Effects of Monetary Policy

8/13/2019 Effects of Monetary Policy

1/28

Using Dollarized Countries to Analyze the Eectsof US Monetary Policy Shocks

Tim Willems y

September 19, 2011

Abstract

Identifying monetary policy shocks is dicult. Therefore, instead of tryingto do this perfectly, this paper exploits a natural setting that reduces theconsequences of shock misidentication. It does so by basing conclusions uponthe responses of variables in dollarized countries. They import US monetarypolicy just as genuine US states do, but have the advantage that non-monetaryUS shocks are not imported perfectly. Consequently, this setting reduces theeects of any mistakenly included non-monetary US shocks, while leaving theeects of true monetary shocks unaected. Results suggest that prices fallafter monetary contractions; output does not show a clear response.

JEL-classication: E52; E31; C32

Key words: Monetary policy eects; Price puzzle; Structural VARs; Iden-tication; Block exogeneity

1 Introduction

Since monetary policy is typically not executed in an erratic fashion, identifyingrandom disturbances to the monetary instrument (the so-called "monetary policyshocks") is dicult. For this reason, the present paper has a dierent focus than

I thank Hal Cole, Martin Eichenbaum, Wouter den Haan, Alexander Kriwoluzky, John Leahy,Bartosz Ma ckowiak, Arturo Ormeo, Vincent Sterk, Christian Stoltenberg, Sweder van Wijnbergen

and participants at the 2011 SED Meeting in Ghent for useful comments and discussions. GunnarPoppe Yanez provided excellent research assistance. Any errors are of course mine.yDepartment of Economics, University of Amsterdam, Valckenierstraat 65-67, 1018 XE Amster-

dam, The Netherlands. E-mail: [email protected]. Tel.: 0031-20-5257159. Fax: 0031-20-5254254.

1


2/28

the standard structural VAR exercise. Instead of trying to nd the perfect shockidentication scheme, this paper starts by asking: if shock identication is so dicult,cant we nd a natural setting that reduces the consequences of the almost inevitable

misidentication of monetary shocks?The natural setting that I exploit to this end is the existence of dollarized coun-tries that import US monetary policy, without being perfectly integrated with the USeconomy. Consequently, non-monetary US shocks do not survive the transmissionprocess to these client economies undamaged, and the setting works a bit like anideal lter that reduces the eects of any mistakenly included non-monetary shockson dollarized country variables, while leaving the eects of the true monetary shocksuntouched.

The fact that shock identication is dicult, might explain the presence of someongoing debates in the structural VAR literature. Next to the fact that there is noconsensus on the eects of monetary shocks on output, many studies nd that pricesincrease after a monetary contraction, which goes against the predictions of moststandard macroeconomic theories (such as the New Keynesian one). Even thoughthis price response can be rationalized through the working capital channel, 1 it isgenerally referred to as "the price puzzle ".

Currently, there are three popular explanations for the positive response of pricesto an interest rate increase. First, some have argued that the working capital channelindeed is important and that prices do go up after a monetary tightening. If this istrue, then the price puzzle is not a puzzle and incorporation of the cost channel intostandard macroeconomic models seems desirable. 2

Second, it has been argued that the price puzzle reects the fact that the esti-mated VAR contains less information than available to the monetary authority. Theidea is that when the monetary authority knows that ination is about to arrive andcontracts in response, prices will still rise, but by less than they would have withoutthe contraction. Sims (1992) tries to correct for this by introducing a commodityprice index in the VAR and shows that this decreases the puzzle. Recently, thissolution has however been questioned. Hanson (2004) for example fails to nd acorrelation between the ability of variables to forecast ination and the ability toreduce the price puzzle (a similar point is made in Thapar (2008)). Moreover, Han-

1With a working capital channel in place, rms need to borrow funds in order to be able to payfor their production factors. Consequently, the interest rate becomes a determinant of real marginalcosts. Cf. Van Wijnbergen (1983), who obtained the price puzzle - avant la lettre - in such a model.

2This development has actually started already: Barth and Ramey (2001, p. 199-200) state that"cost-side theories of monetary policy transmission deserve more serious consideration". Christiano,

Eichenbaum and Evans (2005) did this by adding a working capital channel to their model. Ravennaand Walsh (2006) discuss how the cost channel aects the optimal monetary policy.

2


3/28

son shows that including commodity prices in the VAR does not work for an earlysample period, running from 1959 to 1979. These ndings thus suggest that eitherthe price puzzle is not a puzzle (it is just the working capital channel at work), or

that there is a dierent problem.In this light, it might also be the case that the necessary identifying restrictionsare not met in practice and play a distorting role. A popular way of identifying VARsis to assume that some variables do not respond to certain shocks within the period.But as for example argued in Canova and Pina (2005) and Carlstrom, Fuerst andPaustian (2009), these inertial restrictions may not hold in reality as a result of whichshocks can be misidentied and resulting IRFs (such as the one for prices) can getthe wrong sign.

This paper focuses at this last problem. Hereby, it tries to shed light on thequestion whether economic theory should take the price puzzle seriously, or whetherit is just an artifact of shock misidentication. As I acknowledge the potential of the

working capital channel to explain the price puzzle, I do not take a prior positionon what the price eects of monetary policy shocks should be . This distinguishes myapproach from the sign restriction procedure employed by for example Uhlig (2005).That approach dismisses the cost channel and identies a contractionary monetaryshock as a shock that (among other things) does not increase prices.

To circumvent the use of sign restrictions, this paper takes advantage of a con-venient natural setting: it uses output and price data from dollarized countries (alllocated in Latin America). Previous studies have always looked at responses of USvariables to analyze the eects of US monetary shocks. That approach, however,does not use all available information. In particular, it neglects the fact that we canalso look at responses of variables in dollarized countries. Economically, they arenot so dierent from genuine US states as they import US monetary policy shocks just as normal US states do. After all, dollarized countries also use the US dollaras legal tender (just like, say, Idaho does), without having the possibility to deviatefrom US monetary policy, as there is no local currency to de- or revalue. Conse-quently, standard open economy considerations on the international transmission of monetary shocks do not play a role here. Instead, from this papers perspective, thedollarized countries can just be seen as US states that are not represented in theFederal Reserve System.

Taking this geographical detour has three advantages. Firstly, the econometricrestrictions that follow from this setting enable one to analyze the eects of US

monetary shocks in the client economies, without imposing inertial or sign restrictionson the variables of interest.Secondly, the working capital channel is generally believed to be more important

3


4/28

for the dollarized countries than it is for the US (as short-term bank nancing playsa bigger role in the former). Consequently, dollarized countries are a good candidateto test whether this channel is really capable of generating a price increase after

a monetary tightening. If we would nd that these countries are free of the pricepuzzle, this would therefore be a strong result.Finally, and most importantly, basing conclusions upon the responses of vari-

ables in dollarized countries makes this papers ndings less prone to the majorconcern any structural VAR exercise has to face: misidentication of the US mon-etary shock. This is the case because the dollarized countries that are going to beconsidered (Ecuador, El Salvador and Panama) are only imperfectly integrated withthe US economy. In particular, the economies of Ecuador and El Salvador are onlymoderately open in terms of trade-to-GDP ratios, 3 which is probably a result of thefact that these countries were rather late with decreasing their trade barriers ( cf.Sachs and Warner (1995)). In addition, to the extent that the considered dollarized

economies do trade internationally, most of it takes place with other countries thanthe US. 4 Consequently, non-monetary US shocks can be expected to produce onlyrather limited output- and price uctuations in these countries - especially at shorthorizons. 5

Lindenberg and Westermann (2010) investigated this issue empirically and theyindeed nd that Latin America does not share its business cycle with the US. 6 Thissuggests that these cycles are not driven by the same shocks, or that the shocks areonly transmitted with a delay. By looking at the already dollarized economies of El Salvador and Panama, they are also able to reject the hypothesis that business

3As reported by the CIA World Factbook, the ratio of exports (imports) to GDP equaled 0.250

(0.249) for Ecuador in 2009. For El Salvador, these numbers are 0.183 and 0.318 and for Panamathey equal 0.441 and 0.523. To compare: for a textbook open economy, such as Singapore, theseratios are 1.550 and 1.358, while the corresponding US numbers equal 0.073 and 0.110.

4According to the Factbook, 66 percent of Ecuadorian exports (73 percent of their imports)went to (came from) other countries than the US in 2009. For El Salvador, these numbers are 56percent for exports and 70 percent for imports. Finally, Panama exported 82 percent (imported 88percent) of their total to (from) non-US trading partners.

5Non-monetary shocks are typically transmitted through the time-consuming trade channel, as aresult of which they need a while to arrive at a dierent region. Monetary shocks, on the other hand,are transmitted fully and quickly through nancial markets. This observation has a long-standingtradition in international economics, going back to at least Dornbusch (1976).

6In particular, they report that the correlation in growth rates between the US and El Salvador(Panama) equals only 0.23 (0.12). Ecuador is not included in their study, but in own calculations

this correlation equals 0.30. To compare: for Canada and Mexico the correlations with US growthrates equal 0.77 and 0.73, respectively. A similar result is reported by Alesina, Barro and Tenreyro(2002): they also show that output comovement between the US and the dollarized countries underconsideration is not that high.

4


5/28

cycle synchronization is "endogenous", as they nd that the business cycles of thesecountries do not show a bigger comovement with the US cycle than those of non-dollarized countries. In line with this, Canova (2005) even nds that non-monetary

US shocks do not tend to produce signicant output or price uctuations in LatinAmerica at all.So all this suggests that even if the identied "monetary policy shock" includes

some non-monetary US components, the consequences of this mistake are containedin the dollarized countries as the transmission of these non-monetary US disturbancesto Latin America is not instantaneous and perfect.

Although this approach is certainly not awless (for that one needs to nd adollarized economy that is fully shielded from non-monetary US shocks, which thecountries I look at probably arent), it works at least a bit like an ideal lter, as itreduces (or at the very minimum: delays) the eects of any non-monetary US shockson client country variables, while leaving the eects of the true monetary shocks

unaected.The results of this exercise are univocal: when one analyzes the eects of a con-

tractionary US monetary shock through dollarized economies, prices in all clientcountries fall quickly and signicantly - so the price puzzle disappears. Quantita-tively, the data suggest that prices in the economies considered were pretty exibleover the sample period. Output does not show a clear response, so monetary neu-trality cannot be rejected.

2 Identifying monetary policy shocks

To identify structural shocks, one has to start with the estimation of a reduced formVAR with p lags for a vector of variables Z t :7

Z t = p

Xi=1 B i Z t i + u t ; (1)with Z t = X 01;t ; R t ; X

0

2;t0

. Here, X 01;t is a (k1 1)-vector whose contemporaneousvalues are assumed to be in the information set of the central bank. X 02;t (of dimension(k2 1)) on the other hand contains those variables whose contemporaneous valuesare not in the information set of the monetary authority (its lags are though).

The B -matrices in equation (1) can be estimated by OLS, after which one cancalculate the reduced form errors u t , with E u t u 0t = . However, there is nothing that

7This section draws upon Christiano, Eichenbaum and Evans (1999).

5


6/28

ensures that the residuals are contemporaneously uncorrelated, as a result of whichwe cannot give them a structural economic interpretation. We can do this once wehave transformed the variance-covariance matrix of the residuals into a diagonal one.

This can be achieved by premultiplying the reduced form (equation (1)) with A

0,a (k k)-matrix that captures the contemporaneous relations between the variablesin Z t . This gives us the structural form representation of (1):

A0Z t = p

Xi=1 A i Z t i + " t ; (2)with Ai = A0B i ; i = 1 ;:::;p; and "t = A0u t . "t is a (k 1)-vector of structural

(that is: uncorrelated) shocks with E " t " 0t = I . It then follows that = A 10 A1

00

.As shown in Christiano, Eichenbaum and Evans (1999, henceforth CEE), it is

possible to identify the monetary policy shock (under certain assumptions, to be

discussed below) by assuming that A0 has the following, lower-triangular structure:

A0 =266664

A0;11(k 1 k1 )

0(k1 1)

0(k1 k 2 )

A0;21(1 k1 )

A0;22(1 1)

0(1 k 2 )

A0;31(k 2 k1 )

A0;32(k2 1)

A0;33(k2 k 2 )

377775

(3)

A matrix that has this form, is the Choleski decomposition of .Next to the fact that the structure imposed by (3) is consistent with the infor-

mational assumption made about the central bank before, it also implies that themonetary policy instrument R t (taken to be the federal funds rate, as in Bernankeand Blinder (1992)) has no contemporaneous impact on the variables in X 1;t : it canonly aect these variables with a lag. Any variables in X 2;t on the other hand, canbe aected within the period.

A popular way to identify monetary policy shocks, employed by CEE (1999), is toassume that X 1;t includes output and prices, while the X 2;t -vector is either assumedto be empty or contains monetary aggregates. Economically, this implies that themonetary authority is able to react to changes in output and prices within the period,while the output and price eects of monetary shocks can only show up after oneperiod.

Alternatively, one can also go down the lines of Sims and Zha (2006) and assume

the opposite by allowing for impact eects of monetary shocks, but assuming thatcontemporaneous values of output and prices are not in the Feds information set.This assumption is motivated by the fact that gathering and processing the necessary

6


7/28

data takes time, as a result of which monetary policy may not be able to respond tochanges in these variables within the period.

The degree of realism of both restrictions can however be questioned: the majority

of existing theoretical models imply that output and prices already respond on impactof a shock (which the CEE-scheme does not allow for), while the availability of now- and forecasts makes the Sims-Zha approach debatable. As Canova and Pina(2005) show, the imposition of both of these type of restrictions may lead to shockmisidentication, as a result of which IRF coecients can get the wrong sign (whichmay remind some of the price puzzle).

Summarizing, there thus seems to exist a trade-o: on the one hand, one could usethe CEE-scheme, which gives the monetary authority the most realistic informationset, but this comes at the cost of having to assume that other variables in the VARcannot respond to these shocks within the period. On the other hand, one couldalso follow the Sims-Zha approach (which does allow for impact eects of monetary

shocks) but this comes at the expense of having to restrict the monetary authoritysinformation set.

As will be argued in the following section, dollarized countries can oer a solutionto this conundrum by essentially allowing us to combine the best of both approaches.

3 Dollarized countries to the rescue

To address the issues set out in the previous section, this paper does not look at theresponses of US output and US prices to federal funds rate shocks (henceforth referredto as Y US ; P US and RUS , respectively), but uses their counterparts from dollarized

countries instead ( Y D and P D ). This approach has two advantages. First, it allows usto combine the dominant existing approaches in a convenient way: we can identify USmonetary shocks in the US block of the system (with the contemporaneous values of Y US and P US in the Feds information set), while we can analyze the impact of theseshocks by looking at the responses of output and prices in the dollarized economies,without imposing inertial or sign restrictions on these variables of interest.

Second, inferring from responses of dollarized country variables makes the proce-dure less vulnerable to misidentication of the monetary shock in the US block of thesystem. As Canova (2005) for example reports, the transmission of non-monetary USshocks to Latin American countries is far from perfect. This implies that if the iden-tied "monetary policy shock" contains a non-monetary component in the US blockof the system (for example because the imposed inertial restrictions are not met inpractice), that non-monetary component does not survive the transmission processto the client country undamaged and, ideally, only the eects of the true monetary

7


8/28

shock remain. In this sense, the present paper thus rests with the idea that perfectshock identication is probably not possible, and moves on to the exploitation of anatural setting that reduces the consequences of shock misidentication instead.

Related to this, the setting considered also has the ability to "convert" endogenouspolicy responses to non-monetary disturbances, into monetary policy shocks. To seethis, consider the following example: 8 say that the US is hit by a non-monetary shock,in response to which the Fed immediately raises the interest rate. At this stage, USvariables are moved by both the direct eect of the non-monetary shock, as well asby an indirect eect, being the endogenous response of the monetary authority. Butas this response took place within the period, the sum of the two eects will look likea monetary policy shock to the observer in standard VAR-specications, which iswrong as there never was a monetary shock in reality (only the endogenous responseto the non-monetary shock). However, output and prices in dollarized countries areunlikely to be moved within the period by the non-monetary US shock. Hence, the

direct eect gets killed in the transmission process and the full, unexpected interestrate increase is a true monetary policy shock from the perspective of the dollarizedcountries. So where standard VAR exercises need random variations in the monetarypolicy instrument, the current setting is able to provide us with monetary shocks(through the eyes of the client countries) even if the Fed just responds in a perfectlypredictable way to unpredictable US developments (under the proviso that the latterare not immediately transmitted to the dollarized economies).

As noted before, I am aware that the client economies considered are probablynot fully insulated from non-monetary US shocks (especially at longer horizons), andthat next to those, there may also be aggregate world-wide shocks aecting both theclient countries and the US simultaneously. It seems reasonable to assume, however,that the exploited setting at least reduces the consequences of shock misidenticationsomewhat (a conjecture that will be investigated in greater detail in Section 5).Moreover, I show in the Appendix that the results are robust to the inclusion of acommodity price index, which could be seen as an indicator of aggregate economicconditions.

When estimating the system, this paper exploits the fact that the US economycan be taken to be exogenous to the dollarized economies (as the latter are all small).In particular, one can make use of the fact that the Fed does not pay any attentionto economic conditions in these client countries when designing its monetary policy.Building upon Lastrapes (2005), the remainder of this section sets out an estimation

and identication strategy that exploits this convenient natural setting.8I owe this example to Wouter den Haan.

8


9/28

3.1 Estimation

Let Q t be a vector stochastic process that is assumed to be generated by:

C 0Q t =q

Xi=1 C i Q t i + t (4)Here t is a normalized, white noise vector process such that E t 0t = I . To

simplify notation, I suppress the constant and linear time trend that are also includedin the analysis. Since the data on dollarized countries are not seasonally adjusted,I also include quarterly dummies in the client country block to remove any possibleseasonality in these series.

The corresponding reduced form of this model is:

Q t =q

Xi=1F i Q t i + vt ; (5)

with F i = C 10 C i ; i = 1 ;:::;q , vt = C 1

0 t and E vt v0

t = = C 10 C 1

00

.In estimating the reduced form, I impose two over-identifying restrictions. First,

regarding the absorption of shocks, I assume that the correlation between outputand prices in the client countries is solely due to their joint dependence on the USvariables. 9 This makes it possible to estimate the system eciently by OLS (seeLastrapes (2005)). As one could debate the reasonableness of this assumption, Ishow in the Appendix that the results are robust to dropping this restriction (inwhich case one must estimate the system by seemingly unrelated regressions).

Second, with respect to the emission of shocks from dollarized countries, I assumethat US variables are block exogenous with respect to the variables in the dollarizedcountries. That is: whatever happens in the client country is assumed to have noimpact on the US economy, as the former is too small to aect the latter.

Econometrically, these assumptions amount to the following: organize Q t suchthat Q t = Q Dt ; QUS t

0

, where QDt = ( Y Dt ; P Dt )0 and QUS t = ( Y US t ; P US t ; R US t )

0.10Block exogeneity of Q US t with respect to Q Dt then implies that all C -matrices will beupper-triangular. That is:

C i = C i; 11 C i; 12

0 C i; 22, i = 0 ;:::;q (6)

9That is: the equation for Y D

(in casu P D

) only contains its own lags (and lags of Y US

; P US

and R US , of course). So lagged values of P D do not enter the equation for Y D and vice versa.10 Note that the empirical results that are to follow are invariant to the ordering of the output

and price variables within each block, as I am only going to look at a shock to R US t .

9


10/28

The assumption regarding the absorption of shocks implies that C i; 11 is diagonal.Partitioning the VAR into a dollarized country- and a US-block gives:

QDtQ US t =

q

Xi=1 F

i; 11 F

i; 120 F i; 22 QD

t iQ US t i + v

1;tv2;t (7)

Equivalently, the associated variance-covariance matrix can be partitioned into:

= 11 12012 22

As set out in Hamilton (1994, pp. 309-313), block exogeneity of Q US t with respectto Q Dt implies that (7) can be re-parameterized and separated into two independentparts, least squares estimation of which is fully ecient:

Q Dt =q

Xi=1 F i; 11 QDt i +

q

Xi=0 G i QUS t i + et

Q US t =q

Xi=1 F i; 22QUS t i + v2;t ;

where:

G 0 = 12 122G i = F i; 12 G 0F i; 22 , F i; 12 = G i + G 0F i; 22 ; i = 1 ;:::;q

E et e0

t = 11 12 122 0

12

3.2 Identication

Once the system is estimated, one still has to identify the structural form (given byequation (4)). The vector moving average representation of the latter is:

Q t = ( C 0 C 1L ::: C pL p) 1 t = H 0 + H 1L + H 2L 2 + ::: t (8)

Equivalently, the vector moving average representation of the reduced form (equa-tion (5)) equals:

Q t = ( I F 1L ::: F pL p

)1

vt = ( I + J 1L + J 2L2

+ :::)vt (9)The goal now is to identify the H -matrices in equation (8), as these represent the

systems impulse response functions ( H k = @Qt + k@ t ). This can be done as follows.

10


11/28

Using that vt = C 10 t in equation (9) and comparing with (8) shows us that:

H 0 = C 10 (10)H k = J k H 0; k = 1 ; 2;::: (11)

By Hamilton (1994,p. 260) the solution to the inverted lag polynomial in (9) is:

J 0 = I (12)J k = F 1J k 1 + ::: + F pJ k p; k = 1 ; 2;::: (13)

Moreover, given the imposed restrictions, we know that H 0;11 = C 10;11 and thatH 0;21 = 0 . Through equation (10) one then obtains that = H 0H 00. Partitioning

this matrix as before (and using that H 0;21 = 0

) yields: 11 12

0

12 22 =

H 0;11 H 0

0;11 + H 0;12H 0

0;12 H 0;12H 0

0;22H 0;22H

0

0;12 H 0;22H 0

0;22 (14)

From (14), one can fully identify H 0 - and hence H k by using (11)-(13). Note thatblock exogeneity of QUS implies that H 0;22 (which is where the US monetary policyshock lives) can be identied in isolation from the QD block. Consequently, onecan copy the attractive feature of the CEE-approach and allow for contemporaneousvalues of US variables in the Feds information set by assuming that H 0;22 is lowertriangular. However, as set out in Section 2, this approach does not allow for acontemporaneous impact of US monetary shocks on US prices and output, whichis hard to defend in reality. But this is where the dollarized countries come inhandy: because the Fed does not pay any attention to conditions in these countrieswhen designing its monetary policy, there is no need for an inertial assumption here.Consequently, we can just feed the US monetary policy shock (fully located withinthe US block of the system) through the estimated system for the dollarized economy,without imposing inertial or sign restrictions in the latter.

However, this is still no guarantee that the US monetary policy shock is identiedcorrectly. But as argued before, the fact that the dollarized countries considered areonly imperfectly integrated with the US economy reduces the role played by anyincluded non-monetary disturbances, which makes the route taken by this paper less

prone to misidentication of the monetary shock in the US block of the system.

11


12/28

4 What are the eects of monetary policy shocks?

By now, we are in the position to analyze the eects of US monetary shocks on out-put and prices in dollarized economies. Quarterly output and price data are availablefor three of the largest dollarized countries: Ecuador (dollarized since 2000q1), ElSalvador (dollarized since 2001q1) 11 and Panama (dollarized since 1904, data avail-able since 1996q1). For these countries, the left panels of all gures that are to followdisplay the IRFs of output and prices to a one standard deviation, contractionaryUS monetary shock. Note that these IRFs are obtained without imposing any re-strictions on these variables. For comparison purposes, the right panels of all gurescontain the IRFs of US prices and output to the same shock - a procedure that doesneed to impose a zero response on impact under the current identication scheme.

Following Bernanke and Mihov (1998), I use data on the implicit GDP deatorand real GDP as proxies for US prices and output, respectively. The federal funds

rate is taken to be the monetary policy instrument. All US data are from the St.Louis Fed website. For the dollarized economies the GDP deator is not availableand the CPI (taken from the IMFs IFS database) is used instead. 12 Data on realGDP is obtained from the central banks of Ecuador and El Salvador and from theInstituto Nacional de Estadstica y Censo for Panama. All series (except the federalfunds rate) are logged before they enter the VAR.

Results are based upon a VAR(2), where the lag-length was selected by SchwarzsInformation Criterion. The reported 95 percent condence bands are obtained via aMonte Carlo procedure (with 5,000 replications), in which articial data was gener-ated by bootstrapping the estimated residuals. As shown in the Appendix, resultsare robust to dropping the time trend and to adding M2 or a commodity price indexto the VAR.

Ecuador Figure 1 shows the result of this exercise for Ecuador, where the VARwas estimated on data running from 2000q1 to 2010q3. Its right panel conrms thending of many earlier studies in this literature: following a monetary tightening,

11 Prior to ocial dollarization in 2000, Ecuador was already eectively dollarized as its residentshad started to use the US dollar for daily transactions in the mid-1990s and made increasinguse of US dollar denominated loans and deposits since then (Beckerman, 2001). The former ElSalvadorian currency (the coln) had already been pegged to the US dollar in 1993. So in practiceboth Ecuador and El Salvador had already been importing US monetary policy for some yearsbefore they dollarized ocially in the early 2000s.

12 This only strengthens this papers nding that the price puzzle disappears once one analyzesthe eects of monetary shocks through dollarized countries, as Hanson (2004, p. 1390) reports thatthe price puzzle is most severe when the CPI is used to measure the price level.

12


13/28

the US price level increases signicantly in a very persistent way, which is hard toreconcile with existing theories. 13 In addition, US output goes up as well, so onecould also speak of an "output puzzle" here.

Figure 1: IRFs to a contractionary monetary policy shock for Ecuador and the US.

But as shown in Canova and Pina (2005), these puzzling results may just bedue to shock misidentication - caused by the fact that the imposed zero impactrestrictions, necessary to obtain these IRFs, are not met in practice.

As set out before, including data from a dollarized country allows us to analyzethe eects of this shock without imposing the zero response on impact, while it alsohelps to lter out the eects of any non-monetary elements of the identied shock.The results of this exercise are shown in the left panel of Figure 1.

Three things are to be noted about this. First, the puzzling price response of the standard approach is now reversed: the Ecuadorian price level falls quickly andsignicantly after a monetary contraction. Second, although output does not move

13

As noted before, the price puzzle can be rationalized through the working capital channel, butthis channel only predicts a short-lived increase in the price level after a contractionary shock, whichis inconsistent with the picture shown here.

13


14/28

signicantly, the point estimate indicates that it is depressed for about a year fol-lowing the contraction. Finally, observe that the negative response of prices is quitelarge: after three quarters, prices have for example already fallen by 0.45 percent.

This suggests that prices were rather exible in Ecuador over the sample period,which might explain why output does not show a clear and signicant response.

El Salvador Figure 2 contains the results of the same exercise, now applied toEl Salvador (in which case the dataset runs from 2001q1 to 2010q3).

Figure 2: IRFs to a contractionary monetary policy shock for El Salvador and theUS.

The right panel again shows the puzzling picture that emerges for the US vari-ables, where both prices and output increase after a monetary tightening. However,if one looks at the IRFs of the dollarized country, one concludes that prices fall af-ter a contractionary monetary shock. Although GDP is depressed for the rst twoquarters, it goes up afterwards and even shows an output puzzle at longer horizons. 14

14 One should however keep in mind that this papers procedure is less suited for analyzing theresponses at longer horizons, as part of the possibly included non-monetary US shocks might havespilled over to the client countries by then.

14


15/28

The latter nding may be reminiscent of the results in Uhlig (2005): he ndsthat there are quite a few "contractionary monetary shocks" (identied with signrestrictions on US prices and monetary variables) that do not lead to a subsequent

fall in output, which leads him to conclude that "contractionary monetary policyshocks do not necessarily seem to have contractionary eects on real GDP" (p. 385).

Panama The results of applying the procedure to Panama (estimated on datarunning from 1996q1 to 2010q3) are depicted in Figure 3.

Figure 3: IRFs to a contractionary monetary policy shock for Panama and the US.

The ndings are largely in line with those for Ecuador and El Salvador: prices fallafter a monetary tightening (although the response is never statistically dierent fromzero), while output again does not show a signicant response (the point estimate isinsignicantly negative on impact, after which it becomes insignicantly positive inthe third quarter after the shock).

Summary The results for the dollarized countries thus strongly suggest thatthe price level falls after a monetary contraction. This result is robust in a sense

15


16/28

that it shows up across all countries as well as across dierent specications (seethe Appendix). Output, on the other hand, does not show such a clear response:the results for Ecuador suggest that output is depressed (insignicantly) for about

a year after a monetary contraction, El Salvador shows an output puzzle, while the(insignicant) result for Panama lies somewhere in between. Therefore, one cannotreject monetary neutrality based on these ndings. In this light, the results are quitesimilar to those reported by Faust, Swanson and Wright (2004) and Uhlig (2005),as they also nd that contractionary monetary shocks do not seem to have clearcontractionary eects on real GDP.

Lessons for dollarized countries Next to the fact that Figures 1-3 tell ussomething about the eects of monetary policy shocks in general, they are also in-formative for countries that are (considering to become) dollarized. First, the resultsshow no evidence for an important role of the working capital channel in the analyzedeconomies. This suggests that these countries do not have to be greatly concernedwith possible stagationary eects of monetary contractions, as for example warnedfor in Cavallo (1977) and Van Wijnbergen (1982).

Second, the analysis indicates that dollarized economies should be prepared forlarge spillovers from US monetary shocks on their price level. There is no evidencefor a clear spillover eect on output, however. This might be due to the fact thatprices in the client countries seem to have been quite exible over the sample period,as a result of which these countries were close to a situation of monetary neutrality.

5 Why are the results dierent?

A key question the reader is probably left with at this stage, is why the IRFs to thesame monetary policy shock look so dierent for the US and the dollarized countries.A rst and easy explanation is that the economies of the US on the one hand andEcuador, El Salvador and Panama on the other, are so dierent, that they respondin completely opposite ways to a monetary shock. This would be the case if theworking capital channel would be more important for the US, than it is for thesedollarized countries. This is however hard to imagine. First, if anything, the workingcapital channel is probably more important for emerging economies than it is for theUS, as short-term bank nancing tends to be more important in the former (Van

Wijnbergen, 1982: p. 134).15

Second, the fact that the responses of US variables15 Also see Rabanal (2007): he estimates a DSGE-model on US data using Bayesian methods

(allowing for a role for the working capital channel), but nds that the posterior probability of

16


17/28

are dicult to reconcile with any existing theory, makes them hard to believe ( cf.footnote 13).

A second possible reason for the dierences is shock misidentication. As shown

by Carlstrom, Fuerst and Paustian (2009, henceforth CFP), the "monetary policyshock" identied via the CEE-scheme may very well include non-monetary compo-nents. In particular, they show algebraically that in a New Keynesian model, theCEE-procedure actually identies a weighted combination of the true innovation inmonetary policy and a negative technology shock (the latter is CFPs explanation forthe price puzzle). However, for all dollarized countries for which data are available,the results do not show any evidence for this type of shock confusion. 16 After all,in all client economies, the identied shock (i) increases the federal funds rate, (ii)decreases the price level, and (iii) has no clear eect on output.

Apart from a contractionary monetary shock, it is hard to think of a shock thatcan have these three properties. In particular, CFPs negative technology shock

would increase the price level, which is inconsistent with (ii). Additionally, a negativetechnology shock is inconsistent with (iii), as one would then expect a clear negativeeect on output. On the other hand, the lack of a clear response in output to amonetary shock is in line with a standard model in which prices are rather exible(as suggested by (ii)).

What may play a role here are the ndings of the earlier cited study by Canova(2005), who reports that non-monetary US shocks do not seem to produce signicantoutput or price uctuations in the client economies considered. Consequently, evenif there are other disturbances present in the identied "monetary policy shock", thefact that the non-monetary components of this shock do not seem to survive thetransmission process undamaged, acts like a convenient lter that leaves us more orless with what we are interested in.

If one has a strong faith in the ltering capacity of the employed setting, anyattempt to identify the US monetary shock (currently done via the Choleski approachas in CEE (1999), to allow for contemporaneous values of output and prices in theFeds information set) is not essential. After all, if non-monetary US shocks areindeed not transmitted to the dollarized countries, then the reduced form innovationsto the Feds policy rule should produce very similar responses in these countries.

observing an increase in ination after a monetary tightening is zero - thereby suggesting that theworking capital channel is not that important for the US. Results by Gobbi and Willems (2011)(briey discussed in 6 of this paper) conrm this.

16If anything, the IRFs of output and prices in especially El Salvador look more like the responseto a positive technology shock. This is, however, not only inconsistent with what is to be expected

from theory ( cf. CFP (2009)), but also with the response of the US price level as that should fallafter a positive technology shock, which it does not.

17


18/28

And as can be veried by looking at Figure 4, this is indeed the case: if onegives a one standard deviation, reduced form "monetary policy innovation" (whichcertainly is a combination of all sorts of structural shocks) to the dollarized countries,

the IRFs of especially Ecuador and El Salvador are very similar to the original ones.For Panama, the dierences between the various IRFs are a bit larger, but this isconsistent with the fact that Panamas economy is more open than those of Ecuadorand El Salvador and therefore less capable of ltering out non-monetary US shocks.

Figure 4: IRFs in dollarized countries to a "Choleski identied" US monetarypolicy shock and to an innovation to the Feds reaction function.

6 Are the results informative for the US?

As set out in the Introduction, this paper makes no attempt to identify the USmonetary policy shock in the US itself; instead, it exploits the ltering capacity of dollarized countries to analyze the eects of monetary shocks in the latter. Theobvious cost of this approach is that one may be left wondering to what extentthe ndings for the dollarized countries considered carry over to the US economy.

18


19/28


20/28

Dierences in other rigidities (such as those in the labor market), may also play arole in this respect.

Related work by Gobbi and Willems (2011) however suggests that the present

papers conclusions do tend to carry over to the US economy itself - both qualita-tively and quantitatively: when they identify US monetary shocks by putting signrestrictions on prices in the client countries only (thus leaving the responses of USoutput and prices unrestricted), they nd that US prices tend to fall as well after amonetary contraction. This suggests that the working capital channel does not playa signicant role in the US either. In line with this papers results for the dollarizedcountries, US real GDP also does not show a clear response in their study.

7 Conclusion and directions for future research

This paper has presented an alternative way of analyzing the eects of US monetarypolicy shocks. The approach is akin to the exploitation of a natural experiment- formed by the fact that there exist countries that use the US dollar (and henceimport US monetary policy shocks), while being only imperfectly integrated withthe US economy (as a result of which non-monetary US shocks are not transmittedfully and instantaneously). 19 Consequently, basing conclusions upon the responsesof variables in these dollarized economies, makes the procedure less vulnerable tomisidentication of the monetary shock in the US block of the system. In this sense,the natural setting that this paper exploits thus works a bit like a convenient lter.

Results obtained in this way are univocal: all dollarized economies are free of the so-called price puzzle, as prices fall immediately after a monetary contraction.

The working capital channel thus does not seem to have played a major role overthe sample period in any of the analyzed client economies. This is a strong resultas many previous studies have emphasized the potential importance of this channelfor exactly these economies. Hereby, this paper gives support to the sign restrictionprocedure, as it provides empirical evidence that prices indeed fall immediately aftera contractionary monetary shock (which tends to be a key identifying assumption inthis literature).

Quantitatively, results indicate that the price eects of monetary shocks are largeand show up quickly. This suggests that prices were relatively exible in the dol-larized economies over the sample period. Consistent with this, monetary policy

19 The question why these countries chose to dollarize despite the fact that they do not seem toform an optimum currency area with the US ( cf. Lindenberg and Westermann (2010)) goes beyondthe scope of this paper.

20


21/28


22/28

Christiano, L.J., M. Eichenbaum and C.L. Evans (2005), "Nominal Rigidities andthe Dynamic Eects of a Shock to Monetary Policy", Journal of Political Economy ,113 (1), pp. 1-45.

Dornbusch, R. (1976), "Expectations and Exchange Rate Dynamics", Journal of

Political Economy , 84 (6), pp. 1161-1176.Faust, J., E.T. Swanson and J.H. Wright (2004), "Identifying VARS Based on

High Frequency Futures Data", Journal of Monetary Economics , 51 (6), pp. 1107-1131.

Gobbi, A. and T. Willems (2011), "Identifying US Monetary Policy Shocks ThroughSign Restrictions in Dollarized Countries", mimeo, University of Amsterdam.

Hamilton, J.D. (1994), Time Series Analysis , Princeton, NJ: Princeton UniversityPress.

Hanson, M.S. (2004), "The Price Puzzle Reconsidered", Journal of Monetary Economics , 51 (7), pp. 1385-1413.

Lastrapes, W.D. (2005), "Estimating and Identifying Vector Autoregressions Un-der Diagonality and Block Exogeneity Restrictions", Economics Letters , 87 (1), pp.75-81.

Lindenberg, N. and F. Westermann (2010), "How Strong is the Case of Dollariza-tion in Central America? An Empirical Analysis of Business Cycles, Credit MarketImperfections and the Exchange Rate", mimeo , University of Osnabrck.

Morand, F. and M. Tejada (2008), "Price Stickiness in Emerging Economies:Empirical Evidence for Four Latin-American Countries", Universidad de Chile, Doc-umentos de Trabajo No. 286.

Rabanal, P. (2007), "Does Ination Increase after a Monetary Policy Tightening?Answers Based on an Estimated DSGE Model." Journal of Economic Dynamics and Control , 31 (3), pp. 906-37.

Ravenna, F. and C.E. Walsh (2006), "Optimal Monetary Policy with the CostChannel", Journal of Monetary Economics , 53 (2), pp. 199-216.

Sachs, J.D. and A. Warner (1995), "Economic Reform and the Process of GlobalIntegration", Brookings Papers on Economic Activity , 1, pp. 1-118.

Sims, C.A. (1986), "Are Forecasting Models Usable for Policy Analysis?", Federal Reserve Bank of Minneapolis Quarterly Review , 10 (1), pp. 2-16.

Sims, C.A. (1992), "Interpreting the Macroeconomic Time Series Facts", Euro-pean Economic Review , 36 (5), pp. 975-1011.

Sims, C.A. and T. Zha (2006), "Does Monetary Policy Generate Recessions?",

Macroeconomic Dynamics , 10 (2), pp. 231-272.Thapar, A. (2008), "Using Private Forecasts to Estimate the Eects of MonetaryPolicy", Journal of Monetary Economics , 55 (4), pp. 806-824.

22


23/28

Uhlig, H. (2005), "What Are the Eects of Monetary Policy on Output? Resultsfrom an Agnostic Identication Procedure", Journal of Monetary Economics , 52 (2),pp. 381-419.

Van Wijnbergen, S. (1982), "Stagationary Eects of Monetary StabilizationPolicies", Journal of Development Economics , 10 (2), pp. 133-169.Van Wijnbergen, S. (1983), "Interest Rate Management in LDCs", Journal of

Monetary Economics , 12 (3), pp. 433-452.

9 Appendix

9.1 IRFs when estimated by SUR

In my baseline estimation, I follow the method set out in Lastrapes (2005) and assumethat the correlation between output and prices in the client countries is solely dueto their joint dependence on the US variables. If one drops this assumption, C i; 11in equation (6) can no longer assumed to be diagonal, which implies that laggedvalues of P D now also enter the equation for Y D and vice versa. In that case, OLS-estimation is no longer ecient and one should estimate the system by seeminglyunrelated regressions (SUR). This was done to generate the IRFs in Figure A1.

Figure A1: IRFs estimated via Lastrapes (2005) and via SUR.

23


24/28


25/28

Figure A3: IRFs for Ecuador and the US when M2 is included.

Figure A4: IRFs for El Salvador and the US when M2 is included.

25


26/28

Figure A5: IRFs for Panama and the US when M2 is included.

9.4 IRFs when an index of commodity prices is included

Finally, Figures A6-A8 show that conclusions are also not aected by the additionof the IMFs commodity price index to the VAR. Moreover, one can see from thesegures that the inclusion of commodity prices does not solve the US price puzzle forthe samples currently at hand.

26


27/28

Figure A6: IRFs for Ecuador and the US when a commodity price index is included.

Figure A7: IRFs for El Salvador and the US when a commodity price index isincluded.

27


28/28

Figure A8: IRFs for Panama and the US when a commodity price index is included.

28

Documents

Effects of Monetary Policy