1 presents the breakdown of the measures. Out of 99 measures, 82 are tightening measures and 17 are loosening measures. Most of the measures are legally-binding (54 measures) and do not include sanctions (51 measures). In the empirical analysis we will assess whether the effectiveness varies across these types of measures.

Figure 1.
Figure 1.

Number of Lending Restriction Measures by Types and Across Country Groups

Citation: IMF Working Papers 2019, 045; 10.5089/9781498300872.001.A001

Source: Budnik and Kleibl (2018), IMF staff calculations.Note: Reported are macropmdential lending restriction measures (microprudential measures are not included).

Deployment of tightening measures has picked up following the global financial crisis (Figure 2). By contrast, more loosening measures were implemented before the crisis compared to the post-crisis period. The country-specific distribution of measures (Figure 3) suggests that 18 countries have deployed tightening measures and 9 countries have deployed loosening measures. Tightening measures were particularly frequently deployed in CEEC countries, while the distribution of loosening measures is relatively flat across countries.

Figure 2.
Figure 2.

Number of Tightening and Loosening Lending Restriction Measures: Over Time

Citation: IMF Working Papers 2019, 045; 10.5089/9781498300872.001.A001

Source: Budnik and Kleibl (2018), IMF staff calculations.Note: The vertical red line indicates the start of the global financial crisis (2008q4).
Figure 3.
Figure 3.

Number of Tightening and Loosening Lending Restriction Measures: Across Countries

Citation: IMF Working Papers 2019, 045; 10.5089/9781498300872.001.A001

Source: Budnik and Kleibl (2018), IMF staff calculations.

Figures 4 presents the dynamics of target variables: house price and credit growth.10 The range of growth rates in both variables varies widely across countries, suggesting that country fixed effects should be used to capture country-specific unobserved heterogeneity. Also, both variables have taken a sharp dip following the global financial crisis suggesting that the dynamics of both variables is affected by common factors. The empirical analysis should include time fixed effects to control for these common factors.

Figure 4.
Figure 4.

Dynamics of House Price and Credit Growth

Citation: IMF Working Papers 2019, 045; 10.5089/9781498300872.001.A001

Source: Bank for International Settlements, International Financial Statistics, and IMF staff calculations.Note: The sample includes 28 EU countries. Reported are the median (blue line) and 10–90 percentile interval (grey area). The vertical red line indicates the start of the global financial crisis (2008q4).

IV. Empirical Analysis

In this section, we provide empirical evidence on the relationship between lending restriction measures and target variables (house prices and credit) in 28 EU countries over 1990–2018. We start by assessing the overall effectiveness of lending restriction measures, then provide evidence on the possible asymmetry between tightening and loosening measures, and finally check if the effectiveness varies across country groups and types of measures.

A. Baseline Specification: Do Lending Restrictions Affect House Prices and Credit?

In the first step, we assess the overall effectiveness of lending restriction measures using local projections methods (Jorda 2005). The baseline empirical specification takes the following form:

yi,t+hyi,t1=αih+γth+βhMi,t+Σn=1Nθk,nhXk,i,tn+ϵi,th(1)

where i denotes countries, t denotes time, h=[0, …,16] denotes the projection horizon, N denotes the number of lags, y denotes the log of real credit or real house prices, Mit is the number of lending restriction measures implemented by country i in period t,11 X is a matrix of k lagged dependent and control variables (lending restriction measure, GDP growth, change in monetary policy rate, crisis dummy), and ε is the i.i.d. error term. Regressions include country fixed effects (ai) to control for country-specific unobserved heterogeneity and time fixed effects t) to control for common shocks affecting all countries simultaneously.

The coefficient of interest is βh. It is expected to be negative consistent with the hypothesis that tightening (loosening) of macroprudential measures has been associated with a reduction (increase) in house prices and credit in quarters that followed up the measure.

Figure 5 presents the estimates of coefficient βh for the baseline specification (see also Tables 45). For both house prices and credit, the coefficients are largely negative. This is consistent with the effectiveness hypothesis and suggests that target variables have tended to decline following implementation of macroprudential measures relative to a no-implementation scenario. However, coefficient estimates are imprecisely estimated in the near term and become significant only after three years (quarter 12), peaking at -1.5 percent. The weaker association and significance in the near term could be due to the upward bias mentioned above but could also indicate that the impact of the measures takes time to materialize. Another important caveat is associated with the effectiveness of the measures over time, since most of measures were implemented following the global financial crisis (Figure 2).

Figure 5.
Figure 5.

Results: Response of Target Variables to Lending Restriction Measures

Citation: IMF Working Papers 2019, 045; 10.5089/9781498300872.001.A001

Source: Bank for International Settlements, International Financial Statistics, Budnik and Kleibl (2018), and IMF staff calculations.Note: Reported are βh coefficients from baseline specification ([1]). Filled circles indicate significance at 10 percent confidence level (robust standard errors). Measures are implemented in period 0.
Table 3.

Variables and Their Sources

article image
Table 4.

Estimation Results: Baseline Specification (House Prices)

article image
Source: Bank for International Settlements, International Financial Statistics, Budnik and Kleibl (2018), Lo Duca and others (2017), and IMF staff calculations.Note: Estimations are performed using the local projections methodology. Robust standard errors are in brackets. *, **, and *** denote significance at the 10, 5, and 1 percent levels, respectively.
Table 5.

Estimation Results: Baseline Specification (Credit)

article image
Source: Bank for International Settlements, International Financial Statistics, Budnik and Kleibl (2018), Lo Duca and others (2017), and IMF staff calculations.Note: Estimations are performed using the local projections methodology. Robust standard errors are in brackets. *, **, and *** denote significance at the 10, 5, and 1 percent levels, respectively.

Lagged control variables have expected signs: (i) changes in monetary policy rates have a negative lagged association with house prices and credit, (ii) real GDP growth has a positive association with house prices and credit, and (iii) crisis dummy has a negative association (except 2 lags in credit regressions).

B. Is the Impact Symmetric Across Tightening and Loosening Measures?

After establishing association between lending restriction measures and target variables, we would like to assess whether this association is symmetric across tightening and loosening measures. We use the following empirical specification to address this question:

yi,t+hyi,t1=αih+γth+βMThMTi,t+βMLhMLi,t+Σn=1Nθk,nhXk,i,tn+ϵi,th(2)

where the main difference from the baseline specification is that we introduce two dummies reflecting tightening and loosening lending restriction measures; MT takes the value 1 for tightening episodes and 0 otherwise, while ML takes the value 1 for loosening episodes and 0 otherwise.

The coefficients of interest are βhMT and βhML: the former is expected to be negative (tightening measures are associated with a decrease in house prices and credit), while the latter positive (loosening measures are associated with increase in house prices and credit).

Figure 6 presents the estimates of coefficients βhMT and βhML for this specification (see also Tables 67). For both target variables, there is evidence of asymmetry: loosening measures tend to have a stronger association compared to tightening measures.12 This asymmetry could be driven by leakages due to regulatory arbitrage that tend to hamper the effectiveness of tightening measures but do not affect the loosening measures (BIS 2018). The leakages in response to tightening measures could occur through a shift of customers to: (i) non-bank credit institutions that are not subject to the same level-playing field in terms of macroprudential regulation as banks (Reinhardt and Sowerbutts 2015), or (ii) foreign bank branches that are subject to macroprudential regulation of home authorities (Aiyar and others 2014).13 Nevertheless, the results should be interpreted with caution given a relatively small number of observations for loosening measures.

Figure 6.
Figure 6.

Results: Asymmetric Response of Target Variables with Respect to Tightening and Loosening Measures

Citation: IMF Working Papers 2019, 045; 10.5089/9781498300872.001.A001

Source: Bank for International Settlements, International Financial Statistics, Budnik and Kleibl (2018), and IMF staff calculations.Note: Reported are βhMT and βhML coefficients from specification ([2]). Filled circles indicate significance at 10 percent confidence level (robust standard errors). Measures are implemented in period 0.
Table 6.

Estimation Results: Asymmetric Effects of Tightening and Loosening Measures (House Prices)

article image
Source: Bank for International Settlements, International Financial Statistics, Budnik and Kleibl (2018), Lo Duca and others (2017), and IMF staff calculations.Note: Estimations are performed using the local projections methodology. Robust standard errors are in brackets. *, **, and *** denote significance at the 10, 5, and 1 percent levels, respectively.
Table 7.

Estimation Results: Asymmetric Effects of Tightening and Loosening Measures (Credit)

article image
Source: Bank for International Settlements, International Financial Statistics, Budnik and Kleibl (2018), Lo Duca and others (2017), and IMF staff calculations.Note: Estimations are performed using the local projections methodology. Robust standard errors are in brackets. *, **, and *** denote significance at the 10, 5, and 1 percent levels, respectively.