Journal Issue
Share
Article

The End of An Era? the Medium- and Long-Term Effects of the Global Crisison Growth in Low-Income Countries1

Author(s):
Chris Papageorgiou, Andrew Berg, Catherine Pattillo, and Nicola Spatafora
Published Date:
September 2010
Share
  • ShareShare
Show Summary Details

I. Introduction

Low-income countries (LICs) as a group have enjoyed relatively rapid growth in recent years. Since 1995, for example, sub-Saharan Africa has grown faster than developed countries, after many years of poor average performance (IMF, 2008 and Collier et al, 2008). This growth has not been sufficient to put LICs on the path to meeting most of the Millennium Development Goals (MDGs), but it has reduced poverty and supported better health and education outcomes in many countries (IMF-World Bank, 2010).

Estimates for 2009 suggest that the global financial crisis had already substantially slowed growth in most developing countries, thrusting millions back into poverty and setting back efforts to achieve the MDGs. If, in addition, the crisis has longer-run implications, that is, if it knocks countries off their track of solid medium-long-term growth, it will be a much greater disaster. The question is especially pressing insofar as the growth resurgence since the mid-1990s has been associated with generally supportive external conditions: strong global growth, stable or rising commodity prices, and increasing inflows of external capital. Thus, in considering the implications and policy response to the current crisis, it is important to consider the risks to sustaining medium-term growth.

In principle, a temporary negative shock to external demand or the terms of trade in a standard neoclassical growth model would be followed by a quick reversion to the steady state level of income, implying a growth “bounce-back” and benign transitory effects. However, history is not very optimistic that LICs can uniformly escape global shocks without absorbing long-lasting damage to both growth and welfare.2 Over the past few decades, a LIC’s growth rate in one decade has generally been a poor predictor of its growth rate during the next decade, while many policies and country characteristics are more stable (Easterly et al, 1993). One influential view is that, as Easterly et al. put it: “shocks, especially shocks to the terms of trade, are an important determinant of growth over 10-year periods, and that they can help account for low [growth] persistence.”

An emerging empirical literature has also shown that growth down-breaks, or periods of a severe growth slowdown, are more common than previously thought and are crucial in understanding the medium- to long-run growth process in LICs.3 A related literature shows that countries that suffered spells of real income stagnation were more likely to be poor (see e.g., Reddy and Minoiu, 2009). Moreover, we know that crises can result in sharp declines in investment in education and health, declines that can potentially have long-lasting effects (Benhabib and Spiegel, 1994; Krueger and Lindah, 2001). Finally, there is an extensive theoretical literature that explores the possibility of growth nonlinearities that may result in LICs falling into prolonged periods of underdevelopment, commonly known as poverty traps. Nonlinearities in growth have been highly influential in shaping the thinking of both growth theorists and empiricists in recent years. The work on multiple-growth regimes and the world income distribution suggests that there may be growth factors strong enough to overcome the decreasing marginal productivity of the neoclassical production function, thereby producing persistent underdevelopment and income divergence across countries.4

Whether there will be a persistent negative growth effect of the crisis on LICs depends crucially on the nature of the shock, its transmission mechanism, and the policy response. For example, is the shock transmission mechanism the same as in the emerging markets and advanced economies? How does its dynamic path compare to previous global crises? The growth effect will also be a function of country-specific characteristics; that is, the ability of a country to absorb the shock quickly based on sound market fundamentals, favorable initial conditions, structural reforms and prudent procyclical policies.

Transmission mechanisms from the global crisis shock seem to vary considerably across countries. Advanced economies have primarily suffered a financial/banking crisis, and much ongoing research is devoted to understanding this type of shock. Most developing countries were primarily hit by an external demand (ED) effect, although some, notably fuel exporters, were also hit by a terms of trade (TOT) and perhaps to a lesser extent a foreign direct investment (FDI) effect.5 From a methodological point of view, this difference is quite important because these external shocks are more familiar to LICs than the financial shock is to advanced countries, therefore more credibly permitting a historical analysis of the effects in LICs.

This paper puts the current crisis in historical perspective and examines the prospects for medium- to long-run growth in LICs. Although the uncertainties are enormous, and the amount of light that recent history can shed is limited, conditional answers are possible. In the analysis that follows, we focus on ED, TOT and FDI as the three main transmission mechanisms of the crisis impacting LICs, although FDI data are a constraining factor.

The analysis will be based on a collection of exercises, each tackling the question at hand from a slightly different angle. The first is a simple event study in which we illustrate the growth paths of past crises and compare these to outcomes and projections for the current crisis. The second and third exercises focus on the medium-run effects of the crisis. Specifically, we employ an impulse-response method, followed by an analysis based on 5-year growth panel regressions. We view these two approaches as complementary. While regression analysis is the traditional gold standard in hypothesis testing especially using cross-country data, panel regressions may not fully capture the considerable variation in the data across time. Employing impulse response functions to examine the recovery from shocks makes full use of the within-country variation. Our last exercise is concerned with the longer-run implications of the crisis, using recently-developed methods to capture possible sharp structural down-breaks in growth rates.

II. Past and Current Global Shocks

The first exercise compares the growth experiences of LICs with those of the rest of the world economy, in past global crises. In particular, we consider three past global crises—1975, 1982 and 1991—and the current crisis, 2009. Current projections imply a more rapid recovery of growth in LICs than has been experienced in past global crises. Compared to past global crises, the current crisis is distinguished by the severity of the downturn (Berg et al., 2010) and the synchronization between LICs and global cyclical growth movement (Imbs, 2010).

In past global crises, LICs have tended to recover more slowly than the rest of the world (Figure 1, top panel). However, the current WEO forecasts imply a more rapid V-shaped recovery path out of the recession compared to previous crises. It could be that the different nature of the shocks faced by LICs in the current crisis is consistent with a more rapid recovery. A companion paper (Berg et al. 2010) looks more specifically at the plausibility of 2010-2011 forecasts, from the perspective of a relatively high frequency regression model. Here, we explore some of the medium- and long-term issues that frame the question.

Figure 1.GDP per Capita, TOT and ED Growth in Past and Current Crises

Note: The top panel plots the average per capita GDP growth in the world and in LICs while the bottom panel plots the TOT and ED growth in LICs 5 years before and 5 years after the global crises (centered at zero on the horizontal axis) of 1975, 1982 and 1991, and the current crisis. Also shown in dashed lines are WEO projections until 2013.

Next, we compare the TOT and ED growth effects of past global crisis (Figure 1, bottom panel). TOT growth is defined as growth of terms of trade for goods while ED growth is defined as trade partner real GDP growth weighted by exports to all partner countries.6 It is notable that, unlike in previous crises where TOT growth moved sharply downward relative to ED growth, in the current crisis it is ED that has resoundingly declined, while TOT growth continued at around the historical average rate. This transmission channel is also highlighted in IMF (2009a) and more formal growth regression analysis in Berg et al. (2010). The evidence below, however, is that like TOT shocks, ED shocks also have sharply negative short-to-medium-term output effects.7

III. Growth in the Medium-Term: An Impulse-Response Analysis

What happens to output over the medium term following TOT and ED shocks? Does the path of output per capita remain below its pre-crisis trend and if so for how long? Do growth rates recover to their pre-crisis levels as suggested by the neoclassical growth models or could shocks derail growth permanently? Until recently, the emphasis on the medium term growth impact of shocks has been limited, with the notable exceptions of Boyd, Kwak, and Smith (2005) and Cerra and Saxena (2008). With the current crisis, interest in the topic has surged. For instance, Chapter 4 of the Fall 2009 World Economic Outlook concentrated on medium-term developments following financial crises in advanced, emerging, and developing economies over the past 40 years.8

In our second exercise we employ impulse-response-function analysis, which involves using an autoregressive model of output growth rates augmented by crisis dummies, as in Cerra and Saxena (2008).9 Using panel data for a broad set of developed and developing countries, Cerra and Saxena documented that political and financial crises (i.e., wars, banking or currency crises) are not typically followed by high-growth recovery phases, either immediately following the trough, over several years of the subsequent expansion, or even over the complete expansion that follows a recession. Thus, when output drops following crises, it tends to remain well below its previous trend.

Using data from a panel of LICs we examine whether TOT and ED shocks have historically been associated with severe output losses and whether such output losses have been persistent. Specifically, we test the statistical relationship between growth and TOT and ED shocks by estimating a univariate autoregressive model in growth rates, which accounts for the nonstationarity of output and serial correlation in growth rates. The impulse response functions to each shock are shown with a one-standard-error band drawn from a thousand Monte Carlo simulations. The top and bottom panels in Figures 2a present impulse responses of output loss, measured as the percentage change from a linear growth trend, to a TOT shock and an ED shock, respectively.10

Figure 2a.Impulse Response of Output Loss in LICs to TOT and ED Shocks

Note: The top and bottom panels present impulse responses of output loss in LICs, measured as the percentage change from a linear growth trend, to a TOT shock and an ED shock, respectively. The solid line is the mean of output loss, and the dashed line reflects one standard deviation from the mean.

The key stylized facts that emerge from the analysis are as follows: The impact on output is negative and highly persistent under both types of shock. The medium-term output losses following ED shocks are particularly substantial. Output losses continue to rise without a sign of a reversal even 7 years after an ED shock, mounting to a cumulative loss of over 6 percent of GDP. As indicated by the dashed lines measuring the 90 percent confidence band, the average decline relative to trend is statistically significant. The output-loss path eventually becomes flat as growth tends to eventually return to the pre-crisis rate, but after a decade of lower growth and a substantial loss of output.

This may seem at first a surprising result given the neoclassical growth model’s prediction of rebound to the steady state. However, it is broadly consistent with similar impulse responses to different types of shocks (e.g., financial crises, Fall 2009 WEO; political crises, Cerra and Saxena, 2008). Although the mechanisms under which such output loss could persist after an ED shock merit careful consideration and future research, a reasonable hypothesis is the plausible interactions between ED shocks and private and public investment decisions or policy responses. For example an ED shock could result in a drastic shift in public investment from tradable goods to non-tradable goods or a policy change to reduce exposure to trade.

We replicate the impulse response analysis for Sub-Sahara African (SSA) countries given special interest in how the crisis may be affecting this region.11 As shown in Figure 2b the main results obtained from the LIC sample extend to SSA. One notable difference is that TOT shocks seem to have had a larger and more persistent effect than ED shocks in SSA, relative to the rest of LICs. This may be due to the fact that many countries in SSA are commodity and particularly fuel exporters and therefore more prone to TOT shocks.

Figure 2b.Impulse Response of Output Loss in SSA to TOT and ED Shocks

Note: The top and bottom panels present impulse responses of output loss in SSA, measured as the percentage change from a linear growth trend, to a TOT shock and an ED shock, respectively. The solid line is the mean of output loss, and the dashed line reflects one standard deviation from the mean.

It is important to note that a key assumption of the Cerra-Sexana VAR method is that countries will eventually return to the pre-crisis growth trends. To examine whether this is a reasonable assumption for our analysis we plot histograms for TOT and ED reporting average growth for five years following a crisis relative to the pre-crisis trend. Figure 3 confirms that mean reversion is a reasonable assumption as the median of LICs considered in these exercises tends to revert back to its pre-existing growth trend in the 5 years following the shock.12 There exists considerable variation around the median change in the five-year growth rate, with some countries topping their pre-crisis growth trends and others ending up with substantially lower growth.

Figure 3.Post-crisis Growth Relative to Pre-crisis Trend

Note: Histograms report five-year average growth ending in t+5 relative to pre-crisis trend, where crisis begins at period t.

This raises the question of whether, at least for some important subset of countries, the stationarity assumption for the growth rate is violated. We return to that question in the final section, when we look for breaks in growth.

IV. Growth in the Medium-Term: Panel Regression Analysis

While the impulse response analysis allowed us to examine the dynamic effects of shocks, we are also interested in the medium-term average effects of shocks in a framework that allows for an extended set of variables including, for example, policy interactions. Therefore, next we employ 5-year panel growth regressions as an alternative approach to investigating the impact of shocks on medium-term per capita GDP growth.13 In particular, our estimation results are based on panel GMM regressions in which the main explanatory variables are the three shocks: the change in ED, the change in TOT, and the change in the ratio of FDI to GDP. Other controls include the lagged output growth and a full set of country- and year-specific fixed effects. The full sample covers 87 non-advanced economies, including both LICs and MICs. The sample excludes fuel-exporters since these countries’ growth experience has been heavily influenced by external demand for fuel commodities.

Although omitted variable bias is a source of concern in most cross-country regression estimation, it is less so under our regression specification. This is because we are happy to assign to our shock variables any of their variation that is correlated with Barro/Solow-type variables (which we do implicitly by omitting the latter variables) on the grounds that our shock variables are plausibly exogenous to other omitted growth determinants, such as policies, institutions, investment rates, and so on, at least at the sorts of horizon of interest to us here.14

Results are presented for “All” non-advanced non-fuel countries, non-fuel LICs, and non-fuel non-LICs (Table 3). The comparison is intended to provide some insights regarding the differential effects of these shocks in LICs and middle-income countries.

For the nonfuel LICs subsample, the coefficient estimate on ED growth is positive and highly significant, indicating a positive impact on medium-term growth (Table 1, column 2). While the coefficient estimates on TOT and FDI for LICs using the entire time period in our sample are insignificant, they are highly significant for the entire sample and for non-LICs along with the coefficient estimates for ED (columns 1 and 3, respectively). Columns 4-9 present results from splitting the sample into the periods before and after 1989 (the median year in our sample). Coincidentally, growth increased dramatically for the period after 1989 in most LICs. It is interesting to note that most of the effect of TOT and ED growth for LICs has been driven by variation in the period after 1989 (see columns 5 and 8). Even more notable is that in the post-1989 sample the FDI coefficient becomes positive and significant.15 This may not be surprising given that FDI flows to LICs have become sizable only in the last decade or so.16

Table 1.Panel GMM Growth Regressions
Entire time periodBefore 1989After 1989
VariablesAllLICsNonLICsAllLICsNonLICsAllLICsNonLICs
Lagged Growth-0.245***

(0.061)
0.216***

(0.077)
-0.217**

(0.089)
-0.610***

(0.106)
-0.524***

(0.100)
-0.683***

(0.173)
-0.269***

(0.056)
-0.197**

(0.080)
-0.079

(0.053)
Growth in Terms of Trade0.103**

(0.044)
0.098

(0.060)
0.096***

(0.027)
0.030

(0.028)
0.028

(0.045)
0.044

(0.027)
0.136**

(0.059)
0.148**

(0.071)
0.187***

(0.037)
Growth in External Demand2.050***

(0.281)
1.843***

(0.471)
2.177***

(0.242)
0.667**

(0.316)
0.670

(0.598)
0.421

(0.295)
1.958***

(0.295)
1.340**

(0.525)
1.879***

(0.270)
Lagged Change in (FDI/GDP)0.668***

(0.178)
0.308

(0.216)
0.965***

(0.184)
0.451

(0.631)
-0.486

(0.745)
1.464***

(0.437)
0.725***

(0.186)
0.664*

(0.345)
0.531**

(0.227)
Observations5222812411789286344189155
Number of countries874839854738874839
Notes: Robust standard errors in parentheses. ***, **, and * denote statistical significance at 1, 5, and 10 percent levels, respectively. All specifications were estimated by panel data with year fixed effects, and 5 year averages over 1979-2009.
Notes: Robust standard errors in parentheses. ***, **, and * denote statistical significance at 1, 5, and 10 percent levels, respectively. All specifications were estimated by panel data with year fixed effects, and 5 year averages over 1979-2009.

Next, we ask the question, how does the projected average annual per capita growth obtained from using our coefficient estimates from the regression analysis presented in Table 1 compare with the actual WEO growth projections?17 Given that most of the effect of TOT and ED growth for LICs has been driven by variation in the post-1989 period, we use the coefficient estimates based on the period 1989-2009 (Table 1, column 8), to produce a projection for average growth for the period 2010-2014.18 We calculated our 5-year growth regression based projections for 2010-2014 to be 3.3 percent per year, which matches exactly the WEO’s projection for the same period. Our projected growth highlights that WEO’s medium-run projected recovery path out of the recession is par to the recent historical patterns, as suggested by our regression analysis.

We now investigate how macroeconomic policies may amplify or moderate the effects of shocks on growth. We first present simple illustrations of the bivariate relationship between a select sample of policy variables prior to the shock and GDP growth 5 years after the shock, giving the data a chance to speak with the minimum of auxiliary assumptions. This analysis is then extended to a multivariate regression analysis in which we interact the policy variables with the shock variables TOT, ED and FDI. Figures 4a and 4b plot pre-crisis government balance-to-GDP ratio, debt-to-GDP ratio, exchange-rate regime, and international reserves accumulation (measured in months of imports) against post-crisis growth, where the former variables are calculated as the average over the 5 years preceding the crisis and the later as the annual average GDP per capita growth over the 5 years following the crisis.

Figure 4a.Initial Conditions and Post-crisis Medium-term Growth

Note: Pre-crisis debt was calculated as the difference between t-5 and t, where t is the crisis year. Post-crisis growth is the annual GDP per capita growth over the 5 years after the crisis.

Figure 4b.Initial Conditions and Post-crisis Medium-term Growth

Note: Pre-crisis reserves were calculated as the difference between t-5 and t, where t is the crisis year. Post-crisis growth is the annual GDP per capita growth over the 5 years after the crisis.

The top set of plots (left ED, right TOT) in Figure 4a show evidence of a positive, albeit weak, correlation between government balance and medium-run post-crisis growth, implying that countries with higher government deficits prior to a TOT or ED shocks experience slower growth in the aftermath. The bottom set of plots from the top present a negative relationship between the change in the debt-to-GDP ratio over the five preceding the crisis year, and the subsequent 5-year annual growth rate. This suggests that countries that build more debt in the run-up to the crisis tend to have lower medium-term post-crisis growth. This finding is evident under both ED and TOT shocks. The top set of plots in Figure 4b shows the relationship between a measure of the flexibility of the exchange rate regime (“1” corresponds to fixed- and “5” corresponds to flexible-exchange-rate regime) and medium-term growth. In this case too, there is some evidence that a more flexible exchange regime is associated with higher medium-term growth after a TOT shock. A positive but weaker relationship also exists under ED shocks. Finally, the bottom set of plots illustrates that the level of international reserves prior to either TOT or ED shocks does not seem to have much of an effect on GDP growth in the medium run, at least in the bilateral relationship presented.

Some of these bivariate results—in themselves only suggestive—hold up in a multivariate regression context. In what follows we extend the analysis using multivariate growth regressions in which interaction terms between shocks and the four policy variables are included. We focus our discussion on interactions with the ED shock, seemingly the primary shock impacting LICs in the 2007-2009 global crisis. Table 2, column 2 presents the results of an interaction regression specification in which the government balance-to-GDP ratio is interacted with the three shock variables. The coefficient estimate capturing the direct effect of this variable on growth obtains the expected sign but it is not statistically significant. More importantly though, the interaction term with an ED shock is negative and significant implying that countries with lower deficits may be better equipped to dampen some of the effects of the shock on growth.

Table 2.Panel GMM Growth Regressions with Interactions
(1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)
VariablesAllLICsNonLICsAllLICsNonLICsAllLICsNonLICsAllLICsNonLICs
Lagged Growth-0.596***

(0.067)
0.284***

(0.077)
-0.092**

(0.037)
-0.112*

(0.059)
0.055

(0.070)
-0.062

(0.052)
-0.353***

(0.067)
-0.326***

(0.082)
-0.170***

(0.056)
-0.337***

(0.057)
-0.279***

(0.089)
-0.190***

(0.054)
Growth in Terms of Trade0.126***

(0.042)
-0.013

(0.063)
0.154***

(0.053)
0.129

(0.089)
0.028

(0.116)
0.234***

(0.060)
0.084

(0.077)
0.095

(0.086)
0.069

(0.084)
-0.029

(0.048)
-0.078

(0.053)
0.109

(0.106)
Growth in External Demand0.731*

(0.400)
-0.244

(0.267)
0.653*

(0.346)
0.928***

(0.325)
-0.584

(0.356)
0.350

(0.623)
2.065***

(0.369)
1.739***

(0.577)
2.374***

(0.292)
1.855***

(0.275)
1.295**

(0.507)
2.127***

(0.390)
Lagged Change in (FDI/GDP)-2.257

(1.390)
0.098

(0.336)
0.405

(0.405)
0.361

(0.348)
-1.219**

(0.521)
-0.398

(0.453)
-0.291

(0.529)
-0.373

(0.658)
1.608**

(0.756)
0.838

(0.521)
2.046*

(1.093)
0.382

(0.828)
Central government balance over GDP

(1 lag)
-0.263*

(0.141)
0.200

(0.152)
0.523***

(0.144)
Growth in Terms of Trade * lagged government balance over GDP0.005

(0.005)
-0.006

(0.011)
-0.001

(0.005)
Growth in external demand * lagged government balance over GDP0.077**

(0.031)
-0.065**

(0.032)
-0.129***

(0.040)
Lagged Change in (FDI/GDP) * lagged government balance over GDP-0.465**

(0.229)
-0.012

(0.045)
-0.017

(0.017)
Debt over GDP

(1 lag)
0.014

(0.014)
-0.000

(0.015)
-0.078

(0.057)
Growth in Terms of Trade * Debt over GDP (lag 1)-0.001*

(0.001)
-0.000

(0.001)
-0.001

(0.001)
Growth in external demand * Debt over GDP (lag 1)0.000

(0.004)
0.006*

(0.004)
0.016

(0.014)
Lagged Change in (FDI/GDP) * Debt over GDP (lag 1)-0.001

(0.003)
0.008*

(0.004)
0.010**

(0.004)
Exchange rate regime (fixed=1 floating =0)0.774

(1.791)
-1.429

(2.662)
1.919

(1.405)
Growth in Terms of Trade * Exchange rate regime0.060

(0.097)
-0.009

(0.130)
0.076

(0.099)
Growth in external demand * Exchange rate regime-0.067

(0.480)
0.286

(0.708)
-0.147

(0.372)
Lagged Change in (FDI/GDP) * Exchange rate regime1.365**

(0.689)
1.277

(1.151)
-0.827

(0.949)
Reserves5.768***

(1.676)
7.192***

(2.765)
2.428

(4.477)
Growth in Terms of Trade * Reserves0.118***

(0.045)
0.167

(0.128)
-0.029

(0.103)
Growth in external demand * Reserves-1.015**

(0.444)
-1.111

(0.766)
0.076

(1.197)
Lagged Change in (FDI/GDP) * Reserves0.206

(0.592)
-3.952

(2.919)
1.270

(1.013)
Constant-1.631

(1.320)
1.275

(1.425)
1.259

(1.195)
-3.202***

(1.150)
2.466*

(1.372)
3.134

(2.529)
-6.714***

(1.398)
-7.084***

(2.007)
-6.710***

(1.242)
-6.369***

(1.069)
-4.320**

(2.047)
-6.790***

(1.713)
Observations125161137275137138213108105212109103
Number of countries654337783939753837814437
Notes: Robust standard errors in parentheses. ***, **, and * denote statistical significance at 1, 5, and 10 percent levels, and 5 year averages over 1989-2009.
Notes: Robust standard errors in parentheses. ***, **, and * denote statistical significance at 1, 5, and 10 percent levels, and 5 year averages over 1989-2009.

Next we incorporate an interaction term of the shocks with the debt-to-GDP ratio in the regression specification and find that, consistent with theory, the interacted term’s coefficient estimate is positive and significant (Table 2, Column 5). This suggests that accumulation of large debt-to-GDP ratios could amplify the effects of ED shocks on growth. For the exchange rate regime variable, the direct effect on growth obtains a negative but insignificant coefficient estimate (Column 8). The coefficient estimate on the interaction with ED is positive (but not precisely estimated), consistent with the hypothesis that more flexible exchange rate regimes can buffer the effects of an ED shock on growth. Finally, there is a positive and large direct effect of international reserves on growth (Column 11). More importantly, the interaction coefficient estimate is negative (significant at the 14 percent level) which implies that countries with a higher stock of reserves are more capable of reducing the effects of an ED shock on growth. We also used the four interaction specifications to recalculate the 5-year growth projections for 2010-2014 obtaining values slightly lower but not statistically different from the WEO projections.

In summary, the regression results reinforce the impulse response findings showing economically significant effects of TOT and ED shocks on growth in the post-1989 period, consistent with most literature that emphasizes that LICs pushed for reforms to liberalize their economies starting in the early to mid-1990s. Regression results also show that FDI shocks played a significant role on growth. Annual average growth projections for 2010-2014 based on our regression coefficient estimates match closely the WEO projections.

Finally, we have investigated interaction effects between our three shocks and several policy variables. Both an illustrative bivariate and a multivariate regression analysis provide evidence that lower budget deficits, lower debt, more flexible exchange rate regimes, and higher international reserves help LICs to dampen the effects of an ED shock on medium-term growth.

V. Growth in the Longer-Run: Growth Down-Breaks Analysis

The second and third exercises above were designed to examine the response of growth over the medium term. Our last approach looks at longer-run possible implications of the crisis. Both previous approaches implicitly assume that the growth process itself is not affected by the shock. In other words, it is assumed that growth will return to its pre-crisis trend in the medium run. The previous analysis does not focus on what is perhaps the most telling source of variation in the underlying growth data for LICs. As Easterly et al. (1993) first pointed out, growth performance tends to be highly unstable, a “stop and go” process. Only a few countries have experienced consistently high growth rates over periods of several decades.

The more typical pattern is that countries experience phases of growth, stagnation, or decline of varying length (Pritchett 2000). As convincingly argued in Hausmann, Pritchett, and Rodrik (2005), “Standard growth theory, whether of the neoclassical or the endogenous variant, suggests that our best bet for uncovering the relation between growth and its fundamentals is to look for instances where trend growth experiences a clear shift.” However, while output paths in the advanced countries tend to be reasonably steady, in developing countries they are often characterized by “mountains, cliffs, and plains” (Pritchett, 2000).

Next we present evidence from graphical analysis on the association between TOT and ED shocks and growth down-breaks. Growth down-breaks, broadly defined as extended periods of markedly slow growth, are a striking feature of the development process. Recent work by Hausmann, Pritchett, and Rodrik (2005), Berg, Ostry, and Zettelmeyer (2008), and Jones and Olken (2008) use this new approach to understand the differential growth experiences of rich and poor countries.19 In this section we use a set of growth down-breaks—identified by the econometric methodology of Berg et al.20 for the period 1960-2009—to examine whether growth decelerations are associated with TOT and ED shocks.21 The exercises below examine possible associations between GDP growth decelerations (growth down-breaks) and TOT and ED shocks from two different angles.

First, we considered the number and conditional probability of growth down-breaks during the period leading up to, and following a major negative TOT or ED shock.22 The idea here is to see whether GDP growth down-breaks coincide or follow large TOT and ED shocks (Figures 5a and 5b). One pattern that emerges from is that persistent negative TOT shocks have often coincided with growth down-breaks.23 However, persistent negative ED shocks have shown no association with growth down-breaks. Why would persistent negative TOT shocks be more likely to have quasi-permanent negative effects on growth? One explanation is that countries that are hard-hit by TOT shocks find it difficult to adjust, even after a few years. The factors that have (usually) produced the commodities in question cannot easily switch to other uses, such as satisfying domestic demand or finding other export markets. Thus, the resulting decline in foreign income could squeeze imports and activity persistently, thus impeding productive activities throughout the economy. In contrast, adjustment to ED shocks may be less severe as resources could be directed more flexibly towards domestic demand. This remarkable observation suggests that if indeed the current crisis primarily has affected low-income countries through ED and not through TOT, there may be more reason for hope for a smoother recovery.24

Figure 5a.GDP Growth Decelerations and TOT Shocks

Figure 5b.GDP Growth Decelerations and ED Shocks

Note: The left panels plot the number of GDP growth down-breaks in a large sample of low-income countries (excluding transition economies) during the periods leading up to, and following, a large persistent terms of trade shock (year t+0 on the horizontal axis). The right panels convert the number of breaks to a conditional probability given the total number of down-breaks identified by our methodology. A large persistent TOT shock is defined as the worst 10 percent of the distribution of all TOT shocks, measured as the difference of the average 3 year TOT growth before and after period t. The right panel is the same, except that the shock is to external demand, measured as partner country real growth weighted by export shares.

Second, the top and bottom panels of Figures 6 plot the behavior of TOT and ED in the period leading up to, and following growth downturns (year 0 on the horizontal axis). This exercise differs from the previous one in that the main focus is on episodes of down-breaks rather than episodes of large TOT or ED shocks. The idea here is to see whether there is any association between sharp decreases in GDP growth and growth of TOT and ED. Once again the pattern that emerges is that TOT growth tends to decrease sharply in the run-up to growth decelerations, providing suggestive evidence that sharp TOT growth declines may lead to a sustained period of slow growth. On the contrary, ED growth shows virtually no co-movement with a growth downturn. This observation is consistent with the previous exercise.

Figure 6.TOT vs. ED around Periods of Growth Decelerations in LICs

Note: The top and bottom panels plot the behavior of TOT and ED, respectively, in the period leading up to, and following, growth downturns (year 0 on the horizontal axis) in LICs.

The results from the two exercises relating growth down-breaks with shocks were confirmed when the sample is restricted to SSA countries (Figure 7). One additional observation from this exercise is that the growth down-breaks experienced by SSA countries are sharper and seem to occur after a period of steady rise in growth.

Figure 7.TOT vs. ED around Periods of Growth Decelerations in SSA

Note: The top and bottom panels plot the behavior of TOT and ED, respectively, in the period leading up to, and following, growth downturns (year 0 on the horizontal axis) in SSA.

VI. Conclusions

Although the short-run growth prospects for emerging and developing countries appear to be positive (see Berg et al. 2010), there is still a question on the extent to which the current shock has longer-run implications that may knock LICs off their track record of solid medium-term growth. History is not encouraging that LICs can uniformly escape global shocks without absorbing long-lasting damage to growth and welfare. In past crises, it has often taken several years for LICs to bring growth rates back into positive territory. Could this be pattern be different in the 2009 global crisis? This paper has attempted to answer this question by several econometric methodologies to analyze historical data in a panel of countries.

On the positive side, based on the history of growth decelerations, our results suggest that ED shocks are not associated with sharp declines in output growth. Given existing evidence that LICs were primarily impacted by such a shock, the exercise assigns a low probability that many LICs will suffer from persistently low growth due to the crisis. However, our impulse response analysis shows that there seem to be quite substantial and highly persistent output losses associated with TOT and ED shocks in the medium-run.

Panel growth regressions re-enforced the impulse response findings that show economically significant effects of TOT, ED and FDI shocks in the medium-run. Finally, by using simple illustrations and by extending the regression analysis to include interaction terms we investigated how macroeconomic policies may amplify or moderate the effects of the three shocks on growth. It was shown that countries with lower budget deficits, lower debt, more flexible exchange rate regimes, and higher international reserves are more likely to dampen the effects of an ED shock on growth.

These conclusions are too broad-brush to do more than inform country-specific policy recommendations. These empirical exercises are only rough guides to current circumstances. First, there has always been substantial heterogeneity in the response of countries’ growth to large negative terms of trade and external demand shocks, and it is important to avoid overemphasizing average reactions to average shocks. Second, the current shock is different in many ways. Clearly there is a need for vigilance and prudent policy to prevent a protracted slowdown in some countries. Nonetheless, this paper suggests some cause for optimism about recovery of LICs to their pre-crisis growth rates.

References

    AntoshinS.A.Berg and M.Souto2008Testing for Structural Breaks in Small SamplesIMF Working Paper 08/75 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    AzariadisC. and J.Stachurski2007Poverty Traps” in Handbook of Economic GrowthVol. 1Chapter 5ed. by P.Aghion and S. N.Durlauf(The Netherlands: Elsevier Science).

    • Search Google Scholar
    • Export Citation

    BowlesS.S. N.Durlauf and K. R.Hoff2006Poverty TrapsRussell Sage Foundation.

    BaiJ. and P.Perron1998Estimating and Testing Linear Models with Multiple Structural ChangesEconometrica66 pp. 4778.

    • Search Google Scholar
    • Export Citation

    Ben-DavidD. and D. H.Papell1998Slowdowns and Meltdowns: Postwar Growth Evidence from 74 CountriesReview of Economics and Statistics80 pp. 561571.

    • Search Google Scholar
    • Export Citation

    BenhabibJ. and M.Spiegel1994The Role of Human Capital in Economic Development: Evidence from Cross-Country DataJournal of Monetary Economics34 pp. 143173.

    • Search Google Scholar
    • Export Citation

    BergA.C.PapageorgiouC.PattilloM.SchindlerN.Spatafora and H.Weisfeld2010Global Shocks and their Impact on Low-Income Countries: Lessons from the Global Financial Crisisdraft working paper (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    BergA.D.J.Ostry and J.Zettelmeyer2008What Makes Growth SustainedIMF Working Paper 08/59 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    BoydJ. H.S.Kwak and B.Smith2005The Real Output Losses Associated with Modern Banking CrisesJournal of Money Credit and Banking37 pp. 977999.

    • Search Google Scholar
    • Export Citation

    CerraV. and S. C.Saxena2008Growth Dynamics: The Myth of Economic RecoveryAmerican Economic Review98 pp. 439457.

    CollierP.B.Goderis and A. E.Hoeffler2006Shocks and Growth: Adaptation, Precaution and Compensation” (unpublished; Oxford University).

    • Search Google Scholar
    • Export Citation

    CollierP.B.Goderis and A. E.Hoeffler2009Structural Policies for Shock-Prone Developing CountriesOxford Economic Papers61 pp. 703726.

    • Search Google Scholar
    • Export Citation

    CollierP.B.Goderis and A. E.Hoeffler2010Does Aid Mitigate Shocks?Review of Development Economicsforthcoming.

    CollierP.J. W.GunningB. J.Ndulu and S. A.O’Connell2008Harnessing Growth Opportunities: How Africa Can Advance” in The Political Economy of Economic Growth in Africa 1960-2000ed. by B. J.NduluS. A.O’ConnellR. H.BatesP.Collier and C. C.Soludo(Cambridge, Massachussets: Cambridge University Press).

    • Search Google Scholar
    • Export Citation

    Dabla-NorrisE.J.HondaA.Lahreche and G.Verdier2010FDI Flows to Low-Income Countries: Global Drivers and Growth ImplicationsIMF Working Paper 10/132 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    DrummondP. and G.Ramirez2009Spillovers from the Rest of the World into Sub-Saharan African CountriesIMF Working Paper 09/155 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    EasterlyW.M.KremerL.Pritchett and L. H.Summers1993Good Policy or Good Luck? Country Growth Performance and Temporary ShocksJournal of Monetary Economics32 pp. 459483.

    • Search Google Scholar
    • Export Citation

    FurceriD. and A.Mourougane2009The Effect of Financial Crises on Potential Output: New Empirical Evidence from OECD CountriesOECD Economics Department Working Paper No. 699 (Paris: Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation

    GuptaS.C.Pattillo and K.Carey2006Sustaining and Accelerating Pro-Poor Growth in Africa (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    HausmannR.F.Rodríguez and R.Wagner2006Growth CollapsesCID Working Paper No. 136.

    HausmannR.L.Pritchett and D.Rodrik2005Growth AccelerationsJournal of Economic Growth10 pp. 303329.

    JonesB. F. and B. A.Olken2008The Anatomy of Start-Stop GrowthReview of Economics and Statistics90 pp. 582587.

    ImbsJ.2010The First Global Recession in DecadesIMF Economic Reviewforthcoming (Washington: International Monetary Fund).

    IMF2008Regional Economic Outlook: Sub-Saharan AfricaSpring2008 (Washington).

    IMF2009aRegional Economic Outlook: Sub-Saharan AfricaFall2009 (Washington).

    IMF2009bThe Impact of the Financial Crisis on Low-Income Countries—An Update (Washington).

    IMF-World Bank2010The MDGs after the CrisisGlobal Monitoring Report (Washington).

    KehoeT. and D.Levine1993Debt-Constrained Asset MarketsReview of Economic Studies60 pp. 865888.

    KruegerA. B. and M.Lindahl2001Education for Growth: Why and for Whom?Journal of Economic Literature39 pp. 11011136.

    Pisani-FerryJ. and B.van Pottelsberghe2009Handle with Care! Post-Crisis Growth in the EUBruegel Policy Brief No. 2009/02.

    PritchettL.2000Understanding Patterns of Economic Growth: Searching for Hills among Plateaus, Mountains, and PlainsWorld Bank Economic Review14 pp. 221225.

    • Search Google Scholar
    • Export Citation

    ReddyS. G. and C.Minoiu2009Real Income Stagnation of Countries 1960-2001Journal of Development Studies45 pp. 123.

    RodrikD.1999Where Did All the Growth Go? External Shocks, Social Conflict, and Growth CollapsesJournal of Economic Growth4 pp. 385412.

    • Search Google Scholar
    • Export Citation

    SoutoM.S.Antoshin and A.Berg2008Testing for Structural Breaks in Small SamplesIMF Working Paper 08/75 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
Appendix I. List of Countries and Subsamples used in the Quantitative Analysis
LICsNon LICs
AfghanistanAlgeria *
AlbaniaArgentina
Angola *Belarus
Azerbaijan *Bosnia and Herzegovina
BangladeshBotswana
BeninBrazil
BoliviaBulgaria
Burkina FasoChile
BurundiChina
CambodiaColombia
CameroonCosta Rica
Central African RepublicCroatia
Chad *Dominican Republic
Congo, Dem. Rep. ofEcuador *
Congo, Rep. *Egypt, Arab Rep.
Côte d’lvoireEl Salvador
EritreaEstonia
EthiopiaGabon *
Gambia, TheGuatemala
GeorgiaHungary
GhanaIndonesia
GuineaIran, Islamic Rep. *
Guinea-BissauJamaica
HaitiJordan
HondurasKazakhstan *
IndiaKuwait *
KenyaLatvia
Kyrgyz RepublicLebanon
Lao PDRLibya *
LesothoLithuania
MadagascarMacedonia
MalawiMalaysia
MaliMauritius
MauritaniaMexico
MoldovaMorocco
MongoliaNamibia
MozambiqueOman *
MyanmarPanama
NicaraguaParaguay
NigerPeru
Nigeria *Philippines
PakistanPoland
Papua New GuineaRomania
RwandaRussian Federation *
SenegalSaudi Arabia *
Sierra LeoneSerbia
Sri LankaSouth Africa
Sudan *Swaziland
TajikistanSyrian Arab Republic
TanzaniaThailand
TogoTrinidad and Tobago *
UgandaTunisia
UzbekistanTurkey
VietnamTurkmenistan *
ZambiaUkraine
United Arab Emirates *
Uruguay
Venezuela *
Note: All countries in sample are classified as LICs and non LICs. Non-fuel countries are marked with stars. The dataset used in this paper is available by the authors upon request. For a data appendix see Berg et al. (2010).
Note: All countries in sample are classified as LICs and non LICs. Non-fuel countries are marked with stars. The dataset used in this paper is available by the authors upon request. For a data appendix see Berg et al. (2010).
Appendix II. List of Down-breaks Episodes in our Sample
countryyearcountryyearcountryyear
Albania1988Estonia1990Namibia1981
Angola1975Estonia2005Nicaragua1978
Armenia1971Gabon1977Nigeria1972
Armenia1990Georgia1990Oman1977
Armenia2005Guatemala1981Oman1986
Azerbaijan1974Haiti1982Papua New Guinea1975
Azerbaijan1990Haiti1999Paraguay1982
Bangladesh1973Honduras1969Peru1988
Bangladesh1979Hungary1979Philippines1957
Belarus1961Iran, Islamic Rep.1977Philippines1981
Belarus1978Jamaica1973Poland1980
Belarus1991Jordan1968Romania1989
Bolivia1978Jordan1983Russian Federation1961
Botswana1990Kazakhstan1974Russian Federation1975
Brazil1981Kazakhstan1990Russian Federation1991
Bulgaria1982Kuwait1996Saudi Arabia1982
Bulgaria1990Kyrgyz Republic1971Sierra Leone1975
Burundi1972Latvia1974Sierra Leone1995
Cameroon1971Latvia1989South Africa1982
Cameroon1987Latvia2005Tajikistan1990
Chile1972Libya1969Thailand1997
Congo, Dem. Rep. of1975Lithuania1974Togo1970
Congo, Dem. Rep. of1990Lithuania1991Trinidad and Tobago1983
Costa Rica1979Lithuania2005Tunisia1973
Croatia1980Macedonia, FYR1980Turkmenistan1974
Croatia1990Macedonia, FYR1990Turkmenistan1991
Côte d’lvoire1979Mauritania1972Turkmenistan2005
Ecuador1974Mexico1982Uganda1970
Egypt, Arab Rep.1979Moldova1974Ukraine1974
El Salvador1979Moldova1991Ukraine1991
Eritrea1995Mongolia1990Uzbekistan1974
Eritrea2000Mozambique1975Uzbekistan1991
Eritrea2001Mozambique1982Zambia1970
Estonia1974Myanmar1986Zimbabwe1999
Note: Growth breaks are produced using the methodology developed in Souto, Antoshin, and Berg (2008) and subsequently used in Berg, Ostry, and Zettelmeyer (2008).
Note: Growth breaks are produced using the methodology developed in Souto, Antoshin, and Berg (2008) and subsequently used in Berg, Ostry, and Zettelmeyer (2008).
1We thank David Romer for detailed comments and Philippe Aghion, Hugh Bredenkamp, Paul Collier, Era Dabla-Norris, Ricardo Hausmann, Daniel Leigh, Ayhan Kose, Saul Lizondo, Jonathan Ostry, Martin Schindler, Antonio Spilimbergo, Steven Tokarick, and Hans Weisfeld and participants of the “Managing Volatility and Increasing Resilience in LICs” conference, Washington DC, April 2010, for discussion and suggestions. Freddy Cama, Lisa Kolovich, and Manzoor Gill provided excellent research assistance.
2Notable contributions on the effects of shocks in LICs include Collier, Goderis, and Hoeffler (2006), and Collier and Goderis (2009, 2010).
4See the edited book by Bowles, Durlauf, and Hoff (2006), and the Handbook of Economic Growth Chapter by Azariadis and Stachurski (2007) for literature review, and more specifically the debt trap model in Kehoe and Levine (1993).
5Although changes in remittances could be another possible transmission mechanism, their effect during the global economic crisis were quite mixed; while remittances decrease in some countries, they increased in some others (a notable example is Pakistan where the increase was substantial). Moreover, their likely endogeneity to recipient-country events is difficult to handle.
6An alternative definition weighing for trade shares was considered but not used as it drastically reduced our sample size, rendering most exercises imprecisely estimated.
7FDI data were not sufficient to produce a similar plot.
8IMF (2009b), Furceri and Mourougane (2009) and Pisani-Ferry and van Pottelsberghe (2009) also look at the medium-term output effects of banking crises for different subsamples or case studies.
9Daniel Leigh very helpfully provided his Stata code and invaluable input.
10The shock dummy variables for TOT and ED are defined as the crisis periods that belong to the left tail of the TOT and ED growth distributions, respectively. The left tail is based on a one standard deviation of the sample that excludes extreme values (below the 1st and above the 99th percentiles), to mitigate the effects from measurement error. Results are qualitatively similar to two alternative shock definitions considered.
11The impulse response analysis was also performed using several other subsamples, including all non-advanced countries, non-LICs, commodity- and non-commodity exporters.
12Of course, the maintained assumption that growth eventually returns does not drive how long it will take, or whether it bounces back so that it is temporarily above trend. The above results speak to those questions.
13A similar estimation methodology was followed by Drummond and Ramirez (2009).
14Our approach makes much more sense for low-income countries than for others. One reason is that, for advanced countries, we might worry that common shocks (such as global shocks to productivity growth) could produce co-movements in output not driven by trade linkages. Clearly, this is a risk in our context as well. However, given that trade between low-income countries is relatively unimportant, the common shocks that would be problematic would be those that jointly affect rich and poor countries but are not mediated through output, the terms of trade, or FDI flows (which we also control for). These shocks would seem to be unlikely.
15This result is consistent with Dabla-Norris et al. (2010) who focus on the effects of FDI on LICs.
16We have checked the robustness of these results to alternative specifications and subsamples. Using trade weighted TOT and ED shocks reduced our sample by more than 30 percent rendering most coefficient estimates imprecisely estimated. Furthermore, to consider concerns regarding the unreliability of FDI data in LICs due to measurement error, we examined and confirmed robustness of our TOT and ED results by dropping the FDI shock from the baseline regressions.
17Given our five-year-panel structure, we cannot readily ask whether the crisis of 2008/2009 would have been well forecast by our regression. This question is better addressed in a shorter-horizon framework, as in Berg et al. (2010).
18Predicted growth is calculated as the sum of the contributions of the four regressors (lagged growth, external demand, terms of trade, and lagged difference in FDI/GDP). To calculate the contribution for each variable, we multiply the regression coefficient by the WEO predicted average growth of each of the four regressors.
19Two early precursors of the current work on growth spells are Ben-David and Papell (1998), and Pritchett (2000), both of which employed novel econometric methods to identify shifts in growth performance.
20Berg et al. (2008) in turn follow Antoshin, Berg, and Souto (2008) who identify “growth spells” by modifying the procedure pioneered by Bai and Perron (1998) to determine sample-specific critical values, as is appropriate when the time dimension is 30 years or less.
21In Appendix II we list all the episodes of down-breaks in economic growth since 1960 that we are able to detect by using the aforementioned statistical methodology.
22In our baseline we consider a large permanent shock defined as the 10 percent of the left tail of the distribution of all permanent shocks measured as the difference of the average 3 year TOT and ED growth before and after period t. We have also considered a large transitory shock – the 10 percent of the left tail of the distribution of the difference in annual TOT and ED growth. In addition we considered shocks that resemble the intensity of the current TOT and ED shocks (the difference of the growth in TOT and ED between 2007 and 2009). For ED, given that the current shock is the largest that most LICs have seen in current history and therefore we could not identify more than a handful of such shocks in our sample, we considered lower intensity shocks by taking fractions of the current shock. For TOT we consider the intensity of the current shock which was quite mild in historical terms. Results using the last two alternative definitions of the shock are consistent with our baseline plots and are available upon request.
23This is consistent with the results of Berg et al. (2008) on terms of trade shocks and the ends of growth spells.
24The definitions of “persistent” and “large” can be found in the notes to the table. It turns out that large negative external demand shocks such as those experienced by many countries in 2009 are not unprecedented for many LICs. In the sample used for figure 5, there were 68 instances in which countries faced external demand shocks larger than they faced in 2009 (assuming WEO projections for the out-years).

Other Resources Citing This Publication