Emerging market fluctuations: What makes the difference?

INEC-02771; No of Pages 17 Journal of International Economics xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Journal of International Economics journal homepage: www.elsevier.com/locate/jie

Emerging market fluctuations: What makes the difference? Constantino Hevia ⁎ Department of Economics, Universidad Torcuato di Tella, Argentina

a r t i c l e

i n f o

Article history: Received 26 February 2011 Received in revised form 16 May 2014 Accepted 16 May 2014 Available online xxxx JEL classification: E32 F41 Keywords: Business cycles Small open economy Country spreads Financial frictions

a b s t r a c t Aggregate fluctuations in emerging countries are different from those in developed countries. Using data from Mexico and Canada, this paper decomposes these differences in terms of reduced form shocks that affect aggregate efficiency and distort the decisions of households about how much to invest, consume, and work in a standard model of a small open economy. The decomposition exercise suggests that most of these differences are explained by fluctuations in aggregate efficiency, distortions in labor choices over the business cycle, and distortions in intertemporal consumption choices. Successful models for emerging markets fluctuations should include primitive shocks and frictions that generate these features. Models with financial frictions in the form of working capital constraints, possibly augmented with endogenous collateral constraints, are consistent with these findings. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Aggregate fluctuations in emerging countries differ from those in developed (small-open) countries. This paper quantifies a set of reduced form shocks, or wedges, that account for these differences by extending the ‘Business Cycle Accounting’(BCA) methodology advocated by Chari et al. (2007) (CKM) to an open economy setting. At a conceptual level, shocks and frictions in most structural models drive a wedge between marginal rates of substitution and marginal rates of transformation. Based on this insight, I estimate these wedges using data from Mexico and Canada and the equilibrium decision rules of a frictionless small open economy augmented with five stochastic reduced form shocks (the prototype economy). At face value, these shocks can be interpreted as total factor productivity (the efficiency wedge), as labor and investment taxes (the labor and investment wedges), as fluctuations in real interest rates (the country spread wedge), and as government consumption (the government consumption wedge). I study the statistical properties of these wedges and their contribution to aggregate fluctuations in Mexico and Canada by feeding them back into the model one at a time or in combination. The decomposition exercise leads to the following findings. First, aggregate fluctuations in Mexico are mostly driven by the combined contribution of the efficiency, labor, and country spread wedges. On the other hand, the efficiency and labor wedges account for most business ⁎ Av. Figueroa Alcorta 7350, Ciudad de Buenos Aires, C1428BCW, Argentina. Tel.: +54 11 5169 7310. E-mail address: [email protected].

cycle fluctuations in Canada, although country spread and investment wedges contribute somewhat to fluctuations in investment, the trade balance, and consumption. Second, fluctuations in the country spread wedge account for the qualitative differences between Mexico and Canada: the excess volatility of consumption over output and the highly countercyclical trade balance in Mexico. Third, the investment wedge plays a minor role in Mexico's business cycles. And fourth, the government consumption wedges play a negligible role in both countries. To check the robustness of the results, I apply the methodology using Korean data and find that, as in Mexico, the combined contribution of efficiency, labor, and country spread wedges accounts for most aggregate fluctuations. Yet, there is a difference in the relative importance of the wedges, with the labor wedge playing a more prominent role in Korea, while the efficiency wedge playing a more prominent role in Mexico. The decomposition of aggregate fluctuations into reduced form wedges does not identify primitive shocks and frictions. Indeed, different structural models could induce movements in the same wedge or a single structural shock could induce movements in several wedges. Instead, BCA methodology measures the sum of the impact of all structural shocks on each reduced form wedge, and then measures the marginal effect of each wedge on aggregate fluctuations. As noted in CKM, an advantage of this approach is that it does not require making a priori assumptions to identify structural shocks. Yet, the information obtained with BCA is useful for model development: a successful model should induce reduced form wedges similar to those estimated based on the prototype economy, and these wedges should account for fluctuations in the economic aggregates. Therefore, one can interpret the BCA methodology

http://dx.doi.org/10.1016/j.jinteco.2014.05.002 0022-1996/© 2014 Elsevier B.V. All rights reserved.

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

2

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

as a method to restrict the set of structural models consistent with the data. One conclusion of this exercise is that Real Business Cycle (RBC) models with just productivity shocks are unable to explain a key business cycle fact in emerging countries. Namely, that shocks that introduce a wedge between the equality of the marginal product of labor with the marginal rate of substitution between consumption and leisure account for a large fraction of fluctuations in Korea and Mexico. RBC models with just productivity shocks do not distort the labor– consumption margin.1 Second, the observation that investment wedges play a minor role does not mean that financial frictions are irrelevant; it means that financial frictions should manifest themselves primarily as fluctuations in efficiency, labor, or country spread wedges. Mendoza (2010) proposes a model along these lines where a combination of financial frictions in the form of collateralized working capital constraints and capital adjustment costs drive persistent fluctuations in these key wedges. Section 6 illustrates the mapping between Mendoza's model and the prototype open economy. Interestingly, a binding collateral constraint always induces a decline in the government consumption wedge, the opposite of the finding in Chari et al. (2005). The extension of BCA methodology to a prototype open economy allows me to study two intertemporal disturbances that are relevant for small open economies: wedges that affect the intertemporal allocation of foreign debt (through country spread wedges in the Euler equation for bonds), and wedges that affect the intertemporal allocation of capital (through investment wedges in the Euler equation for capital). This modification is important because it is precisely the country spread wedge, not the investment wedge, what drives the excess volatility of consumption over output and the countercyclical trade balance-toGDP ratio in Mexico. Moreover, this modification also explains why a binding collateral constraint in Mendoza's model induces a negative government consumption wedge in the prototype economy instead of a positive one, as in Chari et al. (2005). Recent related papers are Kehoe and Ruhl (2009), Garcia-Cicco et al. (2010), Chang and Fernández (2013), and Lama (2011). Kehoe and Ruhl (2009) find that a two sector model with labor reallocation frictions, variable capital utilization, observed interest rate spreads and TFP, and a binding credit limit is able to explain the performance of the Mexican economy during the 1994–1995 crisis. Interestingly, these frictions and shocks induce a set of reduced form wedges that are consistent with the results in this paper. In independent work, Lama (2011) uses a version of BCA to study output drops in six Latin American countries and claims that a shock similar to the country spread wedge plays a negligible role. Section 5.4 discusses why this paper reaches a different conclusion.2 The paper is organized as follows. Section 2 discusses the empirical regularities associated with aggregate fluctuations in emerging and developed countries. Sections 3 and 4 describe the prototype small open economy and the BCA methodology used to decompose aggregate fluctuations. Section 5 applies the BCA methodology using data from Mexico, Canada, and Korea. Section 6 describes a structural model with financial frictions that is consistent with the findings and Section 7 concludes. An online Appendix covers additional results and proofs.3

1 Kydland and Zarazaga (2002) argue that a model with just productivity shocks explains successfully the performance of Argentina during the 1980s; Aguiar and Gopinath (2007) argue that a model with permanent productivity shocks explains the statistics in Table 1. 2 Garcia-Cicco et al. (2010) use long time series data from Argentina and Mexico to estimate a model with temporary and permanent productivity shocks. They find that models with just productivity shock miss the behavior of the trade balance and favor a model with stochastic interest rates and an endogenous country spread that depends on the level of foreign debt. Chang and Fernández (2013) estimate a model with permanent and temporary productivity shocks, interest rate shocks, and working capital constraints. They find that the model with the working capital constraint and interest rates shocks provide a better fit than a model with permanent productivity shocks. 3 The online Appendix can be found in https://sites.google.com/site/constantinohevia/.

2. Empirical regularities It has been thoroughly documented that business cycles in emerging economies are different from those in developed economies (Neumeyer and Perri, 2005; Aguiar and Gopinath, 2007). Besides being substantially more volatile, aggregate fluctuations in emerging economies seem to be qualitatively different from those in developed countries: consumption is more volatile than output in the former but less volatile in the latter, and the share of the trade balance on output is highly countercyclical in emerging countries but less so in developed countries. Since the 2000s, however, emerging economies were less prone to suffer the type of crises that they used to suffer in the past. Therefore, one could question whether the documented differences between emerging and developed economies are driven by the inclusion of data from one or two crisis episodes. In this section I revisit these regularities using updated quarterly data and analyze the conjecture that a few crisis episodes could be behind these differences. To have comparable results, I use the sample of emerging and developed economies chosen by Aguiar and Gopinath (2007) with the exception that Chile replaces Ecuador in the sample of emerging economies.4 Table 1 displays business cycle statistics for a group of 13 emerging and 13 developed countries. Each series was filtered using the Hodrick–Prescott (HP) filter with a smoothing parameter of 1600 and expressed as percentage deviations from trend, except for the ratio of the trade balance to GDP, which was expressed as simple deviations from trend—using the Band-Pass filter gives similar results (see the online Appendix). These statistics suggest that the empirical regularities documented in the previous literature do not depend on the particular sample period used to compute them. First, GDP and the ratio of the trade balance to GDP are more volatile in emerging than in developed countries (over 70 and 220%, respectively). Second, consumption is, on average, more volatile than output in emerging countries but less volatile in developed countries. And third, there is a large negative correlation between GDP and the ratio of the trade balance to GDP in the group of emerging countries (− 0.51) compared to that in the group of advanced countries (−0.14). Furthermore, note that the individual statistics for Mexico and Canada, the prototype emerging and developed countries used in the empirical analysis below, are broadly consistent with the experience of the average emerging and developed country, respectively. Table 2 reports statistics for the group of emerging countries dividing the sample between crisis and no-crisis periods. Panel A rewrites the averages from Table 1. Panels B and C divide the sample into crisis and no-crisis periods. A crisis is defined as a drop in GDP from peak to trough of at least 9% and the crisis is defined to be over when GDP recovers 50% of its peak-to-trough drop.5 The two panels differ in the definition of the beginning of the crisis. In panel B, the crisis is defined to begin the quarter after the peak in economic activity. In panel C, the crisis is defined to begin 5 quarters before the peak in economic activity. The latter definition is chosen because it maximizes the difference between the crisis and no-crisis statistics. Ten emerging market crises are identified by the above criterion: one in Argentina, Korea, Malaysia, Philippines, and Thailand; two in Mexico; and three in Turkey. The statistics in the rows labeled “No crisis,” were computed dropping the HP-filtered data in each individual country corresponding to the crisis window. Next, I computed the corresponding statistics for each country and averaged the result across emerging economies. The statistics displayed in the row “Crisis (pooled)” were computed by pooling the data from the 10 crises episodes. This procedure is valid because HP-filtered data is centered around zero and all the statistics displayed in Table 2 are based on contemporaneous 4 Ecuador revised its National Accounts during the 2000s leading to very different set of stylized facts. See the online Appendix for the construction of the data and the sources. 5 Results are robust to changing the definition of the crisis (output drops between 7 and 10%) and of the recovery (using 75 and 100% of the initial output drop).

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

3

Table 1 Business cycles in emerging and developed economies.

Emerging economies Argentina Brazil Chile Israel Korea Malaysia Mexico Peru Philippines Slovakia South Africa Thailand Turkey Mean Developed economies Australia Austria Belgium Canada Denmark Finland Netherlands New Zealand Norway Portugal Spain Sweden Switzerland Mean

σ(y)

σ(tb/y)

σ(c)/σ(y)

σ(x)/σ(y)

ρ(tb/y, y)

4.06 (0.69) 1.52 (0.24) 1.92 (0.23) 1.87 (0.15) 2.18 (0.42) 2.69 (0.58) 2.50 (0.27) 2.02 (0.19) 2.58 (0.61) 2.45 (0.48) 1.63 (0.16) 3.77 (0.68) 3.33 (0.38) 2.50

2.71 (0.52) 0.94 (0.07) 1.98 (0.35) 2.14 (0.19) 1.68 (0.43) 4.22 (0.93) 1.67 (0.33) 1.49 (0.15) 2.34 (0.17) 3.15 (0.58) 1.53 (0.22) 3.66 (0.76) 1.68 (0.20) 2.24

1.15 (0.03) 1.06 (0.26) 1.12 (0.03) 1.63 (0.26) 1.42 (0.15) 1.54 (0.13) 1.26 (0.06) 0.91 (0.11) 0.42 (0.04) 0.94 (0.21) 1.26 (0.07) 0.93 (0.10) 1.19 (0.12) 1.14

3.42 (0.18) 3.67 (0.12) 3.86 (0.15) 5.27 (0.65) 3.45 (0.16) 5.49 (0.36) 3.93 (0.60) 4.62 (0.52) 5.83 (0.72) 3.82 (0.87) 3.47 (0.29) 4.37 (0.28) 3.42 (0.28) 4.20

−0.91 (0.03) −0.55 (0.10) −0.62 (0.08) 0.06 (0.13) −0.69 (0.17) −0.58 (0.19) −0.73 (0.10) −0.62 (0.14) −0.19 (0.15) −0.10 (0.12) −0.49 (0.12) −0.54 (0.17) −0.69 (0.07) −0.51

1.25 (0.23) 1.22 (0.21) 1.07 (0.10) 1.48 (0.17) 1.56 (0.17) 2.28 (0.38) 1.31 (0.15) 1.71 (0.26) 1.32 (0.12) 1.57 (0.17) 1.24 (0.16) 1.73 (0.25) 1.20 (0.09) 1.46

0.95 (0.09) 0.73 (0.07) 0.77 (0.05) 0.94 (0.08) 0.96 (0.08) 1.30 (0.12) 0.62 (0.04) 1.27 (0.13) 1.64 (0.16) 1.07 (0.07) 0.79 (0.04) 0.88 (0.12) 0.95 (0.09) 0.99

0.78 (0.10) 0.80 (0.19) 0.66 (0.09) 0.76 (0.08) 1.21 (0.08) 0.84 (0.11) 0.85 (0.12) 0.94 (0.07) 1.23 (0.14) 1.14 (0.11) 1.14 (0.05) 0.83 (0.16) 0.49 (0.07) 0.90

3.75 (0.50) 2.05 (0.25) 3.46 (0.50) 4.51 (0.21) 3.97 (0.46) 2.86 (0.40) 3.11 (0.13) 3.56 (0.54) 4.62 (0.60) 3.58 (0.40) 3.55 (0.21) 4.86 (0.55) 2.68 (0.28) 3.58

−0.41 (0.09) 0.34 (0.15) 0.03 (0.14) 0.07 (0.25) −0.41 (0.14) −0.02 (0.31) −0.01 (0.10) −0.19 (0.11) 0.07 (0.13) −0.41 (0.14) −0.68 (0.07) −0.19 (0.09) 0.37 (0.21) −0.11

This table reports moments of HP-filtered data. σ(∙) and ρ(∙,∙) denote standard deviation and correlation coefficient, respectively. The variables y, c, x, and tb denote output, consumption, investment, and the trade balance. GMM-based standard errors reported in parentheses.

comparisons that do not involve any time dependence. As expected, data tend to be more volatile during crisis relative to tranquil times, especially under the criterion in Panel C. Yet, the important point to note is that even during tranquil times, consumption is more volatile than output and the trade balance is highly countercyclical, much more than in developed countries. To sum up, results in this section suggest that previous findings do not depend on the particular time span used to compute the business cycle facts. Results are robust to including data from the 2000s and early 2010s (a period with few emerging market crises) and to splitting the sample between crisis and no-crisis times.

faces in world capital markets (the country spread wedge), and as government consumption (the government wedge). Wedges in period t depend on the history st. The efficiency wedge is denoted by A(st); the labor wedge by τl(st); the investment wedge by τx(st); the country spread wedge by z(st), and the government consumption wedge by g(st). A representative household has preferences over contingent sequences of consumption, c(st), and hours worked, l(st), represented by the expected utility function ∞ X X t¼0

3. A prototype small open economy This section describes a small open economy model with incomplete asset markets augmented with five stationary reduced form shocks, referred to as wedges. Time is denoted by t = 0,1,2… and the state of the economy at period t by st. Let st = {s0, s1, …, st} denote the history of the state until time t and π(st) the probability of st as of time zero. The wedges are interpreted at face value as productivity shocks (the efficiency wedge), as labor income and investment taxes (the labor and investment wedges), as fluctuations in the interest rate that the economy

       t t t t t π s ; β ð1 þ ηÞ U c s ; l s

ð1Þ

st

where 0 b β b is a subjective discount factor and η is the growth rate of the population. Here and throughout this section, all variables are expressed in per capita terms. Households own the stock of capital and are able to issue one period bonds traded in international financial markets. Each bond is a contract to deliver one unit of the consumption good in the next period in exchange of q(st) units today. I decompose the discount price as        t ⋆ t t q s ¼ q s exp −z s ;

ð2Þ

Table 2 Empirical regularities: crisis vs. no-crisis times. σ(y)

σ(tb/y)

σ(c)/σ(y)

σ(x)/σ(y)

ρ(tb/y, y)

2.50

2.24

1.14

4.20

−0.51

2.40 4.15

2.12 3.31

1.13 1.09

4.32 4.21

−0.49 −0.54

2.17 5.24

1.88 4.38

1.10 1.15

4.56 3.99

−0.35 −0.80

A. All sample B. Beginning of crisis one quarter after peak No crisis Crisis (pooled) C. Beginning of crisis 5 quarters before peak No crisis Crisis (pooled)

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

4

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

where q⋆(st) is interpreted as the price of a risk-free bond and z(st) as a country spread factor. Fluctuations in z(st) introduce a wedge between the marginal rate of substitution between consumption today and the next period, and the marginal rate of transformation in an economy that faces the relative price q⋆(st). Households face the budget constraint                 t t t t−1 t t t x s þb s ¼ 1−τ l s w s l s c s þ 1 þ τx s       t t−1 t þv s k s þT s     t t þ ð1 þ ηÞb s q s :

         U l st t t t t −  t  ¼ A s F l s ð1 þ γÞ 1−τl s ; Uc s

Here, k(st − 1) is the stock of capital available for production in period t, b(st − 1) is the stock of foreign debt maturing in period t, x(st) is investment, w(st) is the wage rate, v(st) is the rental rate of capital, τx(st) is a tax on investment expenditures, τl(st) is a labor income tax, and T(st) is a lump-sum transfer. Because the government has access to lump-sum transfers, I assume from now on, and without loss of generality, that all foreign debt is held by the households. The initial stock of capital and debt are given by k − 1 and b − 1. Competitive firms rent capital and labor from the households to produce consumption goods with the constant returns to scale technology          t t t−1 t t ; ð1 þ γÞ l s ; y s ¼A s F k s

    X  tþ1 t   tþ1  t t ; π s js U c s q s Uc s ¼ β

ð8Þ

ð9Þ

stþ1 jst

  t   1 þ τx s  t  U c st Gx s   8 9 tþ1  = X  tþ1 t   tþ1 <  tþ1   tþ1  1 þ τ x s tþ1   ¼β A s Fk s þ π s js U c s Gk s : : ; Gx stþ1 tþ1 t s

js

ð10Þ ð3Þ

where γ is the rate of labor augmenting technical progress. These firms choose capital and labor to maximize profits, given by               t t−1 t t t t t t−1 ; ð1 þ γÞ l s −w s l s −v s k s : A s F k s Households own a technology to produce capital goods given by the production function        t t−1 t ð1 þ ηÞk s ¼ G k s ;x s :

An equilibrium allocation of the prototype economy with initial conditions k − 1 and b − 1 is a path for output y(st), consumption c(st), labor l(st), the trade balance tb(st), investment x(st), capital k(st), and foreign bonds b(st) that satisfies the technological constraint Eq. (3), the capital accumulation Eq. (4); the feasibility condition Eq. (5); the trade balance Eq. (6), where q(st) is given by Eq. (2) and q⋆(st) satisfies Eq. (7); and the optimality conditions

The term π(st + 1|st) is the probability of st + 1 conditional on st; Uc(st) and Ul(st) are the marginal utility of consumption and labor; Fk(st) and Fl(st) are the marginal product of capital and labor in the final goods technology; and Gk(st) and Gx(st) are the marginal product of capital and investment in the capital goods technology, all in history st. In particular, Eq. (8) summarizes the intratemporal labor–consumption choice and the demand for labor, Eq. (9) is the intertemporal first order condition with respect to foreign debt, and Eq. (10) summarizes the intertemporal Euler equation with respect to capital.

ð4Þ 4. The decomposition of business cycles

Feasibility in the final good sector requires           t t t t t t c s þ x s þ ð1 þ γ Þ g s þ tb s ¼ y s ;

ð5Þ

where (1 + γ)tg(st) is government consumption and tb(st) represents the trade balance. The latter, in turn, is given by         t t−1 t t tb s ¼ b s −ð1 þ ηÞb s q s :

ð6Þ

Small open economy models with exogenous interest rates have a unit root in equilibrium quantities. Because the unit root complicates the numerical approximation of the equilibrium, the model is rendered stationary by imposing the following debt elastic discount price   1 2 0 3 b st 1 ⋆   ¼ 1 þ r þ ψ4exp@   −φA−15; q⋆ st ð1 þ γ Þy st

ð7Þ

where r⋆ is a world interest rate, φ is the steady state debt to output ratio, and ψ N 0 is a sensitivity parameter. Because in Eq. (7) b(st) and y(st) refer to aggregate variables, households do not internalize the sensitivity of the discount price to changes in b(st)/y(st). Nevertheless, when calibrating the model I set ψ to 0.001, which implies that movements in the debt to output ratio have a small impact on q⋆(st).6

6 Schmitt-Grohe and Uribe (2003) consider other methods to induce stationary and show that they imply similar business cycle dynamics.

The decomposition methodology consists of three steps: the calibration of the model; the measurement and estimation of a stochastic process for the wedges; and the use of counterfactual experiments to measure the contribution of the wedges to aggregate fluctuations. 4.1. Counterfactual experiments The Business Cycle Accounting methodology decomposes the movements in economic aggregates in terms of movements in one or more wedges. To measure the contribution of the wedges to aggregate fluctuations, I simulate counterfactual economies in which one or more wedges take their measured values and the remaining wedges are set to constants. The following experiment, for example, measures the contribution of the labor wedge to aggregate fluctuations. Suppose that we know the stochastic process followed by the state st, we observe the history st, and we know the mappings A(st), τl (st ), τx (st ), z(st), and g(st). I construct a counterfactual economy as follows: given the true stochastic process for st and history st, let     the labor wedge be as in the actual economy, τ^l st ¼ τl st , and map the other wedges to a constant, for instance, their average ^ ðst Þ ¼ A; τ^x st  ¼ τx ; ^zst  ¼ z; and g^ðst Þ ¼ g for all t. The convalues: A tribution of the labor wedge to aggregate fluctuations is measured by comparing the time series generated by the counterfactual economy with the data. The contribution of the other wedges, in isolation or in combination, is measured in a similar way. Note that, in doing these counterfactual experiments, I change the mappings from st to the wedges but keep the same process and realized values for st across

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

experiments. As CKM note, this ensures that expectations of the fluctuating wedges are identical to those in the prototype economy. We therefore need to estimate st and π(st). To estimate st and π(st), I follow Chari et al. (2007) and assume that the state follows a five dimensional stationary autoregressive process stþ1 −s ¼ P ðst −sÞ þ εtþ1 ;

ð11Þ

where s is the mean of st, P is the matrix on lagged values, and εt + 1 is an i.i.d. Gaussian process with mean zero and covariance matrix Σ. Thus, the history of shocks st is summarized by the current state st. To estimate the state, I assume that there is a one to one mapping from the wedges in the prototype economy to st. Since the observation of the wedges uniquely determines the state, let st ≡ (logAt, τlt, τxt , zt , logg t ) without loss of generality. Therefore, the problem of measuring st is reduced to the problem of measuring the wedges. 4.2. Calibration of the model Each period in the model represents one quarter. Preferences are given by

U ðc; lÞ ¼

h i ρ 1−ρ 1−σ c ð1−lÞ −1 1−σ

5

countries, differences in observed long-run averages of the observable variables can be attributed to differences in the steady state wedges and not to preference parameters.8 In this representative small open economy (denoted with a superscript ‘r’), the productivity and population growth rates γr and ηr are 1 and 2%, respectively, and the average country spread, zr , is 3%, all mear sured on an annualized basis. The average efficiency wedge, A , is normalized so that the steady state level of output is 1; the labor and investment wedges are zero, τrl ¼ τrx ¼ 0; and the government consumption wedge, g r , implies a share of government consumption on output of 20%. To find the depreciation rate δ, I use Eqs. (4) and (10) evaluated at the steady state and the assumption τrx ¼ 0 to find δ¼

    r  r  r  ⋆ α 1− 1 þ η 1 þ γ þ ðx=yÞ 1 þ r exp z −1 : α−x=y

Using the parameters of the representative economy and a steady state investment to output ratio of 22% gives δ = 0.057.9 I choose ρ to induce a long-run labor supply of l ¼ 1=3. Evaluating Eq. (8) at the steady state and using τ rl ¼ 0 gives ! 1 1−α 1−l : ¼1þ c=y ρ l

;

where σ N 0, 0 b ρ b 1, and the time endowment is normalized to 1. The production function for consumption goods is given by AF(k, l) = Akαl1− α and the one for capital goods, by

The steady state value of c=y follows by evaluating Eqs. (5) and (6) at the steady state and using the target values for b=y; x=y, and g=y. This procedure gives ρ = 0.29. Finally, evaluating Eq. (9) at the steady state gives the calibrated value β = 0.986.

 2 Gðk; xÞ ¼ x þ ð1−δÞk−0:5ϕ x=k−x=k k;

4.3. Estimation of the model and measurement of the wedges

ð12Þ

where x=k is the investment–capital ratio in a balanced growth path and ϕ N 0. To solve for the equilibrium of the model, all trending variables dated t are normalized by (1 + γ)t. Throughout the paper, a ‘bar’ above a variable refers to its normalized steady state value. The parameters η, γ, ϕ, s, P, and Q are country-specific and their estimation is described in the next section.7 The remaining parameters are calibrated to match the long-run moments of a representative small open economy. The parameter σ, which measures the curvature of the period utility, is set at 2; the capital share in output, α, takes the value 0.32; the debt elasticity parameter ψ, the value 0.001; and the world interest rate r⋆ is set at 4% on an annualized basis. These are standard values used in the open economy literature (Schmitt-Grohe and Uribe, 2003; Neumeyer and Perri, 2005; Aguiar and Gopinath, 2007). The steady state external debt to output ratio, φ, is set at 50% on an annualized basis— Reinhart et al. (2003) report an average external debt to output ratio of 44% for emerging countries with history of default and 54% for industrial countries. It remains to calibrate the parameters β, ρ, and δ. I set these parameters by matching steady state quantities derived from the model with the long-run moments of a representative small open economy. In particular, I evaluate the appropriate steady state conditions at the proposed moments and solve for the parameters that make the steady state of the model consistent with them. This is somewhat different from the usual approach of using data from a single country to calibrate the parameters. I follow the alternative approach for two reasons. First, several steady state quantities depend on the value of the wedges at the steady state. Thus, to calibrate the parameters one needs to take a stand on the average value of the wedges. And second, by using a single calibration for all 7 The parameter ϕ is related to average price of capital and does not affect steady state quantities. CKM set ϕ so that the elasticity of the price of capital with respect to the investment to capital ratio is 25%, a value consistent with U.S. estimates. Because I could not find reliable estimates of ϕ for the group of emerging countries, I chose to estimate it on a country-by-country basis.

The data used for the estimation of the parameters (η, γ, ϕ, s, P, and Σ) are quarterly observations on gross domestic product, investment, the trade balance, government consumption expenditures (all in real terms), and labor input. Data from Mexico covers the period 1985:1–2011:3; data from Canada, the period 1976:1–2011:3. Labor input is defined as average hours worked per worker multiplied by total employment and divided by total hours available for work. (See the online Appendix for details.) In the model, output fluctuates around the trend [(1 + η)(1 + γ)]t. The parameter η is measured as the average growth rate of the working–age population and γ is calibrated as the residual in the equation (1 + output growth) = (1 + γ)(1 + η). Data on output, investment, the trade balance, and government consumption dated t were exponentially detrended by [(1 + η)(1 + γ)]t and further normalized by average detrended output, so that the resulting output series has a sample mean of 1. There are no degrees of freedom for estimatingz. Indeed, Eqs. (9) and (10) in steady state imply that   ⋆ ρð1−σ Þ−1 ; 1 ¼ 1 þ r expðzÞβð1 þ γÞ

ð13Þ

which, given r⋆, β, γ, ρ, and σ, determines z. The parameters A, τl, τx, g, P, Σ, and ϕ are estimated by the method of maximum likelihood. To that end, the model is log-linearized around the steady state and the likelihood function is evaluated using the Kalman filter and data on output, investment, the trade balance, hours worked, and government consumption expenditures. (See Appendix A for details.) Table 3 reports the estimates. 8 Differences in γ and η across countries could also drive differences in steady state values. Given reasonable values for these parameters, however, their impact on steady state quantities is very small. 9 The average investment rate in Canada, Mexico, and Korea is about 22%.

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

6

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

Table 3 Estimates for Mexico and Canada. s

Matrix Q, where Σ = QQ′

Matrix P on lagged values

A. Mexico estimates, 1985:1–2011:3. (ϕ = 20.42, Log-likelihood = 1908.77) 2 3 2 0:29 0:92 −0:16 0:11 1:41 6 0:24 7 6 −0:05 0:85 0:08 0:39 6 7 6 6 0:24 7 6 −0:06 1:14 0:23 −8:60 6 7 6 4 0:005 5 4 −0:02 −0:05 0:03 1:26 −2:31 0:39 −0:28 0:25 3:88

3 0:07 0:07 7 7 −0:37 7 7 0:03 5 0:87

2

B. Canada estimates, 1976:1–2011:3. (ϕ = 16.98, Log-likelihood = 2993.40) 2 3 2 0:94 0:06 0:03 0:06 0:22 6 0:39 7 6 −0:33 0:67 −0:02 2:60 6 7 6 6 −0:11 7 6 0:38 0:37 0:99 −3:42 6 7 6 4 0:008 5 4 −0:01 0:00 0:00 0:99 −1:63 0:19 0:07 −0:05 −1:16

3 −0:02 −0:01 7 7 0:04 7 7 −0:00 5 1:02

2

1:39 0:00 0:00 0:00 6 −0:10 1:43 0:00 0:00 6 6 0:58 −1:70 3:61 0:00 6 4 −0:13 0:23 −0:24 0:12 0:31 0:44 0:60 −2:32

0:00 0:00 0:00 0:56 6 0:24 0:56 0:00 0:00 6 6 −0:13 −0:62 1:52 0:00 6 4 0:06 0:06 0:02 0:02 0:08 −0:22 0:61 −0:26

3 0:00 0:00 7 7 0:00 7 7 0:00 5 1:00 3 0:00 0:00 7 7 0:00 7 7 0:00 5 1:12

This table reports the estimated parameters of the process for the wedges and capital adjustment cost by the method of maximum likelihood. The entries of the matrix Q are multiplied by 100.

Table 4 Properties of estimated wedges in Mexico. Correlation of wedge with Wedge

σw

Output

Investment

Hours

Trade bal.

Consump.

Efficiency Labor Investment Ctry. spr. Gov. cons.

1.90 (0.25) 2.00 (0.15) 4.62 (0.50) 0.45 (0.03) 0.25 (0.03)

0.94 (0.02) −0.77 (0.03) −0.07 (0.26) −0.54 (0.11) 0.28 (0.11)

0.70 (0.10) −0.60 (0.07) −0.61 (0.13) −0.07 (0.12) 0.14 (0.14)

0.47 (0.08) −0.85 (0.02) −0.34 (0.17) −0.25 (0.10) 0.18 (0.09)

−0.65 (0.09) 0.79 (0.04) 0.21 (0.17) 0.48 (0.08) −0.28 (0.11)

0.87 (0.03) −0.85 (0.02) 0.10 (0.24) −0.70 (0.08) 0.28 (0.10)

Correlation matrix of wedges Wedge

σw/σy

Efficiency

Labor

Investment

Ctry. spr.

Gov. cons.

Efficiency Labor Investment Ctry. spr. Gov. cons.

0.76 (0.04) 0.80 (0.06) 1.85 (0.23) 0.18 (0.02) 0.10 (0.02)

1

−0.56 (0.06) 1

0.04 (0.26) −0.06 (0.20) 1

−0.54 (0.11) 0.65 (0.07) −0.71 (0.10) 1

0.25 (0.10) −0.23 (0.09) 0.15 (0.13) −0.51 (0.11) 1

The wedges were HP-filtered with smoothing parameter of 1600. Column σw reports the standard deviation of the wedges in percentage terms, while column σw/σy reports the standard deviation of the wedges relative to that of output. GMM-based standard errors are reported in parentheses.

Given the parameters, the realized wedges are estimated using the Kalman smoother on the log-linearized model.10 The Kalman smoother computes the expectation of an unobservable state of a linear state space model conditional on all the information in the sample. As byproduct, this procedure gives the best estimates of capital and debt at the initial period. Although I use the Kalman smoother to estimate the realized wedges, some of them are directly observed or follow from static first order conditions. First, the government consumption wedge is observed. Second, given a guess for the initial stock of capital and data on output, investment, and hours, the efficiency wedge can be measured as a Solow residual. Finally, the static labor-consumption condition (8) implies that

  t 1−τl s ¼

0  t 1  t c s 1−ρ @ l s  A  t  ; ρð1−α Þ 1−l st y s

which delivers the labor wedge. Up to the linearization of the last equation, the smoother identifies the same wedges as the direct procedure described above. The advantage of the Kalman smoother is that it provides estimates of the two intertemporal wedges (investment and country spread wedges) and of the initial stocks of capital and debt. 10 Because there are five wedges and five observable variables, variations in the wedges account for all the movements in the data.

5. Results This section applies the decomposition methodology to measure the contribution of the wedges to aggregate fluctuations in Mexico and Canada. I examine fluctuations at the business cycle frequency, the 1995 Mexican crisis, and the 1983 Canadian recession. The main findings are the following. First, except for the government consumption wedge, all wedges are more volatile in Mexico than in Canada. Second, the combined contribution of efficiency, labor, and country spread wedges accounts for most aggregate fluctuations in Mexico. Most fluctuations in Canada are driven by movements in efficiency and labor wedges. The investment and country spread wedges, however, contribute to the behavior of investment, the trade balance, and consumption. Third, fluctuations in the country spread wedge explain the qualitative differences between Mexico and Canada: the excess volatility of consumption over output and the highly countercyclical trade balance in Mexico. Finally, the government consumption wedge does not play any role in either country. To check the robustness of the results, Section 5.5 applies the methodology using Korean data. 5.1. Decomposition of aggregate fluctuations Tables 4 and 5 report summary statistics of the estimated wedges in Mexico and Canada. Comparing the top panels of the tables, we observe that, except for the government consumption wedge, all wedges in Mexico are substantially more volatile than in Canada: the efficiency wedge by over 150%, the labor and investment wedges by over 100%,

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

7

Table 5 Properties of estimated wedges in Canada. Correlation of wedge with 3-7 Wedge

σw

Output

Investment

Hours

Trade bal.

Consump.

Efficiency Labor Investment Ctry. spr. Gov. cons.

0.75 (0.06) 0.95 (0.18) 2.13 (0.11) 0.12 (0.02) 0.36 (0.02)

0.68 (0.06) −0.51 (0.16) −0.70 (0.06) −0.18 (0.23) −0.44 (0.12)

0.48 (0.09) −0.47 (0.15) −0.90 (0.01) −0.35 (0.17) −0.57 (0.06)

0.30 (0.09) −0.73 (0.13) −0.70 (0.05) −0.51 (0.21) −0.52 (0.08)

0.22 (0.12) 0.33 (0.08) −0.08 (0.20) 0.57 (0.07) −0.29 (0.16)

0.41 (0.12) −0.72 (0.06) 0.00 (0.16) −0.51 (0.07) 0.06 (0.15)

Correlation matrix of wedges 3-7 Wedge

σw/σy

Efficiency

Labor

Investment

Ctry. spr.

Gov. cons.

Efficiency Labor Investment Ctry. spr. Gov. cons.

0.50 (0.02) 0.64 (0.08) 1.44 (0.11) 0.08 (0.01) 0.25 (0.03)

1

0.07 (0.10) 1

−0.35 (0.11) 0.15 (0.16) 1

0.43 (0.11) 0.89 (0.03) 0.11 (0.19) 1

−0.18 (0.14) 0.16 (0.11) 0.68 (0.06) 0.10 (0.14) 1

See note in Table 4.

and the country spread wedge is almost four times more volatile in Mexico than in Canada. These magnitudes are in line with the relative volatility of output (Table 1). Furthermore, the lower panel of the table shows that all wedges in Mexico (except for the government consumption wedge) are more volatile than in Canada even when measured relative to the volatility of output. The upper panel of Table 4 shows that, in Mexico, downturns are associated with declines in the efficiency and government consumption wedges, and with increases in the labor and country spread wedges. Interestingly, the investment wedge is acyclical and is also uncorrelated with consumption and the trade balance. In other words, downturns in Mexico are periods with low productivity, with more distortions in the allocation of labor, and more distortions in the intertemporal allocation of consumption. Distortions in investment decisions, as measured by the investment wedge, do not seem to be stronger during recessions. On the other hand, the correlations are different in Canada. First, the investment wedge is highly countercyclical, meaning that investment decisions are more distorted during downturns; second, the country spread wedge is acyclical, although, as in Mexico, it is negatively correlated with consumption and positively correlated with the trade balance; and third, the government consumption wedge is countercylical.11 To measure the contribution of the wedges to aggregate fluctuations, I compare the data with the predictions of simulated economies in which some wedges take their estimated values while others are set to constants. These counterfactual economies are computed by changing the mapping from the state st to the wedges but keeping the measured st and estimated process π(st) unchanged. This transformation amounts to setting a diagonal indicator matrix such that 3 2 3 2 e ω1 logA log Abt 7 6 6 τ^ 7 6 e 6 lt 7 6 τl 7 6 0 7 6 6 τ^ 7 ¼ 6 e 6 xt 7 6 τ x 7 þ 6 0 5 4 0 4 ^z 5 4 e z t 0 e b logg logg t 2

0 ω2 0 0 0

0 0 ω3 0 0

0 0 0 ω4 0

3 32 e 0 s1t −logA 6 0 7 τl 7 7 76 s2t −e 6 0 7 τx 7 7; 76 s3t −e 5 4 0 s4t −e z 5 ω5 s5t −loge g

where st = [s1t, s2t, s3t, s4t, s5t]′ is the state estimated in Section 4.3, the symbol ‘~’ on top of a variable denotes the point at which the inactive wedges are evaluated, and ωi equals 1 if the ith wedge is active and 0 otherwise. For example, an economy in which efficiency, investment, and government consumption wedges are set at their steady e ¼ logA; e τx ¼ state values satisfies the following parameterization: logA τx , loge g ¼ logg, ω1 = ω3 = ω5 = 0, and ω2 = ω4 = 1. Importantly, in these counterfactual economies, households and firms still face as 11 Neumeyer and Perri (2005) and Uribe and Yue (2006) find that in emerging countries market interest rates are negatively correlated with output and positively correlated with the trade balance. Kaminsky et al. (2005) document that fiscal policy is procyclical in developing countries but countercyclical in developed countries.

state variable the vector st, but realize that the wedges are now a different function of the state st. In measuring the contribution of, say, the labor wedge to aggregate fluctuations, two counterfactual economies are illustrative. In the first economy, only the labor wedge moves as estimated and the other wedges are fixed at their average values. In the second economy, the labor wedge is fixed at its average value and the rest of the wedges move as estimated. The first experiment measures the direct contribution of the labor wedge: the closer are the counterfactual time series to the data, the more important is the direct contribution of the labor wedge to aggregate fluctuations. The second experiment measures the contribution of the labor wedge when combined with other wedges: the farther away are the predicted time series from the actual data, the more important is the contribution of the labor wedge, when combined with other wedges, to aggregate fluctuations. Tables 6 and 7 report summary statistics of counterfactual economies with just one wedge (Panel A) and with all wedges but one (Panel B). The second and third columns report, respectively, the standard deviation of counterfactual output relative to that of actual output and the correlation between counterfactual and actual outputs. A combination of wedges successfully explains output fluctuations if predicted output has similar volatility than actual output and if predicted output increases when actual output does. The last four columns report summary statistic of counterfactual economies. In reporting these numbers, I added a geometric trend to the counterfactual series equal to the sum of the estimated productivity and population growth rates and next HP-filtered the resulting series. This procedure replicates the transformation applied to the data. The top panel of Tables 6 and 7 suggests that only the efficiency and labor wedges could account, by themselves, for an important fraction of output fluctuations in both countries. In the case of Mexico, however, these models substantially understate the volatility of consumption and investment, and predict that the trade balance-to-GDP ratio is highly procyclical. In Canada, these models miss the behavior of the trade balance, substantially understate the volatility of investment, but produce output volatility similar to that in the data and are consistent with the smaller volatility of consumption relative to that of output. The investment, country spread, and government consumption wedges cannot explain, by themselves, aggregate fluctuations in either country. In Mexico, the economy with just the investment wedge accounts for roughly 10% of output volatility, greatly overstates the volatility of investment, and misses the negative correlation between output and the trade balance. In the economy with just the country spread wedge, predicted and actual output are negatively correlated, and the correlation between output and the trade balance is almost one. Finally, the economy with just the government consumption wedge accounts for only 3% of the output volatility and misses the volatility

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

8

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

Table 6 Contribution of the wedges to aggregate fluctuations in Mexico. σ ðyo Þ σ ðyÞ

ρ(yo, y)

σ ðco Þ σ ðyo Þ

  ρ tby ; yo

σ

−

−

1.26 (0.06)

−0.73 (0.10)

1.67 (0.33)

3.93 (0.60)

A. Economies with just one wedge. Efficiency 1.24 (0.05) Labor 0.78 (0.05) Investment 0.11 (0.02) Ctry. spr. 0.94 (0.04) Gov. cons. 0.03 (0.01)

0.96 (0.01) 0.78 (0.03) 0.39 (0.14) −0.90 (0.02) −0.24 (0.16)

0.24 (0.02) 0.51 (0.02) 1.63 (0.12) 0.89 (0.03) 0.99 (0.09)

0.98 (0.01) 0.90 (0.02) 0.30 (0.06) 0.99 (0.00) 0.20 (0.15)

2.55 (0.68) 1.29 (1.39) 1.61 (5.29) 5.15 (8.59) 0.23 (11.43)

0.45 (0.06) 0.79 (0.06) 29.34 (3.02) 2.71 (0.07) 1.17 (0.20)

B. Economies with all wedges but one. No efficiency 0.40 (0.03) No labor 0.62 (0.03) No investment 0.96 (0.02) No Ctry. spr. 1.89 (0.03) No Gov. cons. 1.00 (0.00)

−0.48 (0.13) 0.63 (0.07) 0.99 (0.00) 0.98 (0.01) 1.00 (0.00)

2.86 (0.32) 1.72 (0.10) 1.36 (0.09) 0.32 (0.03) 1.27 (0.05)

0.51 (0.10) −0.15 (0.12) −0.52 (0.07) 0.91 (0.03) −0.74 (0.09)

3.85 (0.64) 2.67 (0.36) 2.07 (0.19) 3.75 (0.42) 1.57 (0.31)

9.57 (1.87) 5.65 (0.80) 3.32 (0.16) 1.86 (0.27) 3.93 (0.59)

Data

o

o

  o

tb yo

σ ðxo Þ σ ðyo Þ

σ(∙) and ρ(∙,∙) denote standard deviation and correlation coefficient, respectively. The variables y, c, x, and tb denote output, consumption, investment, and the trade balance. Conterfactual time series are denoted with the superscript o. GMM-based standard errors reported in parentheses.

of the trade balance and the correlation between output and the trade balance. Likewise, in Canada, the model with just the investment wedge accounts for less than 20% of the volatility of output and also completely overstates the volatility of consumption and investment relative to that of output. In the model with just the country spread wedge, predicted output is negatively correlated with actual output, consumption is more volatile than output, and the correlation between output and the trade balance is almost one. Finally, the model with just the government consumption wedge accounts for less than 10% of the volatility of output. Panel B of Table 6 suggests that three wedges are essential to understand business cycles in Mexico: the efficiency, labor, and country spread wedges. Eliminating any of these wedges causes the model to miss the data in some dimension. In the model with no efficiency wedge, predicted output is 60% less volatile than actual output and their correlation is negative, investment is over nine times more volatile than output, and the correlation between output and trade balance-to-GDP ratio is positive. In the model with no labor wedges, predicted output is almost 40% less volatile than actual output, the volatility of consumption and investment is overstated, and the correlation of output with the trade balance-to-GDP ratio is not significantly different from zero. Finally, in the model with no country spread wedges, predicted output is almost 90% more volatile than actual output, consumption is substantially less volatile than output, the volatility of investment is understated, and the trade balance-to-GDP ratio is highly procyclical. The investment and government consumption wedges, on the other hand, can be eliminated from the model without severely affecting its ability to match the data. Interestingly, predicted investment is almost as volatile as in the data even without including the investment wedge.

On the other hand, Table 7 suggests that business cycles in Canada are mostly driven by efficiency and labor wedges, although investment and country spread wedges play some role. The model without investment wedges misses the volatility of investment and overstates the correlation between output and the trade balance. Likewise, the model with no country spread wedge understates the volatility of consumption and investment, and overstates the correlation between output and the trade balance. The government consumption wedge has a negligible role in Canada's business cycles. In summary, these findings suggest that business cycles in Mexico are accounted for by the combined contribution of the efficiency, labor, and country spread wedges. Moreover, the country spread wedge accounts for the qualitative differences between Mexico and Canada: the excess volatility of consumption over output and the highly countercyclical trade balance-to-GDP ratio. On the other hand, while most fluctuations in Canada are accounted for by efficiency and labor wedges, the country spread and investment wedges seem to play a role in the behavior of investment, the trade balance, and consumption as well. 5.2. The 1995 Mexican crisis The left panel of Fig. 1 shows output and the estimated wedges in the 1995 Mexican crisis. Here, output is expressed as (log) percentage deviation relative to its value at the beginning of the recession and the wedges are expressed relative to their values at the beginning of the recession. Output falls 12% in two quarters and remains below trend for several years. The efficiency, labor, and country spread wedges deteriorate throughout the recession, although by 1998, the labor wedge

Table 7 Contribution of the wedges to aggregate fluctuations in Canada. σ ðyo Þ σ ðyÞ

ρ(yo, y)

σ ðco Þ σ ðyo Þ

  ρ tby ; yo

σ

−

o

o

  o

tb yo

σ ðxo Þ σ ðyo Þ

−

0.76 (0.08)

0.07 (0.25)

0.94 (0.08)

4.51 (0.21)

A. Economies with just one wedge: Efficiency 0.94 (0.02) Labor 0.87 (0.10) Investment 0.16 (0.02) Ctry. spr. 0.57 (0.06) Gov. cons. 0.09 (0.02)

0.82 (0.02) 0.54 (0.13) 0.17 (0.07) −0.51 (0.21) 0.13 (0.11)

0.50 (0.05) 0.55 (0.07) 3.80 (0.36) 1.23 (0.06) 1.13 (0.05)

0.93 (0.01) 0.95 (0.02) −0.03 (0.05) 0.98 (0.00) −0.34 (0.10)

1.37 (1.39) 1.05 (3.00) 1.29 (4.78) 1.86 (10.52) 0.22 (8.19)

0.79 (0.09) 0.66 (0.06) 24.45 (2.35) 3.30 (0.08) 1.47 (0.08)

B. Economies with all wedges but one: No efficiency 0.59 (0.04) No labor 0.90 (0.11) No investment 0.98 (0.01) No Ctry. spr. 1.38 (0.11) No Gov. cons. 0.99 (0.01)

0.39 (0.04) 0.58 (0.12) 0.99 (0.00) 0.93 (0.03) 1.00 (0.00)

1.56 (0.15) 0.56 (0.08) 0.72 (0.09) 0.28 (0.03) 0.89 (0.09)

0.03 (0.17) 0.50 (0.10) 0.64 (0.10) 0.89 (0.04) −0.08 (0.25)

1.78 (0.24) 1.53 (0.20) 1.45 (0.12) 1.43 (0.30) 0.95 (0.09)

8.36 (0.50) 4.89 (0.58) 1.96 (0.20) 2.64 (0.39) 4.50 (0.24)

Data

See note in Table 6.

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

9

Evolution of output and wedges Mexico

15

Canada

15

10

10

5

5

0

0

−5

−5

−10

−10

Output Efficiency wedge Labor wedge Investment wedge Ctry. spread wedge

−15

−15 1995

1996

1997

1998

1982

1983

1984

1985

Fig. 1. Output and wedges during the Mexican crisis (left panel) and the Canadian recession (right panel). Output is expressed as (log) percentage deviation and wedges as simple deviations from their values at the beginning of the recession. The country spread wedge is expressed in annualized terms.

fully recovers. Relative to its value in the last quarter of 1994, the country spread wedge increased 6.5 percentage points by mid-1996 and remained roughly constant until 1998. In terms of levels, the implicit interest rate (country spread plus world interest rate) jumped from 4.4 percentage points during the last quarter of 1994 to over 13 percentage points by mid 1996.12 On the other hand, the investment wedge initially increased sharply by almost 15 percentage points in two quarters and improved substantially thereafter; by mid 1996, the investment wedge was 5 percentage points lower than at the beginning of the crisis. That is, while there was a sharp initial distortion in investment decisions, these distortions rapidly disappeared. The solid lines in Fig. 2 represent Mexican data. Output, hours, and investment are normalized to equal 100 at the beginning of the crisis; the trade balance is normalized by the initial level of output and is shown in percentage points. Labor initially drops only 3% but soon recovers, and by 1998 it is over 4% above its initial value. Investment drops abruptly and co-moves closely with output. Finally, the trade balance moves sharply from a deficit of 4% to a surplus of 4% and remains in surplus until 1998. The other lines represent the prediction of counterfactual economies in which active wedges are set to their estimated values and inactive wedges, at their values at the beginning of the crisis. The upper panel of the figure shows the prediction of models with just one wedge. The model with just the efficiency wedge matches the evolution of output reasonably well and predicts a larger drop in labor that that observed in the data. This model cannot explain the drop in investment and misses the behavior of the trade balance. The model with just the labor wedge, on the other hand, predicts a fall in output of about one third of the fall in the data and a fall in hours larger than that in the data. This model, however, captures the recovery and most ups and downs in output and hours, but fails to match investment and the trade balance. The models with just the investment wedge and just the country spread wedge are unable to explain the crisis by themselves. The model with just the investment wedge completely misses the data: it predicts a steady increase in output and labor, and a small initial decline in investment followed by a strong increase since mid1995. The model with just the country spread wedge predicts a large increase in output and labor, a relatively small decline in investment, and a large increase in the trade balance. Note, however, that the country spread wedge is the only wedge that, by itself, seems to be able to

12 The correlation between the implicit and actual interest rates (constructed as in Neumeyer and Perri (2005)) is 0.62. Moreover the implicit interest rate peaks during the crisis (although with a lag relative to the actual rate) and broadly matches the persistent decline in the interest rate observed between the late 1990s and the 2008 financial crisis. See the online Appendix for more details.

drive a substantial and persistent increase in the trade balance, as observed in the data. The lower panel of Fig. 2 displays simulations of economies with all wedges but one. In the model with no efficiency wedge, predicted output and hours increase. The model matches the behavior of investment remarkably well, and predicts an increase in the trade balance substantially larger than in the data. In other words, the efficiency wedge contributes substantially to the behavior of output, labor, and the trade balance, but it does not contribute at all to the dynamics of investment. The model with no labor wedge predicts a decline in output of about two thirds of that observed in the data and a quick recovery; predicted hours increase throughout the crisis; predicted investment matches the data remarkably well; and the predicted trade balance increases more than in the data. These findings suggests that, during the crisis, the labor wedge contributed substantially to the evolution of output and labor, somewhat to the behavior of the trade balance, and nothing to the evolution of investment. Consider next the model with no investment wedges. The lower panel of Fig. 2 shows that eliminating the investment wedge does not substantially affect the ability of the model to match output, hours, and the trade balance. Because the model with no investment wedges is only unable to explain the large drop in investment and its recovery, it can be concluded that, if anything, the investment wedge only contributed to the behavior of investment during the Mexican crisis. The model with no country spread wedge predicts a drop in output and hours much larger than those observed in the data; accounts for about half of the drop in investment, predicting a quick recovery; and a large drop in the trade balance. These findings confirm that the country spread wedge contributed substantially to the Mexican crisis. While its contribution to the behavior of output and labor was important, its key role was to drive the large increase in the trade balance and drop in consumption, for the model without country spread wedges predict movements in the opposite direction in these aggregate variables. In effect, the left panel of Fig. 3 shows that the country spread wedge is the only wedge that is able to account for the large drop in consumption. In sum, these findings are consistent with those of the previous section and suggest that the efficiency, labor, and country spread wedges account for most of the aggregate behavior during the 1995 Mexican crisis. Moreover, the country spread is the main wedge that accounts for the behavior of consumption and, combined with the efficiency wedge, of the trade balance. 5.3. The 1983 Canadian recession The right panel of Fig. 1 displays output and the estimated wedges during the Canadian recession. Output drops by 9% at the trough of

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

10

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

Mexico: models with just one wedge Output

Hours

115

110

110

105

105

100

100 95 95 90 90 85 85

Investment

120

Trade Balance 15

110

10

100

5

90

0

80

−5

70

−10

60

−15

50 1995 Data

1996

1997

Efficiency wedge

1998

1995

Labor wedge

1996

1997

Investment wedge

1998

Country spread wedge

Mexico: models with all wedges but one Output

Hours

115

110

110

105

105

100

100 95 95 90 90 85 85

Investment

120

Trade Balance 15

110

10

100

5

90

0

80

−5

70

−10

60

−15

50 1995 Data

1996

1997

No efficiency wedge

1998 No labor wedge

1995

1996

No investment wedge

1997

1998

No ctry. spread wedge

Fig. 2. Data and predictions of models during the Mexican crisis.

the recession. There is a small decline in the efficiency wedge, which returns to trend by 1984. The labor wedge increases during the recession, mirroring the behavior of output and reaching 5 percentage points at the trough. The investment wedge increases to roughly 13 percentage points and remains at that level for several years. The country spread increases by about 2 percentage points, and remains relatively flat throughout the recession. The solid lines in Fig. 4 represent Canadian data. Investment co-move closely with output and hours until 1983, but then recovers more slowly than these variables. The trade balance,

on the other hand, increases from 0 to over 4 percentage points, and remains in surplus for several years. The upper panel of Fig. 4 shows the prediction of counterfactual economies with just one wedge. The model with the efficiency wedge matches the initial drop in output and hours reasonably well, but it predicts a faster recovery; this model also completely misses the behavior of investment and the trade balance. The model with just the labor wedge matches output and hours remarkably well, although it fails to account for the dynamics of investment and the trade balance. The

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

11

Consumption and counterfactual economies Mexico

Canada 105

100

100

95

90

Consumption No efficiency wedge No labor wedge No investment wedge No ctry. spread wedge

95 85 1995

1996

1997

1998

1982

1983

1984

1985

Fig. 3. Consumption and predictions of models with all wedges but one during Mexican crisis (left panel) and the Canadian recession (right panel).

model with just the investment wedge completely misses output and hours, but matches investment and the trade balance reasonably well. Finally, the model with just the country spread wedge predicts an increase in output and hours, predicts only a small decline in investment and a large increase in the trade balance (surpassing the actual increase) with a quick decline after the peak. The lower panel of Fig. 4 displays the prediction of models with all wedges but one. The model with no efficiency wedge only matches reasonably well the behavior of investment, but completely misses the behavior of output and hours, and predicts an increase in the trade balance much larger than that observed in the data. The model with no labor wedge also misses output and labor, and matches closely the dynamics of investment. The model explains the trade balance for only three quarters, but then predicts a large increase not observed in the data. The model with no investment wedge matches output and hours almost perfectly, but completely misses the behavior of investment and the trade balance. Finally, the model with no country spread wedge predicts a drop in output and hours larger than those observed, explains about two thirds of the decline in investment, but completely misses the dynamics of the trade balance. Moreover, the country spread wedge also seems to contribute substantially to the drop in consumption during the Canadian crisis (right panel of Fig. 3). Overall, these findings suggest that most of the drop in output and labor was due to the combined contribution of efficiency and labor wedges. The investment wedge only contributed to the behavior of investment and the trade balance. In addition, the modest increase in the country spread wedge seems to have acted to avoid larger drops in output and labor than those actually observed, and contributed to the dynamics of consumption, the trade balance, and somewhat to that of investment.13

country spread wedge does not play any important role in aggregate fluctuations. Importantly, Lama estimates an extremely flat (almost constant) country spread wedge for the six countries he analyzes. This is in sharp contrast with the results in this paper, where the country spread wedge not only fluctuates substantially, but is also the main reduced form shock that can account for the excess volatility of consumption over output and the highly countercyclical trade balance in Mexico. Several differences between the papers could explain the opposite conclusions. First, Lama uses annual data and short time series; for example, Mexico's data only covers the period 1991–2006. This is of concern because fluctuations in aggregate variables are muted at an annual frequency and because wedges are persistent and, therefore, long time series are preferred. Second, Lama uses data on employment while I constructed data on labor input that considers not only changes in the extensive margin (employment) but also in the intensive margin (hours worked). This is important in the case of emerging economies where large layoff costs could lead to labor adjustments mostly on the intensive margin. Finally, and most importantly, Lama assumes that each wedge follows an independent autoregressive process. This assumption is inconsistent with the equivalence results in the BCA methodology because shocks in many structural models induce correlated reduced form wedges (see Section 6). I re-estimated the model imposing diagonal P and Q matrices, replicating Lama's framework. The constrained model generates an extremely flat country spread wedge leading to the conclusion that country spread wedges do not play any important role in explaining the behavior of the trade balance or consumption during the Mexican crisis. Thus, it seems that the constraint imposed in the stochastic process for the wedges – a constraint that is statistically rejected – leads Lama to conclude that the country spread does not play any fundamental role in aggregate fluctuations.

5.4. Discussion 5.5. Decomposition of fluctuations in Korea In independent work, Lama (2011) uses a version of BCA to study output drops in a group of Latin American countries. Lama concludes that only efficiency and labor wedges are important to understand fluctuations in these countries, and claims that a shock similar to the

13 Results are robust to using a quadratic approximation to the policy functions. The only difference is that the investment wedge seems to play some role in the recovery of output and hours in Canada and, to a lesser extent, in Mexico. I also performed a number of sensitivity analysis of the previous findings: I estimated a model with extreme capital adjustment costs based on Mexican data, and I replicated all the above experiments using the quasi-linear preferences advocated by Greenwood et al. (1988). With minor caveats, all results are consistent with the previous findings. See the online Appendix.

This section applies the decomposition methodology to Korea. Table 8 shows the properties of the estimated wedges (top panel) and the prediction of counterfactual economies with all wedges but one (bottom panel). Note that the volatility and the sign of the correlation of the wedges with the macroeconomic aggregates in Korea are mostly consistent with those in Mexico, in particular for those which are statistically different from zero. An important difference, however, is in the volatility of the labor wedge in Korea, which is about 80% more volatile than in Mexico. Likewise, while in Mexico the efficiency wedge is positively correlated with hours worked, in Korea this correlation is virtually zero. As argued below, these observations reflect a difference in the

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

12

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

Canada: models with just one wedge Output

105

Hours 105

100

100

95

95

90 90 85

Investment

110

Trade balance

100

10

90

5

80

0

70 −5 60 −10 1982 Data

1983

1984

Efficiency wedge

1985

1982

Labor wedge

1983

1984

Investment wedge

1985

Country spread wedge

Canada: models with all wedges but one Output

105

Hours 105

100

100

95 95 90 90 85

Investment

110

Trade balance 10

100 90

5

80

0

70

−5

60 −10 1982 Data

1983

1984

No efficiency wedge

1985 No labor wedge

1982

1983

No investment wedge

1984

1985

No ctry. spread wedge

Fig. 4. Data and predictions of models during the Canadian recession.

relative importance of the efficiency and labor wedges in Korea. Yet, the dimensions in which Korea looks different from Mexico are also dimensions where it looks different from Canada as well. Consider next the predictions of models with all wedges but one. As in the case of Mexico, results are consistent with the view that efficiency, labor, and country spread wedges account for most fluctuations in Korea. Compared with Mexico, however, there are some differences

in the relative importance of these wedges. The labor wedge plays a major role in Korea: the predictions of the model with no labor wedges fail to match most business cycle moments, like the excess volatility of consumption over output and the negative correlation between the trade balance and output. The labor wedge does not have such a prominent role in Mexico. In contrast, the efficiency wedge plays a smaller role in Korea than in Mexico—the contribution of this wedge, however,

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

13

Table 8 Decomposition exercise in Korea. Correlation of wedge with Wedge

σw

Output

Investment

Hours

Trade bal.

Consump.

Efficiency Labor Investment Ctry. spr. Gov. cons.

1.67 (0.18) 3.57 (0.78) 1.25 (0.17) 0.48 (0.06) 0.22 (0.03)

0.71 (0.06) −0.70 (0.17) 0.27 (0.20) −0.29 (0.22) 0.08 (0.08)

0.58 (0.12) −0.57 (0.25) −0.04 (0.35) −0.01 (0.36) 0.18 (0:09)

−0.04 (0.29) −0.89 (0.03) 0.15 (0.20) −0.17 (0.22) 0.08 (0.11)

−0.46 (0.15) 0.72 (0.17) −0.40 (0.21) 0.47 (0.21) −0.19 (0.06)

0.51 (0.12) −0.82 (0.12) 0.47 (0.15) −0.52 (0.15) 0:17 (0.05)

Counterfactual experiments

Data





σ ðyo Þ σ ðyÞ

ρ(yo, y)

σ ðco Þ σ ðyo Þ

ρ

−

−

1.42 (0.15)

−0.69 (0.17)

Economies with all wedges but one. No Efficiency 0.94 (0.13) No labor 1.07 (0.26) No investment 1.25 (0.04) No Ctry. spr. 1.31 (0.07) No Gov. cons. 1.00 (0.00)

0.52 (0.22) 0.11 (0.25) 0.94 (0.02) 0.93 (0.03) 1.00 (0.00)

1.54 (0.33) 1.08 (0.30) 1.42 (0.17) 2.39 (0.69) 1.36 (0.08)

o

tb yo

; yo

σ

0.53 (0.05) 0.67 (0.05) −0.23 (0.19) −0.00 (0.12) −0.70 (0.17)

  o

tb yo

1.68 (0.43)

σ ð xo Þ σ ð yo Þ

3.45 (0.16)

2.26 (0.21) 2.26 (0.41) 6.85 (1.13) 4.78 (0.97) 1.68 (0.43)

3.73 (0.36) 2.74 (0.21) 8.79 (0.82) 6.17 (1.78) 3.54 (0.15)

See notes in Tables 4 and 6.

Output, wedges, consumption, and counterfactual economies during the Korean crisis Output and wedges

Consumption and models with all wedges but one

25

105 Labor wedge

20

100

15

95

10

Ctry. spread wedge

90 5

Investment wedge

85

0 Consumption No efficiency wedge No labor wedge No investment wedge No ctry. spread wedge

80

−5 Efficiency wedge

75

−10 Output

70

−15 1998

1999

2000

1998

1999

2000

Fig. 5. The left panel reports output and the measured wedges. The right panel displays consumption and the prediction of models with all wedges but one.

is not trivial as the model with no efficiency wedges fails to match the countercyclical trade balance. Finally, the model without country spread wedges misses many of the business cycle facts in Korea, like the volatility of the trade balance and investment, and the correlation between the trade balance and output.14 Figs. 5 and 6 display the contribution of the wedges to the 1997 Korean crisis. The most notable distortion during the Korean crisis is a large and persistent increase of the labor wedge, of almost 25 percentage points at its peak. The country spread wedge also increased substantially, reaching almost 10 percentage points by the first quarter of 1998, gradually declining to its pre-crisis level by 2000.15 The efficiency wedge also shows an important drop, of about 8 percentage points at its trough. In contrast, the investment wedge showed an improvement, declining about 5 percentage points by 1998, although rapidly returning to its pre-crisis level. The predictions of the counterfactual economies in

14 The large standard error of the estimate does not allow us to say much about the contribution of the country spread wedge to the excess volatility of consumption over output in Korea. 15 The correlation between the implicit and actual interest rates in Korea is 0.56. Moreover, the implicit interest rate captures well the increase in spreads during the crisis. See the online Appendix.

Fig. 6 suggest that, as in the case of Mexico, the combined contribution of efficiency, labor, and country spread wedges account for most of the Korean crisis. Note that the improvement of the investment wedge contributed to the behavior of investment and the trade balance, but acting as a buffer rather than as an amplification mechanism. In sum, results for Korea are also consistent with the view that the combined contribution of efficiency, labor, and country spread wedges accounts for most fluctuations. The labor wedge in Korea, however, plays a more prominent role than in Mexico. This is also reflected in the dynamics of consumption during the crisis (Fig. 5): of all counterfactual experiments, the one with no labor wedges provides the worst match of consumption. 6. A structural interpretation of the wedges The wedges of the prototype model do not have a structural interpretation. Chari et al. (2002, 2005, 2007), however, prove that a large class of structural models with interpretable primitive shocks and frictions induce reduced form wedges similar to those in a prototype closed economy. The mapping from the class of models to the prototype economy, however, is not one to one. More than one structural model could induce variations in the same reduced form wedge, or a single structural

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

14

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

Korea: models with just one wedge Output

105

Hours 110

100

100

95 90

90

85 80 80 70

75

60

70

Investment

180

Trade Balance 20

160 140

10

120

0

100

−10

80 −20

60

−30

40 20

−40 1998 Data

1999

1998

Efficiency wedge

Labor wedge

1999

Investment wedge

Country spread wedge

Korea: models with all wedges but one Output

105

Hours 110

100 95

100

90

90

85 80 80 70

75 70

60

Investment

180

Trade Balance 20

160 140

10

120

0

100

−10

80 −20

60

−30

40 20

−40 1998 Data

1999 No efficiency wedge

1998 No labor wedge

1999

No investment wedge

No ctry. spread wedge

Fig. 6. Data and predictions of models during the Korean crisis.

shock could induce movements in more than one wedge. Thus, the BCA methodology does not identify primitive shocks and frictions. BCA, however, limits the set of models consistent with the data: a model is not consistent with the data if its primitive shocks and frictions induce wedges in the prototype economy that do not move as the estimated wedges and do not contribute to observed fluctuations. Below I describe a structural model with financial frictions that is consistent with the previous findings and, therefore, can be used to interpret the measured wedges.

6.1. A model with financial frictions This section shows that an economy with financial frictions in the form of collateral and working capital constraints is equivalent to a prototype economy with five fluctuating wedges.16 The model, a simplified version of Mendoza (2010), includes most shocks and frictions

16

To avoid clutter, the argument st is kept implicit throughout this section.

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

considered in the literature. The interaction of a collateral constraint with a working capital constraint drives fluctuations and introduces persistence into the five wedges of the prototype economy. Specifically, when working capital loans are collateralized, relatively small structural shocks manifest themselves as large and persistent fluctuations in efficiency, labor, and country spread wedges. In contrast, if working capital loans do not require collateral, a tightening in the collateral constraint does not induce fluctuations in efficiency or labor wedges. Thus, the model needs larger shocks or more stringent working capital requirements to generate fluctuations similar to those observed in emerging countries. Moreover, a binding collateral constraint manifests itself as a drop in the government consumption wedge, which reinforces the contractionary effects of the other wedges. Chari et al. (2005), however, reach the opposite conclusion: using a prototype closed economy, a tightening in the collateral constraint manifests itself as an increase in the government consumption wedge. As shown below, the reason for this difference is the type of prototype economy used to interpret the wedges: while the prototype closed economy predicts an increase in government consumption, the prototype open economy predicts a drop in that wedge. The model is a small open economy with a large number of households who operate their own technology but must rent labor from other households. Output is produced using capital kt − 1, labor elt , and an imported intermediate input zt by means of the production function BtF(kt − 1, elt , zt). The household chooses consumption ct, labor supply lt, investment xt, capital kt, debt holdings bt, inputs zt, and labor demandelt , t to maximize E0 ∑ ∞ t = 0β U(ct, lt) subject to a capital accumulation equation kt = G(kt − 1, xt), a budget constraint

15

 −α  1 α 1−α μ ι t ϕRt pt 3 α 3 Bt α1 þα2 kt−1 lt bt U lt lt U ct ct þ xt þ bt−1 − −

 1 −ϕðR1 −1Þ Rt U ct α 1 þα2 μt ι ϕRt þ ϕðRt −1Þ þ 1 U ct ek l ¼B t t−1 t α

2

1−α

;

ð18Þ

3

  1 −α3 α 1−α α þα pt α 3 Bt 1 2 kt−1 lt

6bt κkt U l7  − lt t 7 ¼ 0; μt6 4R − G þ ιϕRt μ 1 U ct 5 t t xt ι ϕRt þ ϕðRt −1Þ þ 1Þα1 þα2 U ct

ð19Þ

where μ ≥ 0 is the Lagrange multiplier on the constraint (14), α = α1/ (α1 + α2), and " # α3 1=α α 1 þα 2 α 3 Bt 3 e ¼ ðα þ α Þ : B t 1 2 pt ½ϕðRt −1Þ þ 1 þ ιϕRt μ t =U ct 

ð20Þ

6.1.1. Prototype economy In the prototype economy, households have the same preferences as in Mendoza's model, and face the budget constraint b

ct þ ð1 þ τxt Þxt þ bt−1 ¼ ð1−τlt Þwt lt þ vt kt−1 þ T t þ bt =Rt

  ct þ xt þ wtelt þ pt zt þ ϕðRt −1Þ wtelt þ pt zt þ bt−1   ¼ wt lt þ Bt F kt−1 ;elt ; zt þ bt =Rt ;

where the notation is the same as in Section 3, except that now, for simplicity, I assume that there is no productivity or population growth, ignore the sensitivity of the interest rate to the debt-to-output ratio, and denote the country spread wedge by 1/Rbt . The capital accumulation equation is kt = G(kt − 1, xt), and the feasibility constraint is

and a collateral constraint

ct þ xt þ gt þ bt−1 −bt =Rt ¼ At kt−1 lt

  bt =Rt þ ιϕRt wtelt þ pt zt ≤κqt kt :

b

ð1−α ÞBet kαt−1 l−α U lt t ¼ ; U ct ϕðRt −1Þ þ 1 þ ιϕRt μ t =U ct μt þ 1 Rt βEt U ctþ1 ; Rt βEt U ctþ1

ð15Þ

  μ U ct e kα−1 l1−α þ Gktþ1 ; ¼ βEt U ctþ1 α B 1−κ t tþ1 tþ1 t U ct Gxt Gxtþ1

:

ð21Þ

The equilibrium conditions of the prototype economy are summarized by the feasibility constraint, the capital accumulation equation, and the first order conditions U lt α 1−α ¼ ð1−α ÞAt kt−1 lt ð1−τlt Þ U ct b  U ct ¼ Rt βEt U ctþ1  G 1 þ τ xt α−1 1−α U ct ¼ βEt αAtþ1 kt ltþ1 þ 1 þ τxtþ1 ktþ1 : Gxt Gxtþ1 −

Proposition. Equivalence result Given an equilibrium allocation in Mendoza's model {ct, lt, xt, kt, bt, μt}, there are stochastic processes {At, τlt, τxt, Rbt , gt} such that {ct, lt, xt, kt, bt} is an equilibrium allocation of the prototype economy. et , where B et is given by Eq. (20); set labor taxes accordProof. Let At ¼ B ing to 1−τlt ¼

1 ; pt ½ϕðRt −1Þ þ 1 þ ιϕRt μ t =U ct 

ð22Þ

fix τx0 = 0, and let τxt + 1 be defined recursively by

 U ct ¼

1−α

ð14Þ

Here, wt is the wage rate, Pt is the price of the intermediate input, Rt an exogenous gross interest rate, and qt is the price of capital. The budget constraint includes   net interest payments on working capital loans, ϕðRt −1Þ wtelt þ pt zt , because the household is required to pay for a fraction ϕ of the total costs of labor and intermediate inputs before cashing their sales. These loans are taken at the beginning of the period at the gross interest rate Rt and repaid at the end of the period after production is realized. The collateral constraint (14) limits the value of debt to be smaller than a fraction κ of the value of capital, where ι ∈ {0, 1}. When ι = 1 (as in Mendoza, 2010) foreign lenders require that collateral must cover one period debt and gross working capital loans. When ι = 0 (as in Chari et al., 2005) only one period debt must be collateralized. α α Let F ðk; l; zÞ ¼ k 1 l 2 zα3 , where αi ≥ 0 for i = 1, 2, 3 and ∑ 3i = 1αi = 1. The online Appendix shows that the equilibrium of this model is an allocation {ct, lt, xt, kt, bt, μt} satisfying kt = G(kt − 1, xt) and −

α

ð16Þ

ð17Þ

1 þ τ xtþ1 " ! # e kα−1 l1−α þ G αB α−1 1−α Gxtþ1 tþ1 t ktþ1 =Gxtþ1 tþ1 e ð1 þ τ xt Þ−α Btþ1 kt ltþ1 ¼ ; 1−μ t κ=U ct Gktþ1

ð23Þ

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

16

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

define the country spread wedge as  b Rt ¼ Rt 1 þ

μt ; βRt Et U ctþ1

ð24Þ

and set the government consumption wedge according to  −α  1 α 1−α μ ι t ϕRt pt 3 α 3 Bt α1 þα2 kt−1 lt −btþ1 μ t =Rt U l U ct −

−ϕðRt −1Þ lt t ; gt ¼  Rt βEt U ctþ1 þ μ t U ct μt ι ϕRt þ ϕðRt −1Þ þ 1 U ct 1 α 1 þα 2

structural economy, bt/Rbt b bt/Rt. To induce the same allocation, the prototype economy requires less government expenditures. The second, more subtle, reason is only valid when working capital loans are collateralized and captures an asymmetry in the way the working capital constraint affect firms' costs and collateral requirements. Specifically, firms only pay net interest charges on the working capital loans, while collateral must cover the gross amount of these loans. The difference between gross and net amounts is reflected as a negative government consumption wedge in the prototype open economy. 7. Concluding remarks

ð25Þ where all equations are evaluated at the equilibrium allocation of Mendoza's economy. The proof follows by noting that the proposed allocation satisfies the equilibrium conditions of the prototype economy. Eqs. (20) and (22)–(25) summarize the mapping from the structural shocks and frictions in Mendoza's model into the wedges of the prototype economy. As already noted in Chari et al. (2005), Eqs. (20) and (22) show that the working capital constraint translates interest rate shocks into efficiency and labor wedges. The working capital constraint is also manifested as investment wedges through its impact on the effiet (Eq. (23)), and as government consumption wedges ciency wedge B (Eq. (25)).17 Suppose that working capital loans are collateralized (ι = 1) and there is a shock that makes the collateral constraint bind. The increase from μt − 1 to μt N 0 is reflected as a worsening in the efficiency wedge (due to the working capital constraint on foreign inputs (Eq. (20)) and in the labor wedge (due to the working capital constraint on labor, Eq. (22)). As asset prices (qt) decline, the collateral constraint is expected to bind for a number of periods, leading to a persistent drop in these wedges. In addition, the tightening in the collateral constraint also manifests itself as investment wedges (Eq. (23)), as country spread wedges (Eq. (24)), and as government consumption wedges (Eq. (25)). That is, when working capital loans are collateralized, a tightening in the collateral constraint leads to fluctuations in the five reduced form wedges. On the other hand, when working capital loans are not collateralized (ι = 0) a tightening in the collateral constraint does not affect labor or efficiency wedges. Because labor and efficiency wedges are the main drivers of output fluctuations in emerging countries (Section 5), the model will require large interest rate shocks or large values of ϕ to translate interest rate shocks into sufficiently volatile efficiency and labor wedges. Indeed, while Neumeyer and Perri (2005) and Uribe and Yue (2006) need ϕ ≥ 1 to match business cycle facts in emerging countries, Mendoza's model does so with ϕ = 0.26. Finally, note that a suddenly binding collateral constraint leads to a drop in the government consumption wedge (Eq. (25)). In contrast, Chari et al. (2005) show that a binding collateral constraint manifests itself as an increase in the government consumption wedge of a prototype closed economy. To understand the intuition for this result, consider how a binding collateral constraint is translated into the wedges of a prototype closed economy. A suddenly binding collateral constraint implies an abrupt increase in net exports. For a prototype closed economy to implement the same allocation as the structural open economy, net exports must be interpreted as government expenditures. Otherwise, feasibility in the closed economy would be violated. On the other hand, there are two forces that transform a binding collateral constraint into a negative government consumption wedge in the prototype open economy. First, the collateral constraint drives an implicit country spread wedge, Rbt N Rt; therefore, given the same allocation of bonds, the prototype economy raises less money than the 17 Inspecting Eq. (20) suggests that an increase in pt is manifested as a drop in the efficiency wedge. Kehoe and Ruhl (2008, 2009), however, show that a negative terms of trade shock need not produce a decline in measured productivity if one constructs GDP in the model using base year or chain-weighted prices.

Aggregate fluctuations in emerging economies are different from those in developed small open economies. This conventional wisdom is robust to computing the business cycle statistics using recent data, which do not include episodes of emerging market crises, and to splitting the sample between crisis and tranquil times. Even during tranquil times, the main differences between emerging and developed economies remain. To shed light on these differences, this paper has decomposed fluctuations in Mexico and Canada in terms of reduced form shocks that drive a wedge between marginal rates of substitution and marginal rates of transformation relative to a frictionless open economy. This decomposition was done by extending the Business Cycle Accounting methodology of Chari et al. (2007) to an open economy setting. There are three reduced form shocks that, by far, account for most aggregate fluctuations in Mexico: shocks to aggregate productivity (the efficiency wedge), shocks that affect labor markets by distorting the equality between the marginal product of labor and the marginal rate of substitution between consumption and leisure (the labor wedge), and shocks that distort intertemporal consumption choices (the country spread wedge). On the other hand, business cycles in Canada are mostly accounted for by fluctuations in efficiency and labor wedges, although country spread and investment wedges do play some role. Overall, results are robust to a number of sensitivity checks, including estimating the model with different preferences, increasing capital adjustment costs, and using Korean data. The main conclusion of the exercise is that, to understand emerging market fluctuations, one needs a model whose primitive shocks and frictions drive procyclical fluctuations in efficiency wedges and countercyclical fluctuations in labor and country spread wedges. Thus, these findings suggest that RBC models with just productivity shocks (either temporary or permanent) do not provide a successful benchmark to understand fluctuations in emerging economies. A model that is consistent with these findings is Mendoza (2010). In that model, financial frictions in the form of collateral constraints interact with working capital constraints transforming possibly small shocks into large and persistent fluctuations in efficiency, labor, and country spread wedges. One possible extension of this paper, along the lines of Aguiar and Gopinath (2007), is to allow for temporary and permanent shocks to the efficiency wedge. These authors argue that the main difference between emerging and developed countries is the persistence of the productivity shocks they face. It is straightforward to extend the prototype economy along these lines. This extension, however, could prove difficult to implement because the number of parameters to estimate increases substantially, making the maximization of the likelihood function considerably more difficult. Acknowledgments I thank two anonymous referees and the editor, Tim Kehoe, for helpful suggestions that substantially improved the paper and motivated the inclusion of Section 2. I also thank V.V. Chari, Andrés Fernandez, Aart Kray, Norman Loayza, Ellen McGrattan, Andy Neumeyer, Juan Pablo Nicolini, Fabrizio Perri, Demian Pouzo, Claudio Raddatz, Sergio Schmukler, Luis Servén, and Martín Uribe, for helpful comments.

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002

C. Hevia / Journal of International Economics xxx (2014) xxx–xxx

Part of this research was conducted while the author was affiliated with The World Bank. Appendix A. Maximum likelihood estimation

the annealing step. I use two local search algorithms in combination: a Quasi-Newton algorithm with inverse Hessian updating and a Nelder– Mead simplex algorithm. I initialized the estimation at different points to reduce the chances of finding a local optimum.

The model is rendered stationary by dividing all trending variables xt denote the normalized version of xt and x its dated t by (1 + γ)t. Let e steady state value. The model is log-linearized around the steady state and written in the state space form

Appendix B. Supplementary data

X tþ1 ¼ M ðθÞX t þ νtþ1

ðA:1Þ

References

Y t ¼ N ðθÞX t ;

ðA:2Þ

where Xt and Yt are the vectors of state and observable variables in terms of deviations from the steady state, h       i X t ¼ log e kt =k ; log e bt =b ; log At =A ; τ lt −τ l ; τxt −τ x ; zt −z; logðg t =g Þ h   e −tb; log ðg =g Þ; yt =yÞ; logðe xt =xÞ; log lt =l ; tb Y t ¼ logðe t t

the noise process is given by νt + 1 = [0, 0, εt + 1′]′; εt + 1 is the innovation of the process Eq. (11); and the matrices M(θ) and N(θ) are functions of the parameters θ ¼ fs; P; Σ; ϕg. e xt; e lt ; m ft , and gt using data I first construct the empirical analogs of yt; on output, investment, labor, net exports, and government consumption. To do this, I detrend the data using the average growth rate of real GDP and the population. (Hours worked are only detrended using the population growth rate.) Next, given a guess θ, I solve for the log-linearized policy functions M(θ) and N(θ), and construct a time series Yt using the transformed data and the steady state values. The likelihood function is then evaluated using the Kalman filter on the system (A.1–A.2). To induce stationarity, I follow CKM and add a penalty term of 5 × 105 max (λ − 0.995, 0)2 to the likelihood function, where λ is the eigenvalue of P with the largest absolute value. The log-likelihood, a non-linear function of 45 parameters, is difficult to optimize. I first use a simulated annealing algorithm (Kirkpatrick et al., 1983) to obtain a good approximation to the global maximum. Next I initialize a local search algorithm using the estimate from

17

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.jinteco.2014.05.002.

Aguiar, Mark, Gopinath, Gita, 2007. Emerging market business cycles: the cycle is the trend. J. Polit. Econ. 115, 69–102. Chang, Roberto, Fernández, Andrés, 2013. On the sources of aggregate fluctuations in emerging economies. Int. Econ. Rev. 54, 1265–1293. Chari, V.V., Kehoe, Patrick J., McGrattan, Ellen R., 2002. Accounting for the great depression. Am. Econ. Rev. 92 (2), 22–27 (May). Chari, V.V., Kehoe, Patrick J., McGrattan, Ellen R., 2005. Sudden stops and output drops. Am. Econ. Rev. 95 (2), 381–387 (May). Chari, V.V., Kehoe, Patrick J., McGrattan, Ellen R., 2007. Business cycle accounting. Econometrica 75 (3), 781–836 (05). Garcia-Cicco, Javier, Pancrazi, Roberto, Uribe, Martin, 2010. Real business cycles in emerging countries? Am. Econ. Rev. 100 (5), 2510–2531. Greenwood, Jeremy, Hercowitz, Zvi, Huffman, Gregory W., 1988. Investment, capacity utilization, and the real business cycle. Am. Econ. Rev. 78 (3), 402–417 (June). Kaminsky, Graciela L., Reinhart, Carmen M., Vegh, Carlos A., 2005. When it rains, it pours: procyclical capital flows and macroeconomic policies. NBER Macroeconomics Annual, pp. 11–82 (April). Kehoe, Timothy J., Ruhl, Kim, 2008. Are shocks to the terms of trade shocks to productivity? Rev. Econ. Dyn. 11 (4), 804–819. Kehoe, Timothy J., Ruhl, Kim, 2009. Sudden stops, sectoral reallocations, and the real exchange rate. J. Dev. Econ. 89 (2), 235–249 (July). Kirkpatrick, S., Gelatt, C.D., Vecchi, M.P., 1983. Optimization by simulated annealing. Science 220 (4598), 671–680 (May). Kydland, Finn E., Zarazaga, Carlos E.J.M., 2002. Argentina's lost decade. Rev. Econ. Dyn. 5 (1), 152–165 (January). Lama, Ruy, 2011. Accounting for output drops in Latin America. Rev. Econ. Dyn. 14 (2), 295–316. Mendoza, Enrique G., 2010. Sudden stops, financial crises, and leverage. Am. Econ. Rev. 100 (5), 1941–1966 (December). Neumeyer, Pablo A., Perri, Fabrizio, 2005. Business cycles in emerging economies: the role of interest rates. J. Monet. Econ. 52 (2), 345–380 (March). Reinhart, Carmen M., Rogoff, Kenneth S., Savastano, Miguel A., 2003. Debt intolerance. Brook. Pap. Econ. Act. 34 (2003–1), 1–74. Schmitt-Grohe, Stephanie, Uribe, Martin, 2003. Closing small open economy models. J. Int. Econ. 61 (1), 163–185. Uribe, Martin, Yue, Vivian Z., 2006. Country spreads and emerging countries: who drives whom? J. Int. Econ. 69 (1), 6–36 (June).

Please cite this article as: Hevia, C., Emerging market fluctuations: What makes the difference?, J. Int. Econ. (2014), http://dx.doi.org/10.1016/ j.jinteco.2014.05.002