THE INTERACTION OF MONETARY AND MACROPRUDENTIAL POLICIES: BACKGROUND PAPER
1
1
8
8
INTERNATIONAL MONETARY FUND
characteristics. Our regressions also control for monetary policy rates and dummy variables
denoting phases of credit and economic cycles.
17,18
48. The measurement of the effects of policy changes on both financial and aggregate
variables is subject to well-known endogeneity problems. This issue is shared by most existing
studies on the effects of macroprudential policy study (e.g., Lim and others, 2011). When
macroprudential policy responds to credit and asset prices, rather than output, this bias should in
principle be stronger when measuring the effect on credit and asset prices—as does much of the
existing literature—and weaker when investigating aggregate effects—which is the focus here.
Moreover, as long as the problem does not differ across tools considered it may not affect
comparisons across tools in their relative effects. Throughout, we lag all policy variables by one
quarter in an attempt to address endogeneity biases.
19
Nonetheless, we take the results as only
suggestive of the relative strength of the effects across tools, rather than as a reliable measure of
the size of each effect.
20
49. Quarterly data from 2000 to 2011 were used for 36 countries,
21
including 21 emerging
market economies (EMEs) and 15 advanced economies (AEs).
22
Most of the data are collected
from official and commercial sources, such as IFS, central banks, national statistical offices, Haver
Analytics, and CEIC database, being specified along the results. Detailed information on countries
which have used macroprudential policies can be found in Table 2.
17
Variables in the form of dummies are constructed to control for the stages of financial and economic cycles. First, a
credit bust is classified as a stage with either of the two following conditions being satisfied: (i) the deviation from a
HP filtered trend is smaller than 1.5 times its standard deviation; or (ii) the quarterly credit growth rate is lower than a
long-run average by 1.5 times its standard deviation. Second, a recession dummy is equal to one on the quarters
when the output gap, using the HP filter, is negative for previous six consecutive quarters.
18
In addition, interactions between monetary and macroprudential policies are analyzed, but no significant empirical
evidence is found. We created dummy variables indicating whether the monetary stance is tight, or whether it is
loose and estimated interactions between the macroprudential tools and the monetary policy dummies. This result is
in line with results obtained by Dell’Ariccia and others (2012). Coefficients on the interaction terms are unstable and
rarely significant across all macroprudential tools. Similarly we do not find that effectiveness of macroprudential
policy depends on the monetary and FX regime, echoing results already documented by Lim and others (2011).
19
In order to try to reduce the endogeneity problem, one quarter lagged policy variables are used for the main
results, and a robust test is conducted with concurrent variables, which shows similar results.
20
As mentioned in Lim and others (2011), the estimation of a dynamic panel by OLS with country and time fixed
effects will be biased, since by construction there is a positive correlation between the lagged dependent variable
and the unobserved individual level effects. We dropped the lagged dependent variable as a robustness check, and
found that the main results do not change materially.
21
The countries in the sample are as follows: Argentina, Austria, Brazil, Bulgaria, Canada, Chile, China, Colombia,
Croatia, Estonia, Hong Kong SAR, Hungary, India, Indonesia, Ireland, Israel, Italy, Latvia, Malaysia, Mexico,
Netherlands, Norway, Peru, Poland, Romania, Russia, Serbia, Singapore, Slovak Republic, South Korea, Spain, Sweden,
Thailand, Turkey, Uruguay, and U.S.
22
As of September 2012, the number of countries in the sample with each de facto exchange rate regime is as
follows: free floating (10), floating (15), fixed (3) (of which currency board (2) and conventional peg (1)), and others (8)
(of which crawl like (3), other managed (4), and stabilized (1)).