Is it really more dispersed?

Abstract

The ECB’s one size monetary policy is unlikely to fit all euro area members at all times, which raises the question of how much monetary policy stress this causes at the national level. I measure monetary policy stress as the difference between actual ECB interest rates and Taylor-rule implied rates at the member state level. These rates explicitly take into account the natural rate of interest to capture changes in trend growth. I find that monetary policy stress within the euro area has been steadily decreasing prior to the recent financial crisis. Current stress levels are not only lower today than in the late 1990s, they are also in line with what is commonly observed among U.S. states or pre-euro German Länder.

Appendices

Appendices

1.1 Appendix 1: Data descriptions and calculations

My sample includes 11 euro area members (Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal, and Spain), either all 51 U.S. states (counting the District of Columbia as a separate state) or 31 U.S. states on which I have CPI data (Alaska, Arizona, California, Colorado, Connecticut, District of Columbia, Delaware, Florida, Georgia, Hawaii, Illinois, Indiana, Kansas, Kentucky, Massachusetts, Maryland, Maine, Michigan, Minnesota, Missouri, New Hampshire, New Jersey, New York, Ohio, Oregon, Pennsylvania, Texas, Virginia, Washington, Wisconsin, West Virginia). These 31 states represent 84 percent of the U.S. according to their GDP in 2006. All 16 German Länder are included in my sample (Baden-Württemberg, Bavaria, Berlin, Brandenburg, Bremen, Hamburg, Hesse, Lower Saxony, Mecklenburg-Vorpommern, Northrhine-Westphalia, Rhineland-Pfalz, Saarland, Saxony, Saxony-Anhalt, Schleswig-Holstein, Thuringia).

1.1.1 Interest rates

All interest rate data is taken from the International Financial Statistics (IFS). The nominal short-term interest rate for the euro area is given by the EONIA and is provided as quarterly average ranging from 1994q1–2012q4. The Federal Funds Rate for the U.S. and the call money market rate for the Bundesbank are on an annual basis. They range from 1976–2012 and 1973–1999, respectively.

1.1.2 Inflation rates

Inflation rates are calculated on an annualized basis from CPI indices. For the euro area the HICP is provided by the ECB, ranging from 1994q1–2012q4. The CPI data for the U.S. Fed and the Bundesbank are taken from the IFS. They range from 1976–2012 and 1973–1999, respectively. All CPI data for the euro area member states are taken from the IFS as well, ranging from 1998q1–2012q4. The CPI price data for the U.S. is only available for selected metropolitan areas and is provided by the Bureau of Labor Statistics (BLS). I relate these data to the corresponding states in which the metropolitan areas are located. If more than one metropolitan area is located in a particular state, I will take the average of these indices. Some metropolitan areas include counties of several different states. If this happens and I do not have any other series that can be exclusively related to these states, the corresponding data will be assigned equally to these states. The aggregation becomes complicated for those states, on which I have more than one series from metropolitan areas, which are ranging over different states. If this happens, I will take the weighted average of these series. The weights are given by the proportion every metropolitan area contributes to the state according to the population living in the area. Population data is from 2009 and is provided by U.S. Census Bureau. By this method I am able to obtain CPI data for 31 U.S. states. The remaining 19 states are aggregated to one artificial state for which I can calculate a CPI taking the data I have on the 31 states and the national CPI Index for the U.S. Fortunately, the results of my study are not sensitive to how the aggregation of CPI data is conducted. I do not have price data on German Länder.

1.1.3 Output gap measures

Output gap data are derived by a HP filter (with λ=1,600 for quarterly data and λ=6.25 for annual data, see Ravn and Uhlig (2002)) expressing it as the deviation of the logarithm of actual real GDP from its trend. Real GDP data for the euro area is taken from the IFS, ranging from 1990q1–2012q4 (for the ECB the series starts in 1995q1, for Ireland in 1997q1, and Greece in 2001q1). The data will be seasonally adjusted if that has not been already done by the source. For the U.S. these data is provided by the Bureau of Economic Analysis (BEA) and is available for all 51 U.S. states on an annual basis, ranging from 1976–2012. The real GDP of the German Länder comes from the German Federal Statistical Office on an annual basis, ranging from 1973–1999.

1.1.4 Natural rate of interest

From the utility maximization of an infinitely lived household I obtain a standard Euler equation explaining the optimal intertemporal allocation of consumption:²⁴

$$ u_{C}\left(C_{t}\right) = \beta E_{t}\left[u_{C}\left(C_{t+1}\right) \frac{1+i_{t}}{1+\pi_{t+1}}\right], $$

(6)

where β is the subjective discount factor and $u\left (C_{t}\right )$ a utility function depending on the level of real consumption C _t . Market clearing results in C _t =Y _t , with Y _t being real output. Substituting π _t =0 and $Y_{t}={Y^{n}_{t}}$ in Eq. 6 with ${Y^{n}_{t}}$ being the long-term natural level of output, I obtain an expression for the natural rate of interest ${r^{n}_{t}}$:²⁵

$$1+{r^{n}_{t}} = \frac{1}{\beta} E_{t}\left[\frac{u_{C}\left({Y^{n}_{t}}\right)}{u_{C}\left(Y^{n}_{t+1}\right)}\right]. $$

Assuming a constant elasticity of substitution (CES) utility function $u\left (C_{t}\right )=\frac {C_{t}^{1-\sigma }}{1-\sigma }$ and applying the logarithm I finally obtain:

$$1+{r^{n}_{t}} = -\ln\beta +\sigma\left[\ln Y^{n}_{t+1} -\ln {Y^{n}_{t}}\right] -\sigma g_{t}. $$

This equation is in line of Laubach and Williams (2003) and Garnier and Wilhelmsen (2005) describing the natural rate of interest as a linear function of trend growth. When dealing with quarterly data, I multiply the above expression by 4 to obtain annual rates. As the forecast error $g_{t}=\ln Y^{n}_{t+1} -E_{t}\left [\ln Y^{n}_{t+1}\right ]$ turns out to be small, I finally neglect it when computing the natural rate. The calibrated parameters are β=0.99 when dealing with quarterly data, β=0.96 when dealing with annual data and σ=1.4.²⁶ The natural level of output is given by the trend output obtained when calculating the output gap measures.

1.1.5 Additional variables

The country weights used in the calculations of the weighted mean and standard deviations are obtained from nominal GDP data. For the euro area as well as for the U.S. states I calculate these weights using data for 2006 taken from the IFS and the BEA, respectively. For the German Länder weights are calculated for 1998 taking data from the German Federal Statistical Office.

In the GMM equation additional instruments besides the lagged explanatory variables are included. In the estimation for the ECB the commodity price index (including oil prices) is taken from the IMF World Economic Outlook (WEO) and the real exchange rate comes from the IFS. For the U.S., the commodity price index (excluding oil prices) is taken from the WEO, whereas the real exchange rate is taken from the IFS. In the estimation for the Bundesbank the commodity price index (excluding energy prices) and a world oil price index are taken from the OECD. The real exchange rate comes again from the IFS.

Appendix 2: Robustness checks

I first comment on the robustness check for the euro area and proceed with the robustness checks for the comparison of monetary policy stress of different currency areas

1.1 Euro area

This section presents various robustness checks showing that the results discussed in the main text are fairly robust for most countries along a number of important dimensions. In detail, I conduct the following tests:

• 1. I drop the lagged interest rate from my estimation of the policy rule, so that I am now merely using Eq. 2 for calculating the artificial interest rates for every member state.

• 2. The calibration of the natural rate of interest ${r^{n}_{t}}$ is changed by setting σ=1, thus using a log-utility function in the Euler equation.

• 3. Instead of using the contemporaneous (2) to identify starting values for the lagged interest rate in the policy function, I use actual pre-euro interest rate data and take the mean rates of 1998 for every country (for Greece I take the year 2000) as starting value.

• 4. The data sample used in the estimation is extended to 2009q1 to include the financial crisis. However, I still exclude the quarters from 2009q2 onwards because of the zero lower bound.

• 5. In a last step, I calibrate the policy rule instead of estimating it. I use the original Taylor rule and augment it with the natural rate of interest ${r^{n}_{t}}$:

  $$i_{t} = -5.61 +{r^{n}_{t}} +1.5\pi_{t} +0.5 x_{t}.$$

  The coefficient for the inflation rate 1.5 and the output gap 0.5 are set according to Taylor (1993) while the coefficient of the natural rate of interest 1 is taken from the theoretical considerations of Woodford (2001). The original Taylor rule includes only a constant intercept, which is equal to one. Since the rule above includes the time-varying intercept $\bar {i}_{t}\equiv \alpha +{r^{n}_{t}}$, I set α equal to -5.61 so that the mean intercept over the sample for the ECB will be equal to 1.

Figure 8 summarizes the robustness checks for all countries included in my sample. Computing the stress measures for every of the above alternatives and for the baseline model, the figure presents the distribution of these measures by showing their minimum and maximum. For most member states this band is fairly narrow indicating that my results are robust along the discussed alternatives. As my comparison of monetary policy stress uses aggregated measures, in a next step, I analyze how these summary statistics are affected by varying the estimation equation.

Fig. 8

Euro Area: Summarizing Country-Specific Robustness Checks. The graph summarizes the robustness checks by showing the minimum and maximum of monetary policy stress measures obtained with the discussed alternatives in the appendix and the baseline model

Figure 9 contrasts the aggregated monetary policy stress under the baseline scenario with the stress obtained under the above discussed alternatives.

Fig. 9

Euro Area: Robustness of Summary Measures of Monetary Policy Stress. The graph shows the weighted standard deviation of monetary policy stress measures obtained with the discussed alternatives. Every panel includes the results of the baseline model and compares these with the results obtained when dropping the lagged interest rate from the estimation equation (1), using a log-utility function for the natural rate of interest (2), using pre-euro data for identifying starting values (3), extending the estimation sample to include the financial crisis (4), and using a calibrated instead of an estimated rule to compute artificial country-specific rates (5)

Every panel includes the weighted standard deviation of these measures obtained under the baseline scenario together with one or two robustness checks.²⁷

Starting with the first two alternatives, the upper left panel of Fig. 9 shows that neither modification alters the results significantly. Dropping the inertia term merely increases the variability of the policy rate. Changing the calibration of ${r^{n}_{t}}$ does not alter the results at all. This is because the constant included in the estimation equation together with the coefficient of ${r^{n}_{t}}$ absorb changes in the calibration of the discount factor β and the intertemporal elasticity of substitution σ. In the upper right panel, I analyze the impact of alternative assumptions regarding the starting values for the lagged interest rate in the policy function together with the effect of extending the estimation sample. Using actual pre-euro interest rate data as starting values, monetary policy stress is lower in 1999, but starts to align with the stress levels under the baseline scenario soon after. This suggests that the convergence of actual interest rates in the immediate run-up to the euro was out of line with the economic conditions of member states. Prolonging the sample and including the financial crisis in the estimation sample has negligible effects on the aggregated stress level. In a last exercise, I contrast in the lower left panel the baseline scenario with the results obtained with a calibrated rule. While the stress under the calibrated rule also exhibits a clear trend over the period of study, the dispersion fluctuates more strongly as the calibrated rule does not include an element of inertia.

1.1.1 Cross-currency union comparison

I conduct two robustness checks for the cross-currency union comparison showing that the results are robust to the number of states included in a currency area. Starting with the comparison of the euro area with the U.S., I check if there is a law of large number effect in the U.S. results, as the Fed has to balance more states than the ECB. To that end, I compute the aggregated stress measure based on only the largest eleven U.S. states for which I have CPI data.²⁸ Figure 10 depicts the weighted standard deviation of monetary policy stress from 1977 onwards and shows that this does not significantly alter the comparison with the euro area.

Fig. 10

Euro Area and U.S.: Varying Number of U.S. States. The graph shows the weighted standard deviation of monetary policy stress measures for the euro area and the U.S. It compares the results obtained with the baseline model with monetary policy stress for an artificial currency area consisting of the 11 biggest U.S. states (for which CPI data are available)

In a second robustness check, I consider all U.S. states when computing monetary policy stress. In the comparison of results for the euro area, U.S., and Germany I narrow down the model and proxy state-specific inflation with national inflation to make the model comparable across all areas. This means, I am no longer limited to include only 31 U.S. states, on which I have CPI data, but I can distinctively include all 51 U.S. states in my sample. Figure 11 shows that this does not significantly alter the comparison of the currency areas.

Fig. 11

Euro Area, U.S., and Germany: Varying Number of U.S. States. The graph shows the weighted standard deviation of monetary policy stress measures for the euro area, the U.S., and Germany. It compares the results of the baseline model where the U.S. results are obtained by using only a subsample of 31 states (and aggregating the data for the remaining states) with the results obtained when distinctively differentiating between 51 U.S. states