Recent Literature on Trilemma, Definition of MP Independence, and Our Strategy

Frankel et al. (2004) view MP independence as the ability of countries to set their nominal interest rates. Accordingly, countries are considered as enjoying permanent monetary autonomy if their interest rates show limited pass-through from the most relevant foreign rates (as judged from trade and financial market linkage), in the long run. Conversely, countries are thought of having temporary monetary autonomy if it takes relatively long time for local interest rates to fully reflect changes in the most relevant foreign interest rates. This definition offers a simple strategy for empirical studies. By examining the short-term interest rate in a cross-country panel, they find that there is limited scope for permanent monetary autonomy regardless of the exchange rate regime, but floaters show slower transmission of foreign interest rate changes to local rates, and thus floaters are regarded as enjoying more temporary monetary autonomy.

Though such results fit naturally with the common belief that having a floating exchange rate allows the monetary authority more room to maneuver, they were challenged recently by Rey (2013). She identifies strong common movements of cross-border capital flows, the international financial cycle, and claims that these capital flows, together with the leverage of global financial institutions, transmit monetary conditions present in core economies (such as the U.S.) into other economies. Quite importantly, those that let their exchange rates float against core countries are no exceptions. However, this has not been considered the traditional way of looking at MP independence. As Klein and Shambaugh (2013) put it, after recognizing the importance of the international credit cycle: “It is incorrect to say that monetary policy has no effect because there is also a credit channel, just as it would be incorrect to say that monetary policy has no effect because there is also fiscal policy.”

Obstfeld (2015) recognizes that the highly-integrated global financial market complicates the policy-making of central banks by adding financial stability to their objectives. However, he finds that floaters still enjoy more MP independence, in the sense that their short term interest rates are insulated from quick transmission of foreign rate changes, and thus can focus more on addressing domestic objectives.

We will look at MP independence from both of the angles discussed above, one more strictly related to the independence of the monetary policy instrument and the other more related to the lack of foreign influence in the monetary transmission channel. It becomes then essential to be explicit about the two working definition of MP independence adopted in this paper.

First, we focus on the short-term response of domestic local currency policy rates to foreign rates to assess independence. Indeed, policymakers are more interested in their immediate (short-term) ability to make autonomous choices in monetary policy. In the long run policy rates can be expected to co-move in today’s globalized financial world (as also documented in our paper) to the extent that inflation differentials are stationary (which has become more and more the case over the past decades, in part as a consequence of the wider adoption of inflation targeting to anchor inflation expectations; see Davis (2014)).

More precisely, in this first angle, we consider MP independence to be mainly about whether a country’s policy rate reaction to U.S. policy rate is related to policy choices, by analyzing: a) the degree of the short-run response of local policy rates to foreign ones; b) the extent to which such degree depends on a policy choice such as: b1) the desirability to respond to domestic factors, b2) the need to use interest rates to address fear of floating, or b3) the need to have a different monetary policy, for example because of weak business cycle correlation). For peggers, lack of independence would be ascribed simply to a high correlation with foreign rates, while for floaters, such a high correlation would not be considered lack of MP independence if it can be explained by policy choices. Note that although the main focus is on the short term, we also check the response in the long run and at different time horizons, particularly to assess at what horizon independence seems to vanish.

The second angle evaluates MP independence by inspecting the next step in the monetary transmission channel: we focus on how other local interest rates move with the local policy rate as opposed to foreign rates. The intention is to understand whether there is a direct effect of foreign rates on the local rate curve (lack of independence in the monetary transmission mechanism) or such apparent correlation simply reflects the correlation between domestic and foreign policy rates. This helps gauge whether the local policy rate is an effective policy instrument, and is related to Rey (2014) argument that there is strong foreign influence on the domestic transmission channel. For this purpose, we employ a broader coverage of the interest rate structure than other studies (for example Obstfeld, 2015), from 3-month rates to the 10-year nominal government bond yield.

In light of the size of the U.S. economy and the important role it plays in the international financial market, along with the data availability, U.S. rates are chosen as the reference foreign rates. Indeed, the U.S. policy rate explains 70 percent of the first principal component of the policy rates of the largest 30 countries since 2000 (85 percent since 1995, for 17 countries). This strategy is consistent with several academic contributions, such as Rogers, et al. (2014), who argue that U.S. monetary policy spillover to other advanced economies are much larger than the spill-backs.² In light of this strategy, we exclude euro-area countries from the sample, apart from Germany.

Our initial empirical findings are consistent with the literature that floaters have more MP independence in the short run. Yet, the distinguishing feature of our work is that our cross-country study of heterogeneous interest rate responses uncovers a much richer picture than a simple grouping of countries by floaters versus peggers would reveal. By linking such heterogeneous responses to country-specific policy objectives, we are able to conduct our analysis of MP independence on a broader basis rather than simply focusing on the lack of foreign influence. A few recent papers (Aizenman et al. 2015; Edwards 2015) are closely related and offer consistent results, although the methodology makes them more suitable for an assessment of long-term spillovers than monetary policy independence (the analysis is in “levels” as opposed to “changes”).

The rest of the paper is organized as follows. Section II and Section III outline the data and empirical strategies we use, while major results and economic discussions are presented in Section IV. Various robust checks are given in Section V. We conclude with plans for future works in Section VI.

Nesting Two Monetary Reaction Functions

Classical international economic theory asserts that a country with a fixed exchange rate regime and open capital market needs to follow the monetary policy of the foreign country it pegs to, thus the two policy rates may only differ by a risk premium (which should change with domestic characteristics as well as global appetite for the country assets):

Where i is the domestic interest rate, i* is the foreign rate, g() and h() are functions.

Exchange rate flexibility or capital controls should give a country more freedom to tailor its monetary policy to domestic targets, such as inflation, output, etc. Therefore, a Taylor-type inflation targeting rule could be used to characterize such behavior, similarly supplemented by a risk factor.

In reality, very few countries follow these extreme regimes, and most countries are likely to lie in between. It is therefore best to adopt for each country a specification that nests the two monetary reactions functions and then see how results related to available proxies for the various regimes. The nested reaction function would respond to both domestic and external factors, such as the foreign interest rates and the global financial market uncertainty (serving as proxy for global financial uncertainty and appetite for riskier assets).

Estimation Approaches

Thanks to the relative data abundance, we are able to directly work on individual country and perform country-specific regression of the following form:

We regress the local interest rate i of type j in country c on the constant, the U.S. interest rates (I, which will be specified in more details below), the VIX index, and domestic factors (X), such as inflation in the baseline model (and output gap, industrial production, local policy rate, etc. in robustness checks). Our baseline setup favors simple OLS where both dependent and independent variables are expressed in first difference, which captures the short term (one-month horizon) responsiveness of local interest rate to various external and domestic factors (many countries’ interest rates have unit-roots, but are co-integrated with the specified regressors).⁴ In Section V, we perform robustness check via an error-correction model setup where the local interest rate can have different short-term responses to regressors and different converging speeds across countries, but all countries share the same long-term relationship.⁵

We then inspect the distribution of coefficients estimated via the first-stage country-by-country regressions. In particular, we zoom in on the local policy rate regression and examine if MP is conducted differently according to country-specific characteristics, such as whether the exchange rate is fixed or floating, or whether there are capital controls imposed, etc.

Results from the country-specific analysis are then double-checked through the following panel regression estimated for both the full sample and for subsamples distinguished by different regimes of exchange rate flexibility or capital market openness:

Variables shown in the panel regression are the same as in the country-specific regressions. The panel regression sample also has the same coverage as the country-specific regressions, both including only countries with at least five years of data. In order to easily relate and compare the panel coefficients with the country-specific coefficients, we use the Pesaran-Smith mean group estimator in the panel regression.

The Baseline Economic Models

There are two main economic models underlying our estimation, with different focus on the asset substitutability in the international financial market. The Matching Model (USM) emphasizes the international substitutability within asset class, while the Yield Curve Model (USYC) places more weight on the substitutability across asset classes. Both models in the parsimonious baseline share a few common variables, i.e., the financial market uncertainty VIX, and the domestic inflation measured by the year-on-year change in CPI with one-month lag.

For each type of the local interest rates, the Matching Model uses as right-hand-side foreign rate ($I_t^{j,us}$ in Equation (1) or $I_{us,t}^j$ in Equation (2)) the corresponding type of U.S. interest rate, while the U.S. 10-year government bond yield is employed in the local EMBI spread regression. For example, when explaining the interbank rate of country c, we use the U.S. interbank rate. The short-term interest rate in the U.S. flattens out in the post-2009 period, hence the effect of the Fed’s unconventional monetary policy (UMP) on other countries cannot be captured by short-term U.S. rates. For this reason, we include the estimated shadow policy rate in the regression for every local interest rate with maturity up to two years, under the assumption that the effect of UMP would be proxied by the shadow rate.⁶ Analysis of the local 10-year government bond yield and the emerging market sovereign spread is done without the U.S. shadow policy rate, since the effect of UMP should be directly captured by the U.S. long-term interest rate movements.

The Yield Curve Model, on the other hand, focuses more on how the local interest rates respond to movements of the whole U.S. yield curve. We proxy the U.S. yield curve via two terms in the regression (which correspond to ($I_t^{j,us}$ in Equation (1) or $I_{us,t}^j$ in Equation (2)): the U.S. policy rate (the effective federal funds rate) and the differential between the U.S. 10-year government bond yield and the U.S. policy rate, so that changes in the first term represent shifts in the whole yield curve, while movements in the second term reflect the changes in the slope. The two segments of the yield curve are treated as explanatory variables in all regressions, regardless of the type of local interest rate on the left-hand side. Moreover, it is not necessary to include the U.S. shadow rate in the Yield Curve Model since the long-term premium already captures the influence of UMP.

Robustness via Augmented Models

The robustness of the baseline model is then checked via augmented models which encompass various additional variables.

First, the estimated response of local policy rate to U.S. policy rate might simply reflect the fact that both countries are heavily influenced by the same underlying global business cycle, rather than pure U.S. interest rate pass-through. Second, central banks may choose to set interest rates in response to changes in the exchange rate. Third, the time-series movement in sovereign risk premia would not only depend on global risk factors (captured by the VIX), but would also reflect movements in domestic risk factors (captured by country sovereign spreads which are not included in the baseline model as they reduce the sample substantially). To address these concerns, we augment our simple baseline models by including these relevant factors and check the robustness of the results.

The common global cycle is proxied via three variables: international oil price inflation, global inflation and global economic activity, the last two being proxied by the principal component of inflation and of growth in industrial production (IP) of individual countries, respectively.⁷ Then, the lag of the year-on-year change in the nominal effective exchange rate is included in the augmented models to capture the exchange rate pressure faced by the monetary authorities. The sovereign risk premium for emerging markets is proxied by the EMBIG spread, and for all other countries by the five-year credit default swap. When we investigate whether the effect of U.S. rate onto local rates (excluding the policy rate) occurs directly or indirectly via the local policy rate, we also add the local policy rate to the augmented regressions for non-policy rates.

A. Creating a short term (ST) Baseline: USM and USYC Models

This section discusses the results for the country-by-country regressions related to the baseline model in equation (1), and especially the characteristics of the coefficients. The model is mainly estimated in month-on-month differences as we want to focus on the short-term transmission of foreign variables which is more relevant than the long term one for the discussion related to the dilemma-trilemma (as a robustness check, later in Section V, we will compare the country-by-country short-term coefficients with the ones implied by an heterogeneous panel error-correction model). However, we show in Appendix Charts 2 and 3 the level regression for Brazil and Mexico, to indicate how well a model based on these few variables can capture the general pattern of the local interest rates; we also compare short-term and long-term coefficients (see Section IV.B. and the related Appendix Figure 6).

Foreign variables appear particularly relevant. Figure 1 illustrates that foreign variables are very important in explaining the ST changes of local interest rates. The six panels compare the adjusted R-squared in the baseline model and in models solely dependent on domestic factors for the six local interest rates (hence excluding the EMBI spread). Due to the frequent changes exhibited by the interest rates over the one-month horizon, the country-specific model fit as measured by adjusted R-squared is usually moderate at between 0.05 and 0.2. Nonetheless, it is clear that the inclusion of foreign variables (U.S. interest rates and VIX) greatly improves the fit of the model at both the short end and the long end of the local yield curve (as expected, especially at the long end). Local deposit and lending rates show relatively larger influence from domestic conditions, but nonetheless foreign conditions still seem to be highly relevant for many countries in the sample.

View Full Size

Relevance of Foreign Variables (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

The adjusted r-square from the baseline USM model is displayed on the horizontal axis. The vertical axis reports the adjusted r-square from including only domestic factors: blue dots correspond to regressions with only domestic CPI, while red dots include also domestic output gap. Dots appearing below the 45-degree line illustrate that models augmented by foreign variables yield better fit. From upper-left to lower-right, clockwise, the six local rates are policy rate, interbank rate, bank deposit rate, bank lending rate, 2-year government bond nominal yield and 10-year government bond nominal yield, respectively.

• Download Figure
• Download figure as PowerPoint slide

The empirical distribution of short-run coefficients offers insights as to the characteristic of such coefficients. Figure 2 presents such distributions for the country-specific regressions related to the Matching Model (upper panel) and the Yield Curve Model (lower panel), via box plots. The following can be observed:

• The response to the corresponding U.S. rates is somewhat positive for all rates and becomes stronger for longer maturities.⁸ The negative effect of long-term foreign rates on the spread is consistent with a less-than-one effect on long-term local rates (indeed a difference between these two coefficients equal to one would indicate that local currency and foreign currency interest rates move by exactly the same amount). It is worth noting that the U.S. non-policy interest rates react almost one-to-one to the U.S. policy rate even in the short run (those coefficients are reflected in red cross in the bottom left panels);

• VIX has a strongly positive impact on the EMBI spread, but not the local rates, possibly suggesting either that the effect of VIX on exchange rates allows the local rates to remain more insulated, or that risk goes up more for foreign currency issuances when global uncertainty rises as local currency issuances can be monetized;

• The response to U.S. shadow policy rate in the Matching Model is positive for local 2-year government bond yield, implying that the UMP in U.S. helped lower the medium term interest rates not just in the U.S. but also in other countries;

• In the Yield Curve Model, the U.S. long-term premium (the U.S. ten-year government bond yield minus the U.S. policy rate) mainly affects the medium/long term interest rates in other countries, with the expected positive sign;

• All local interest rate considered in the analysis respond positively to domestic inflation in both models, with medium/long term rates less affected.

View Full Size

Effect Across the Curve (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

In USM, the four charts shown in order are coefficients of the U.S. matching rate, the U.S. shadow policy rate, VIX, and the domestic CPI, respectively. In USYC, the four charts are coefficients of the U.S. policy rate, the U.S. ten-year premium over policy rate, VIX and the domestic CPI. The seven bars in each chart denote coefficients from the local policy rate regression (PO), local interbank rate regression (IN), local deposit rate regression (DE), local lending rate regression (LE), local two-year government bond yield regression (G2), local ten-year government bond yield regression (G10) and the EMBIG spread regression (EM). The horizontal line in each bar denotes the median of the empirical distribution, while the dark box represents the 25–75 percentiles. The whiskers of each box mark the maximum and minimum of the empirical distribution, with observations beyond 1.5 times the inter-quartile range below/above the 25/75 percentile being regarded as the outside values and thus excluded. The red crosses in USYC session indicate the estimated coefficients in the U.S. rate regressions.

• Download Figure
• Download figure as PowerPoint slide

B. The Trilemma

In this section, we will evaluate the relevance of the policy trilemma in light of our results on local policy rates presented in the last session.

Do Floating Exchange Rate Regime (and Capital Controls) Buy Interest Rate Independence in the short term (ST)?

The first result is that, in the short run, countries that let their exchange rate float indeed seem to have on average smaller response to the U.S. local policy rate. Figure 3 (left panel) presents the response of the local policy rate to the U.S. policy rate versus the average dummy for floating over the sample, and shows a downward sloping trend line that is marginally significant at 5 percent for both the Matching Model and the Yield Curve Model. Figure 3 utilizes all estimated U.S. policy rate coefficients (whether significant or not in the local policy rate regression), thus offering information from the full distribution of the heterogeneous responses; yet, restricting to significant country-specific coefficients only does not alter the qualitative story.

View Full Size

Floating/Capital Controls Buy Short-Term Independence

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

On the left of Figure 3, the U.S. policy rate coefficient in the local policy rate regression in USM (upper panel) and USYC (lower panel) is plotted against an indicator of floating to U.S., where 1 means a country floats to U.S. for the whole sample period. Value between 0 and 1 suggests that the given country underwent exchange rate regime switch during the sample period. On the right of Figure 3, the upper panel plots the shadow U.S. policy rate coefficient in USM against floating to U.S. indicator, while the lower panel shows the coefficient associated with the U.S. 10Y-policy rate differential in USYC versus the floating to U.S. indicator. In each panel, the trend line is denoted in red with the estimated slope and its p-value given at the bottom.

• Download Figure
• Download figure as PowerPoint slide

Second, peggers show clearly a high correlation with the U.S. policy rates, while floaters show a large heterogeneity of response, encompassing also the absence of foreign influence. The three peggers at the upper-left corner of the chart move, as expected, closely with the U.S. policy rate in the short run. The average pass-through among pure floaters seems to be quite close to nil, demonstrating that their policy rate is indeed less influenced by the U.S. monetary policy.

Third, what is striking in Figure 3 is the highly diverse responsiveness of floaters’ policy rates. What drives such heterogeneity of response? Only if such heterogeneity could be ascribed to a “policy choice” of whether or not to follow U.S. monetary policy, it would be appropriate to interpret the results as supporting MP independence for floaters. We will investigate this point more in details below.

These conclusions drawn from the country-specific estimates are borne out by the mean-group estimator panel regressions (Table 1), where the U.S. policy rate coefficients are strongly positive and significant for pure peggers in both models, slightly positive and significant for countries that have switched exchange rate regime during the sample period (mixed exchange rate regime), and insignificant for pure floaters. At the same time, the response of policy rates to domestic inflation becomes stronger and more significant as we move from pure peggers to pure floaters (see Appendix Figure 4).

Table 1.

Baseline Panel (ST)

*** p<0.01, ** p<0.05, 3 * p<0.1. P-values in parentheses.

Table 1.

Baseline Panel (ST)

MATCHING MODEL
VARIABLES	All	Pure Peg to US	Mixed ER Regime	Pure Float to US	Kctrl	no Kctrl	Pure Float Kctrl	Pure Float no Kctrl
MGE
US policy rt	0.1363*** (0.0074)	0.7990*** (0.0000)	0.1867** (0.0232)	0.0395 (0.5235)	0.1313 (0.1555)	0.1376** (0.0215)	0.0054 (0.9634)	0.0438 (0.5232)
VIX	0.0040*** (0.0014)	0.0053** (0.0441)	0.0059* (0.0524)	0.0030** (0.0269)	0.0054 (0.2185)	0.0036*** (0.0013)	0.0047*** (0.0057)	0.0027* (0.0659)
Linflation	0.1002*** (0.0000)	0.0016 (0.7869)	0.0719*** (0.0013)	0.1287*** (0.0000)	0.0210 (0.2476)	0.1204*** (0.0000)	0.0143 (0.3441)	0.1430*** (0.0000)
US shadow rt	0.0031 (0.8217)	0.0244** (0.0346)	0.0234 (0.4031)	0.0029 (0.8279)	−0.0384 (0.1809)	0.0137 (0.3740)	−0.0731 (0.1077)	0.0124 (0.3523)
Observations	9,371	567	2,520	5,845	1,787	7,584	472	5,373
Number of ifs_code	59	3	18	36	12	47	4	32
YIELD CURVE MODEL
VARIABLES	All	Pure Peg to US	Mixed ER Regime	Pure Float to US	Kctrl	no Kctrl	Pure Float Kctrl	Pure Float no Kctrl
MGE
US policy rt	0.0585 (0.3441)	0.8304*** (0.0000)	0.1380 (0.1733)	−0.0583 (0.4473)	0.1068 (0.3569)	0.0462 (0.5224)	−0.0615 (0.7394)	−0.0580 (0.4908)
VIX	0.0034*** (0.0093)	0.0054** (0.0422)	0.0060* (0.0595)	0.0020 (0.1421)	0.0060 (0.1930)	0.0027** (0.0188)	0.0044** (0.0210)	0.0017 (0.2606)
Linflation	0.1007*** (0.0000)	0.0014 (0.7599)	0.0749*** (0.0003)	0.1279*** (0.0000)	0.0255* (0.0796)	0.1199*** (0.0000)	0.0132 (0.3458)	0.1422*** (0.0000)
US 10y-pol rate	−0.0882*** (0.0002)	0.0379 (0.2501)	−0.0624 (0.2218)	−0.1084*** (0.0001)	−0.0263 (0.6545)	−0.1040*** (0.0001)	−0.0799 (0.3424)	−0.1120*** (0.0003)
Observations	9,683	574	2,623	6,033	1,846	7,837	493	5,540
Number of ifs_code	59	3	18	36	12	47	4	32

*** p<0.01, ** p<0.05, 3 * p<0.1. P-values in parentheses.

View Table

Table 1.

Baseline Panel (ST)

MATCHING MODEL
VARIABLES	All	Pure Peg to US	Mixed ER Regime	Pure Float to US	Kctrl	no Kctrl	Pure Float Kctrl	Pure Float no Kctrl
MGE
US policy rt	0.1363*** (0.0074)	0.7990*** (0.0000)	0.1867** (0.0232)	0.0395 (0.5235)	0.1313 (0.1555)	0.1376** (0.0215)	0.0054 (0.9634)	0.0438 (0.5232)
VIX	0.0040*** (0.0014)	0.0053** (0.0441)	0.0059* (0.0524)	0.0030** (0.0269)	0.0054 (0.2185)	0.0036*** (0.0013)	0.0047*** (0.0057)	0.0027* (0.0659)
Linflation	0.1002*** (0.0000)	0.0016 (0.7869)	0.0719*** (0.0013)	0.1287*** (0.0000)	0.0210 (0.2476)	0.1204*** (0.0000)	0.0143 (0.3441)	0.1430*** (0.0000)
US shadow rt	0.0031 (0.8217)	0.0244** (0.0346)	0.0234 (0.4031)	0.0029 (0.8279)	−0.0384 (0.1809)	0.0137 (0.3740)	−0.0731 (0.1077)	0.0124 (0.3523)
Observations	9,371	567	2,520	5,845	1,787	7,584	472	5,373
Number of ifs_code	59	3	18	36	12	47	4	32
YIELD CURVE MODEL
VARIABLES	All	Pure Peg to US	Mixed ER Regime	Pure Float to US	Kctrl	no Kctrl	Pure Float Kctrl	Pure Float no Kctrl
MGE
US policy rt	0.0585 (0.3441)	0.8304*** (0.0000)	0.1380 (0.1733)	−0.0583 (0.4473)	0.1068 (0.3569)	0.0462 (0.5224)	−0.0615 (0.7394)	−0.0580 (0.4908)
VIX	0.0034*** (0.0093)	0.0054** (0.0422)	0.0060* (0.0595)	0.0020 (0.1421)	0.0060 (0.1930)	0.0027** (0.0188)	0.0044** (0.0210)	0.0017 (0.2606)
Linflation	0.1007*** (0.0000)	0.0014 (0.7599)	0.0749*** (0.0003)	0.1279*** (0.0000)	0.0255* (0.0796)	0.1199*** (0.0000)	0.0132 (0.3458)	0.1422*** (0.0000)
US 10y-pol rate	−0.0882*** (0.0002)	0.0379 (0.2501)	−0.0624 (0.2218)	−0.1084*** (0.0001)	−0.0263 (0.6545)	−0.1040*** (0.0001)	−0.0799 (0.3424)	−0.1120*** (0.0003)
Observations	9,683	574	2,623	6,033	1,846	7,837	493	5,540
Number of ifs_code	59	3	18	36	12	47	4	32

*** p<0.01, ** p<0.05, 3 * p<0.1. P-values in parentheses.

Figure 3 also studies the relationship between interest rate independence and capital controls in the short term. The story appears similar: strong capital controls allow monetary independence (the range of responses is quite close to zero for high capital controls), while countries with open capital account show a large heterogeneity of responses. However, the results are less decisive, probably due to the fact that most countries in our sample imposed only limited capital controls for a short period of time, and such temporary and partial capital controls may not be effective in reality.⁹

When looking at the joint role of exchange rate regimes and capital controls, it is possible to notice that capital controls help peggers retain independence (see Appendix Figure 9). However, capital controls do not seem to matter for the independence of floaters.

Going forward, we will focus more on the independence associated with floating rather than capital controls, both because of the more comprehensive availability of extreme regimes (we have 36 countries classified as pure float and zero countries classified as imposing full capital controls), and because it is more policy relevant: few policymakers nowadays would consider closing their capital account completely in order to achieve monetary policy independence, but the exchange rate regime is more easily on the policy discussion table.

These results prompt two natural questions. If there is short term independence, how long does it last? What explain the large heterogeneity of responses for floaters?

How Long Does Monetary Policy Independence (Under Floating) Last?

Figure 4 shows tentative evidence pointing to a horizon of about one year. Among pure floaters, the estimated response to U.S. policy rate in panel regressions starts getting marginally significant at 10 percent over about ten-month horizon, and strongly significant after one and a half years. When employing country-specific regressions, the distribution of the coefficients suggests similar patterns (see Appendix Figure 5). It is interesting to note that countries that present a higher short-term (1 month) response, also present a higher long-term response (see Appendix Figure 6).

View Full Size

USM Panel Mean-Group Estimates and Confidence Interval (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

We repeat panel regressions for equation (2) using first difference (baseline ST model), two-lag difference (2M), etc., up to twenty-four-lag difference (2Y). The chart plots on the horizontal axis the horizon of the lag-difference, and on the vertical axis the mean-group estimator of U.S. policy rate coefficient in the Matching Model for all countries (in black), for pure peggers (in red surrounded by 90 percent confidence interval) and for pure floaters (in green surrounded by 90 percent confidence interval).

• Download Figure
• Download figure as PowerPoint slide

C. Why Not All Floaters Appear Independent?

Even if floating allows monetary policy independence in the short term, floaters may still “choose” to include foreign interest rates in their monetary reaction function (i.e. empirically would not appear independent). In general, the different response of the policy rate of floaters to foreign rates would depend on various factors such as the degree of priority assigned to domestic objectives, the degree of tolerance for large swings in the exchange rate (fear of floating), or simply the degree of co-cyclicality of the local economy with the foreign one. The investigation of these country-specific factors and their impact on monetary policy making are the main goal of this section.

Higher Priority on Domestic Objectives is Associated with Less Responsiveness to Foreign Rates

Controlling inflation is among the major mandates of central banks all over the world, and in theory, flexible exchange rate regime enables central banks to move their policy rates away from foreign interest rates if domestic inflation steps out of the desired range. We can proxy the degree of priority of the domestic inflation objective via our inflation coefficient in the original local policy rate regression. Restricting our attention to pure floaters in our sample, we indeed find a marginally significant negative relationship between the short-run responsiveness of local policy rate to the U.S. policy rate and to domestic inflation (Figure 5). This suggests that floaters that care more for domestic objectives respond less to foreign rates.

View Full Size

Following U.S. vs. Controlling Inflation (Floaters only, ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Pure floaters are defined as countries with either zero values of Jay Shambaugh “peg” indicator during the whole sample period or base countries that are different from U.S. Figure 5 compares the estimated coefficients on domestic inflation (horizontal axis) and coefficients on the U.S. policy rate (vertical axis), in the baseline country-specific local policy rate regressions.

• Download Figure
• Download figure as PowerPoint slide

Fear of Floating is Associated with More Responsiveness to Foreign Rates

Exchange rate volatility is often a concern of central banks in setting MP, especially in emerging markets with large trade and financial linkages denominated in foreign currency.

Moreover, countries with flexible exchange rate regime seem to treat the policy rate and international reserve as substitutable instruments to maintain exchange rate stability, since floaters with higher foreign interest rate pass-through seem to have relatively lower reserve volatility (Figure 6). Overall, this suggests that fear of floating seems to be correlated with high pass-through of U.S. interest rate changes onto domestic monetary policy.

View Full Size

Interest Rate vs. Reserve Intervention (Floaters only, ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Pure floaters are defined as countries with either zero values of Jay Shambaugh “peg” indicator during the whole sample period or base countries that are different from U.S. Reserve volatility is calculated as the standard deviation of annual change of international reserves (as percent of GDP). Figure 7 compares such reserve volatility (horizontal axis) again with the estimated coefficients of U.S. policy rate in country-specific local policy rate regressions, for both the Matching (USM) and the Yield Curve Model (USYC).

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Do Floaters Need Different MP? (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Pure floaters are defined as countries with either zero values of Jay Shambaugh “peg” indicator during the whole sample period or base countries that are different from U.S. Figure 7 compares economic cycles with the U.S. as measured by real GDP growth (left), by output gap (middle) and by CPI (right). The charts plot economic cycles (horizontal axis) versus coefficients on the U.S. policy rate (vertical axis), in the baseline country-specific local policy rate regressions.

• Download Figure
• Download figure as PowerPoint slide

Stronger Business Cycle Synchronization is Associated with More Responsiveness to Foreign Rates

Another possible reason for high correlation between local rates and foreign rates is that countries—especially those more integrated in the global trade and financial markets—might have similar business cycles to those of major advanced economies. In this case, adopting a similar MP would constitute the “right” policy response to domestic economic fluctuations, even if full independence is warranted by floating exchange rates.

We measure the co-cyclicality of the domestic and the U.S. economy on the basis of three commonly used macro variables: the annual real GDP growth, the annual real output gap and the monthly inflation calculated as the year-on-year change in CPI. Figure 7 presents, again for floaters, three panels with these three measures of co-cyclicality on the horizontal axis, versus the usual responsiveness of local policy rate to U.S. rate on the vertical axis.

Three results are noticeable. First, all three variables suggest a positive correlation between domestic and U.S. business cycles for almost all countries in the sample (almost all dots are in the right quadrant). Therefore, it would be reasonable for floaters to conduct similar MP as the U.S. to some extent.

Second, the positive (and highly significant with respect to inflation cycle) trend line confirms the intuitive idea that floaters with more correlated business cycles with U.S. tend to have also a more correlated monetary policy.

Third, the relation between the co-cyclicality with the U.S. and the responsiveness of the local policy rate to the U.S. rate is larger for CPI, which is not surprising as inflation is generally a priority of most central banks.

When we run a small cross-country regression for the responsiveness to the U.S. policy rate on the proxies for these three arguments (importance of domestic objectives, fear of floating, and co-cyclicality with the U.S.)—controlling for the degree of capital controls, which potentially has an autonomous effect from floating on monetary policy independence—all coefficients have the expected signs (Table 2). Moreover, fear of floating and correlation with U.S. inflation appear as the most robust arguments (although the limited number of observations precludes a definitive conclusion on such ranking):

Table 2.

Why Floaters Have Different Pass-through (Floaters only, ST)

***p<0.01, ** p<0.05, * p<0.1. P-values in parentheses.

Table 2.

Why Floaters Have Different Pass-through (Floaters only, ST)

U.S. Pol. Rate coefficient
	MATCHING MODEL			YIELD CURVE MODEL
K control	0.0777 (0.8368)	−0.3405 (0.3385)	0.2843 (0.3412)	−0.0630 (0.8769)	−0.4532 (0.3417)	0.2511 (0.3908)
Own CPI coefficient	−0.3867 (0.2899)	−0.0569 (0.8802)	−0.2963 (0.3301)	−0.8276 (0.1325)	−0.4400 (0.4526)	−0.7357 (0.1240)
Reserve volatility	−0.0346** (0.0283)	−0.0426*** (0.0040)	−0.0168* (0.0933)	−0.0476* (0.0503)	0.0616*** (0.0032)	−0.0274* (0.0778)
Corr. with US: GDP	0.3088 (0.2458)			0.2440 (0.4112)
Corr. with US: Output G		0.2651 (0.4605)			0.3059 (0.4613)
Corr. with US: CPI			1.1636** (0.0136)			1.3477** (0.0146)
Observations	35	31	35	35	31	35
Adjusted R-square	0.0586	0.0781	0.2816	0.1329	0.1938	0.3517

***p<0.01, ** p<0.05, * p<0.1. P-values in parentheses.

View Table

Table 2.

Why Floaters Have Different Pass-through (Floaters only, ST)

U.S. Pol. Rate coefficient
	MATCHING MODEL			YIELD CURVE MODEL
K control	0.0777 (0.8368)	−0.3405 (0.3385)	0.2843 (0.3412)	−0.0630 (0.8769)	−0.4532 (0.3417)	0.2511 (0.3908)
Own CPI coefficient	−0.3867 (0.2899)	−0.0569 (0.8802)	−0.2963 (0.3301)	−0.8276 (0.1325)	−0.4400 (0.4526)	−0.7357 (0.1240)
Reserve volatility	−0.0346** (0.0283)	−0.0426*** (0.0040)	−0.0168* (0.0933)	−0.0476* (0.0503)	0.0616*** (0.0032)	−0.0274* (0.0778)
Corr. with US: GDP	0.3088 (0.2458)			0.2440 (0.4112)
Corr. with US: Output G		0.2651 (0.4605)			0.3059 (0.4613)
Corr. with US: CPI			1.1636** (0.0136)			1.3477** (0.0146)
Observations	35	31	35	35	31	35
Adjusted R-square	0.0586	0.0781	0.2816	0.1329	0.1938	0.3517

***p<0.01, ** p<0.05, * p<0.1. P-values in parentheses.

D. Do Foreign Rates Generate the Equivalent of a Greenspan Conundrum for Central Bankers?

The transmission from foreign rates to domestic rates creates a potential conundrum for central banks, as it makes more difficult to control their local rate curve and fully control the monetary transmission mechanism. Greenspan had a similar problem in 2004–06 when long-term rates failed to increase while he was raising the policy rate (unlike in the past, when they always responded¹⁰), and termed this situation a “conundrum” during a congressional testimony in 2005. Here we wonder more broadly about the ability of countries to influence the whole curve, not just the long end, also because often in emerging markets short-term and medium-term rates play a bigger role in the monetary transmission mechanism than long rates. The question arises from the strong influence of foreign variables on the local rates that we have documented in previous sections (see for example Figure 2).

Indeed, this question relates very closely to a mirror question: is the effect of foreign interest rates on the local curve occurring via its effect on the local policy rate? If so, then central bankers have full control of their yield curve, and the response of the local curve to foreign rate is simply the reflection of the choice of the monetary authorities to follow foreign rates coupled with their ability to influence local rates. If not, then there is a direct influence of foreign rates on local rates which diminishes the ability of central bankers to influence the local curve.

To study the transmission of foreign interest rate shocks into the domestic yield curve, we modify our baseline models to include the local policy rate in the non-policy rate regressions. Figure 8 presents the estimated effects across the curve with the last column showing the newly added estimates related to the local policy rate. Comparing it with Figure 2, the following observations stand out:

• - The short end of the local yield curve (the interbank rate and the bank deposit and lending rates) does not respond directly to the U.S. policy rate (left panels). Instead, the local policy rate has a substantial impact on these rates (right panels). Hence their association with foreign rate is indirect, via the channel of the local policy rate;

• - The inclusion of the local policy rate does not significantly alter the medium response of the local interbank rate and bank deposit and lending rates to other determinants, i.e. the U.S. shadow rate or the U.S. yield curve, financial market uncertainty (VIX), and domestic inflation (middle three panels);

• - The 2-year and 10-year government bond yields coefficients in Figure 8 do not show a big difference from Figure 2, meaning that local policy rate has limited impact on the long end of the local yield curve, and most pass-through from the U.S. rates is the result of a direct effect. Appendix Figure 7 (which shows that the response of local LT rates to U.S. LT rates—on the vertical axis—is larger than the response to local policy rates—on the horizontal axis—i.e. dots are above the 45 degree line) confirms such result and shows that Greenspan conundrum also holds for most countries in our sample (mainly EMs). This is likely to be due to the strong global financial integration.

• - These results hold also when considering the long-term responses of the local rates to U.S. rates and to the local policy rate.¹¹

View Full Size

Effect Across the Curve with Local Policy Rate (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Figure 8 has the same presentation as Figure 2. Coefficient distribution of local policy rate in the country-specific regressions is shown in the two panels on the right.

• Download Figure
• Download figure as PowerPoint slide

Hence, central banks face a conundrum equivalent to the one of Greenspan in 2004–06 (limited ability to control the long term rates), but such a conundrum does not extend to short-term rates, over which they can exert significant influence.

This finding is quite meaningful for the MP independence debate in the sense that it establishes the linkage between policy rate independence and local short-term rates independence. In other words, if countries manage to maintain a policy rate independent from foreign influences, they gain control of the short end of their local currency rate curve.

This relates closely to Rey (2014) argument that a global credit cycle impedes central banks from operating their monetary transmission mechanism. Our results suggest that central banks retain the control of their monetary transmission mechanism via the short end of the curve (as local policy rates appear to be the conduit for the transmission of the global credit cycle) but not at the long end of the curve (where a “Greenspan conundrum” applies).

A. Do Global Factors and Exchange Rate Alter The Picture?

One possible issue that might distort our conclusion is the common cycle in the global economy. If both the U.S. and local monetary authorities respond in a similar way to the global cycle, our estimate of U.S. rate pass-through might be biased upward. To study whether this is indeed the case, we augment the baseline models by adding the following variables: (1) oil price inflation; (2) global inflation calculated as the first principle component of CPI inflation across our sample of countries; and (3) global output proxied by the first principle component of growth in industrial production across our sample.

On top of the global factors introduced above, we also add the year-on-year change of the nominal effective exchange rate (NEER) to reflect the possibility that monetary authorities may react to the change in the exchange rate when choosing their policy rate. Finally we add the local policy rate, in order to check robustness also versus the arguments presented in section IV.D (see Appendix Figure 8).¹²

The estimated impact of U.S. policy rate on local policy rate is robust to the inclusion of global factors and change in NEER (left panels, first column). Again as discussed in section IV.D, short-term non-bank rates are more prone to indirect pass-through from U.S., while medium-term and long-term rates display much larger direct pass-through. Responsiveness of local interest rates to the VIX index does not seem to be affected by the presence of global variables and NEER.

Interestingly, the local interest rates respond positively to global inflation cycle, and—quite consistently—the estimated inflation coefficient in the local policy rate regression becomes smaller. This suggests that the influence of the global cycle was captured in the baseline regression by the response to inflation, not—as one might have feared—from the response to the U.S. rates. The global industrial production and international oil price inflation on average have little impact on local interest rate, except for the emerging market sovereign spread, in which case they impose negative influence. Local currency interest rates in the average country in our analysis also shows no reaction to NEER changes, though NEER depreciation tends to be negatively related to the sovereign spread.

Since responsiveness of local policy rate to U.S. policy rate remains roughly the same despite the inclusion of global factors and exchange rate risks, our previous discussion on the trilemma remains robust.

B. Baseline versus Error-Correction Model (ECM)

The highly persistent interest rates are consistent with unit roots and cointegration.¹³ For robustness, we estimate an error-correction model widely adopted in the literature when the underlying data show long-term trend. Following Pesaran’s Pooled Mean-Group approach, we assume that individual countries share the same long-run coefficients in the panel regression, while the short-run coefficients as well as the converging speed can differ.

As can be seen in Figure 9, the coefficients estimated in our baseline model are consistent with the short term coefficients implied by the corresponding ECM. The country-specific converging speed parameter (ξ ranges from 0 to −0.15, with an average of about −0.05, which implies a half-life of deviations of about 1 year.

View Full Size

Robustness—Baseline vs. Error-Correction (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Figure 9 compares the coefficients we obtained in the baseline Matching Model (left panel) and Yield Curve Model (right panel) with the corresponding short term coefficients estimated from an error-correction model where countries share the same coefficients associated with the long term trend, but differ in their short term responsiveness and the converging speed to the long term trend. The red line denotes the 45-degree line in each comparison panel.

• Download Figure
• Download figure as PowerPoint slide

C. Adding Output Gap

The output gap, a standard component in the Taylor Rule estimation, does not appear in our baseline models because a robust measure for real GDP at monthly frequency is difficult to obtain for most countries in our sample. In this section, for robustness concern, we smoothly splice the annual output gap data retrieved from the World Economic Outlook (WEO) database and add the constructed monthly series to the regressions. A twelve-month lag is imposed for the output gap to reduce potential endogeneity problems, which are exacerbated by the fact that original data is annual.

In the country-specific regressions, the output gap indeed has a significant coefficient for most countries and, according to the adjusted R-squared, helps improve the fit of the models. However, as shown in Figure 10, adding output gap does not significantly change the response of the local interest rate to the U.S. interest rate (the local policy rate regression is used as an example), as well as to VIX (not shown), in the baseline models. The same results hold in the augmented models.

View Full Size

Robustness—Output Gap (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Figure 10 compares the coefficient of the U.S. policy rate in the local policy rate regression, with (vertical) or without (horizontal) output gap. The left panel refers to the USM model, while the right panel refers to the USYC one. The red line denotes the 45 degree line.

• Download Figure
• Download figure as PowerPoint slide

D. Structural Change during the Recent Global Crisis

As discussed above, the presence of a zero lower bound is addressed by introducing the shadow interest rate in the Matching Model and the U.S. curve in the Yield Curve Model. However, one may nonetheless wonder if restricting the sample to pre-crisis period would affect the results, as post-2009 the Fed switched to unconventional monetary policy. Cutting the sample at December 2008 offers a robustness check (although it somewhat restricts the number of eligible countries). The resulting U.S. policy rate coefficients in the local policy rate regression are plotted in Figure 11 against their counterparts in the full-sample baseline regressions. For both the Matching Model and the Yield Curve Model, the pre-2009 sample shows no systemic bias of the U.S. policy rate pass-through estimated from the full sample. Similar results hold for the VIX coefficients.

View Full Size

Robustness—Pre-2009 vs. Full Sample (ST)

Citation: IMF Working Papers 2016, 075; 10.5089/9781475522921.001.A001

Figure 11 compares the coefficient of the U.S. policy rate in the local policy rate regression, with the full sample (horizontal) and the pre-2009 sample (vertical). The left panel refers to the USM model, while the right panel refers to the USYC one. The red line denotes the 45 degree line.

• Download Figure
• Download figure as PowerPoint slide