Shocked by the world! Introducing the three block open economy FAVAR

Transcription

1 Shocked by the world! Introducing the three block open economy FAVAR Özer Karagedikli Leif Anders Thorsrud November 5, 2 Abstract We estimate a three block FAVAR with separate world, regional and domestic blocks. The domestic block is assumed to be too small to influence the region, and the region is assumed to be too small to influence the world. The model is tested on New Zealand data, where the region is defined as Oceania. We identify three world shocks (world demand, world supply and world interest rate shocks), two region specific shocks (regional demand and regional supply shocks), and a domestic monetary policy shock using two different identification schemes (recursive and sign restrictions). Our main results are as follows: (i) World shocks have significant effect on both the region and the New Zealand economy. (ii) The effects of the world inflation factor (directly or indirectly) are very important in the transmission of international shocks to the New Zealand economy (iii) The regional price factor seems to be especially important for non-tradables inflation in New Zealand. (iv) The responses to a domestic monetary policy shock are all in accordance with economic theory and exhibit none of the well known puzzles. The views expressed in this paper are those of the authors and do not necessarily reflect the views of the Reserve Bank of New Zealand. The authors would like to thank Güneş Kamber, Haroon Mumtaz, and members of the modeling team at the Reserve Bank of New Zealand for valuable comments and fruitful discussions, and one anonymous referee for providing comments on an earlier draft. Any remaining errors are the authors s sole responsibility. Reserve Bank of New Zealand. Ozer.Karagedikli@rbnz.govt.nz Norges Bank and Reserve Bank of New Zealand. leif-anders.thorsrud@rbnz.govt.nz

2 Introduction The last decades have been characterised as the globalisation era. The share of total trade to world GDP has increased significantly, while liberalisation of economic policies and financial markets have boosted financial integration. At the same time different regions of the world experience different growth paths, their own shocks and display heterogeneity in response to global developments. For policy institutions in small open economies it is important to understand how these international developments transmit into the domestic economy. In this paper we propose and develop a three-block factor augmented vector autoregression (FAVAR) with separate world, regional and domestic blocks. The model enables us to describe the transmission mechanisms through which international and regional demand and supply shocks, international interest rate shocks and domestic monetary policy shocks affect a small open economy. We apply the model to New Zealand data, where we have defined the region as Oceania. Investigating the transmission mechanisms and contribution of international shocks to domestic macro variables is high on the research agenda, and there is a large literature on the existence and effects of a world business cycle on different economies. In a seminal paper Kose et al. [23] show that for the period a common world factor accounts for a large part of the business cycle fluctuations in developed economies 2 (examples of related studies include Backus et al. [995], Baxter and Kouparitsas [25] and Kose et al. [28]), while a recent and emerging literature argues that domestic inflation in many countries has become more of an international phenomenon, driven by international factors (see for example Mumtaz and Surico [28], Monacelli and Sala [29] and Ciccarelli and Mojon [25]). 3 Similar global output and price pressures also affects the synchronisation of domestic interest rates, as more countries conduct counter-inflationary monetary policies. The role of international financial shocks and their transmissions across economies The global economy has become more open. In 99 the total value of trade was less than 4 percent of global GDP; by 24 the world economy had grown 5 percent and two-way trade exceeded 55 percent of global GDP ( contents/section6_.htm. 2 Interestingly however, they find the world and regional factor to explain much less of the business cycle fluctuations in New Zealand compared with other developed economies. 3 Karagedikli et al. [2] argue that the co-movement in headline inflation rates found in these studies come from high co-movements in relative prices, which is found to be greater with products that are traded more for example.

3 have been well recognised and also highlighted by the recent global financial crisis. Mumtaz and Surico [29] for example, show how a shock to short term interest rates in the rest of the world affects a broad range of domestic UK variables. Not all variation in domestic variables can be attributed to international developments. The role of regional factors are also found to be important. Forni et al. [27], using a generalised dynamic factor model to data of European economies found that a common European activity factor explains between 35 percent and 96 percent of the variation in country level GDP. Clark and Shin [2], found that a common factor explains a substantial proportion of variations in industrial production of European economies. Hall and McDermott [28] finds the existence of a regional business cycle in the Oceanian region, while Coleman [27] finds a high long term co-movement between the relative prices of tradable, and also non-tradable goods between Australia and New Zealand. 4 Conway [998] found that the shocks from Australia are equally as important for the New Zealand economy as the shocks from the US economy, while Dungey and Fry [23] in a three block structural factor autoregression finds the Japanese economy needs to be taken into account when estimating the effects of the rest of the world on the Australian economy. Traditionally the open and closed economy VAR literature have modelled the dynamics of economic variables by means of a few series. This limited information set approach has tended to produce a few anomalies such as the exchange rate and price puzzles. The pioneering work by Bernanke et al. [25] proposed to augment the standard VAR with latent factors extracted from large data-sets, instead of the limited information set typically used in VAR studies, to circumvent these problems. For a closed economy specification they demonstrate how the FAVAR methodology produces responses to monetary policy shocks that are in accordance with economic theory. Recently the literature has started applying the FAVAR methodology to open economy specifications as well. Boivin and Giannoni [27], Mumtaz and Surico [29], and Baumeister et al. [29] are examples of this approach, and Mumtaz and Surico [29] argue that many of the anomalies typically found in standard open economy VAR analysis are overcome by using the FAVAR methodology. Moreover, an additional advantage of this approach is that instead of estimating and identifying the responses to only a few 4 In a high frequency event study Coleman and Karagedikli [2] find strong response of the New Zealand-US dollar exchange rate to Australian specific activity and monetary policy surprises. 2

4 variables, the FAVAR approach deliver the responses to all the variables included in the data set used to estimate the latent factors. This data set can potentially include hundreds of variables. To our knowledge this is the first paper that utilises the FAVAR methodology to identify world, regional and domestic shocks. 5 Following much of the later literature we use two different identification schemes; recursive identification and sign restrictions 6, and show that by including both a large data set and regional and world factors our model is able to produce results that are in accordance with economic theory and not affected by the typical anomalies found in open and closed economy studies. 7 Our work is closely related to Kose et al. [23] approach of separating out the the effects of the world, region and the country specific factors, and the international FAVAR literature discussed above. In contrast to the work by Kose et al. [23] however, our approach allows us to identify both price and monetary policy shocks in addition to the activity shocks. Our approach also utilises a much bigger domestic data set, which allows for a richer description of the domestic responses to different international and regional shocks. Further, we believe that augmenting the FAVAR proposed by Mumtaz and Surico [29] with regional factors might be important for correctly identifying the heterogeneity of economic development across different regions of the world. This paper is also, loosely, related to the global VAR (GVAR) approach of Dees et al. [27]. The GVAR approach is particularly useful in linking a number of countries to each others by means of smaller VARs for each country. Moreover, it is more convenient when one tries to understands the impacts of shocks that originates in a particular country, while our approach is more convenient as we are trying to understand the effects of the world and the region on the domestic economy. Another branch of the literature has circumvented the omitted variable bias problem, common for regular VAR studies, by utilising large Bayesian VARs (De Mol et al. [28]). 8 This strand of the literature has however focused mainly on forecasting, and identification of structural shocks quickly becomes difficult as the number of variables 5 For the New Zealand economy an identified FAVAR have never been applied neither for a closed or open economy. In fact, few studies have investigated the transmission of international shocks in New Zealand. 6 See for example Canova [25] and Scholl and Uhlig [28]. 7 We use information from nearly 4 variables, where the international and regional block of data accounts for over half the size of the data set. In addition to testing the models structural properties we also investigate the FAVAR s forecasting performance. 8 See Bloor and Matheson [29] for an example using New Zealand data. 3

5 included in the model gets large. Our main findings are as follows: (i) The transmission of shocks between the identified world, regional and domestic factors are all in accordance with the conventional wisdom, (ii) the world activity factor effects the New Zealand economy stronger than what have been found in many earlier studies. Especially we find the effects of the world inflation factor (directly or indirectly) to be a important factor in transmission of international shocks to the New Zealand economy, (iii) regional factors have significant effects on some key domestic macro economic variables such as wages and non-tradable inflation, highlighting the importance of regional factors and shocks, (iv) the responses to a domestic monetary policy shock are all in accordance with economic theory and exhibits no puzzles The remainder of the paper is structured as follows: Section 2 describes the model, the different identification schemes and the estimation procedure. In section 3 we report the results. First we describe the estimated factors, then we show the impulse responses of the international and regional factors as well as some main domestic variables to the identified shocks. We also give a detailed description of the variance decomposition for some of the main domestic variables. In section 4 we describe the sensitivity analysis we have conducted, compare the FAVAR results to a standard VAR approach, and summarise the model s forecasting properties. Section 5 concludes. 2 The model We assume that the world economy is driven by some fundamental unobserved factors and that these factors can be categorised into activity, prices and interest rate factors. Further we believe that regions of the world are affected differently by these factors and also that the regions themselves are affected by some fundamental regional specific factors, which we also categorise into activity and price factors. Finally we believe that the domestic economy is driven by some purely domestic factors. For monetary policy authorities the effect of the monetary policy instrument is of special interest. We thus include this as an observable factor in our model. 9 9 We do not include a regional interest rate factor in the model for two reasons: An additional factor will make the model (especially the dynamic specification of the model) bigger and require estimation of more parameters. We also believe that the domestic and regional interest rate market are very interlinked, and thus would be difficult do separately identify. 4

6 Our factor augmented vector autoregression (FAVAR) follow the general setup of Bernanke et al. [25], and extended to the international economy by Mumtaz and Surico [29]. As already mentioned, the factors are generally unobserved, and have to be inferred from the data. Thus, the model can naturally be represented in a state space form. Following Mumtaz and Surico [29], we specify the transition equation as: [ Ft R t ] = β(l) [ Ft R t ] + u t, () where F t is a set of unobserved world, regional and domestic factors, and R is a observed domestic interest rate factor. β(l) is a conformable lag polynomial of order p and u t is the reduced form residuals. The structural disturbances follow u t = Ω /2 ε t, with ε N(, ) and Ω = A (A ). The observation equation of the system is: X t = Λ F F + Λ R R + e t, (2) where X t is a N x vector, Λ F and Λ R are N x K and N x matrixes of factor loadings. Finally e t is a N x vector of idiosyncratic, zero mean, disturbances. 2. Identification In the two following sections we describe how we identify the FAVAR. Generally, restricting the factor loadings in equation 2, ensures identification of the factors, while restrictions on the lag polynomial and the ordering of the factors in equation, identifies structural shocks, ε t. 2.. Identifying the factors To identify the unobserved factors, the X matrix in the observation equation is portioned into 6 different blocks. Each block consists of either global, regional or domestic data. As already mentioned, we assume that the different blocks of data share some common underlying factors. These unobserved factors will be linear combinations of the different blocks of data, defined by the size and sign of the factor loadings. By restricting the different data blocks in X t, and also the factor loading matrix Λ in equation 2, we argue 5

7 that we can identify the unobserved factors, or the underlying driving forces of the world, regional and domestic economy. In our application the X matrix is portioned as: X t = [ X act X pri X r X act X pri X D X R], where act, pri, r, D and R. are abbreviations for activity, price, interest rate, domestic and domestic interest rate numbers respectively. A * indicates that the block consists of world or global variables, while ** indicates that the variable block consists of regional variables. The unobserved factors in equation are then related to the different blocks of data according to the restricted factor loading matrix: X act X pri X r X act X pri X D X R λ i,act λ i,pri λ i,r = λ i,act λ i,pri λ i,d λ i,r F act F pri F r F act F pri F D R + ɛ, (3) where X is the T x N block portioned matrix. λ is a the factor loading. The i reflects that each variable in the X matrix has its own factor loading, thus i =, 2,..., N. F is the identified unobserved factors, while R is an observed interest rate factor. ɛ is the idiosyncratic error term. The diagonal structure of the international part of the loading matrix above ensures that we can uniquely identify the international unobserved factors as global activity, price and interest rate factors, and regional activity and price factors. Note that we have not put any restrictions on the signs of the factor loadings. The domestic factors, F D, have not been given any economic interpretation. Restricting the domestic factors to rely on specific variables, and thereby identify them as for example inflation or output factors could easily have been done. These extra identification restrictions would however have limited the potential heterogenous responses of the domestic variables to shocks in the transition equation. We are mostly interested in how international shocks affects the domestic economy, and have thus left the domestic factors unrestricted. 6

8 Section 2.3 describes the details of the estimation procedure. Here it is sufficient to note that the unobserved factors are estimated by principal components. To avoid the rotational indeterminacy problem associated with principal component analysis, we use the standard normalisation implicit in the literature and restrict C C/T = I, where C( ) represents the common space by the factors of X in each block of data Identifying the shocks We apply two different identification schemes for the transition equation. These are the standard recursive ordering of the variables, or Cholesky identification, and the later method of sign restrictions directly on the structural shocks. For both identification schemes we assume a block exogenous structure in the transition equation. That is, the β(l) term from equation is restricted according to the scheme illustrated in equation 4 below. Regional (F ) and domestic factors (F ) do not affect the world (F ), regional factors are affected by the world, but not the domestic factors, and finally the domestic factors are affected by both the world, regional and domestic factors themselves. This scheme is in accordance with the relative sizes and the importance of New Zealand and the region. The difference between our approach and Boivin and Giannoni [27] is that we directly identify the worldwide shocks, which would affect simultaneously domestic and international (both world and regional) factors. Ft Ft F t β (L) = β 2 (L) β 22 (L) β 3 (L) β 32 (L) β 33 (L) Ft Ft F t u t u t u t (4) For the recursive identification scheme activity factors are ordered before price and interest rate factors within each block, following the ordering outlined in equation 3, and impulse responses and variance decompositions can be computed using standard VAR techniques. Given the identification of the factors, we argue that we can uncover six different shocks using the Cholesky ordering, namely international activity, price and interest rate shocks, regional activity and price shocks, and also a domestic monetary policy shock. The implementation of the sign restrictions assumes the same ordering of the variables as in the recursive identification scheme, but put additional restrictions on the structural disturbances. Especially we restrict Ω, defined in section 2, to follow the structure: 7

9 u act u pri u r u act u pri u D u R = x x x + + x x x + x x x x x x x x x x x x + ε demand ε supply ε r ε demand ε supply ε D ε R (5) where a + indicates that the parameter must be positive, a restricts the parameter to be negative, x leaves the parameter unrestricted, and finally zero imposes exclusion restrictions. Following this identification scheme we can identify both global and regional demand and supply shocks. A positive structural international demand shock increases global activity, prices and interest rates. A positive supply shock increases global activity, but has a negative impact on international prices and interest rates. International interest rates shocks are restricted to reduce international activity and prices, but rise interest rates. The regional shocks are identified as demand and supply shocks following the same restrictions as for the international demand and supply shocks. The restrictions only affect the regional block itself though. The domestic shocks are mostly left unrestricted, and for both the global demand, supply and interest rate shocks we have left the responses of the domestic variables unrestricted. The zero restrictions follow the block exogenous assumption outlined above. Domestic shocks does not affect the region, while regional shocks does not affect the world. For New Zealand and the region including Australia these are reasonable assumptions given their relative sizes. As thoroughly described in both Bernanke et al. [25], Boivin and Giannoni [27] and Mumtaz and Surico [29], one final point regarding identification should be mentioned: The variables in the domestic block includes both interest rates and different exchange rates and other financial variables. These are all fast moving variables. To correctly identify the domestic monetary policy shock, and justify the ordering of the variables described above, we thus need ensure that the factors extracted from the X D matrix are not the interest rate or any other variable that in theory will react contemporaneously with the interest rate. To do so, we follow the procedure outlined in Boivin and Giannoni [27]. Starting from an initial principal component estimate of F D, denoted by F, we iterate through the following steps:. Regress X D on F and R to obtain λ D and λ R 8

10 2. Compute X D = X D λ R R 3. Estimate F as the first K- principal components of X D 4. Back to We end the iterations when the change in the factor estimate is less than some predefined criteria. 2.2 The data We include 363 variables from 28 different countries in the FAVAR. The data includes variables from the US, UK, Switzerland, Norway, Netherlands, Japan, Italy, France, Finland, Denmark, Canada, Sweden, Spain, Luxembourg, Germany, Belgium, Austria, Ireland, Korea, China, Malaysia, India, Taiwan, Hong Kong, Thailand, Singapore, Australia and New Zealand. The domestic block consist of data from New Zealand, the regional block data from Australia, and the world block includes date from all the other countries. For each of the countries entering into the international block we have used data measuring real activity, prices and interest rates. In the regional block we have only used activity and price measures. 2 Since we are mainly interested in investigating how the different world and regional shocks affect the domestic economy, the variables entering into the domestic block is collected from a much wider pool of series. The domestic data block thus includes a total of different 75 series, measuring activity, prices and disaggregated prices, wages, long and short interest rates, exchange rates, immigration and In this application we have implicitly assumed that the region is not part of the world variables or factors. For a different region, e.g. Europe, this is of course not a valid assumption, and the same identification problem associated with the domestic block and the interest rate arises. The rotation procedure described above can however be used to correctly identify the regional contributions, and justify ordering the world before the region. Thus making the three block FAVAR useful for studies of other regions and domestic economies as well. Compared to the data set used by Mumtaz and Surico [29], our data set includes a much bigger share of variables from the emerging and developed Asian economies. We believe this is important to accurately capture the through but unobserved world factors. 2 The number of series in the world interest rate block is only 8. We have deliberately excluded many of the countries mentioned above from this block due to their more or less fixed exchange rate regime with the US. Because the regional block only includes Australia, the number of price series used is 5. 9

11 expectations. Appendix 6 gives a more detailed description of the variables entering into the model. The model is estimated on quarterly observations from 992:Q2 to 27:Q4. Some monthly series are included in the model, these are aggregated to quarterly series by taking the mean. All non-stationary variables are in quarterly growth rates, and variables affected by seasonality are seasonally adjusted using the X2 ARIMA procedure. To make the estimation of the factors invariant to scale, all variables are standardised prior to estimation. 2.3 Estimation Bernanke et al. [25] investigate two different methods for estimating the state space system in equation and 2; a two step procedure and a joint estimation by likelihood-based Gibbs sampling techniques. Bernanke et al. [25] show that the two procedures produce very similar results. The two step procedure is however simpler, and much less computational intensive. In this application we follow the two step approach. 3 The unobserved factors and loadings are first estimated by principal components. The international activity factor(s) is extracted based on the international activity numbers, the international price factor(s) is extracted based on the international price data, etc. The factor loadings are restricted as described in section 2. and equation 3. In the second step the estimated factors are used as observed variables in a VAR framework. Bai and Ng [22] propose different criteria for how many factors to extract from each block of data. We have not followed a formal statistical procedure in this application. To keep the system at a reasonable size we have extracted only one factor from each of the international data blocks, while we have chosen to use three factors from the domestic block. Given our relatively short estimation sample we estimate the VAR in equation using Bayesian techniques. In our baseline model we restrict the number of lags to 2. Restrictions on the world and regional block of the VAR follow naturally from our identification strategy. We do however not have any priors regarding the dynamics of the domestic block of the VAR, which therefore contains 2 lags of all variables. We apply a independent normal-wishart prior for the VAR and use the Gibbs sampler to derive the posterior distributions of the parameters. The further 3 This is also the route taken in Mumtaz and Surico [29], who note that the panel of data in Bernanke et al. [25] only includes 2 variables, which makes the joint estimation more feasible.

12 avoid the problem of overfitting we adopt a Minnesota-type prior on the coefficients. That is, we set the prior of the first own lag of the dependant variable in each equation equal to its AR() coefficient, while the prior mean for all other variables are set to zero. For the prior variances we adjust for differences in scale between the variables and the lag length of the system according to the following scheme: V i,jj = a p 2 a2σ ii p 2 σjj a3 σii for coeffcients on own lags for coeffcients on lags of variable j i for coeffcients on exogenous variables where the standard errors are derived from AR(p) estimations, where p is equal to the number of lags in the full system. We specify the hyperparameters a, a2 and a3 as.6,.3 and. respectively. The degrees of freedom prior is set to 3, and the prior covariance matrix equals I(n)., where n is equal to the number of equation in the system. Thus, our prior is relatively tight, and imposes a fair amount of shrinkage. Finally, we make iterations of the Gibbs sampler, with 2 iterations used as burn-in. Sign restrictions are implemented with some minor changes to the procedure outlined in Rubio-Ramirez and Zha [29] and modified in Mumtaz and Surico [29]. Specifically we implement the following algorithm for each draw of the reduced form covariance matrix Ω:. Let Omega = P P be the Cholesky decomposition of the VAR covariance matrix Ω, and à = P. 2. Draw an independent standard normal n x k matrix J, where n is the size of the block (e.g. international or regional block) and k is the number of shocks affecting that block according to the block exogenous structure outlined in section 2. and equation 5. Let J = QR be the economy size QR decomposition of J with the diagonal of R normalised to be positive. 3. Compute a candidate structural impact matrix A = à Q, where Q is a N x N identity matrix with Q in the n x k block associated with either the international or regional block in equation Redo step -3 for the next block of international or regional data. The candidate matrix A will have a lower triangular structure for the domestic block, as in the standard Cholesky decomposition, while also satisfying the zero restrictions outlined in equation 5. If the candidate matrix satisfies the sign restrictions, we keep it. Otherwise the procedure above is repeated.

13 3 Results In this section we discuss the estimated world and regional factors, the impulse responses for the identified structural shocks as well as the variance decomposition of some key domestic variables. Irrespective of which identification scheme we use, recursive identification or sign restriction, the model produces very similar results. Thus, we argue that the shocks we have identified are; world demand, supply and interest rate shock, regional demand and supply shocks and the domestic monetary policy shock Identified factors Figure shows the estimated factors; world activity, price, and interest rate factors, regional activity and price factors, and the observed domestic short term interest rate (9-day bank bill) factor. 5. Generally our estimated world factors resembles the factors extracted by Mumtaz and Surico [29]. 4 The separate results from the two different identification schemes are available from authors upon request. 5 We do not show the three estimated domestic (New Zealand) factors, as we do not identify them 2

14 Figure : Identified factors (a) World activity (b) World price (c) World Interest rate (d) Regional activity (e) Regional price (f) 9 Day interest rate Notes: World factors and regional factors are estimated. The 9 Day interest rate is the observed short term interest rate for New Zealand. The world activity factor captures the important features of the world business cycle the last 2 years. In the early 9 s the world was coming out of a recession. The succeeding boom was followed by the Asian crisis which induced a recession in many countries, and thus a downturn in the world business cycle. The world activity factor also captures the 2 slowdown, which was deeper than the other recessions in our sample period. The world price factor inhabits the global co-movement in inflation rates across the world found in earlier studies, such as Mumtaz and Surico [29] and Ciccarelli and Mojon [25]. Particularly striking is the significant upturn at the end of the sample, probably representing a hike in commodity prices. The world interest rate factor represents the co-movement in the short-term interest rates across the globe. The peaks and the troughs in the interest rate factor follows the world price factor cycles rather closely, confirming the inflation fighting mandates of many central banks around the world. One interesting observation is the persistent and very large deviation of the interest rate factor from its historical mean in the early to mid 2s. The regional factors captures the Oceania specific business cycles. Given our definition of the region, and the data used to estimate the regional factors, these are closely related to Australian GDP and inflation. 3

15 The variance explained by the international factors are typically in the range 2-3 percent, while the three domestic factors explain around 4 percent of the variance of the domestic block. Extracting one or more domestic factors only increases the variance explained by 5 percent or less. 3.2 A positive shock to world demand We expect an increase in the world demand to have a positive effect on the New Zealand economy via the trade channel. We would also expect the strong world to have an effect on the New Zealand economy indirectly, via its effect on the regional economy. Figure 2 shows the impulse responses to a positive world demand shock. Panels (a) - (f) in the figure show the impulse responses of the factors (three world factors, two regional factors) and the observed domestic interest rates. 6 The responses of the factors are all in accordance with conventional wisdom and appear to be well identified. An unexpected increase in the world demand is persistent and stays positive for around a year. It leads to a quick and slightly more persistent increase in world inflation. In response to higher world activity and inflation the world interest rate factor increase significantly by.8 percent after a year, and does only return to its original level after about 5 years. In response to the world demand shock, regional activity immediately increase by.4 percent while the regional price factor reaches its peak after almost a year. Finally the domestic interest rate increase in response to the shock. Panels (g) - (o) in figure 2 show the impulse responses of some of the domestic macroeconomic variables to the world demand shock. A rise in headline inflation is caused by an increase in tradable inflation, which initially increases by.2 percent. Inflation expectations also responds positively to the world demand shock. After about one year non-tradable inflation, and thus headline inflation, starts falling. The fall is probably caused by the interest rate response, which peaks after about one year, before it slowly reverts back to normal. The responses of the domestic real variables are interesting. Both GDP and consumption fall. The initial reaction of both variables is close to zero, but after around one year the growth rates are significantly below zero. We speculate this might be coming from two sources: High tradable prices, all else being equal, would create a fall in demand due to income effects, while 6 We do not show or discuss the domestic factors as we leave them unidentified. 4

16 high interest rates puts downward pressure on domestic activity. Higher world prices also increases the costs of imported intermediate inputs, such as oil. The responses of the real variables are supported by the movements in import and export prices as a world demand shock increases import prices more than export prices (not shown), thus worsening the terms of trade. 7 The earlier literature used the US GDP as a proxy for the world GDP. The problem with using the US GDP as a proxy for the world GDP or activity measure is that the US GDP alone may not be sufficient to capture the effects the world activity on world prices. Therefore one may not be capturing the price effect associated with the international activity that we observe in our FAVAR. So a VAR that identifies the shocks to the US GDP is not necessarily identifying an activity shock that is worldwide (or widespread) as it is in our analysis. In short, the shock we identify should be thought of as a shock to some kind of underlying world activity that is common across a number of countries. 3.3 A negative world supply shock A world supply shock may be interpreted as a technology shock or a productivity shock. The technology shocks are usually assumed to be very persistent, whereas productivity shocks may inhabit less persistence. The classic example of this type of shock would be the 97s episode when oil prices rose unexpectedly (e.g. a supply shock) and inflation increased while activity fell all over the world. Panels (a) - (f) in figure 3 show the impulse responses of the factors and the domestic interest rates to a negative world supply shock. Given the relatively low persistence of the shock, we interpret this as a negative productivity shock. The unexpected increase in world inflation leads the world activity factor to fall sharply. It takes up to two years before world activity returns back to normal. World interest rates respond positively, but only by just over. percent. Both the regional activity and price factors follow the responses of the world factors. The New Zealand interest rates increase in response to the negative world supply shock. In sum, these impulse responses are as expected. We believe the impulse responses shown in panels (g) - (o), are in line with the recent emerging literature on the globalisation of domestic inflation rates 7 Also earlier studies document the same pattern of response to an international demand or activity shock, see for example Buckle et al. [27] and Haug and Smith [27]. 5

17 (Mumtaz and Surico [28] and Monacelli and Sala [29]). The negative supply shock causes tradable inflation to increase sharply, which in turn leads to a sharp increase in the headline inflation rate (New Zealand s headline inflation measure includes just under 5 percent tradable products). The non-tradable inflation rate increase only marginally. 8, while the exchange rate depreciates initially in response to the negative world supply shock. This depreciation would exacerbate the increase in tradable inflation even further. The domestic GDP response is very erratic. It increases for a quarter, before it decreases for a one or two quarters. Beyond 3 quarters GDP does not respond significantly to the negative world supply shock. The same pattern of response can be found for domestic consumption, capacity utilisation and unemployment. As was the case for the world demand shock, the main transmission mechanism of the international supply shock to the New Zealand economy seems to be via tradable prices, as opposed to through activity measures. The increase in tradable inflation leads to higher headline inflation, and in return the central bank raises interest rates. As was the case for the international demand shock, the interest rate rise suppresses the real variables. 3.4 A positive world interest rate shock In this section we discuss the effects of a shock that increases the world interest rates by percent on average. This can be interpreted as a special circumstance such as the 28 period where a number of central banks changed interest rates more or less simultaneously. Panels (a) - (f) in figure 4 show the estimated impulse responses of the factors and the domestic interest rates. The world interest rate shock is very persistent. Interestingly, world inflation increases as a response to the world interest rate rise, while the world activity factor initially falls, as expected. The regional price factor responds negatively to the world interest rate rise, and returns to normal after about one year. Finally, the domestic 9-day interest rate increase by around 5 basis points after 3-4 quarters. Panels (g) - (o) in figure 4 show the estimated impulse responses for the domestic New Zealand variables. We expect an international monetary policy shock to have offsetting effects on the domestic economy due to the exchange rate and the foreign demand effects. The New Zealand exchange rate depre- 8 This may partly reflect the influence of the regional price factor on New Zealand s nontradable inflation rate, which we will be discussing more in section 3.5 and

18 ciates in response to the shock. As a result, New Zealand exports increase. The additional demand pressure generated by increased exports can be seen on falling unemployment, increased GDP growth, and increased capacity utilisation numbers. Tradable prices increase as a result of the depreciation in the currency. Also headline inflation increases, but the inflation rate falls as the domestic interest rate responds positively to the international interest rate shock after about year. These responses are all consistent with simple open economy models. 7

19 Figure 2: Impulse responses of factors to world demand shock (a) World Activity (b) World Prices (c) World Interest Rates (d) Regional Activity (e) Regional Prices (f) R9D.6 2 (g) Household consumption.4.4 (h) GDP (i) Inflation expectations (j) Inflation.6 2 (k) Non-tradable inflation (l) Tradable Inflation (m) Terms of trade.2 (n) Import prices.4 (o) Exchange rate Notes: The solid black line is the FAVAR median impulses while the dashed black lines are one standard deviation confidence bands based on the posterior distribution of the parameters. The initial shocks are normalised to %, while we report the responses up to 2 quarters ahead.

20 Figure 3: shock Impulse responses of factors to negative world supply (a) World Activity (b) World Prices (c) World Interest Rates (d) Regional Activity (e) Regional Prices (f) R9D (g) Unemployment.8 (h) House Prices (i) Household consumption (j) Inflation (k) Non-tradable inflation.5 (l) Tradable Inflation (m) Capacity Utilisation (n) Exchange rate (o) GDP.5 2 Notes: The solid black line is the FAVAR median impulses while the dashed black lines are one standard deviation confidence bands based on the posterior distribution of the parameters. The initial shocks are normalised to %, while we report the responses up to 2 quarters ahead.

21 Figure 4: Impulse responses of factors to world interest rate shock (a) World Activity (b) World Prices (c) World Interest Rates (d) Regional Activity.4 (e) Regional Prices.5 (f) R9D (g) Unemployment.5 (h) Exports (i) Household consumption (j) Inflation.8 2 (k) Non-tradable inflation (l) Tradable Inflation (m) Capacity Utilisation (n) Exchange rate (o) GDP Notes: The solid black line is the FAVAR median impulses while the dashed black lines are one standard deviation confidence bands based on the posterior distribution of the parameters. The initial shocks are normalised to %, while we report the responses up to 2 quarters ahead.

22 3.5 Regional demand and supply shocks Figure 5 shows the responses of the regional factors and some selected New Zealand variables to a positive regional demand shock. Firstly, the shock is not very persistent. The regional activity factor grows above normal for just over 2 3 quarters, and the shock causes almost no reaction in the regional price factor. After an unanticipated shock to regional demand, the New Zealand exchange rate appreciates significantly the first year after the impulse. The appreciation of the exchange rate significantly affects tradable inflation, which falls for over a year. The exchange rate response is consistent with recent findings in Coleman and Karagedikli [2]. 9 Further, as a result of the exchange rate appreciation, exports go down 2 and imports go up, and the model produces a very small and short lived expansion in output. While a regional activity boom is thought to be beneficial for the New Zealand economy, our results indicate that the exchange rate effects undermines these positive impulses. However, the short lived increase in New Zealand output is apparently enough to generate non-tradable inflation, which again leads the central bank to raise interest rates. As displayed in figure 6, a negative regional supply shock increases regional prices significantly. The effects of the shock on the regional prices do however not last for more than a year, and the effects on the regional activity factor are minor. Domestic non-tradable inflation increases as a result of the negative regional supply shock and the exchange rate appreciates. The significant domestic non-tradable inflation response is supported by findings in Coleman [27], and will be discussed further in section 3.7. Also wages increase significantly as a response to the negative supply shock. This is not surprising given the large labour mobility within the region. Domestic activity variables show a positive response to the negative regional supply shock. These responses are probably driven by the strong and positive reaction of the terms of trade, which lasts for almost 3 years. These results are also consistent with Neely 9 Applying an event analysis on New Zealand and Australian data Coleman and Karagedikli [2] find that unexpected real economy surprises (e.g. an unanticipated shock to demand) lead to an appreciation of the New Zealand - US dollar exchange rate, and presumably all the New Zealand dollar exchange rates (except the cross-rate between New Zealand and Australia). 2 This is total New Zealand exports. Although one would expect New Zealand dollar to depreciate against the Australian dollar, and hence an increase in exports to Australia, the total exports fall. 2

23 and Rapach [28] which finds that the regional factors explain around 3 per cent of the variation in headline inflation rates in Australia and New Zealand over the period. This country factors become after the regional factors in terms of explaining the overall variation. Overall these regional impulse responses and the transmission of shocks to the domestic economy seem reasonable. The results confirm findings in earlier regional studies, and thus highlights the importance of including regional factors in open economy models as the FAVAR. Although we assumed the region is Oceania for New Zealand, it would still be wise to discuss how we define a region. What defines a region may be an arbitrary choice at times. Most empirical studies would put New Zealand and Australia together as the region of Oceania (Kose et al. [23], Neely and Rapach [28] for example). While this might geographically appropriate, in terms of economic (both trade and financial) linkages, it may not represent the true economic region(s). For example, up until 973 most of New Zealand s exports were imported by the United Kingdom. By 2 the Asian economies make up the largest destination for the New Zealand exports. 3.6 An unanticipated shock to domestic monetary policy As discussed in the introduction, VAR based impulse responses derived from limited information sets typically exhibit some well known anomalies, such as the exchange rate and price puzzles (see e.g. Eichenbaum and Evans [995] and Kim and Roubini [2]). The FAVAR literature proposed by Bernanke et al. [25] argue that the FAVAR methodology is a very useful tool to overcome many of the empirical anomalies associated with identifying a domestic monetary policy shock. Our results, reported in figure 8 confirms this. A percent unexpected increase in the short term interest rates in New Zealand has a significant and negative effect on activity, consumption, investment and capacity utilisation. The effects on consumption and investment peak after three quarters, while the effects on capacity utilisation peaks after five quarters. GDP growth initially falls by.5 percent, but returns to normal after about two and a half years. The exchange rate responds immediately to the unexpected shock to the interest rates, before it gradually depreciates to baseline. This leads to a immediate fall in tradable inflation. 22

24 Non tradable inflation also falls, but not as strongly as tradable inflation. Accordingly, headline inflation also falls. The response in house prices are negative, and lasts for over two years. The exchange rate response is in accordance with the exchange rate overshooting hypothesis. We believe it is the FAVAR s big data set and contemporaneous interaction between the exchange rate and the interest rate that ensures this result. 2 Further, the model does not show any sign of the typical price puzzle, which is also found in other FAVAR literature, such as Boivin et al. [29] and Mumtaz and Surico [29]. 2 A similar argument is used by Bjørnland [29], who identifies a monetary policy shock in four economies (including New Zealand), by using long-run neutrality restrictions on the exchange rate. 23

25 Figure 5: Impulse responses regional demand shock (a) Regional Activity (b) Regional Prices (c) R9D (d) Unemployment. (e) Exports.4 (f) Imports (g) Inflation (h) Non-tradable inflation.5 (i) Tradable Inflation (j) Capacity Utilisation (k) Exchange rate (l) GDP.5 2 Notes: The solid black line is the FAVAR median impulses while the dashed black lines are one standard deviation confidence bands based on the posterior distribution of the parameters. The initial shocks are normalised to %, while we report the responses up to 2 quarters ahead. 24

26 Figure 6: Impulse responses to a negative regional supply shock (a) Regional Activity (b) Regional Prices (c) R9D (d) Unemployment.25 2 (e) Household consumption (f) Wages (g) Inflation.5 2 (h) Non-tradable inflation (i) Tradable Inflation (j) Terms of trade (k) Exchange rate (l) GDP.4 2 Notes: The solid black line is the FAVAR median impulses while the dashed black lines are one standard deviation confidence bands based on the posterior distribution of the parameters. The initial shocks are normalised to %, while we report the responses up to 2 quarters ahead. 25

27 Figure 7: Impulse responses domestic monetary policy shock (a) GDP (b) Unemployment (c) House Prices (d) Household consumption (e) Investment (f) Inflation Expectations (g) Capacity utilisation (h) Exchange Rate (i) R9D (j) Inflation (k) Non-tradable inflation.2.. (l) Tradable Inflation Notes: The solid black line is the FAVAR median impulses while the dashed black lines are one standard deviation confidence bands based on the posterior distribution of the parameters. The initial shocks are normalised to %, while we report the responses up to 2 quarters ahead. 26