This paper investigates the importance of external and domestic shocks in Chile’s highly open economy. Using a Bayesian Vector Autoregression (BVAR) and a two-step multivariate regression, the analysis suggests that both external and domestic factors have been playing an important role, though the relative contribution varied across periods.

Abstract

This paper investigates the importance of external and domestic shocks in Chile’s highly open economy. Using a Bayesian Vector Autoregression (BVAR) and a two-step multivariate regression, the analysis suggests that both external and domestic factors have been playing an important role, though the relative contribution varied across periods.

The Macroeconomic Effects of External and Domestic Shocks in Chile1

This paper investigates the importance of external and domestic shocks in Chile’s highly open economy. Using a Bayesian Vector Autoregression (BVAR) and a two-step multivariate regression, the analysis suggests that both external and domestic factors have been playing an important role, though the relative contribution varied across periods.

A. Introduction

1. For a small and very open economy such as Chile, external shocks tend to have a substantial impact. It is, thus, interesting to ask how much of the business cycle fluctuations in economic activity (e.g., as measured by GDP growth) can be attributed to factors external to the Chilean economy vis-à-vis domestic factors. While this distinction is not always econometrically straightforward, this paper attempts such a decomposition for GDP fluctuations observed over the last two decades.

uA005fig01

Trade Openness 1/

(In percent of GDP)

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Source: World Bank Development Indicators.1/: Trade is defined as exports plus imports.

2. Over the past decade, the Chilean economy experienced large swings in growth. After growing at a fast pace, Chile’s GDP growth has averaged a mere 1.3 percent between 2014Q1 and 2017Q2. This is a very subdued rate—especially in the absence of any outright economic recession or a financial crisis—well below Chile’s historical average growth of about 4½ percent since 1990 and below the growth of most of its main trading partners including the U.S. However, this period of low growth was associated with low copper prices and at the same time a spike in domestic policy uncertainty (both of these factors are likely to have contributed to the fall in business confidence and investment). The subsequent analysis, will try to shed lights on these casual observations.

uA005fig02

GDP Growth and Copper Price

(In percent)

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Sources: IMF 2018 WEO and Bloomberg.

3. In light of the extensive openness of the Chilean economy, several external factors tend to play a significant role. The degree of Chile’s economic openess (i.e., export plus imports as share of GDP) is about 68 percent on average over the 2003–17 period, which is among the highest in the region. Almost 55 percent of Chile’s export value, however, comes from commodities and about 45 percent is related to copper.2 Shocks to U.S. GDP and interest rates also have a prominent impact on Chile’s economic activity, both via trade channels and financial channels. Indeed, the U.S. is Chile’s second largest export market, while Chile’s financial market and banking systems are well integrated in the global financial markets. At the same time, the exchange rate is free to float since the end of the 90s and thus responding to international factors (although with one of the lowest pass-through effects of the region (IMF, 2016)). The Chilean economy is, thus, substantially exposed to fluctuations in the global demand for commodities, especially copper. Since about 50 percent of metal demand (including copper) comes from China, the Chilean economy is, directly and indirectly, exposed to fluctuations in the Chinese economic activity (especially its utility and construction sectors which are the main consumers of copper).3

uA005fig03

Investment and Policy Uncertainty

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Source: Central Bank of Chile.

4. The most noticeable external development over the last 10 years is the behavior of the terms of trade—mostly driven by copper prices. The terms of trade, driven by copper prices, rebounded strongly after the 2008–2009 global financial crisis and peaked in 2011. The terms of trade declined rapidly afterwards, reflecting the fall in the price of copper that started in 2011 and reached a trough in 2016 (offset in part by the oil price collapse in 2014–16). Also, trading partner growth disappointed as the Euro crisis unfolded with GDP in the Euro Area (the third largest Chile’s trading partner) declining each quarter between 2011Q4 and 2013Q1.

uA005fig04

Top 5 Export Markets for Chile, 2017

(In percent of GDP)

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Source: IMF Direction of Trade Database.

5. Domestic factors are usually related in these analyses to the behavior of fiscal and monetary policy but also, in the case of Chile, to domestic policy uncertainty. A proxy for domestic policy uncertainty is used to capture the role of uncertainty associated with the policy discussions and the political and electoral environment, among other things, which tend to affect consumer and business sentiment.

B. Methodology

6. We employ two techniques, a BVAR and a two-step multivariate regression, to disentangle the role of domestic and external factors. The economic literature has often relied on dynamic stochastic general equilibrium models (DSGE) to provide a shock decomposition of GDP or other observable variables (Adolfson and others 2007; among many). The use of a DSGE model, however, is more demanding since it introduces a substantial amount of cross-equation restrictions and, thus, requires to put considerable faith in the ability of the chosen economic model to replicate the actual data-generating process. Hence, in this case, a more flexible approach is preferred, and we employ a BVAR and a two-step multivariate regression approach. In both cases, we use a set of variables to proxy external forces and another set to control for domestic shocks.

7. Both approaches exploit the fact that Chile is a small economy that does not affect global variables. The BVAR exploits the presence of a block with external variables and of a tight prior which indicates that fluctuations in external variables are a priori expected to be only very modestly affected by developments of Chilean domestic variables. Similarly, shocks affecting the external block are allowed to affect Chile’s set of domestic variables, but not vice versa. The BVAR, thus, exploits a partial identification strategy where the goal is simply to disentangle external and domestic shocks without being more specific on the contribution of each type of external or domestic shock.4 Overall, the benefit of the BVAR approach is a reasonably good identification of the relative role of external versus domestic factors (it also better captures the dynamics imbedded in the data). The multivariate regression approach uses a two-step procedure where in the first step domestic variables are orthogonalized by regressing them on external variables. This method has the advantage of simplicity, and indeed it lends itself to a more refined assessment of the contribution of each variable. However, it is less effective at a clean identification of external versus domestic factors (especially the latter), and presents a residual which cannot be intepreted as external or domestic: notably it relies on the reasonable assumption that external shocks are not affected by domestic variables, while domestic variables will attempt to explain the remaining volatlity (i.e. after controlling for external shocks). Note also, that the contribution of each variable may mask correlation across variables, for example the role of copper may be in part captured by the China growth variable.

8. Both approaches are based on the same data sample. The analises are perfomed using quarterly data from 1999 to 2017. The external block includes global variables relevant for Chile as discussed above—i.e., U.S. and China real GDP (the latter interacted with its share of world trade to account for its growing role), real copper prices, the U.S. 10-year yield, we include also Chile’s real mining output, which is considered mostly exogenous to domestic developments. A beta- convergence factor (i.e., the GDP per-capita difference between Chile and the U.S.) has been introduced in both exercises as a exogenous process as suggested by the growth literature (Sala-i- Martin, 1996). The block of domestic variables, instead, includes real GDP (the variable of interest), real investment, the 3-month interest rate on BoC’s notes, the Peso-USD real exchange rate, and domestic policy uncertainty in Chile.5,6 The fiscal balance to GDP is also introduced in the multivariate regression; to limit the number of parameters to estimate, it is not included in the BVAR, but a robustness exercise shows that results would be similar when including it. The choice of domestic variables strikes a balance between parsimony and the need to capture developments in the aggregate domestic demand including the ones induced by movements in the Peso and interest rates.

C. Findings

9. Both external and domestic factors have played an important role in the post crisis period and recent slowdown but with different timing. Figure 1 and the text chart show the results of the BVAR approach in terms of historical contributions of GDP growth into external and domestic factors, as well as the convergence factor (capturing the secular trend). The sum of all contributions, by construction, equals the observed GDP growth in the BVAR approach.7 Results point to a strong negative contribution to GDP growth of external factors especially during the global crisis and since the end of 2015. In particular, they were a drag on growth from mid-2015 to mid-2017 owing to lower trading partner growth in 2015 and the decline in copper prices in 2016. The 2017Q1 is attributed to the sizeable drop in mining output (due to the mining strike) which is classified as exogenous in the analysis. Since mid-2017, external factors have been strongly contributing to the growth recovery, owing also to a rebound in copper price. Domestic factors also contributed both to the contraction during the global crisis and the subsequent recovery. They negatively affected growth afterwards, particularly in 2014, and more mildly since 2016, partly reflecting policy uncertainty. Finally, the convergence factor and internal BVAR cyclical dynamics contribute by about half a percentage point to the reduction in potential growth from the pre-crisis period to 2018Q1 (the most recent data point in the sample).

uA005fig05

BVAR: Contributions to Growth

(In percent, four quarter moving average)

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Source: National sources, 2018 IMF WEO, Bloomberg, and IMF staff calculations.

10. To some extent, the results from the multivariate approach are consistent with the BVAR findings. However, the presence of the residual makes them more difficult to interpret, as it cannot be easily ascribed to an external versus a domestic component (Figure 2 and text chart). The importance of both external and domestic shocks during the global crisis is confirmed. Also, domestic shocks were a drag on growth in 2014, while external shocks were a positive contributor since 2017. However, the model is unable to explain most of the 2016–17 growth decline, as visible from the important contribution of the residual (which captures unidentified factors).

uA005fig06

Multivariate: Contributions to Growth

(In percent)

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Source: National sources, 2018 IMF WEO, Bloomberg, and IMF staff calculations.

D. Concluding Remarks

11. Both external and domestic factors play an important role in driving business cycle fluctuations in Chile as well as low frequency trends. The two methodologies tend to offer a similar picture for some key patterns. Both external and domestic factors were strognly relevant in explaining growth during and after the global crisis. When looking at the more recent period, we see domestic factors as a drag on growth particularly in 2014, and external factor as helping the recent rebound since 2017. It is also worth noting that income convergence with advanced economies has been playing a persistent role in lowering growth towards advanced economy averages as Chile increases its physical and human capital and approaches to the technological frontier.

Appendix I. Technical Appendix

A. The Bayesian VAR

1. We assume the following specification for the Bayesian VAR model:

Π(L)yt=Φdt+et

Where yt is a p-dimensional vector of time series at time t, dt is a q-dimensional vector of deterministic trends or other exogenous variables. Π(L) = Ip- Π1L1 – … – ΠkLk is the matrix of coefficients to be estimated where L is the back-shift operator such that Lkyt = yt-k. We assume that yt is a stationary process and et ~ N(0,Σ) with independence between time periods. Let’s also assume that the vector yt can be understood as combination of two vectors of size n1 and n2 such that yt = (y1t;y2t).

2. Bayesian inference requires a prior distribution on Σ, Π1, …, Πk and Φ. The prior on Σ is

p(Σ)Σp+12.

Let Π = (Π1, …, Πk)′, the prior for vec(Π) is

vec(Π)N(θπ,Ωπ)

similarly for Φ.

3. In the analysis we have defined y1t as the external block which includes the following variables: the U.S. and China GDP (log-diff), copper prices (logs), the U.S. 10-year yield, and Chilean mining output (log-diff). The domestic block, instead, includes domestic Chilean variables such as investment (log-diff), the 3-month interest rate, the exchange rate (logs), and credit-to-GDP (logs). Defining Π = [Π11, Π12; Π21, Π22] and, consistently, θπ = [θπ11π12π21π22], and similarly for Σ, we have assumed that θπ12 and Σ12 have their elements equal to 0.001. This guarantees a high exogeneity tightness for the external block of variables. The rest of the parameters follow the Minnesota prior (Litterman 1986).

4. Finally, the BVAR matrix of structural shocks is computed assuming a Cholesky decomposition where the external block is ordered first. A historical decomposition is then performed. A historical decomposition answers the following question: what would the data have looked like if a subset of the shocks had been zero throughout the estimation period? We, thus, shut down all domestic (external) shocks (and remove the exogenous converge term), to compute the contributions of external (domestic) shocks. We start the decomposition in 2008 since the early part of the sample is usually affected by the choice of the initial conditions.1

B. The Two-Step Multivariate Approach

5. The methodology builds on Yeyati and Williams (2012) and its further improvements in de la Torre and others (2013) and subsequent editions of the Semiannual report of the World Bank’s Office of the Chief Economist for Latin America and the Caribbean. The domestic variables are

CHLt=c+β*CHNINTt+γ*G7t+γ*US10yrt+λ*TRCRBt+δ*MPRt4+θ*FBALt4+κ*DPUt+π*CONVt+ϵt(2)

where CHLt is Chile’s YoY real GDP growth rate, CHN INTt is China’s YoY real GDP growth rate interacted with its share of world trade, G7t is the G7’s YoY real GDP growth rate, US10yrt is the 10-year U.S. Treasury yield, TRCRBt is the YoY growth rate of the Thomson Reuters CRB commodity index, and et is the error term. The orthogonalized variables are: the lagged residual from the monetary policy rate regression (MPRt-4), the lagged residual from the central government’s fiscal balance regression (FBALt-4), the residual from the domestic policy uncertainty index regression(DPUt). Finally, CONVt is the convergence term. The text table shows the regression coefficients.

article image
Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

C. Figures

Figure 1.
Figure 1.

Bayesian VAR. Chile’s GDP Growth Decomposition

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Figure 2.
Figure 2.

Two-Step Multivariate Regression. Chile’s GDP Growth Decomposition

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

The BVAR includes an external block with global variables and Chile’s mining output while the domestic block includes Chile’s domestic variables (see text and Appendix).

Figure 3.
Figure 3.

Regression Variables

Citation: IMF Staff Country Reports 2018, 312; 10.5089/9781484383919.002.A005

Sources: Haver, Bloomberg, 2018 WEO and IMF staff calculations.

References

  • Adolfson, Malin, et al., 2007, “Bayesian Estimation of an Open Economy DSGE Model with Incomplete Pass-Through,” Journal of International Economics, Vol. 72(2), pp. 481511.

    • Search Google Scholar
    • Export Citation
  • Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans, 1999, “Monetary Policy Shocks: What have we Learned and to what end?”, Handbook of Macroeconomics, Vol. 1, pp. 65148.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2016 Regional Economic Outlook, Western Hemisphere Department, April (Washington).

  • Kilian, Lutz, 2009, “Not all Oil Price Shocks are Alike: Disentangling Demand and Supply Shocks in the Crude Oil Market,” American Economic Review, Vol. 99(3), pp. 105369.

    • Search Google Scholar
    • Export Citation
  • Levy Yeyati, Eduardo, and Tomas Williams, 2012, “Emerging Economies in the 2000s: Real Decoupling and Financial Recoupling,” Journal of International Money and Finance, Elsevier, Vol. 31(8), pp. 210226.

    • Search Google Scholar
    • Export Citation
  • Sala-i-Martin, Xavier X., 1996, “The Classical Approach to Convergence Analysis.” The Economic Journal, pp. 101936.

  • Smets, Frank, and Rafael Wouters, 2007, “Shocks and Frictions in U.S. Business Cycles: A Bayesian DSGE Approach,” American Economic Review Vol. 97(3), pp. 586606.

    • Search Google Scholar
    • Export Citation
  • de la Torre, Augusto, Eduardo Levy Yeyati, and Samuel Pienknagura, 2013, “Latin America and the Caribbean as Tailwinds Recede: In Search of Higher Growth,” LAC Semiannual Report, World Bank, Washington, DC.

    • Search Google Scholar
    • Export Citation
  • Villani, Mattias, 2009, “Steady-State Priors for Vector Autoregressions,” Journal of Applied Econometrics Vol. 24.4, pp. 63050.

1

Prepared by Andrea Pescatori and Martin Sasson.

2

Commodities are defined as the sum of mining, agriculture and silviculture, and fishing. Their 2017 shares of exports are 54.9 and 43.9, respectively, not dissimilar to their 2003–2017 average. The value of copper exports as share of GDP peaked at 25 percent in 2007.

3

The domestic power sector is the largest copper consumer in China, due to investment in power network upgrades and rural power network renovation. The residential construction sector is also a key driver of copper demand. The relevance of China, however, has clearly increased over time. In the empirical analysis we, thus, weight China GDP growth by its share of world trade (see Figure 3).

4

The BVAR, for example, does not attempt to distinguish between domestic supply and demand shocks. Partial identification strategies have a long-standing tradition in the VAR literature, one of the first examples is in Christiano and others (1999).

5

All variables are in log-differences but for the 3-month interest rate, domestic policy uncertainty, the fiscal balance- to-GDP, and credit-to-GDP which are introduced untransformed in levels, while the real effective exchange rate is in logs. For robustness, in the BVAR, we have also introduced the fiscal balance over GDP (in levels), replaced domestic policy uncertainty with credit-to-GDP (in levels), replaced U.S. GDP with G7 GDP (in PPP US$), and the copper price with the World Bank’s commodity price index. Results are qualitatively unchanged and quantitatively very similar.

6

Chile’s data are from Banco Central de Chile, copper price data are from Bloomberg and divided by U.S. CPI, China and U.S. data are from HAVER. The domestic policy uncertainty index comes from the Economic Policy Uncertainty website: http://www.policyuncertainty.com/chile_monthly.html.

7

Given the large volatility of q/q growth rates, results are presented as a 4-quarter moving average of the q/q growth rates employed in the BVAR. This also eases the comparability with the other methodology that adopts y/y growth rates.

1

As standard in the literature (see Kilian 2007, for example), we assume the first point to be the VAR steady state.

Chile: Selected Issues Paper
Author: International Monetary Fund. Western Hemisphere Dept.