II. Monetary Policy transmission and the new policy framework17
A. Introduction and Summary
56. After a quarter century of monetary targeting, the Swiss National Bank (SNB) adopted a new monetary policy framework in January 2000. The new framework defines an explicit inflation objective; specifies that monetary policy decisions should be based on the medium-term inflation forecast; and designates a short-term interest rate as the operating target. Although the framework differs substantially from monetary targeting, its adoption could be considered the culmination of a gradual, decade-long shift in emphasis from quantity indicators to other variables in the pursuit of price stability.
57. The eminent role of the medium-term inflation forecast in monetary policy decisions in the new framework focuses attention on two basic questions: how does monetary policy influence prices, and what are the output costs attached over the forecast horizon—in other words, how does monetary transmission operate over the medium term in Switzerland? To shed some light on these questions, this chapter examines the relationship between monetary policy (summarized by interest rates) and economic developments (captured by the behavior of output, inflation, exchange rate, and price) in Switzerland in the 1983-99 period. It offers three main findings:
Monetary policy influences the economy with long lags and large uncertainties.
Both the exchange rate and the credit channel are important ingredients of the transmission mechanism, in line with the fact that Switzerland is a small open economy with a sophisticated financial sector.
Monetary policy appears to have been successful in anchoring inflation expectations.
58. The first finding confirms conventional wisdom, and suggests that a three-year horizon for the inflation forecast is broadly appropriate. The second finding indicates that a synthetic indicator of the monetary policy stance could benefit from including information on interest rates, exchange rates, and credit market developments. The third finding provides some support that increased transparency might generate benefits for the efficiency of monetary policy in controlling inflation.
59. After a brief review of the main features of the monetary policy framework in Section B, Section C presents results of the empirical analysis, while Section D offers some policy conclusions.
B. The New Monetary Policy Framework in an International Perspective
60. The Swiss economy has been operating under a floating exchange rate regime since the collapse of the Bretton Woods system in 1973, with an independent monetary policy that has been geared towards the objective of price stability. However, the SNB repeatedly emphasized that it retained the option to offset undesirable exchange rate developments, should it become necessary.18 Correspondingly, foreign exchange market developments temporarily became the driving force of monetary policy on several occasions, namely, during 1978-79 (appreciation of the franc), in 1981-82, in 1987 (at the time of the U.S. stock market crash), in 1992-93, in 1996, and in 1998 (during the Asian crisis).
61. Until January 2000, the SNB conducted its monetary policy in the framework of a monetary aggregate target.19 Prior to 1990, annual target growth rates were set, first for Ml, then from 1979 for base money. Beginning in 1990, annual targets were replaced by a medium-term (5-year) growth rate objective. The shift to a medium-term target was triggered by a series of velocity shifts beginning in 1988, and the perception that the relationship between the targeted monetary aggregate (monetary base) and prices had become unstable. During the 1990s, various other quantity indicators (e.g., M3) were also considered as supplementary indicators. Although the rhetoric of monetary targeting was retained, the prominence of indicators other than monetary aggregates (e.g., interest rates) appears to have increased steadily in the formulation of monetary policy.20
62. The 2000 monetary framework abandoned monetary targeting and set in place a regime that most resembles inflation targeting. However, contrary to some other countries that adopted similar monetary policy frameworks (e.g., New Zealand, Sweden, or the U.K.), the change in the monetary policy regime is not a result of poor performance or an outright failure by the previous framework. In this respect, the new framework represents evolution rather than revolution. Further, there has been no abrupt shift in policy objectives, and no particular reason to believe that the policy reaction function has changed substantively, Rich (2000). In addition, given the SNB’s long and relatively successful track record of independent monetary policy (Figures II-1 and II-2), Swiss policymakers adopted the new regime equipped with a considerable stock of credibility.
Figure II-1.Switzerland: Inflation Rate in Selected Countries
Sources: OECD Analytical Database; and IMF, World Economic Outlook.
Figure II-2.Switzerland: Actual Inflation and Medium-Term Inflation Goal
Sources: IFS; and staff calculations.
63. The SNB describes the main features of the new framework as (i) an explicit definition of the price stability objective; (ii) a central role for medium-term inflation forecasts in monetary policy decisions; and (iii) an operational target range for the three-month Swiss franc interest rate.21 As Table II-1 illustrates, some of these features are shared with the monetary framework of the ECB, and with countries that practice inflation targeting. In particular, all the comparator countries have an explicitly defined inflation objective and publish inflation forecasts. In the pursuit of their objective, all countries use a short-term interest rate as an operating target. However, details vary substantially across the different frameworks – for instance, New Zealand and the UK do not target headline CPI, but a price index excluding the direct effect of interest rates; several comparators have a point rather than a range objective for inflation; and different interest rates serve as operating targets in different countries.
|Switzerland||Canada||ECB||New Zealand||Sweden||United Kingdom|
|Objective||price stability||price stability||price stability||price stability||price stability||price stability|
|Institutional commitment||no||yes||“first pillar”||yes||yes||yes|
|Inflation rate||headline CPI||core inflation||headline HICP||CPIX||headline CPI||RPIX|
|Inflation level||<2%, but positive||2% with tolerance of ± 1%||<2%, but positive||0-3%||2% with tolerance of ± 1%||2.5%|
|Horizon||medium term||2 years||medium term||1½ to 2 years||2 years||2 years|
|Frequency||4x per year||2x per year||2x per year staff forecast||4x per year||3x per year||4x per year|
|Operating target||3-mo SwF LIBOR; 100 bp. range||overnight rate; level||3-mo repo rate; level||official cash rate; level||1-week repo rate; level||2-week repo rate; level|
|Decision maker||Gov. Board||Gov. Board||Executive Board||Governor||Gov. Board||MPC|
|Accountability||Central bank||Accountable to||Accountable to||Governor||Central bank||“Open letter”|
|accountable to||Federal||European Parl.,||personally||accountable to|
|Federal||Government||Council of Min,||accountable||Parliament;|
64. A further and more substantial difference between Switzerland and the ECB on one hand, and the inflation targeter comparators on the other hand, lies in the area of transparency and accountability.22 In particular, formal mechanisms to ensure transparency of monetary policy decisions, and enhance the accountability of policymakers are in place and are being spelled out in ever-increasing detail in the inflation targeting countries. This is not by accident, as the mechanisms that secure transparency and accountability are viewed by some analysts as an integral part of the framework.23 The corresponding mechanisms remain largely informal in the case of Switzerland and the ECB.
65. Transparency can yield economic benefits by strengthening the central bank’s credibility and helping to build up or preserve its reputation. As a transparent central bank can be held accountable more easily, transparency could also contribute to correcting the “democratic deficit” that arises from delegating monetary policy to an independent agency.24
However, transparency is not a free lunch, nor is more information necessarily better. First, producing, processing, packaging and transmitting information is costly. Second, bogus transparency in the form of redundant or confusing information can be counterproductive.25 For these reasons, practices vary across countries.
66. While all central banks in Table II-1 are transparent regarding their objectives, there are substantial differences with respect to other aspects of transparency. Regarding economic transparency, the divide is between the SNB and the ECB on one hand and inflation targeting countries on the other. The latter publish detailed inflation reports, compared with a very compact few pages for Switzerland, and none (until recently) for the ECB. Procedural transparency is mainly influenced by the institutional setup. Sweden and the U.K. publish minutes of meetings of their monetary policy decision making bodies, complete with voting records of members. This would be unnecessary in New Zealand, where the responsibility rests with the Governor. Decisions in the other three countries are consensus based, hence publishing minutes is perceived to have little value added. In contrast, policy transparency is present in all six countries, as policy decisions and likely future actions are communicated to the public. Similarly, operating targets and procedures are well-defined. However, information provided on details of monetary policy implementation (such as market intervention) varies across countries.
67. With respect to accountability, the central banks in Table II-1 are in some form responsible to their respective governments or parliaments. In addition, monetary policy in the U.K. and Sweden is subject to the open letter system, whereby the governor of the central bank is required to explain the reasons in an open letter to the chancellor whenever inflation deviates by more than 1 percent from the target. In New Zealand, the governor of the central bank is directly responsible for monetary policy outcomes.
C. The Monetary Transmission Mechanism in Switzerland
68. In an attempt to quantify broad policy tradeoffs, this section examines the effects of monetary policy on output and inflation. Although the empirical analysis does not yield particularly sharp or surprising insights, the following tentative conclusions can be offered:
The empirical relationships linking main macroeconomic variables do not appear stable over the past 25 years. However, they display somewhat more stability in the more recent period.
Lags in the transmission mechanism are relatively long and variable, but comparable to those found for other countries.
The exchange rate plays a significant role in the transmission of interest rate shocks.
The relative importance of domestic real interest rates and the rate of real appreciation for real GDP growth is probably 2:1 to 3:1.
Monetary policy inflation targets appear to have been fairly credible and successful in anchoring inflation expectations.
Lags and channels—an unrestricted VAR approach
69. To obtain a broad picture of the monetary transmission mechanism in Switzerland, in particular, to gauge the length of lags from policy shocks to output and inflation and to get a sense of the relative importance of various transmission channels, a small unrestricted vector autoregression (VAR) model is estimated. The model links five variables, and summarizes the economy’s observed responses to various shocks.
70. The VAR model includes the output gap, the deviation of inflation from its longer run target26, the credit to potential GDP ratio, the short-term interest rate, and the nominal effective appreciation rate. Data sources and the construction of the variables are discussed in detail in the Appendix. The methodology and the choice of variables are fairly standard in the empirical literature on monetary transmission: similar specifications were estimated e.g., by Leeper et al (1996) and Bagliano and Favero (1998) for the U.S., by Ramaswamy and Sløk (1997) and Clements et al (2000) for EU countries, and by Morsink and Bayoumi (1999) for Japan.
71. In this reduced form model, short-term interest rates (three-month Swiss franc LIBOR) are assumed to represent monetary policy. The rationale for this choice is that the SNB’s current monetary framework employs this interest rate as an operating target, therefore it should be included in the empirical analysis if relevant lessons are to be learned.
72. Shocks to short-term interest rates are transmitted to output and inflation via three channels: (i) directly; (ii) through their effect on the availability of credit; and (iii) through the exchange rate. The domestic financial sector can magnify or dampen the effects of a policy impulse on output and inflation, as financing constraints in the economy become more or less binding. The existence of this channel can be justified by the presence of liquidity constrained agents or by balance sheet effects (for example, Bolton-Freixas (2000) or Gertler et al (2000)). The exchange rate channel is an obviously important ingredient of monetary transmission in a small, open economy like Switzerland.
73. The quarterly VAR model is estimated with two lags over the 1983-99 period, and includes changes in oil prices, German short-term interest rates, and two dummies as exogenous variables.27 The dummies are as follows: a quarter dummy for 1988Q1; and a shift variable for the 1989-90 cyclical peak. The first dummy coincides with a large drop in interest rates following the introduction if the Swiss Interbank Clearing System. A possible rationale for including the second dummy—a shift variable for the cyclical peak—is the relatively short sample period that includes only one peak. Empirically, the peak dummy improves the description of output dynamics. In principle, all variables should be stationary, and neither casual observation of the series nor formal unit root tests indicate otherwise. Errors are orthogonalized using the Choleski decomposition, with the variables ordered as listed above. The ordering corresponds to the assumed reaction speed of the variables to shocks—the output gap is assumed to be the most sluggish, while the exchange rate reacts the quickest.
74. The sample period is chosen to begin in 1983 for several reasons. First, studies (for instance, Belongia (1988), Peytrignet and Fischer (1991)) have found that structural changes took place in the money demand relationship in Switzerland at the beginning of the 1980s. It is likely that these changes lead to modifications in the monetary transmission mechanism as well. Second, eyeballing the data reveals that the output gap series was considerably more volatile prior to this period (Figure II-3), indicating a change either in measurement or in economic structure (including the nature of shocks). The results are broadly invariant to extending or shortening the sample period, but become unstable when the sample is extended to include the 1970s, and completely fall apart when the estimation is carried out excluding the 1990s.
Figure II-3.Switzerland: Output Gap, 1970-99
Sources: OECD Analytical Database; and staff calculations.
75. The impulse responses of the endogeneous variables to a “typical” (one standard deviation) short-term interest rate shock are fairly intuitive (Figure II-4).28 As the first panel indicates, an increase in the short-term interest rate has an initial perverse effect on output, and starts to have a contractionary effect on the output gap only about 1 year later.29 The output response bottoms out after 2 years. Inflation responds faster (the effect reaches a trough within one year), but then the effect of an interest rate shock becomes positive. This counterintuitive response of inflation to interest rate shocks has often been found in the literature and has been dubbed the “price puzzle”. Although the price puzzle might partially be explained by the presence of interest related items in the inflation variable, such as rents30, usual explanations include imperfect identification of policy induced interest rate shocks, and incomplete specification of the policymaker’s information set in the model.31 Higher interest rates also reduce credit and appreciate the exchange rate. Credit is scarcest after about 1½ years. The exchange rate appreciates nearly 1:1 instantaneously, but the maximum effect on the appreciation rate is not reached until after 1½ years.
Figure II-4.Switzerland: Impulse Responses to an Interest Rate Shock
Source: IMF staff calculations.
76. Other impulse responses are also fairly intuitive:
Financing shock32 (Figure II-5) is expansionary, with long-lasting effects on output, but also leads to higher inflation with a short lag. Interest rates rise, but effects on the rate of exchange rate appreciation are not significant.
A shock to the appreciation rate feeds through to lower output and lower inflation relatively fast, with the effect at maximum after 1½years in both cases (Figure II-6). In response, interest rates decline and credit supply contracts.
Figure II-5.Switzerland: Impulse Responses to a Financing Shock
Source: IMF staff calculations
Figure II-6.Switzerland: Impulse Responses to an Exchange Rate Shock
Source: IMF staff calculations
77. The impulse responses of the short-term interest rate to the various shocks are consistent with countercyclical changes in the monetary policy stance (Figure II-7). While the short-term interest rate increases in response to output, inflation, and financing shocks, an exchange rate shock (higher nominal appreciation rate) triggers a decline in interest rates. Further, interest rate shocks are persistent in international comparison.
Figure II-7.Switzerland: Response of Interest Rates to Shocks
Source: IMF staff calculations.
78. The estimated effects of an interest rate shock are quite sensitive to the sample period (Figure II-8). The output response bottoms out after 1½ to 2½years, with faster reactions obtained for the more recent sample periods. Although the transmission lags are longer than for instance in the U.S. (where the output response is usually estimated to bottom out within 1½years), they are not out of line with lag lengths found for some small European economies. The size of the estimated effects, however, is small in international comparison.33 The inflation response obtained from data for the more recent period shows a decline in inflation due to higher interest rates.
Figure II-8.Switzerland: Effects of a Unit Shock to Interest Rates on Output Gap and Inflation
Source: Fund staff calculations.
79. “Shutting down” either the exchange rate or the credit channel of transmission dampens the impulse response of both output and inflation (Figure II-9).34 In particular, when the exchange rate channel is eliminated, no significant output response to an interest rate shock is detected. This may indicate the presence of an “expenditure switching” effect, with foreign demand making up for weaker domestic demand when domestic interest rates rise but the exchange rate does not change. Similarly, closing the credit channel leads to a less pronounced and less speedy transmission from interest rate shocks to output.
Figure II-9.Switzerland: Effects of a Unit Shock to Interest Rates With and Without the Exchange Rate and Credit Channels
Source: Fund staff calculations.
80. The VAR’s estimated impulse responses give some indication as to what weight to assign to domestic interest rates, the exchange rate, and credit developments when describing the degree of tightness in monetary conditions. Figure II-10 shows the output effects of 1 percentage point contractionary shocks to these three factors. After about 6 quarters, the effects of an interest rate shock are about twice as large as the effects of an exchange rate shock. Over this horizon, a (negative) financing shock has an output effect comparable to an exchange rate shock. This would suggest giving interest rates, exchange rates, and credit the respective weights 2:1:1 when constructing an indicator of monetary conditions.35 However, identifying shocks is not feasible in practice, and monetary condition indices usually rely on the total change in interest and exchange rates. Weighting these two factors 2:1 to 3:1 would appear to be broadly consistent with the findings. Augmenting such an MCl with an indicator of credit developments would in general amplify measured changes in monetary conditions. For example, an MCI based only on interest and exchange rates would show an increase of about 4 percent between 1993 and late 1996, while an “augmented” MCI would register an increase of about 8 percent.
Figure II-10.Switzerland: Output Effects of 1 Percent Contractionary Shocks
Source: Fund staff calculations.
A more structured look at output, inflation, and monetary policy
81. The impulse responses obtained in the previous subsection motivate imposing some restrictions on the links between key macroeconomic variables. In particular, the VAR results point to a simple policy feedback rule connecting interest rates to output, inflation, and the exchange rate. This, along with more parsimonious equations that can be given behavioral interpretations—an output process (an aggregate demand relationship) and an inflation process (an aggregate supply relationship)—complete a small system that helps shed further light on the monetary transmission process.
82. The output process links output growth (measured as quarterly change in log GDP) to lags of the output gap, external market growth, real appreciation, and real interest rates. In addition to these variables, two dummies are included: a dummy for 1991Q1 (coincident with the German unification), and a dummy variable spanning the 1989–90 cyclical peak. Table II-2 presents the regression results. The coefficients have the expected sign, and the negative effect of the real interest rates on growth is significant at the 10 percent level (with p-value 0.06). On the other hand, the negative effect of real appreciation is not statistically significant (p-value = 0.19). The magnitude of the estimated effects is consistent with the findings based on a VAR—a unit increase in the real interest rate has about twice the effect of a unit real appreciation on output after 5 quarters.
Dependent variable: Real GDP growth
Sample: 1983:1 1999:4
Included observations: 68
|Variable 1/||Coefficient||Sth. Error||t-statistic||p-value|
|Output gap (-1)||-0.09||0.05||-1.78||0.08|
|Market growth (-1)||0.06||0.04||1.33||0.19|
|Change in exchange rate (-2)||-0.04||0.02||-1.55||0.13|
|Change in exchange rate (-3)||-0.01||0.02||-0.33||0.74|
|Change in exchange rate (-4)||-0.01||0.02||-0.52||0.61|
|Real interest rate (-2)||-0.27||0.09||-3.09||0.00|
|Real interest rate (-3)||0.35||0.11||3.26||0.00|
|Real interest rate (-4)||-0.21||0.08||-2.78||0.01|
|91 Q1 dummy||-0.02||0.01||-2.94||0.00|
|R-squared||0.50||Mean dependent var||0.00|
|Adjusted R-squared||0.41||S.D. dependent var||0.01|
|S.E. of regression||0.00||Akaike info criterion||-7.63|
|Sum squared resid||0.00||Schwarz criterion||-7.27|
83. The inflation process models four-quarter inflation as a function of the economy’s cyclical position, the inflation objective, nominal appreciation, oil price increases, and its own lags.36 As Table II-3 shows, all variables have the expected sign. The null hypothesis that the long-term coefficient on the inflation objective variable is equal to one cannot be rejected at conventional significance levels. Furthermore, leads and lags of the inflation goal are insignificant. Coefficient estimates remain similar when the sample is extended back to 1976; the major difference appears to be in the estimated greater persistence of inflation in the longer sample.37
Dependent variable: Inflation
|Sample: 1983:1 1999:4 Included observations: 68||Sample: 1983:1 1999:4 Included observations: 68||Sample: 1976:1 1999:4 Included observations: 96|
|Variable A23||Coefficient||Sth. error||t-statistic||p-value||Coefficient||Sth. error||t-statistic||p-value||Coefficient||Sth. error||t-statistic||p-value|
|Inflation target (-4)||0.05||0.10||0.49||0.62|
|Inflation target (4)||0.04||0.08||0.57||0.57|
|Change in exchange rate (-1)||-0.03||0.02||-1.53||0.13||-0.03||0.02||-1.59||0.12||-0.03||0.02||-1.86||0.07|
|Change in exchange rate (-2)||0.00||0.02||0.20||0.84||0.00||0.02||0.20||0.84||0.01||0.02||0.59||0.56|
|Change in exchane rate (-3)||-0.04||0.02||-1.90||0.06||-0.04||0.02||-1.95||0.06||-0.02||0.02||-0.99||0.33|
|Change in exchange rate (-4)||0.00||0.02||0.16||0.87||0.00||0.02||0.27||0.79||0.02||0.02||1.01||0.32|
|Oil price inflation||0.01||0.00||4.02||0.00||0.01||0.00||4.23||0.00||0.01||0.00||4.22||0.00|
|S.E. of regression||0.00||0.00||0.00|
|Sum squared resid||0.00||0.00||0.00|
84. The unit long-run coefficient on the inflation objective could be interpreted as evidence that monetary policy was highly credible and successfully anchored inflationary expectations in the sample period. An alternative representation of this idea is provided by Table II-4. This table presents estimates of an error correction like specification for the change in inflation. The change in inflation was regressed on the lagged deviation of inflation from the objective, the change in the objective, lagged changes in the output gap, appreciation rate and oil price inflation, and its own lags.38 When monetary policy is credible, the inflation objective is expected to act as an attractor (the coefficient on the deviation from the objective is negative), and a change in the inflation objective produces an “announcement effect” (the coefficient on the change in the objective is positive). As Table II-4 shows, neither of these hypothesis are falsified by the data.
Dependent variable: Change in inflation
|Sample: 1983:1 1999:4 Included observations: 68||Sample: 1976:1 1999:4 Included observations: 96|
|Variable||Coefficient||Sth. error||t-statistic||Prob.||Coefficient||Sth. error||t-statistic||Prob.|
|Deviation of inflation from target (-1)||-0.08||0.04||-1.85||0.07||-0.06||0.04||-1.70||0.09|
|Change in target (-4)||0.10||0.08||1.37||0.18||0.15||0.08||1.94||0.06|
|Change in output gap (-1)||-0.04||0.09||-0.46||0.65||0.06||0.05||1.11||0.27|
|Change in output gap (-2)||0.24||0.09||2.78||0.01||0.01||0.05||0.21||0.84|
|Change in nominal effective appreciation rate (-1)||-0.02||0.02||-1.21||0.23||-0.01||0.01||-1.18||0.24|
|Change in nominal effective appreciation rate (-2)||0.00||0.02||0.29||0.77||0.00||0.01||0.30||0.77|
|Change in inflation (-1)||0.44||0.10||4.55||0.00||0.40||0.08||4.95||0.00|
|Change in oil price inflation||0.01||0.00||3.41||0.00||0.01||0.00||4.68||0.00|
|S.E. of regression||0.00||0.01|
|Sum squared resid||0.00||0.00|
85. Regression results in Table II-5 characterize a simple policy feedback rule. In line with the previous discussion, short-term interest rates are taken to represent monetary policy, and a Taylor-rule like relationship39 is estimated, linking the interest rate to the current output gap, the future deviation of inflation from the objective, and the deviation of the nominal exchange rate from trend.40 In addition, a constant and a dummy for 1988Q1 are also included. All stochastic right hand side variables are clearly endogenous, so the equation is estimated using their lags as instruments. Coefficient estimates indicate a positive correlation with future inflation; a negative correlation with the exchange rate; and no correlation with the output gap. The first two findings are consistent with Cuche (2000), who also finds a strong feedback from the output gap to indicators of monetary policy.
Dependent Variable: Short term interest rate
Sample(adjusted): 1983:1 1998:4
Included observations: 64 after adjusting endpoints
|Variable 1/||Coefficient||Sth. error||t-statistic||Prob.|
|Nominal interest rate (-1)||0.76||0.07||11.27||0.00|
|Deviation of inflation from target (+4)||0.64||0.21||2.99||0.00|
|Deviation of inflation from trend||-0.09||0.04||-2.42||0.02|
|S.E. of regression||0.01|
Instruments include two lags of interest rates, gap, and nominal effective exchange rate; four lags of the deviation of inflation from the objective; the inflation objective; a dummy for 1988Q1; and a constant.
Instruments include two lags of interest rates, gap, and nominal effective exchange rate; four lags of the deviation of inflation from the objective; the inflation objective; a dummy for 1988Q1; and a constant.
86. However, this open economy Taylor rule is likely to be an overly simplistic description of Swiss monetary policy. As discussed in Section B, the SNB actively managed the exchange rate in certain periods. Correspondingly, Dueker and Fischer (1996) argue that the conduct of monetary policy in Switzerland can be well characterized by a switching rule, whereby policy responsiveness to domestic variables and to the exchange rate varies, depending on whether the central bank is in “manager” or “floater” mode. To obtain a valid picture of how output, inflation, and exchange rate expectations are linked to the interest rate, a policy feedback rule could be estimated along these lines.
D. Lessons for Monetary Policy
87. The empirical findings of Section C provide some basis for offering two lessons for monetary policy. First, in light of the relatively long and uncertain lags in the monetary transmission mechanism, the current 3-year horizon for monetary policy appears appropriate.41 Second, enhanced transparency would appear useful for two reasons. On the one hand, communicating the SNB’s inflation forecast to the public in a more effective manner would help maintain the SNB’s credibility, and there could be an added benefit from anchoring inflation expectations more tightly.42 In this regard, more detailed discussion of inflation forecasts could be useful, and—in keeping with the SNB’s practice so far—publication of point forecasts complete with the discussion of risks, rather than range forecasts, would be preferable. On the other hand, large uncertainties associated with the short-term workings of the monetary transmission machine would call for careful ex ante explanation and ex post evaluation of policy decisions. This could also contribute to a learning process gradually reducing the uncertainties.
88. The SNB, similar to other central banks, uses an eclectic approach when deriving its policy decisions: a range of indicators and models as well as non-model-based aids are combined to analyze the current economic situation and choose the appropriate monetary policy action. As simple empirical models have limited chances to fully capture the interactions between macroeconomic variables and policy instruments (as illustrated by the lack of sharp conclusions in this chapter), this approach remains fully justified.
A. Sources and Definitions
89. Quarterly data on real GDP were obtained from the OECD Analytical Database. The quarterly output gap was constructed by using the Hodrick-Prescott filter (Figure A1).
Figure A1.Switzerland: Output Gap, Inflation, Credit, Interest Rate, and Nominal Appreciation
Source: IMF staff calculations.
90. Target inflation for the post-1984 period was compiled based on information from the SNB’s Quarterly Bulletin. The Bulletin is available from 1984, and the medium-term inflation rate can be inferred for most years. Together with the monetary target, the inflation rate consistent with the monetary target over the medium run is usually reported. In the years when the medium-term inflation rate target was reported as a range, the midpoint of the range was chosen. When “target inflation” was not available from the Bulletin, Figure 2 in Dueker and Fischer (1996), p.99 was used to fill in the gap. This latter source was used to generate the series for the pre-1984 period.
91. The SNB’s data are indicative of policy intentions, and are based on information available to the policymaker prior to the year for which the target is set. Thus, they are clearly exogenous. In contrast, data from Dueker and Fischer (1996) are estimated policy parameters and hence might not be exogenous. As a result, estimates for the 1976-99 period should be treated with caution.
92. The inflation objective series is based on annual information, and changes in a lumpy fashion in the first quarter of the year by construction. This might introduce spurious correlations when seasonal effects are present. However, inclusion of quarterly dummies made no difference to any of the results presented in the paper.
93. The CPI series was obtained from the IFS (line 64). The series for inflation deviations was defined as the difference between actual inflation (4-quarter change in log CPI) and the longer run inflation target. The series for seasonally adjusted inflation deviation and for inflation deviation excluding rents are similarly defined, based on the corresponding CPI series from the Federal Statistical Office.
94. The credit variable is based on domestic credit outstanding to the private sector from the IFS (line 32d). As the credit-to-GDP ratio was strongly trending in the sample period (see Figure A1), a detrended (with a broken linear trend) version was used in the empirical work.
Short-term interest rates
95. Short-term interest rates are 3-month LIBOR rates and 3-month Swiss franc euro rates from the IFS (lines 60a and 60b). The two series are nearly identical, but LIBOR rates are not available prior to 1978.
Effective exchange rates
96. The nominal effective exchange rate series was constructed based on IFS and DOTS data (CPI and nominal exchange rate information; and trade weights, respectively). The nominal effective exchange rate is defined as a trade-weighted average (with smoothed moving weights) of the bilateral exchange rates (normalized to 1995 Q1 = 100) vis-à-vis OECD countries. The effective exchange rate is expressed in effective currency/Swiss franc, s.t. an increase indicates appreciation. The appreciation rate is defined as the 4-quarter log change in the nominal effective exchange rate. The corresponding real effective exchange rate and the real effective appreciation rate series are based on relative CPIs.
Oil prices, German and Swedish variables
97. Oil prices are expressed in U.S. dollars and are taken from the IFS (line 76a). German and Swedish short-term interest rates as well as Swedish CPI, exchange rate, and domestic credit are also from the IFS, while the source of the Swedish real GDP data is the OECD Analytical Database.
B. Specification Issues
Specification of the baseline VAR.
98. The baseline VAR model contains five endogenous and five exogenous variables.
deviation of 4-quarter inflation from medium-term target
credit to potential GDP ratio (detrended)
short-term interest rate
change in nominal effective exchange rate
German short-term interest rates
4-quarter change in oil prices
dummy for 1988 Q1 (takes the value 1 in 1988 Q1, zero otherwise)
dummy for the 1989-90 cyclical peak (takes the value 1 from 1989 Q1 through 1990 Q4, zero otherwise)
99. Of the endogenous variables, the output gap captures developments in the real economy, inflation deviation—trivially—captures price developments, the third variable is a summary statistic for financial sector developments, the forth variable characterizes monetary policy, and the last one describes effects from foreign exchange markets.
100. Including German short-term interest rates as an exogenous variable is justified on the grounds that Switzerland is a small, open economy, which has operated under a floating exchange rate regime under the conditions of international capital mobility during the period examined. Under these circumstances, world interest rates (captured by German interest rates here) are bound to influence domestic rates. Oil prices are included to control for work-wide supply shocks. The 1988 Q1 dummy coincides with a large drop in interest rates following the introduction if the Swiss Interbank Clearing System. The cyclical peak dummy is included to improve the description of output dynamics.
101. All variables are measured in logs, except for interest rates which are expressed as fractions. Inflation deviation and the change in the nominal effective exchange rate are defined as 4-quarter log changes.
102. The ordering of the endogenous variables is as listed. Assuming that output and prices react the most sluggishly follows conventional wisdom. Ordering the exchange rate last assumes that international financial markets react to shocks to all other variables within the same quarter. Considering that exchange rates are forward-looking asset prices, this assumption is also relatively uncontroversial. The relative ordering of credit and interest rates is more debatable, as these two variables are related to quantities and prices on the same market. Credit is ordered higher in the baseline specification for three reasons. First, interest rates are given a policy interpretation and it is assumed that monetary policy reacts to financial market developments within the same quarter. Second, credit outstanding is a stock variable and hence is assumed to react relatively slowly to interest rate shocks. Third, previous research43 has found some evidence of a non-zero reaction time.
103. The above system was estimated for the 1983Q1-1999Q4 period including 2 lags of the endogenous variables.44 The choice of lag length is not invalidated by the Akaike and Schwarz criteria, and is in line with the empirical literature. Figure A2 traces out the estimated impulse responses. Changing the ordering of the credit and the interest rate variable produces impulse responses which do not differ substantially from the baseline.
Figure A2.Switzerland: Impulse Responses from the Baseline Specification
Source: IMF staff calculations.
104. The estimates and impulse responses are also fairly robust to changes in the definitions of inflation and the exchange rate. For example, the baseline specification measures inflation deviation based on “raw” CPI for reasons of comparability with the studies by Ramaswamy and Sløk (1997) and Clements et al (2000). As inflation is defined as 4-quarter change in the log CPI, the lack of seasonal adjustment might not present a serious problem. However, using seasonally adjusted CPI data makes little change to the results. Nor were the impulse responses very different if the VAR was reestimated using an inflation measure that excludes rents (which represent over 20 percent of the CPI and are indexed to interest rates). Finally, the results were broadly similar when the VAR was reestimated using the bilateral Swiss franc/Deutsche Mark instead of the effective exchange rate.
BaglianoFabio and CarloFavero1998 “Measuring Monetary Policy with VAR Models: An Evaluation” European Economic Review Vol. 42. pp. 1069–1112
BelongiaMichael1988 “Stability of Swiss Money Demand: Evidence for 1982-87” SNB Quarterly Bulletin Vol. 6. No. l. pp. 68–74.
BlejerMarioAlain IzeAlfredo Leone and Sergio Werlangeds2000 “Inflation Targeting in Practice” International Monetary FundWashington DC.
BoltonPatrick and Xavier Freixas2000 “Corporate Finance and the Monetary Transmission Mechanism” MimeoIMF
BriaultCliveAndrew Haldane and Mervyn King1996 “Independence and Accountability” Bank of England Working Paper No. 49.
ClaridaRichardJordi Gali and Mark Gertler1997 “Monetary Policy Rules in Practice: Some International Evidence” NBER Working Papers No. 6254.
ClementsBenedictZenon Kontolemis and Joaquim Levy2000 “Monetary Policy Under EMU: Differences in the transmission mechanism” MimeoIMF
CucheNicolas2000 “Monetary Policy with Forward-Looking Rules: The Swiss Case” Studienzentrum Gerzensee Working Paper No. 00–10
DharShamikDarren Pain and Ryland Thomas2000 “A Small Structural Empirical Model of the UK Monetary Transmission Mechanism” Bank of England Working Paper No. 113.
DuekerMichael and Andreas Fischer1996 “Inflation Targeting in a Small Open Economy: Empirical results for Switzerland” Journal of Monetary Economics Vol. 37. pp. 89–103
EttlinFranz1994 “Economic Activity Effects of Monetary Policy Through the Interest Rate and Excahnge Rate Channels. A reduced Form Model with an Application to Switzerland.” Paper presented at the International Symposium on Economic Modelling held at the World BankWashington DC22–24June1994.
EttlinFranz1995 “Simulation Analysis of the Transmission of Monetary Policy in Switzerland” BIS
GeraatsPetra2001 “Why Adopt Transparency? The Publication of Central Bank Forecasts” European Central Bank Working Paper No. 41
GersbachHans and Volker Hahn2001 “Should the Individual Voting Records of Central Bankers be Published?” Economic Research Centre of the Deutsche Bundesbank Discussion Paper No. 02/01
GertlerMarkSimon Gilchrist and Fabio Massimo Natalucci2000 “External constraints on monetary policy and the financial accelerator” MimeoIMF
LeeperEricChristopher Sims and Tao Zha1996 “What does monetary policy do?” Brookings Papers on Economic Activity No. 2 pp. 1–78.
MartinBen and Chris Salmon1999 “Should Uncertain Monetary Policy-Makers Do Less?Bank of England Working Paper No. 99
MorsinkJames and Tamim Bayoumi1999 “A Peek Inside the Black Box: The Monetary Transmission Mechanism in Japan” IMF Working Paper WP/99/137
NelsonEdward2000 “UK Monetary Policy 1972-97: A Guide Using Taylor Rules” Bank of England Working Paper No. 120.
OrphanidesAthanasiosRichard. PorterDavid ReifschneiderRobert Tetlow and Federico Finan1999 “Errors in the Measurement of the Output Gap and the Design of Monetary Policy” Federal Reserve Board FEDS Papers No. 1999–45.
PeytrignetMichel and Andreas Fischer1991 “Agregats Monetaires Suisses: Ml Exogene ou Endogene?” SNB Quarterly Bulletin Vol. 9. No. 3. pp. 248–281.
RamaswamyRamana and Torsten Sk>k1997 “The Real Effects of Monetary Policy in the European Union: What are the Differences?” IMF Working Paper No. WP/97/160
RichGeorg1989 “Geldmengeziele und Sweizerische Geldpolitik: eine Standortbestimmung” SNB Quarterly Bulletin Vol. 7. No. 4. pp. 345–360.
RichGeorg1997 “Monetary Targets as a Policy Rule: Lessons from Swiss Experience” Journal of Monetary Economics Vol. 39. No 1. pp. 113–141.
RichGeorg2000 “Monetary Policy without Central Bank Money: A Swiss Perspective.” Paper presented at the World Bank conference on the Future of Monetary Policy.
RickettsNicolas and David Rose1995 “Inflation, Learning, and Monetary Policy Regimes in the G-7 Economies” Bank of Canada Working Paper No. 95–6.
SimsChristopher1986 “Are Forecasting Models Usable for Policy Analysis?” Quarterly ReviewFederal Reserve Bank of Minneapolis,10: 2–16
SvenssonLars1998a “Open Economy Inflation Targeting” NBER Working Paper No. 6545.
SvenssonLars1998b: “Inflation Targeting as a Monetary Policy Rule” Sveriges Riksbank Working Paper No. 62.
TetlowRobert and Peter von zur Muehlen2000 “Robust Monetary Policy with Misspecified Models: Does Model Uncertainty Always Call for Attenuated Policy?” Federal Reserve Board FEDS Papers No. 2000-28.
Prepared by Kornélia Krajnyák.
“The Bank will react appropriately to any unexpected developments, which would prove detrimental to the Swiss economy without losing sight of the goal of price stability”, Quarterly Bulletin 1997/2, p. 166. Also, the SNB “[retains] the option of deviating from its monetary course in the event of serious disruptions in the financial markets”, Quarterly Bulletin 1996/4, p.295.
Topics of the studies published in the SNB Quarterly Bulletin are probably an indicator of this process. Between 1990 and 1996, all but one volume contained at least one article on a topic closely related to monetary targeting (e.g., money demand, monetary aggregates, etc.). However, other topics also begin to surface (transparency in 1993, inflation targeting in 1994, etc.), and by 1997, none of the articles are closely related to monetary targeting.
“Transparency” of monetary policy making is defined as open, clear, and honest communication of (i) policy objectives (goal transparency); (ii) economic information such as the central bank’s assessment of the current economic situation and its forecasts complete with the appropriate data, models, and methods (economic transparency); (iii) procedures; (iv) policy decisions; and (v) operations. Similar interpretations are used by Winkler (2000), Gersbach-Hahn (2001), and Geraats (2001). “Accountability” is understood to be as the monitoring and ex post evaluation (complete with possible penalties) of the central bank’s policymaking activity.
See for instance Svensson (1998a).
An example is publishing the voting records of uninformed central bankers in Gersbach and Hahn (2001).
By construction, the inflation objective variable incorporates features of both the inflation goal and the inflation forecast.
The Akaike and Schwarz information criteria do not indicate that longer lag length is warranted, and the results remain broadly stable when the system is reestimated with longer lag length. The Appendix discusses the specification issues in more detail.
The full set of impulse responses is reported in the Appendix, Figure A2.
The effect on the growth rate turns negative earlier, with a lag of 3 quarters (at the point when the output gap starts declining).
Rents, which comprise over just 20 percent of the Swiss consumption basket, are indexed to interest rates. Although excluding interest rate induced variation in rents from the CPI would be desirable, excluding rents altogether is more problematic, as it skews CPI inflation towards non-domestic components. In the event, it turns out that the impulse functions are not greatly changed if rents are excluded from the CPI.
See for instance Sims (1986) and Leeper et al (1996). In the first case (improper identification), the impulse response could trace the reaction of prices to velocity shocks. In the second case (incomplete specification), information about future inflation available to the policymaker when making her monetary policy decision, but not included in the model, would produce the mirage of inflationary effects of a contractionary monetary policy shock.
An example of such a shock is a financial innovation which facilitates monitoring of borrowers. This is likely to ease credit constraints immediately but feed through to (measured) potential output with a delay. Another example is a shock to real estate prices, which decreases collateral and tightens financing constraints.
For instance Ramaswamy and Sløk (1997), and Clements, Kontolemis and Levy (2000) find that maximum output response occurs after 1½–2½ quarters for European countries. Real output response ranges between 0.4–1 percent, which is larger than for Switzerland. However, these estimates are based on different (level) specifications of the VAR model.
This experiment involves recalculating the impulse responses after excluding the variable in question from among the endogeneous variables and including its lagged values among the exogenous variables.
Strictly speaking, the weight of interest rate shocks should be based on the impulse response excluding the exchange rate channel.
Including quarterly dummies made no difference to the estimates.
This is consistent with Ricketts and Rose (1995).
Change in inflation is measured as d pi = pi-pi(-1)=(p - p(-4))-(p(-1)-p(-5)), where p denotes log CPI. Results from a similar specification using pi-pi(-4) as the dependent variable and appropriately defined RHS variables are qualitatively similar, but errors in that regression are serially correlated. Including quarterly dummies makes no difference in either case.
In recent years, estimating Taylor rules has developed into a minor industry. Some examples are Clarida, Gali, and Gertler (1995) for the G-3; Nelson (2000) for the U.K; and Cuche (2000) for Switzerland.
The nominal exchange rate trend is derived using an HP filter. In this equation, the level, rather than the change of the exchange rate is included, reflecting the evidence that periodically, monetary policy reacted to the level of the exchange rate.
Policy implications of uncertainties associated with the transmission mechanism are not clear-cut, cf. for instance Orphanides et al (1999) and Tetlow et al (2000). Intuitively, when the source of uncertainty is mismeasurement, then observed macroeconomic variables contain limited information for the policymaker and thus trigger relatively weak policy response. In contrast, when there is uncertainty regarding the model, more aggressive policy response might be called for.
As the inflation objective variable used in Section C shares features with both the inflation goal and the inflation forecast (see Appendix), the empirical results can be interpreted in favor of both goal and economic transparency.
Some of the literature estimates similar specifications as a vector error correction mechanism for the level of output, prices, exchange rates etc (Dhar et al (2000) represents a recent example). This route was not followed here because estimating cointegrating relationships over a time period little longer than 15 years requires a considerable stretch of imagination.