Macroprudential Policies and Housing Price
A New Database and Empirical Evidence for Central, Eastern, and Southeastern Europe
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Several countries in Central, Eastern and Southeastern Europe used a rich set of prudential instruments in response to last decade’s credit and housing boom and bust cycles. We collect detailed information on these policy measures in a comprehensive database covering 16 countries at a quarterly frequency. We use this database to investigate whether the policy measures had an impact on housing price inflation. Our evidence suggests that some—but not all—measures did have an impact. These measures were changes in the minimum CAR and non-standard liquidity measures (marginal reserve requirements on foreign funding, marginal reserve requirements linked to credit growth).

Abstract

Several countries in Central, Eastern and Southeastern Europe used a rich set of prudential instruments in response to last decade’s credit and housing boom and bust cycles. We collect detailed information on these policy measures in a comprehensive database covering 16 countries at a quarterly frequency. We use this database to investigate whether the policy measures had an impact on housing price inflation. Our evidence suggests that some—but not all—measures did have an impact. These measures were changes in the minimum CAR and non-standard liquidity measures (marginal reserve requirements on foreign funding, marginal reserve requirements linked to credit growth).

I. Introduction

Despite much interest among policymakers at the global level since the onset of the recent financial crisis, the econometric evidence on the effectiveness of macroprudential policies (MPPs) available to date is limited, as Galati and Moessner (2011) point out in their recent survey. In Central, Eastern and South-Eastern Europe (CESEE), a significant number of countries went through large and synchronized credit and housing boom-bust cycles during the last decade and macroprudential policies were actively used, thus the region seems fertile ground for an investigation of the effectiveness of these policies.2 In some CESEE countries policymakers did not attempt to curb credit expansion through macroprudential policies while in others many instruments were deployed, including capital requirements, loan classification and provisioning rules, reserve or liquidity requirements, and credit eligibility criteria.3 In some cases, policies were tightened late, when the cycle had already turned. In others yet, policies were relaxed during the expansion for exogenous reasons, notably the pressure or desire to harmonize regulation upon joining the European Union. When policymakers took action, they did it through different instruments and with different intensity. This experimentation probably reflected different macroeconomic conditions and institutional settings, but also, possibly, the lack of a well-established rulebook for the use of macroprudential policies. In any case, to the advantage of the researcher, the experience of the CESEE is very rich in terms of policy actions. Our objective in this paper is to contribute to the policy debate on the usefulness macroprudential policies by exploiting this rich regional experience using a systematic and quantitative approach to the assessment of the effectiveness of macroprudential policy tools.

An important contribution of our paper is the construction of a comprehensive database at a quarterly frequency of all the major prudential measures—grouped into 29 categories—that were adopted in sixteen CESEE countries from the late 1990’s or early-2000’s to end-2010. 4To the best of our knowledge, information at this level of detail in a cross-section of countries has not been available to date and we hope that this effort will be useful to future researchers. In addition, for the purposes of our own quantitative analysis, we also devise scoring rules to quantify each measure’s intensity over time and across types.

The specific question we ask in this paper is whether MPPs were a significant determinant of housing price inflation in CESEE during the last decade. The reason for our focus on housing price inflation as a source of systemic risk is twofold. First, a large literature (summarized recently in Crowe et al., 2011) emphasizes the dangers of asset price bubbles and the linkages between housing booms and financial instability episodes. The amplitude of the housing cycle in the CESEE region was spectacular, with countries such as the three Baltic countries (Estonia, Latvia, and Lithuania) witnessing housing price inflation in the range of 120–160 percent between the first quarter of 2004 and the first quarter of 2007. Second, focusing on housing prices rather than domestic credit (as has been done in some of the literature) allows us to avoid a significant measurement problem. Because foreign-exchange denominated or indexed loans are very common in the CESEE region, changes in the stock of credit (expressed in domestic currency) are strongly affected by valuation effects associated with exchange rate movements.5 Unfortunately, the currency breakdown of domestic credit aggregates which is necessary to correct for these valuation effects is available only for some countries or short time periods. Thus, truly meaningful series of quarterly real credit growth are not widely available about half of the countries we are focusing on. At the same time, we acknowledge that housing price data also have drawbacks, such as uneven quality and cross-country comparability as well as, for some countries, short time series. We also acknowledge that demand by foreign investors was significant in some market segments in several CESEE countries during the boom years and therefore that in those cases housing price dynamics responded to some extent to shifts in foreign investors’ interest and access to foreign financing.6,7

In line with the empirical literature (e.g., Malpezzi, 1999, Capozza et al., 2002, Egert and Mihaljek, 2007), we model housing price dynamics using an error correction model in which a long-run relationship between housing prices and output per capita exists. It turns out the estimated elasticity we obtain from the regressions is about one, making our model equivalent to one where a measure of housing affordability - the ratio of housing prices to income per capita- is included as a determinant (as in Igan and Loungani, 2012). As for the short-run, our evidence suggests that one type of capital measure (changes in the minimum capital adequacy ratio (CAR) as well as two types of non-standard liquidity measures (changes in marginal reserve requirements on foreign funding and changes in marginal reserve requirements linked to credit growth) had an impact on housing price inflation. The economic significance of their effect is meaningful. For example, a change by one percentage point of the minimum CAR has on average a cumulative effect of 8.5 percent on housing prices after four quarters. This compares with mean quarterly real housing price inflation of 0.92 percent in our sample. We do not find robust evidence that changes in standard average reserve requirements, provisioning rules, or eligibility criteria (loan-to-value ratio, debt-service-to-income ratio) had any significant effect.

We also study whether the effect of each of the three types of MPPs mentioned above was different depending on whether the policies were tightened rather than loosened or depending on whether the change in policy occurred during the expansionary phase of the cycle rather than during the contraction. We find that the three policies had a significant impact when tightened and that only changes to the minimum capital adequacy ratio had a significant impact when eased. The four policies had a strong impact during the boom years, while the impact during the bust was less robust.

In interpreting these results, it is important to recognize the limitations of our methodology. In particular, the endogeneity of the policy measures to macro-financial developments—for example if policymakers tighten MPPs in anticipation of an increase in housing price inflation—is likely to bias the estimates of policy impact downwards, leading us to conclude that some measures were ineffective. In addition, some measures may be calibrated so as not to be immediately binding, so their effect may be discernible only after several quarters. Finally, measures may have been anticipated and their effects may have occurred before the implementation date. These are limitations common to most studies that do not rely on “clinical experiments” for policy evaluation, and they certainly apply to our paper as well.

Among the few recent contributions to the econometric literature on the effectiveness of MPPs, some are more supportive of average reserve requirements, provisioning rules and eligibility requirements than ours. Tovar et al. (2012) find that average reserve requirements and a composite of other types of macroprudential policies had a moderate and transitory effect on credit growth and played a complementary role to monetary policy rates in a panel of five Latin American countries during 2004–11. Jiménez et al. (2012) find that dynamic provisioning requirements in Spain helped smooth the credit cycle and supported credit supply in bad times. Igan and Kang (2011) find that the adoption of maximum loan-to-value (LTV) and debt-service-to-income (DTI) ratios in Korea in the second half of the 2000s was successful in slowing down housing price inflation and the growth of transaction volumes. Craig and Hua (2011) find that curbs on LTVs and stamp duties on property transactions helped slow down property price inflation in Hong Kong S.A.R. Wong et al. (2011) offer evidence of LTV effectiveness in reducing delinquencies after property busts in a few Asian economies (including Hong Kong S.A.R.). Lim et al. (2011) find that several instruments (LTV, DTI, credit growth ceiling, foreign currency lending ceiling, reserve requirements, dynamic provisioning, countercyclical capital requirements) reduce the procyclicality of credit and/or bank leverage in a panel of 49 countries during 2000–10. Dell’Ariccia et al. (2012) construct a composite measure of six MPPs (differential treatment of deposit accounts, reserve requirements, liquidity requirements, interest rate controls, credit controls, and open foreign exchange position limits) and find that stricter MPPs reduce the incidence of credit booms and decrease the probability that booms end badly.

In the CESEE region, the only two available econometric studies focus on Croatia. Galac (2010) finds that credit growth limits (i.e. marginal reserve requirements related to credit growth) were successful in reining in domestic private sector credit growth but that they did not reduce total private sector credit growth because domestic credit was substituted by cross-border credit. Kraft and Galac (2011) fine-tune Galac’s analysis by breaking down the private sector into households and corporations and find that the credit growth limits were effective in slowing down household credit, but not corporate debt (because of the circumvention through cross-border loans). Both papers also find that marginal reserve requirements on foreign funding were instrumental in building banks’ capital buffers. Our finding about the effectiveness of marginal reserve requirements on foreign funding and marginal reserve requirements linked to credit growth is therefore consistent with these analyses.

The effectiveness of MPPs has also been studied using the event analysis methodology or through narratives. The results of Tovar et al. (2012)’s and Lim et al. (2011)’s event analyses are consistent with the econometric results mentioned above. Pereira da Silva and Eyer-Harris (forthcoming) find that making risk-weights on certain types of consumer loans contingent on loan-to-value and maturity had the desired effect on the flow, maturity and interest rates of these loans in Brazil. Terrier et al. (2011) describe a wide variety of MPP instruments that have been used in Latin America without systematically analyzing their effectiveness. As to the CESEE region, a series of World Bank Policy Research Working Papers published in 2011 describes the experience with macroprudential policies of the Czech Republic, Estonia, Macedonia, Poland, and Turkey.8, 9 These papers generally argue that macroprudential policies implemented during the boom helped improve the resilience of the banking system during the bust. Dimova, Kongsamut, and Vandenbussche (forthcoming) analyze through a large number of event analyses the experience of the four Southeastern European countries (Bulgaria, Croatia, Romania, and Serbia) that were most active in using MPPs in the CESEE region. Their conclusions are consistent with ours and those of Kraft and Galac (2011): the strictest measures—including credit growth limits and strict capital ratios—had a noticeable impact on credit growth, the composition of credit and/or housing prices.

The rest of the paper is structured as follows. The next section contains a description of the housing price and MPPs data. Section III presents the empirical model, regression results are discussed in Section IV, and Section V concludes. Two appendices contain further details on data sources and scoring rules used to quantify the intensity of prudential policy measures.

II. A First Look at the Data

In this section, we preview the main data series used in the empirical analysis and explain what the MPP database covers and how it was constructed.

A. Housing Prices

We compile housing prices data from the BIS, national statistical offices, local and international real estate companies, and the Central Bank of Albania. All in all, we manage to gather quarterly housing price series for 16 CESEE countries covering different time periods, generally beginning in the early 2000s.10 When several data series are available for one country, we choose the longest one.11 The series are not fully harmonized across countries as they sometimes cover different types of residential real estate or different geographical entities within a country, but this is the only way to have a reasonable coverage along both the cross-country and the time dimensions. In our econometric analysis below, the inclusion of country fixed effects will help deal with possible concerns raised by this cross-country heterogeneity in types of real estate. We deflate all nominal series with the national CPI and then seasonally adjust all real housing prices series. Details on data availability, sources, and coverage are provided in Table A1 in Appendix 1.

Real housing prices developments differed substantially across countries in the CESEE region over the sample period. While our data show a pronounced boom and bust cycle over the last decade in the Baltic countries and Ukraine, real house price inflation was more contained in other countries such as Croatia, Serbia, and Slovenia (Figure 1).

Figure 1.
Figure 1.

Selected CESEE Countries: Seasonally-Adjusted Real Housing Price Index, 1997:Q1–2011:Q1

Citation: IMF Working Papers 2012, 303; 10.5089/9781475550139.001.A001

Sources: BIS housing price statistics, Global Property Guide, Central Bank of Albania, FHB, REAS, Reidin, IFS, and authors’ calculations.

B. Fundamental Macroeconomic and Demographic Variables

Following the literature, we hypothesize that the three fundamental variables driving real housing prices are real income per capita, real interest rates, and working-age population. Because foreign currency lending is widespread in most of the countries in our sample, we include both a domestic currency interest rate and a foreign currency effective interest rate. Most macroprudential policies are expected to affect lending rates through intermediation spreads; therefore we use interest rate variables that are related to the liability side of banking systems’ balance sheets rather than lending rates. Since some countries in our sample do not have a monetary policy rate (e.g. because they have a currency board arrangement), we use the domestic deposit rate as our measure of the domestic currency interest rate. For the foreign currency interest rate, we use the Fed Funds rate in countries that are partially dollarized (Russia, Turkey, and Ukraine) and the ECB policy rate in all others. Swiss franc mortgages are widespread in Croatia, Hungary, and Poland, but we do not add a second foreign currency rate in the regressions in order to economize on degrees of freedom. To construct our effective interest rate, we adjust the series by the year-on-year appreciation of the local currency against the dollar or the euro, as applicable.12 We seasonally adjust the GDP per capita series, as we did for the housing price series. Demographic variables are often included as determinants of housing prices in the literature. Following Igan and Loungani (2012), we use the year-on-year change in working-age population. Some authors have included mortgage credit growth (or total credit growth) in the list of determinants of housing price inflation. However, because prudential measures affect housing price inflation through credit, including a measure of credit as a control variable is not appropriate in our setup as it would obscure the relationship we are interested in. We do not include a measure of construction costs either, for lack of available data. Table A2 in Appendix 1 contains further details on macroeconomic and demographic data sources.

C. Macroprudential Policies

The main hypothesis we want to test is that housing price inflation is affected at least temporarily by policies (other than interest rates) that indirectly affect the cost and availability of bank credit in general and mortgage credit in particular. We refer to these policies as “macroprudential policies”, though some of them are sometimes used as traditional monetary policy instruments (e.g., standard reserve requirements).

Data sources

We construct a novel dataset of macroprudential measures in 16 CESEE countries at a quarterly frequency for the purposes of performing the analysis presented in this paper. To do so, we exploit a wide variety of sources. Our main sources are documents posted on national central banks’ or national banking supervisors’ websites such as annual reports, inflation reports, financial stability reports, prudential regulations, press releases, as well as IMF Staff Reports and Financial System Assessment Program documents. We cross-check this information with that contained in country-level studies mentioned in the introduction and in specific chapters of the book edited by Enoch and Őtker-Robe (2007). We keep track of all prudential measures that we deem most relevant for credit supply in general and, through retail and mortgage lending, housing prices. We strive to collect information for time periods covering at least those for which housing prices data are available in each particular country. It is important to point out that some of our MPP measures only capture changes in the policy stance from the beginning of the sample, because we have no way of measuring and comparing across countries the initial “tightness” of some types of prudential regulation. In any case, we only use changes in policies, not their “levels” in the regressions.

In parallel to the MPP database, we compile information on fiscal and other regulatory policy measures that are directly relevant to the real estate market and household borrowing, such as changes in mortgage interest payments deductibility or the inclusion of non-bank credit institutions into the regulatory perimeter, whenever such information was present in the sources listed above or in “Taxation trends in the European Union” published yearly by Eurostat.13

Categorization

We compile data on twenty-nine categories of prudential measures, which we gather into five groups: capital measures, provisioning measures, liquidity measures, loan eligibility requirements, and other quantitative restrictions. We discard moral suasion measures, which were used almost universally according to the documents we consulted, because we view them as weak policy instruments that are unlikely to have any measurable impact. Information on the use of the various measures is provided in Figure 2 (the mapping between the name of a measure and its full description is provided in Appendix 2, while the mapping between measures and the countries that implemented them is provided in Appendix 3).

Figure 2.
Figure 2.

Number of Macroprudential Policy Changes in the Dataset

(by category of measure)

Citation: IMF Working Papers 2012, 303; 10.5089/9781475550139.001.A001

Notes: See Appendix 2 for a definition of the variables. Data for “Other bank regulatory measures” are not shown on the Figure.Source: Authors’ calculations.

Capital measures affect the amount or type of capital that banks must hold and consist of twelve different types of measures that change the following regulatory parameters: minimum CAR; minimum target CAR; minimum CAR related to credit growth; definition of regulatory capital; maximum ratio of loans to households relative to capital; maximum ratio of loans in foreign currency to capital; risk-weights used in the calculation of risk-weighted assets for mortgage loans (in local and foreign currency) or loans to households (in local or foreign currency) or on corporate loans (in foreign currency) or on bank exposures exceeding a threshold related to credit growth. Among this category of measures, changes in the minimum CAR, changes in risk-weights on mortgages, changes in capital eligibility, and changes in the ratio of household loans to capital were used most frequently (see the top 2 panels of Figure 2)

Provisioning measures consist of changes in the rules for general provisions, and changes in the rules for specific provisions on domestic currency loans or foreign currency loans. While the use of general provisioning is limited in the countries in our sample, changes in specific provisioning rules have not been infrequent.

Liquidity measures cover prudential measures related to reserve requirements or liquidity ratios: minimum reserve requirement ratios for demand deposits in domestic currency or in foreign currency; the definition of the base used to calculate reserve requirements and the minimum reserve requirement ratios for liabilities other than demand deposits; marginal reserve requirements on foreign borrowing (i.e., reserve requirements imposed only on increments in the stock of foreign borrowing); special reserve requirements on liabilities of banks arising from issued securities; marginal reserve requirements related to credit growth; liquidity ratios; and foreign currency liquidity requirements. Changes in average reserve requirements were by far the most commonly used instrument in our dataset. Marginal reserve requirements related to credit growth were used in two countries (Bulgaria and Croatia) while the other three of the other four liquidity measures were used only in one country (Croatia), which explains the low frequency of their use.

Loan eligibility requirements consist of four different types of measures: a maximum loan-to-value ratio for local currency loans or foreign currency loans; and a maximum debt-service-to-income ratio for domestic currency loans or foreign currency loans. These measures were used only sparsely in the CESEE region, as can be seen in the bottom right panel of Figure 2.

“Other quantitative restrictions” measures consist of limits on the amount of foreign currency lending as a share of total lending, whether in stock or flow terms, including outright bans on certain types of lending. Our dataset contains four observations for that category of measures and they all belong to the bust period.

Quantifying the strength of the policy measures

From the descriptions of the policy measures, we proceed to code numerically the strength of changes in the regulation in each category to capture their relative variation, both over time and across categories of measures. We believe that this approach is preferable to one used commonly in the emerging literature on MPP effectiveness that relies only on dummy variables to capture changes in regulation. We acknowledge that our approach involves judgment to a large degree but it is the logical consequence of the observation that policy measures vary in intensity and that both financial prices and quantities can be expected to react to this intensity. To take an example, a 1 percentage point change in reserve requirement rates cannot be expected to have the same impact as a 10 percentage point change. It is a very challenging task to capture interactions between various prudential policies—for example the interaction between reserve requirements and liquidity requirements—and we do not attempt to do so here.14

For regulation that can be summarized in a simple number (e.g. maximum or minimum ratios), our rule is to use a simple linear transformation of that number. For regulation that involves a small (but greater than one) number of variables (e.g. risk-weights on mortgages that are conditional on the loan-to-value ratio), we use a formula that takes into consideration all variables. For more complex cases, we use a rule where a tightening (resp. loosening) would increase (resp. decrease) an index summarizing the strength of the regulation by a fixed amount (0.25, 0.5, or 1 depending on the measure). Since we are only interested in the effect of the change in the various categories of regulation, the level of our measure of the strength of regulation is irrelevant and can be arbitrarily set to an arbitrary value (e.g. zero) during the quarter preceding the start date of our data sample.

As an example, for changes in the minimum CAR, the score is simply the quarterly change in the minimum ratio. This rule yields a score of zero during times when the minimum ratio is constant and a score of two during a quarter when a country moves from an eight percent to a ten percent minimum ratio. For across-the-board changes in risk-weights on mortgages, we first compute for each quarter the difference between risk-weights on domestic currency mortgages in the actual regulation and in the Basel capital standards (Basel I or Basel II) otherwise used in the country, then divide this number by 25, and then take the quarterly change in that series. This rule yields a value of two when a country operating under Basel I deviates from the standard by implementing a risk-weight of 100 (instead of 50) on mortgages. For changes in risk-weights on foreign currency mortgages relative to those in domestic currency mortgages, we first compute the difference between risk-weights on mortgages in foreign currency and those on mortgages in domestic currency, then divide this number by 50, and then take the quarterly change in that series. This rule yields a score of one during a quarter when a penalty of 50 percentage points is imposed on mortgages in foreign currency. The full list of the rules we apply is provided in Appendix 2.

By summing the scores across all categories, we obtain a summary representation of the intensity of the change in prudential regulation in each quarter of the sample period in each country (Figure 3). Positive values indicate a tightening and negative values an easing of prudential regulation. Then, by taking the cumulative sum of quarterly changes, we obtain a representation of the cumulative change in the macroprudential policy stance during the boom and bust (Figure 4).

Figure 3.
Figure 3.

Selected CESEE Countries: Quarterly Changes in Strength of Prudential Regulation, 1997:Q1–2011:Q1

Citation: IMF Working Papers 2012, 303; 10.5089/9781475550139.001.A001

Source: Authors’ calculations.
Figure 4.
Figure 4.

Selected CESEE Countries: Cumulative Changes in Strength of Prudential Regulation, 1997:Q1–2011:Q1

Citation: IMF Working Papers 2012, 303; 10.5089/9781475550139.001.A001

Source: Authors’ calculations.

There are clear differences among countries in terms of their policy “activism.” In a number of countries (Czech Republic, Russia, Slovakia, Slovenia) hardly any MPP measures were taken, despite considerable housing price inflation in some cases. In other countries, prudential regulation displays a clear countercyclical pattern (e.g. Bulgaria, Croatia, and Serbia), and in others yet it even appears to be mildly procyclical at times (e.g. Latvia and Lithuania in 2004:Q4, Romania in 2007:Q1, when some prudential policies were relaxed upon joining the European Union). Hungary displays procyclical policy during the downside of the cycle, as the authorities started tightening prudential regulation during the recession (in the beginning of 2010) as the drawbacks of excessive reliance on foreign currency debt became clear following the sharp depreciation of the forint.

Table 1 shows summary statistics for the dependent variable, macroeconomic control variables, and the individual MPPs.15

Table 1.

Summary Statistics

article image
Source: Authors’ calculations.

III. Econometric Model and Choice of Specification

We start our econometric analysis by checking the order of integration of these series. The Maddala and Wu (1999) panel unit root test indicates that both the log of real GDP per capita and the log of real housing prices are I(1) variables. One of the two Westerlund (2007) ECM panel cointegration tests rejects the null of no cointegration between these two variables while the other does not.16 We proceed by modeling housing price dynamics in our sample in an error correction framework, where changes in the log of (seasonally adjusted) real house prices are explained by lagged changes in the log of (seasonally adjusted) real GDP per capita, lagged changes in the domestic currency real interest rate, lagged changes in the effective foreign currency real interest rate, lagged changes in MPPs, and an error correction term. We include country fixed effects to account for time-invariant country-specific characteristics captured by intercepts in the short-run and the long-run equations, and include time dummies to account for common shocks across the region. The latter include conditions in global capital markets that would influence capital flows to CESEE. We later check that our key results also hold when the error correction term is not included.

Our panel is unbalanced. For most countries, the sample period starts in the early 2000’s but for Romania, Slovakia and Turkey housing price data availability is a constraint and the sample only starts later. The sample ends in 2011:Q1.

Ideally, we would want to run regressions including all individual policy variables, since all of them can potentially affect housing prices. In addition, from the point of view of a policymaker it is important to know which specific measures are effective. However, a regression including all individual MPPs would exhaust most or all of our degrees of freedom, so we need to pare down the number of MPPs that enter separately in the regression. To this end, we run some preliminary regressions including the first two lags of each policy variable and the first two lags of an aggregate index of the remaining MPP changes constructed as the sum of the scores for each of these individual measures. In addition, to further economize on degrees of freedom we drop the second lag of the change in real GDP and the second lag of the change in the real interest rate, which are insignificant across all specifications. Thus, for each MPP variable x, we estimate the following equation using the fixed effects estimator: 17

Δhi,t=φ(hi,t1θyi,t1)+Σj=12(ρjΔhi,tj)+α1Δyi,t1+α2Δri,t1+α3Δri,t1*+α4Δ4wpi,t+Σj=12(βjΔxi,tj+γjΔCi,tjx)+Σj=14ukrj+δi+μt+εi,t

where the subscripts i and t represent a country and a time period respectively, δ is a country dummy variable, μ is a period dummy variable, Δ is the first difference indicator, Δ4 is the four-quarter difference indicator, h is the log of real housing prices, y is the log of real GDP per capita, r is the domestic real interest rate, r* is the effective foreign currency real interest rate (defined as the foreign currency policy rate minus the rate of year-on-year CPI inflation and minus the y-o-y appreciation of the local currency against its natural cross), wp is the log of the working age population, Cx is a control variable aggregating all MPPs other than x as well as relevant tax and non-bank regulatory policy measures. To account for possible nonlinear effects of the devaluation in Ukraine in the last quarter of 2008, we include four dummies (ukr1, ukr2, ukr3, and ukr4) corresponding to the three periods following the devaluation. The α’s,β’s,γ’s,ρ’s, θ, and φ are coefficients to be estimated, and ε is an error term.

These preliminary regressions allow us to identify a core set of policy variables which seem to have a significant impact on housing prices. We then estimate a baseline regression with all the variables in this core set included separately and the rest included as an aggregate.

To check whether the impact of measures might be different depending on whether the policy is being tightened or eased, we also estimate an equation where the coefficient of a policy variable is allowed to differ when it represents a tightening or an easing of the policy.

We are also interested in whether the effects of the policy measures vary based on the phase of the economic cycle. Therefore in an alternative specification we allow the coefficients of the policy to vary depending on whether the economy is in a boom or a bust. Capital inflows to CESEE accelerated in late 2002 and came to a sudden stop once the U.S. financial crisis spilled over to CESEE in full force after mid-September 2008 (IMF, 2010). We thus define the boom period to run from 2002:Q4 to 2008:Q3.18 The bust period runs from 2008:Q4 until the end of our sample, i.e. 2011:Q1.

IV. Estimation Results

A. Preliminary Regressions

Table 2 presents results of the preliminary regressions, which include the change in each MPP one at a time together with the change of an aggregate of the rest of the policy changes. In addition to individual measures, we also look at changes in selected combinations of individual measures of the same type (e.g. a combination of all risk-weight measures). As explained above, the regressions also include time and country dummies and real per capita GDP growth and real interest rates (these coefficients are not reported). The MPPs are grouped in the same five categories as discussed above: bank capital measures, provisioning rules, liquidity measures, borrower eligibility criteria, and other quantitative restrictions. The estimated coefficients for the prudential measures are negative if a tightening (easing) in prudential regulation is followed by a deceleration (acceleration) in housing prices.

Table 2.

Macroprudential Policies and Housing Prices -- Preliminary Regressions

article image
Note: The dependent variable is the log difference of the real housing price index. The regressions include time and country fixed effects. P-values in parentheses.*, **, and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent confidence levels respectively.Source: Authors’ calculations

Among capital measures, changes in the minimum CAR appear to significantly affect housing prices in the expected direction in the first quarter following the change in policy. The effect during the second quarter is in the expected direction as well, but it is marginally below conventional significance levels. Changes in maximum ratio of household lending relative to capital were also followed by changes in housing price inflation in the expected direction, though the coefficient for the second lag is positive and insignificant. Changes in risk weights on loans to households used in the computation of capital requirements, on the other hand, do not seem to have consistent effects on house price growth.

We also find little evidence that changes in provisioning rules, whether related to general or specific provisions, and whether across the board or for foreign currency loans only, had any impact on housing price inflation. Since standard provisioning rules do not bind until loans start to become non-performing, which does not happen on a significant scale until after the cycle has turned, it might be that the tightening effect of measures related to specific provisions only materialized with a longer lag (and might have been pro-cyclical if the measures were not reversed during the bust). It is perhaps more surprising to find that changes in rules for general provisions, which are closer to a truly dynamic provisioning system where provisions are built even against performing exposures, do not show a more robust effect, but we have only a very small number of observations of that type of measure in the sample.

Changing average reserve requirements also does not seem to have had an effect on housing price inflation, nor does changing the foreign currency liquidity ratio, a Croatia-specific instrument, or changing the liquidity regulation. In several cases, changes in the reserve requirements rate on demand deposits took place at the same time as changes in the base, with the two changes working in opposite directions. If we combine changes in the rate and changes in the base into a composite measure for reserve requirements, then the coefficients remain insignificant. A possible explanation for this lack of significance is that reserve requirements may have been used to sterilize foreign exchange intervention following a surge in capital inflows. In this case, the policy might have simply have forestalled a further acceleration of credit (and housing prices) rather than caused a deceleration. In other words, endogeneity bias might be particularly strong for reserve requirements, if this policy was used as the “first line of defense” to counter excessive credit market froth. It may also be, as discussed in Gray (2011), that reserve requirements may have been used for various—and sometimes contradictory—objectives across countries. In addition, it may be that changes in reserve requirements were in some cases made concurrently with changes in other monetary instruments such as central bank bills and we do not account for the latter. Furthermore, reserve requirements are a multidimensional instruments, and we do not capture some of these dimensions such as the eligibility of some assets (e.g. cash in vault) to meet the requirements, or variations in averaging rules. While changes in “plain vanilla” average reserve requirements seemed to have had little impact, more unorthodox measures, i.e. marginal reserve requirements on foreign borrowing and marginal reserve requirements on credit in excess of a certain threshold are both associated with a significant changes in housing prices in the “right” direction.

Turning to eligibility measures, coefficients are generally insignificant, suggesting that these measures did not have much of an impact in CESEE, in contrast with the findings for some East Asian countries. Coefficients on across-the-board LTV and DTI measures have the right sign, but coefficients on stricter eligibility requirements for foreign currency borrowers don’t.19 A composite of all loan eligibility measures yields a significant coefficient in the right direction for the second lag, but the coefficient on the first lag is positive and insignificant. Since these measures were implemented only in a handful of cases in our sample, we do not wish to draw too firm conclusions from this lack of statistical significance overall.

Based on the results in the preliminary regression, we select the three MPPs for which the regression coefficient has the expected negative sign for both lags and is statistically significant for at least one lag and include them as separate regressors in the baseline specification.20 These policies are changes in the minimum CAR (used in Croatia, Latvia, Lithuania, Romania, Serbia, and Ukraine), marginal reserve requirements related to foreign borrowing (used in Croatia), and marginal reserve requirements related to credit growth (used in Bulgaria and Croatia). The other MPPs together with relevant tax and non-bank regulatory policies are included as an aggregate index as a control variable.21 Each of the three MPPs in the core set was changed between seven and nine times during the sample period.

B. Baseline Regression

Column (1) in Table 3 shows estimation results for our baseline regression, which includes the same control variables as the preliminary regressions as well as the three MPP variables in our core set, i.e. minimum bank CAR, marginal reserve requirements on foreign funding, and marginal reserve requirements on credit growth, and an aggregate of all other policies.

Table 3.

Prudential Policies and Housing Prices -- Baseline Regression

article image
Notes: The dependent variable is the log difference of the real housing price index. The regressions include time and country fixed effects. P-values in parentheses. *, ** and, *** denote statistical significance at the 10 percent, 5 percent and 1 percent confidence levels respectively.Source: Authors’ calculations

Going through the explanatory variables in Table 3 from top to bottom, we see that the long term effect of per capita GDP is positive, as expected, albeit not significant (upper panel). The estimated coefficient is close to one, suggesting housing prices and GDP per capita co-move one-for-one in the long run, keeping housing affordable. The error correction coefficient, which measures the speed at which deviations from the long-term equilibrium self-correct, is negative and highly significant. Both autoregressive terms are significant, showing that housing price inflation is persistent. Surprisingly the coefficient estimates for our set of macroeconomic and demographic fundamentals (lagged changes in per capita GDP and interest rates, changes in working-age population) are not significant.22

With respect to the MPP policy variables, changes in the minimum CAR, changes in marginal reserve requirements related to foreign borrowing, and changes in marginal reserve requirements related to credit growth are all significant, consistent with the results of the preliminary regressions. Moreover, the estimated coefficients are very similar to those obtained in the preliminary regressions. The estimated coefficient for the aggregate of all other policy measures is not significant, suggesting that other policy measures had, on average, no measurable effect on the housing cycle, at least not during the first two quarters following their implementation.

To assess the economic magnitude of the effects, we compute the dynamic multipliers tracing out the response of housing price inflation to changes in each of the three MPPs in the core set over the following ten quarters (Figure 5). The charts also report 95 percent confidence intervals. In each of the policy experiments the MPP index is increased by one point, which has a different interpretation depending on the index (again, see Appendix Table 2) but in each case corresponds to a policy change of a plausible magnitude.

Figure 5.
Figure 5.

Dynamic Multiplier of Shock to Selected Macroprudential Policies

Citation: IMF Working Papers 2012, 303; 10.5089/9781475550139.001.A001

Note: Each shock represents an increase by one unit in the intensity of the policy variable. The cumulative change in house prices is shown on the vertical axis (in percent). Time (in periods) is on the horizontal axis. Source: Authors’ calculations.

Based on the estimated coefficients, an increase in the minimum capital requirement by one percentage point would have its maximum impact after four-to-five quarters, when housing prices are 8.5 percent lower than they would have been without the policy change. Subsequently, the effect starts to die down and after ten quarters the cumulative decline is of 4.5 percent. Standard errors around this point estimates, however, are quite large, indicating that a precise quantification of the magnitude of the effect is not possible within our sample and empirical framework. Nonetheless, these results suggest that the countercyclical capital buffer under Basel III, which can reach up to 2.5 percentage points (corresponding to a score of 2.5 in our scoring system), could potentially have a large impact on housing price dynamics in the short and medium term. Therefore the recent experience in CESEE provides support for actively using the countercyclical capital buffer to address concerns about housing bubbles.

Turning to the other MPPs in the core set, the effect of marginal reserve requirements on foreign borrowing, while negative and econometrically significant, is relatively small: an increase by 20 percentage points lowers housing prices by only 3 percent at the peak. Finally, a one-point increase in the index of marginal reserve requirements on excessive credit growth (a relatively large increase since the maximum observed in the sample is 1.18 points) lowers housing prices by 8 percentage points after one year. 23

Given that the estimated long-run relationship between housing prices and GDP per capita is not statistically significant in the baseline regression and, as mentioned above, one of the two Westerlund (2007) panel tests does not reject the null of no cointegration between housing prices and GDP per capita, we run an alternative regression that includes the same variables as the baseline except for the error-correction term. As shown in Column (2) of Table 3, the significance and the order of magnitude of the effects of the four MPPs in the core set are consistent with those obtained under the baseline.

C. Are the Effects of MPPs Asymmetric?

We further explore whether the effects of MPPs in the core set differ depending on whether they are loosened or tightened, or depending on the different parts of the cycle when they occur.

In the regressions in the middle panel of Table 4, we re-run the baseline specification allowing for separate coefficients for tightening and loosening of the MPPs; the top panel of the table reports the coefficients in the baseline regression for ease of comparison. The results show that for the minimum CAR, the effect on housing prices was stronger and more prolonged when the regulation was loosened (as in Latvia, Lithuania and Romania during the boom) than when it was tightened.24 For the other two types of measures, only tightening seems to have had a robust effect, but the coefficients are bigger and more statistically significant than in the baseline.

Table 4.

Macroprudential Policies and Housing Prices: Are the Effects Asymmetric?

article image
Note: The dependent variable is the log difference of the real housing price index. The regressions include time and country fixed effects.P-values in parentheses. *, **, and *** denote statistical significance at the 10 percent, 5 percent and 1 percent confidence levels respectively.Source: Authors’ calculations.

Before reading too much into these asymmetries, however, we should point out that they may reflect different endogeneity biases. For instance, if capital requirements tended to be tightened in response to an expected acceleration in housing prices while they tended to be loosened for exogenous reasons (i.e., harmonization with EU minima following EU entry), the endogeneity bias would be much smaller for the loosening coefficient than for the tightening coefficient.

Distinguishing between boom and bust periods, we observe that the effects are very strong and significant during the boom period for the three measures. In particular, they are much more significant during the boom than during the whole sample period for the marginal reserve requirement on foreign funding. Accordingly, the effects during the bust are not very robust, although they go in the expected direction when both lags’ coefficients are combined.

V. Conclusions

After the 2008–09 global financial crisis, preventing credit and housing price booms has become a major priority for policy-makers in both advanced and emerging market countries. To this end, policy instruments beyond those in the conventional macroeconomic policy toolkit are being considered, and many countries are in the process of developing an institutional framework to use them on a regular basis. The interest around these new potential “macroprudential” instruments is so far not matched by the quantity of empirical evidence supporting their effectiveness. Furthermore, as the list of possible macroprudential instruments is long, the question of which levers should be used is equally important.

This paper has attempted to shed light on these questions taking advantage of the experience in CESEE countries during the boom-bust cycles of the last decade. Using a novel database we constructed on twenty-nine types of macroprudential policy changes, we have tested whether changes in these policies affected housing price inflation in the last decade.

We found that some macroprudential policies did have an impact, while others did not. In particular, raising the minimum CAR was followed by a significant deceleration in housing prices. This finding bodes well for one of the main macroprudential tools introduced as part of the Basel III reforms, i.e. the countercyclical capital buffer. An equally important result, especially against the background of the current debate on maximum harmonization in the context of the EU’s new Capital Requirements Directive and the future banking union, is the finding that allowing banks to hold less capital (typically following EU entry) was followed by a sizeable acceleration in housing prices. While we do not find that standard reserve requirements had an impact on housing price inflation, marginal reserve requirements targeting specific excesses, such as those related to credit growth or to foreign funding were found to have an effect.

This study is not a perfect policy experiment: endogeneity may bias coefficients downwards, thus some tighter policies that appear ineffective may do so only because they were adopted when policymakers anticipated accelerating housing price inflation. Effects may have become visible only after the two-quarter horizon we use in our empirical framework, or may have taken place as soon as the policy change was announced and before its actual implementation. The coding of the intensity of the various policy measures, a complicated process relying on subjective judgment, may also be less-than-fully adequate in some cases. Nonetheless, we believe that the evidence this study provides is informative and can be useful to policymakers.

As to future research, an interesting avenue would be to explore alternative dependent variables to capture the boom, such as construction activity, real estate transaction volumes, or credit growth once sufficiently detailed data on the currency composition of loans becomes available.

Appendix 1. Data Sources

Table A1.

Housing Price Data

article image
Notes: CB=Central Bank; NSO=National Statistical Office; BIS=Bank of International Settlements.FHB, GPG, REAS and Reidin are private companies providing real estate services.
Table A2.

Macroeconomic and Demographic Data Sources

article image

Appendix 2: Description of Macroprudential Policy Measures and the Scoring Rules to Measure Their Intensity

article image
article image
article image

Appendix 3: Macroprudential Policy Measures and Countries Where They Were Used

article image

References

  • Abiad, A., E. Detragiache, and T. Tressel, 2010, “A New Database of Financial Reforms,” Staff Papers, International Monetary Fund, Vol. 57, pp. 281302.

    • Search Google Scholar
    • Export Citation
  • Bakker, B., and C. Klingen, ed., 2012, How Emerging Europe Came Through the 2008/09 Crisis: An Account by the Staff of the IMF’s European Department (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Banai, A., J. Király, and M. Nagy, 2011, “Home High Above and Home Deep Down Below—Lending in Hungary” World Bank Policy research Working Paper No.5836.

    • Search Google Scholar
    • Export Citation
  • Blackburne E.F., and M.W. Frank, 2007, “Estimation of Nonstationary Heterogeneous Panels,” The Stata Journal, Vol. 7(2), pp. 197208.

    • Search Google Scholar
    • Export Citation
  • Capozza, D., P. Henderschott, C. Mack, and C. Mayer, 2002, “Determinants of Real House Price Dynamics”, NBER Working Paper No 9262.

  • Celeska F., V. Gligorova, and A. Krstevska, 2011, “Macroprudential Regulation of Credit Booms and Busts—The Experience of the National Bank of the Republic of Macedonia,” World Bank Policy research Working Paper No.5770.

    • Search Google Scholar
    • Export Citation
  • Craig, S., and C. Hua, 2011, “Determinants of Property Prices in Hong Kong S.A.R.: Implications for Policy,” IMF Working Paper 11/227 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Crowe, C., G. Dell’Ariccia, D. Igan, and P. Rabanal, 2011, “How to Deal with Real Estate Booms: Lessons from Country Experiences,” IMF Working Paper No. 11/91 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Dell’Ariccia, G., D. Igan, L. Laeven and H. Tong, B. Bakker, and J. Vandenbussche, 2012, “Policies for Macrofinancial Stability: How to Deal with Credit Booms,” IMF Staff Discussion Note 12/06 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Dimova, D., P. Kongsamut, and J. Vandenbussche, forthcoming, “The Effectiveness of Macroprudential measures: The Experience of South Eastern Europe”, IMF Working Paper (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Enoch, C., and I. Ötker-Robe, eds., 2007, Rapid Credit Growth in Central and Eastern Europe – Endless Boom or Early Warning, 1st edition (Houndmills: Palgrave Macmillan).

    • Search Google Scholar
    • Export Citation
  • Egert, B., and D. Mihaljek, 2007, “Determinants of House Prices in Central and Eastern Europe,” BIS Working Paper No. 236 (Basel: Bank for International Settlement).

    • Search Google Scholar
    • Export Citation
  • Frait, J., A. Gersl and J. Seidler, 2011, “Credit Growth and Financial Stability in the Czech Republic,” World Bank Policy Research Working Paper No.5771.

    • Search Google Scholar
    • Export Citation
  • Galac, T., 2010, “The Central Bank as Crisis-Manager in Croatia—A Counterfactual Analysis,” Croatian National Bank Working Paper 27 (Zagreb: Croatian National Bank).

    • Search Google Scholar
    • Export Citation
  • Galati, G., and R. Moessner, 2011, “Macroprudential Policies—A Literature Review,” BIS Working Paper No. 337 (Basel: Bank for International Settlement).

    • Search Google Scholar
    • Export Citation
  • Gray, S., 2011, “Central Bank Balances and Reserve Requirements,” IMF Working Paper 11/36 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Igan, D., and H. Kang, 2011, “Do Loan-to-Value and Debt-to-Income Limits Work? Evidence from Korea,” IMF Working Paper No. 11/297 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Igan, D., and P. Loungani, 2012, “Global Housing Cycles,” IMF Working Paper 12/217 (Washington: International Monetary Fund).

  • International Monetary Fund (IMF), 2010, Regional Economic Outlook: Europe—Building Confidence, October (Washington).

  • Kenc, T., M. I. Turhan, and O. Yildirim, 2011, “The Experience with Macro-Prudential Policies of the Central Bank of the Republic of Turkey in Response to the Global Financial Crisis,” World Bank Policy Research Working Paper No.5834.

    • Search Google Scholar
    • Export Citation
  • Kraft, E., and T. Galac, 2011, “Macroprudential Regulation of Credit Booms and Busts–The Case of Croatia,” World Bank Policy Research Working Paper No. 5772.

    • Search Google Scholar
    • Export Citation
  • Kruszka, M. and M. Kowalczyk, 2011, “Macro-Prudential Regulation of Credit Boom and Busts—The Case of Poland,” World Bank Policy Research Working Paper No.5832.

    • Search Google Scholar
    • Export Citation
  • Jiménez, G., S. Ongena, J-O. Peydro, and J. Saurina, 2012, “Macroprudential Policy, Countercyclical Bank Capital Buffers and Credit Supply: Evidence from the Spanish Dynamic Provisioning Experiments,” European Banking Center Discussion Paper No. 2012-011.

    • Search Google Scholar
    • Export Citation
  • Lim, C. H., F. Columba, A. Costa, P. Kongsamut, A. Otani, M. Saiyid, T. Wezel, X. Wu, 2011, “Macroprudential Policy: What Instruments and How to Use Them? Lessons from Country Experiences,” IMF Working Paper 11/238 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Maddala, G.S. and S. Wu, 1999, “A Comparative Study of Unit-Root Tests with Panel Data and a New Simple Test,” Oxford Bulletin of Economics and Statistics, Vol. 61, pp. 63152.

    • Search Google Scholar
    • Export Citation
  • Malpezzi, S., 1999, “A Simple Error Correction Model of Housing Prices,” Journal of Housing Economics, Vol. 8, pp. 2762.

  • National Bank of Poland, 2006, Financial Stability Report 2006.

  • Nier, E., J. Osiński, L. Jácome, and P. Madrid, 2011, Institutional Models for Macroprudential Policy, IMF Staff Discussion Note 11/18.

    • Search Google Scholar
    • Export Citation
  • Pereira da Silva, L., and R. Eyer Harris, forthcoming, “Sailing through the Global Financial Storm: Brazil’s Experience with Monetary and Macroprudential Policies to Lean Against the Financial Cycle and Deal with Systemic Risks,” IMF Working Paper (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Persyn, D., and J. Westerlund, 2008, “Error Correction Based Cointegration Tests for Panel Data,” Stata Journal, Vol. 8 (2), pp. 23241.

    • Search Google Scholar
    • Export Citation
  • REAS (in cooperation with Jones Lang LaSalle), 2008-2012, Residential Markets in Central European Capitals.

  • Sutt, A., H. Korju, and K. Siibak, 2011, “The Role of Macro-Prudential Policies in the Boom and Adjustment Phase of the Credit Cycle in Estonia,” World Bank Policy Research Working Paper No.5835.

    • Search Google Scholar
    • Export Citation
  • Terrier, G., R. Valdés, C. Tovar, J. Chan-Lau, C. Fernández-Valdovinos, M. García-Escribano, C. Medeiros, M.-K. Tang, M. Vera Martin, and C. Walker, 2011, “Policy Instruments to Lean against the Wind in Latin America,” IMF Working Paper 11/159 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Tovar C., M. Garcia-Escribano, and