WPS7386


 Policy Research Working Paper                        7386




Should Latin America Save More to Grow Faster?
                                Augusto de la Torre
                                    Alain Ize




 Latin America and the Caribbean Region
 Office of the Chief Economist
 August 2015
Policy Research Working Paper 7386


  Abstract
 A widely shared view holds that there is no policy-exploit-                         competitive real exchange rate; and an endogenous saving
 able causal connection from saving to growth because                                channel, whereby saving follows growth but in a less than
 domestic saving is fully endogenous, optimally determined,                          perfectly elastic manner, thereby amplifying the effects of
 or substitutable by foreign saving. Yet, abandoning these                           the first two channels. Broad-based econometric evidence
 assumptions, which are questionable in the real world of                            supports all three channels and suggests that Latin America,
 frictions, leads to three channels through which domestic                           a historically low growth-low saving region, would ben-
 saving may promote growth: a real interest rate channel,                            efit from boosting its saving rate, especially in countries
 whereby saving reduces the cost of capital (the sovereign                           with recurrently weak balance of payments and persistent
 risk premium) and enhances macro sustainability; a real                             domestic demand pressures on the non-tradable sector.
 exchange rate channel, through which saving leads to a more




  This paper is a product of the Office of the Chief Economist, Latin America and the Caribbean Region. It is part of a larger
  effort by the World Bank to provide open access to its research and make a contribution to development policy discussions
  around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors
  may be contacted at adelatorre@worldbank.org and aize@worldbank.org.




          The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
          issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
          names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
          of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
          its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                        Produced by the Research Support Team
                       Should Latin America Save More to Grow Faster?


                                  Augusto de la Torre and Alain Ize 




JEL classification codes: O1, O4, E2

Keywords: saving, current account deficits, growth, real exchange rate determination,
exchange rate appreciation or depreciation, sovereign risk premium, Latin America





  Both authors work for the World Bank. Augusto de la Torre (Adelatorre@worldbank.org) is Chief Economist
for the Latin America and The Caribbean Vice-Presidency. Alain Ize (Alain.Ize@gmail.com) is a senior
consultant. This paper benefitted from the advice of Laura Chioda and Ha Nguyen on econometric matters; the
comments of Eduardo Cavallo, Eduardo Fernandez Arias, Anton Korinek and Aart Kray; discussions with Cesar
Calderon and Luis Serven; and the research assistance of Nicolas Kohn, Magali Pinat, Lucas Rusconi, Martin
Sasson, Tanya Taveras, and Matias Vieyra. The views in this paper are entirely those of the authors and do not
necessarily represent the views of the World Bank, its executive directors, and the countries they represent.
1.     Introduction

         The debate as to whether saving matters for growth, or is just a corollary of growth, is
an old and familiar one. Much of the original empirical debate has centered on causality, found
to be flowing mostly from growth to saving, rather than the other way around (Carroll and Weil,
1993). Yet, a more recent strand of literature has detected robust evidence indicating that
countries that rely on external saving, rather than domestic saving, grow less (Aizenman, Pinto
and Radziwill, 2004; Prasad, Rajan and Subramanian, 2007). Conversely, foreign saving has
been found to flow more to countries with lower growth prospects, as per the Gourinchas and
Jeanne (2013) puzzle. Both observations suggest that domestic saving matters for growth.

         But does it? Logic dictates that a policy-exploitable causal connection between saving
and growth should not exist in an idealized world where at least one of the following three
conditions holds: (i) saving is perfectly growth-elastic, hence does not constrain growth (the
perfect elasticity condition); (ii) foreign saving (a current account deficit) offsets any shortfall
in domestic saving, leaving growth unaffected (the perfect substitutability condition); or (iii)
there is no scope for policy because private saving is constrained efficient (the perfect efficiency
condition). Yet, in the real world all three conditions are questionable. First, the perfect
efficiency condition may fail to obtain on account of un-internalized externalities. Second, even
when saving follows growth (a well-established fact), it may not do so in a perfectly elastic
manner, such that the growth-induced saving response to an autonomous increase in investment
fully finances it. Third, even if foreign saving emerges swiftly to fill up a domestic saving gap,
it may not be a perfect substitute on account of frictions.

        The failure to meet the above conditions opens up the space for three channels through
which saving can causally affect growth. The first channel, the real interest rate (IR) channel,
relates to the cost of capital. Once sovereign default risk is introduced, decentralized agents
can under-save (relative to what would be socially optimal) because they do not take into
account the adverse impact of their individual borrowing on the collective cost of foreign debt
or the negative externalities in the event of a balance of payments crisis. Hence, a low saving
rate can affect growth by undermining macroeconomic sustainability. The second channel, the
real exchange rate (ER) channel, relates to external competitiveness. By inducing a persistent
real exchange rate appreciation, a switch in the composition of saving, from domestic to foreign,
can slow growth by shrinking the relative size of the tradable sector. Decentralized economic
agents can under-save in this case because they fail to internalize the technological spillovers
and other positive learning externalities associated with a thriving tradable sector. The third
channel, the endogenous saving (ES) channel, obtains when saving follows growth but in a less
than perfectly elastic manner. In this case, which we will refer to as the “normal” or “not on
steroids” ES, the ES channel amplifies the impact on growth that a change in saving would
have via any of the other two channels.

        These channels have distinct macroeconomic signatures, thereby leading to several
testable patterns. The IR channel dominates when the low saving rate leads to current account
deficits that threaten balance of payments viability, thereby raising the sovereign risk premium.
In this case, countries that save less should have a higher risk premium and grow less, despite
having a more competitive real exchange rate. Instead, the ER channel dominates when a low
saving rate pushes domestic demand above national income, putting pressure on the supply of
non-tradables and widening the current account deficit but without affecting the sovereign risk
premium. In this case countries that save less should have a less competitive real exchange rate
and grow less. Finally, the ES channel dominates when saving turns sufficiently growth-elastic


                                                 2
(the ES on steroids case). In this case, the ER and IR channels cease to operate because growth
becomes self-propelled and saving is no longer a constraint.

        The channels can interact and operate in sequence, becoming dominant at different
points in time. Thus, a period of IR channel dominance may be followed by improvements in
macro-financial policy that set the conditions for a return of the ER channel. In turn, a
prolonged period of current account deficits and uncompetitive real exchanges under the ER
channel can bring back IR dominance.

        We test for such patterns empirically based on a simultaneous equations model that
focuses on medium-term equilibrium relationships and uses an extensive set of controls to
correct for structural as well as policy diversity across countries. The model is driven by a set
of elasticities linking domestic saving, investment, the real exchange rate, the sovereign risk
rating, and growth. Using data for 119 countries over the 1980-2012 period, these elasticities
are estimated through both structural-form ordinary least squares (OLS) and reduced-form
instrument-based OLS regressions. A benchmarking methodology is used to explore the macro
patterns associated with the channels, ensuring that countries can be properly compared and
that deviations from benchmark can be interpreted as being largely driven by differences in
policy or policy-driven institutions.

        The empirical results broadly support the IR and ER macroeconomic patterns. In
particular, under-saving correlates robustly with real exchange rate over-valuation (all relative
to benchmark), which suggests a statistical dominance of the ER channel for the entire sample.
On average, countries that under-save have over-valued real exchange rates, rely less on
external demand (are less export oriented), grow slower, and invest less. However, the
countries that deviate the most from the ER pattern tend to fit the IR pattern, which associates
under-saving with low growth despite undervalued real exchange rates. As expected, the under-
savers and over-savers in this latter group differ considerably as regard their country risk
ratings, with the under-savers having the worst ratings, the over-savers the best. We also find
support for a normal (hence not dominant) ES channel. The latter mainly accrues from
corporate saving, by far the most growth-elastic component of domestic saving.

       While structural OLS estimates yield limited impacts of saving on growth, the reduced-
form instrumented estimates, when available (only some elasticities could be retrieved due to
the high endogeneity in the system), point toward substantially higher impacts. Moreover, the
estimated strength of these effects is much greater for countries with current account deficits
in the middle-income stages of development. Hence, domestic saving mobilization policies
make more sense for middle-income countries with weak balance of payments.

         Applying the analysis to Latin America, we find that the region has been a clear outlier
relative to the world average in that the IR channel played a uniquely important role during the
crisis times of the 1980s and 1990s. During the more recent commodities boom period that
ended around 2011, the region mostly re-aligned itself along a more typical ER pattern, with
much improved sovereign risk ratings and substantial real exchange rate appreciations. After
2011, however, the falling terms of trade have tended to realign the region along an IR direction,
as several low-saving commodity-exporting Latin American countries have experienced
substantial declines in country ratings accompanied by large real exchange rate depreciations.

      The rest of this paper is structured as follows. Section 2 discusses the three saving and
growth channels, relating them to the literature, and sets out testable hypotheses. Section 3


                                                3
models these channels in a simple way, paving the way for the econometric work presented in
Section 4. Section 5 explores macroeconomic patterns consistent with the channels using a
benchmarking approach. It also locates and contrasts the Latin American region and individual
countries within the region with respect to these patterns. Section 6 concludes with a brief
discussion of policy implications.

2.       The three channels linking saving and growth and testable hypotheses

        Saving would not matter for growth if either one of three conditions hold. Under the
perfect substitutability condition, all that matters for growth are investment prospects; whether
investment is financed by domestic or foreign saving is immaterial. Higher foreign saving alters
the distribution of growth dividends (from local residents to foreigners) but does not affect
growth itself. Instead, under the perfect elasticity condition, domestic saving does not constrain
growth because the saving required for growth automatically becomes available as needed, that
is, saving responds “on steroids.” Finally, under the perfect efficiency condition, the saving
decisions of private agents are optimal given the constraints and the government has no
comparative advantage to improve the outcome; hence, while domestic saving may be a
binding constraint on growth, there is no role for policy.1

        Departing from any of these three conditions has major implications. Dropping the
perfect substitutability condition opens the door to the real interest rate (IR) and real exchange
rate (ER) channels linking saving to growth. Dropping the perfect elasticity condition gives
rise to the normal (not “on steroids”) endogenous saving (ES) channel, through which the
action of the other two channels becomes amplified. In this section we review each of these
channels in light of the relevant literature and set out some key testable hypothesis. We start
with the most widely recognized, hence less controversial, ES channel.

         (a) The ES channel

         In the conventional neoclassical growth model, since output growth is a function of the
rate of technological change, it makes sense to consider how saving may endogenously respond
to changes in expected growth. Under rational inter-temporal consumption smoothing with
perfect foresight, higher expected growth (hence higher consumption tomorrow) should raise
consumption today, thereby reducing saving instead of increasing it (Gourinchas and Rey,
2013). However, this result can be reversed in various, plausible ways. One way is to introduce
uncertainty, so that precautionary savers save more aggressively in order to maintain an optimal
wealth to income ratio when their income grows faster (Leland, 1968; Zeldes, 1989). An
alternative is to make saving a function of expected productivity: if productivity is envisioned
to increase in the future, agents save and invest more today in order to reap the benefits of
higher returns tomorrow (Busso, Fernández, and Tamayo, 2015).



1
  Under the perfect efficiency condition, while the government would not be justified in directly influencing
private saving decisions, it could still foster growth through policies that improve the enabling environment, thus,
indirectly affecting saving. By expanding the opportunity set of households and firms, a better enabling
environment should enhance saving decisions. Note also that deviations from social optimality could also reflect
incomplete markets (as in the case of overlapping generation models), private time inconsistency (as in the case
of hyperbolic discounting), or public time inconsistency. In the latter case, the introduction of overly generous
social safety nets could lead to under-saving if private agents rely unduly on the state to support them in old age
but the state fails to mobilize the fiscal saving required to uphold its promises to the elderly.


                                                         4
        An overlapping generations life cycle setting can accentuate the private saving response
to growth, as growth raises the income of the middle-agers who save more than both the young
and the old (Modigliani, 1986). Also, consumption can lag income growth (i.e., saving can lead
growth) due to habit formation (Campbell and Cochrane, 1999). On the firms’ side, as income
and profits expand, corporate saving can rise as firms limit dividend distribution to mobilize
internal finance (Fazzari et al, 1988) or increase output prices relative to wages (Lewis, 1954;
Kaldor, 1958). Finally, a more controversial strand of literature (Rowthorn, 1982) extends the
Harrod-Domar, Keynesian-type constructs to support the ES on steroids view that domestic
saving responds to investment on a one-to-one basis.

        There is broad empirical support for the ES channel. In panel regressions, output growth
is found to be a significant determinant of private saving (Loayza, Schmidt-Hebbel and Serven,
2000). In Granger causality studies, growth generally causes saving (Carroll and Weil, 1993).2
Countries undergoing growth transitions are found to end up with permanently higher saving
rates (Rodrik, 2000) and saving accelerations are found to be mostly preceded (rather than
followed) by stronger output growth (Ebeke, 2014) or total factor productivity growth (Busso,
Fernández, and Tamayo, 2015). Consistent with this perspective, Guariglia, Liu and Song
(2011) and Yang, Zhang and Zhou (2011) find that the Chinese growth acceleration of the past
quarter of a century was what drove increases in corporate saving.

        As far as we know, the critical condition for self-propelling growth (the ES on steroids
claim that an autonomous increase in expected productivity generates a marginal increase in
domestic saving sufficient to fully cover the increase in investment triggered by the rise in
productivity) has never been tested (nor, for that matter, discussed). Arguably, however, the
very existence of balance of payments tensions and crises provides an indirect indication that
the shortages of domestic saving that occur in the real world matter. Thus, a strict ES on steroids
view of the world would be ruled out if domestic saving rises with growth but not enough to
avoid widening current account deficits. By potentially undermining balance of payments
sustainability, a less than perfectly elastic saving response to growth makes room for the IR
channel, to which we now turn, to become operative.

         (b) The IR channel

         In the closed economy Ramsey-Cass-Koopmans inter-temporal optimization model
with exogenous technological change, agents choose how much to save by comparing the
marginal return on saving (the marginal product of capital) to the marginal cost of saving (the
rate of time preference). A lower rate of time preference implies higher saving, which raises
the growth rate of consumption and output during the transition to the economy’s steady state.
Once there, higher domestic saving raises income but no longer affects growth. However, once
technological change is endogenized (Romer, 1986), saving becomes relevant for growth even
in the steady state. Furthermore, as long as private agents do not internalize the learning
externalities from higher saving (hence higher investment), the perfect efficiency condition
fails to hold, thereby opening a role for policy.

       This policy role vanishes in a financially open economy, however, if domestic and
foreign saving are perfect substitutes. In this case, the uncovered interest rate parity condition

2
  The failure to find clear empirical backing for saving causing growth may simply reflect data limitations as few
countries have engaged in massive efforts to boost their saving rate in a way that clearly stands out in the data in
the form of sufficient within-country variation over time.


                                                         5
holds and the world (real) rate of interest becomes the opportunity cost of saving. Through an
inflow of foreign saving, the marginal product of capital immediately adjusts to equalize the
world cost of saving with the world rate of return on saving. Hence, the rate of growth of
consumption immediately adjusts to its steady state level (the world’s growth rate of
productivity) while the rate of growth of local output immediately adjusts to equal the local’s
growth rate of productivity (Gourinchas and Rey, 2013). There is therefore under these
conditions no link from domestic saving to growth, nor any role for saving promoting policies.
This conclusion remains basically unchanged when technological change is endogenous
(Turnovsky, 2000).

         However, once sovereign default risk is introduced, reflecting for instance collateral
constraints, domestic and foreign savings are no longer perfect substitutes. Beyond some
optimal level, substituting domestic saving with foreign saving raises the cost of capital (as
measured by the country risk premium), thereby reducing steady state growth (Turnovsky,
2000). Thus, saving matters for growth in the presence of credit constraints. Moreover, foreign
borrowing introduces an un-internalized externality. Agents fail to take into account that as
they individually increase their amount of borrowing, they raise the country’s debt-to-capital
ratio, raising the cost of debt to all. Thus, there is a role for policy as well.

        This purely “ex-ante” (pre-crisis) characterization of the IR channel needs to be
complemented from an “ex-post” (post-crisis) perspective. In an imperfectly open economy
with a non-tradable sector, the real exchange rate will depreciate as the country risk premium
rises, reflecting an increased risk of a balance of payments crisis and mounting concerns
regarding the associated costs, which might include inflation, financial repression, and an
economic slump. If and when the crisis hits, access to foreign borrowing is lost, which is
possibly compounded by capital flight, and the cost of capital rises sharply. Again, individual
agents fail to internalize the crisis externalities, which further opens the role for policy. Note
that, by raising the marginal return on domestic saving, the crisis may thus push domestic
saving back up.

        The IR channel is at the core of an ample body of literature, going back to the analysis
of the links between foreign debt accumulation and balance of payments crises (Eaton and
Gersovitz, 1981), and subsequently connected to sudden stops (Calvo, 1998; Mendoza, 2010),
debt sustainability (Reinhart and Rogoff, 2013), and growth (Cerra and Saxena, 2008). A more
recent strand of literature has modeled balance of payments crisis as coordination failures
caused by collateral constraints and un-internalized pecuniary externalities that translate into
under-saving (Jeanne and Korinek, 2010). There is also an emerging literature linking saving
in foreign assets to macroeconomic volatility (Fogli and Perri, 2015).

       (c) The ER channel

        Switching the spotlight from the marginal cost of saving to its marginal return brings
up the ER channel. Growth rises when higher (lower) saving durably depreciates (appreciates)
the real exchange rate, thereby raising (lowering) the marginal rate of return of investing in the
relatively capital intensive tradable sector. The ER channel has two legs, the first linking saving
to the real exchange rate, the second the real exchange rate to growth.

       Starting with the first leg, with less than perfect factor mobility across firms, sectors or
borders, lower saving (hence higher domestic demand) must clearly appreciate the real
exchange rate. While an excess demand for tradables can be resolved solely via quantities (a


                                                6
widening of the current account deficit as imports increase) at given world prices, an excess
demand for nontradables should raise their price relative to that of tradables (Dornbusch, 1980).
Over time, however, the real exchange rate appreciation should induce a supply response via a
reallocation of factors from the tradable to the nontradable sector (Corden, 1981). Hence, the
permanence or durability of the real appreciation crucially depends on the elasticity of the
factors of production and their mobility across firms, sectors, or borders.

        In a two-factor setting (physical capital and labor) with perfect cross-sector labor
mobility and cross-border capital mobility, the equilibrium real exchange rate does not
appreciate in response to a fall in domestic saving. Instead, it is strictly determined by Balassa-
Samuelson-type supply side effects (Rogoff, 1992). This ceases to be the case, however, when
factor reallocation across sectors is slow and costly, due for example to adjustment costs for
investment (Morshed and Turnovsky, 2004), or if sector-specific human capital is added as a
factor of production (Brock and Turnovsky, 1993). In this latter case, even when physical
capital can be freely borrowed across borders (thereby equalizing marginal rates of return),
reflecting learning frictions, human capital must be accumulated locally, sector by sector. Thus,
in either case, reflecting induced scarcities, an increase in aggregate demand can durably
appreciate the real exchange rate. Irreversibilities and path dependence (hysteresis) can further
prolong these effects. For all such reasons, the impact of saving on the real exchange rate can
be durable enough to make saving relevant.

         On the empirical side, the first leg of the ER channel gets support from an early strand
of literature (Bergstrand (1991), De Gregorio et al. (1993)) that documents the importance of
demand factors as determinants of real exchange rates. At the same time, a large body of more
recent literature (Banerjee and Duflo, 2005; Hsieh and Klenow, 2009) finds a very wide
dispersion of productivities and returns across firms and sectors in developing countries, prima
facie evidence that factors are slow to reallocate. Importantly, however, given that productive
resources should gradually re-allocate to their best uses, the real exchange rate should
eventually return to its long-run Balassa-Samuelson equilibrium. A predictable and durable
link between saving and the real exchange rate can therefore only remain present (hence be
detected in the data) in a stochastic world of sluggish factor reallocation where recurrent saving
shocks keep throwing the real exchange rate away from its long-run equilibrium. As we will
argue in Section 5, such saving fluctuations should naturally arise in a world buffeted by
domestic or external macroeconomic shocks.

         When it comes to the second leg of the ER channel, the key paper is Rodrik (2008),
which finds that countries with more depreciated real exchange rates have larger tradable
sectors and grow faster.3 Rodrik posits that tradables are special in that they produce more
growth-enhancing positive externalities than non-tradables that are not necessarily internalized
by individual economic agents, thus opening the space for policy action.4 A parallel strand of
literature (Hausmann et al., 2005; Berg et al., 2008) finds that growth spurts are more likely to

3
 Notice that higher domestic saving should also be naturally associated with stronger current accounts, greater
export orientation and, hence, a higher ratio of external to domestic demand. The logic of the ER channel therefore
also implies the existence of a link between growth and the composition (external versus internal) of aggregate
demand.
4
  Tradable sector externalities can also take the form of learning-by-investing spillovers in a setting where the
tradable sector is more capital intensive than the non-tradable sector, as in Korinek and Serven (2010). Berg and
Miao (2010) found some evidence in support of tradable sector externalities. However, the identification of greater
positive externalities in tradables and the role of externalities in the link between the real exchange rate and growth
remain elusive to date (see, for instance, Giles and Williams, 2000; and Harrison and Rodriguez-Clare, 2009).


                                                          7
occur in countries with more competitive exchange rates. Finally, in a somewhat similar vein,
Levy Yeyati et al. (2013) find that countries that pursue exchange intervention policies geared
at keeping or enhancing external competitiveness display better growth performance, although
the transmission channel between the exchange rate and growth is via higher investment, rather
than via increased exports.5

        Be it as it may, the ER channel, particularly the link between saving and the real
exchange rate, has not yet been fully explored in the theoretical growth literature. Korinek and
Serven (2010) and Itskhoki and Moll (2014) develop models where the relative productivities
of the tradable and non-tradable sectors (i.e., the real exchange rate) in a Romer-type
endogenous growth model become functions of aggregate demand. Because externalities are
not internalized, private agents save and invest too little, the tradable sector is too small, and
the economy grows less than optimally.6 Yet, neither of these two papers directly focuses on
saving itself (in both cases, the first-best intervention is a direct investment subsidy or a tax on
nontradables rather than a saving boost).7 Nonetheless, both papers implicitly recognize that
raising saving can be a less distortionary, hence second-best, approach to boosting growth.

         (d) Testable hypotheses

       The fact that the three channels linking saving and growth have theoretically distinct
macroeconomic signatures leads to empirically testable patterns. Where the ER channel
dominates, countries that save less should have a less competitive real exchange rate, be less
export oriented (hence, have a higher ratio of external to internal demand), and grow less.
Where the IR channel dominates, countries that save less should have a higher sovereign risk
premium, and grow less, despite having a more competitive real exchange rate. Finally, for the
ES channel not to be on steroids, domestic saving should rise with growth but not enough to
avoid current account deficits.

       At the same time, we would expect the ER and IR channels to interact and operate in
sequence. Thus, a period of IR channel dominance may be followed by improvements in
macro-financial policy that set the conditions for a return of the ER channel. In turn, a
prolonged period of current account deficits and uncompetitive real exchanges under the ER
channel can bring back IR dominance.

        Some important non-linearities are also to be expected. In particular, the IR channel
should become operative only when current accounts have deteriorated sufficiently to
undermine balance of payments sustainability. Thus, where high saving rates have already led
to current account surpluses, a further increase in the saving rate should have very limited

5
  A closely related literature focuses on the resource curse (Benigno and Fornaro, 2013) and the Dutch disease
(Van der Ploeg, 2010). It shows that broader access to foreign capital or a natural resource boom can slow down
growth by inducing a real exchange rate appreciation that promotes the development of the non-tradable sector at
the expense of the tradable sector. Typically, this literature does not contemplate long-term capital accumulation
and does not focus on saving per se. However, it has similar implications for saving as the ER channel. Indeed,
an increase in exports of natural resources can result in a decline of domestic saving, once the resulting decline in
the stock of (non-renewable) natural resources is properly accounted for (World Bank, 2006).
6
 Key to the results in both papers are assumptions limiting the cross-border equalization of returns, through
agency frictions in the case of Itskhoki and Moll (similar to those in Aghion et al, 2009), and a self-imposed
assumption of a financially-closed economy in the case of Korinek and Serven.
7
 Korinek and Serven focus on international reserve accumulation. Itskhoki and Moll establish a link between
saving, the real exchange rate, and growth in a latter part of their paper, where they introduce a nontradable sector.


                                                          8
effects on growth. Conversely, where domestic savings are weak enough to generate chronic
current account deficits, an increase in the saving rate, by reducing the risk of default, should
have much larger beneficial effects on growth.

3.         The analytical framework

        To help analyze empirically how the three above channels interact, identify their
imprints, and explore the above hypotheses while keeping things as simple as possible, we
posit a highly stylized macro model that focuses on medium term equilibria.8 While the lack of
a dynamic framework is an important limitation (more on this below), it greatly simplifies the
estimation. At the same time, to facilitate comparisons across a broad and heterogeneous set of
countries, we develop a model-based benchmarking methodology that fully exploits the
systemic linkages across macroeconomic variables.

           (a) A minimalist model of saving and growth

         The demand side of the model is captured by a domestic (IS-type) and external (interest
rate parity) equilibrium conditions (the signs above the variables denote the direction of the
first partial derivatives):
                                        
                    SF  I (e,  )  SD (e, g )                                (1)

                                    
                       *   (r )                                           (2)

        In the IS equation the current account (foreign saving, SF) equals the difference between
investment, I, and domestic saving, SD. Both investment and domestic saving are functions of
e, the real exchange rate (the ER channel). In addition, investment also depends on the real
interest rate, , which is the sum of the world interest rate, *, and , the sovereign risk
premium (the IR channel). The latter moves inversely with r, the sovereign risk rating.
Domestic saving also depends on the rate of growth, g (the ES channel).

        The supply side of the model is a reduced-form growth equation that collapses stocks
(capital) and flows (investment), based on a time-invariant capital-output ratio; in this case,
growth is a simple function of investment, the real exchange rate (the ER channel) and the
country rating (the IR channel):9
                               
                     g  g( I , e, r )                                         (3)

        The model is closed with two reduced-form functional expressions; the first sets the
real exchange rate as a function of SF, the current account (the ER channel) and r, the sovereign
rating (the IR channel); the second links the country rating to the current account deficit:


8
  We leave aside all dynamics, including short-term Keynesian output dynamics around full employment or
exchange rate dynamics around the interest rate parity condition, medium-term dynamics such as those underlying
consumption boom-appreciation cycles, and longer-term dynamics such as the exchange rate appreciations
resulting from the build-up of foreign assets under current account surpluses.
9
    See Appendix A for a simple derivation.


                                                      9
                                   
                    e  e(S F , r )                                                  (4)

                              
                    r  r (S F )                                                     (5)

        Replacing  in the investment function by r, using (2), and SF in the real exchange rate
and country rating equations by I - SD, using (1), the model can be linearized for the endogenous
variables and expressed as a function of a set of other control variables (the C’s) that may
include policy and structural country characteristics as well as domestic or external shocks
(more on this below):
                                               
                     e   e ( I  S D )   e r  Ce                                (6)
                          
                     r   r ( I  S D )  Cr                                        (7)
                                       
                     I   I e   I r  CI                                          (8)
                                       
                    SD   g   S  C S                                             (9)
                                          
                     g   I   g e   g r  Cg                                    (10)

         The coefficients of the endogenous variables in this system are structural elasticities
(signs as marked above the coefficients) that relate to each of the three channels (see Figure 1).
The betas (e, I, s and g) are associated with the ER channel and denote the responsiveness,
respectively, of: (i) the real exchange rate to net saving; (ii) investment to the real exchange
rate; (iii) domestic saving to the real exchange rate; and (iv) growth to the real exchange rate.
The gammas (e, r, I, and g) are associated with the IR channel, with analogous
interpretations as those for the previous set of elasticities. Alpha () is associated with the ES
channel, denoting the responsiveness of domestic saving to growth. Finally, delta () is a
general structural elasticity linking growth to investment, which reflects productivity.

           The growth impact of an autonomous increase in gross saving:10

                                  g     A
                                                                                   (11)
                                  SD    

                              A  ( e   e r )( g   I )   r ( g   I )   (12)

                                1   A   e I   e S                          (13)

         Equation (12) breaks down all the possible means through which an increase in saving
can affect growth. The first two terms jointly measure the effect of changes in saving on growth
through the real exchange rate. According to the first term, by reducing the current account
deficit, a gross saving boost depreciates the real exchange rate, via e and the ER channel. But
by improving the country’s risk rating (r), it also appreciates the real exchange rate, via er
and the IR channel. Thus, the net impact on the real exchange rate of an increase in saving

10
     See Appendix A for derivations.


                                                              10
ultimately depends on the relative strengths of the ER and IR channels, with the ER channel
dominating the IR channel when e+er > 0. According to the second term of A, an exchange
rate depreciation boosts growth, both directly via g (by promoting the positive growth
externalities of investing in the tradable sector), and indirectly via the interaction of I (the rise
in investment due to changes in external competitiveness) and   the impact of capital
accumulation on growth). The third and fourth terms of A also interact and jointly measure the
growth impact of changes in saving through the sovereign risk rating. An increase in saving
improves the country rating, as measured by r, which in turn positively affects growth, both
indirectly (by reducing the interest rate, thereby raising investment), via the interaction of I
and , and directly (by reducing or eliminating the negative growth externalities of balance of
payments crises), via g.

        There are three feedback loops operating through the  term in equation (13). First, an
increase in the rate of growth leads to higher gross saving, which sets into motion a multiplier
effect coming from the ES channel, further boosting saving and growth via  (the term).
The two additional feedback loops operate within the ER channel. According to the second
feedback loop, a decline in gross saving has a multiplier effect by appreciating the real
exchange rate, which, in turn (through wealth and income effects) tends to further reduce
saving and stimulate consumption (the eS term). In the third feedback loop, by appreciating
the real exchange rate, the decline in saving reduces investment, thereby reducing the excess
of investment over domestic saving, which partly offsets the effects of the initial depreciation
(the eI term).

       Finally, the impact on the real exchange rate and the current account of an autonomous
increase in investment is given by:

        e                   ( I  SD ) ( e   e r )(1   )
            ( e   e r )                                            (14)
        I                       I                 

       Assuming e+er > 0, it follows from this expression that, if and when  > 1, the real
exchange rate depreciates in response to a boost in investment as the supply of gross saving
rises more than demand (net saving rises). This is the ES on steroids case, where domestic
saving no longer matters for growth.

        Figure 1 provides a graphical depiction of this model, where the transmission channels,
interactions, and feedback loops can be grasped easily and in one go.

        (b) Benchmarking methodology

        In matrix form, the system (6)-(10) can be expressed as:


                          Xtk  QXtk  RYt k  SZtk  Ft k               (15)

                          Ztk  R 'Yt k  F 'tk                          (16)

where Xtk is the vector of endogenous macro variables for country k at time t; Yt k a vector of
identifiable country-specific fundamentals; Ztk is a vector of identifiable policy choices that are


                                                     11
themselves partly endogenous to the country’s fundamentals; and Ft k is a vector of residuals
that includes domestic and external stochastic shocks, as well as all remaining unidentified
fundamentals or policy choices. Replacing Ztk from (16) in (15) and solving for Xtk :

                            Xtk  ( I  Q )1[( R  SR ')Yt k  ( Ft k  SF 'tk )]   (17)

                                                           ˆ k , such that:
           This yields a set of policy-neutral benchmarks, X t


                             ˆ k  ( I  Q )1 ( R  SR ')Y k
                             X                                                       (18)
                               t                           t


       In turn, from (18), the following set of policy-neutral gaps,                        , is inferred as the
expected value of a country’s deviation from benchmarks:

                                                                                     (19)

         Thus, the benchmarks indicate where a country is expected to be for a given
endogenous variable, taking into account its level of economic development, its non-policy-
related structural characteristics, and the average policies and policy-related institutions of its
peers. The gaps reflect country specificities, which can be unidentified endowments,
preferences or policy choices embedded in the F term (in all cases omitted variables), or
identified policy deviations from the average policies of peers embedded in the SF’ term. To
the extent that these omitted variables primarily reflect preferences or policy choices (rather
than unidentified endowments), saving gaps become the result of policy choices (that can
presumably be altered), rather than irreversible facts of life. In this case, on at least a first
approximation, the gaps can be interpreted as reflecting the deviation of a given country’s
policies and policy-related institutions from the average of peer countries. Note also that
because the gaps are linearly related through the Q matrix, the cross-correlations between them
reflect the elasticities embedded in the model. In particular, the saving and real exchange rate
gaps should be negatively correlated if the elasticity of the exchange rate with respect to
changes in the current account deficit is positive and significant.

4.         Econometric estimates

        We first estimate each structural equation using ordinary least squares (OLS). Because
cross-equation correlations between the endogenous variables are not accounted for, these
estimates are biased. Moreover, they do not prove causality. To circumvent both of these
problems, we subsequently use reduced-form instrumented estimates. Reflecting the high
endogeneity, these estimates could be calculated only for a limited subset of elasticities. Even
so, results suggest that the impact of saving on growth could be substantial.

           (a) Data and specification

        We use country-level yearly data for the period 1981-2012 and a sample of 119
countries, including low, middle and high-income countries.11 To better capture medium-term
equilibria, we run all the regressions on three-year averages instead of on the original annual

11
     See Appendix B for data definitions, statistics and sources.


                                                            12
data. However, to check for consistency and robustness, we repeat these estimates using
dynamic ordinary least squares (DOLS) with yearly observations.12 In addition, we control for
short-term disequilibria through macro and financial crisis dummies (see below). To control
for correlation of error terms, we use country clustering; to capture worldwide shocks, we use
time fixed effects. To allow for possible asymmetries according to countries’ balance of
payments position (and, hence, to capture traces of nonlinearities), we introduce current
account deficit dummies in the estimation of e and r. To test for heterogeneity between
countries at different levels of development, we divide the sample of countries by income (low,
middle, high) and run separate sets of regressions for each income class. Finally, to test for
differential saving propensities across sectors (households, corporates, public and external) we
also present the results of saving regressions broken down by sectors, based on a smaller
sample of about 50 countries.

        To make countries comparable, we first control for the level of economic development
(as proxied by per capita GDP), which, importantly, also controls for the more systematic
component of the Balassa-Samuelson effect in the real exchange rate regression.13 We then
include as additional controls key structural and institutional features (trade and capital
openness, demographics, dependence on natural resources, dependence of foreign-earned
income as determined by net unrequited transfers, quality of institutions, country size); fiscal
policy (the fiscal balance and public sector consumption); exposure to external shocks (terms
of trade changes, safe haven effects); 14 and the country’s record of macroeconomic and
financial stability, as proxied by the incidence of inflationary or external debt crises. We
distribute these controls across the five structural equations based on reasonable priors.

        To eliminate the valuation biases that terms of trade changes introduce in nominal
macro variables, we use the UN database that provides national accounts in real terms (i.e.,
where each component of aggregate demand is deflated by its own price deflator). 15 To
compare real exchange rate levels over time and across countries (rather than real exchange
rate changes over time for the same country), we use as a measure of the real exchange rate the
World Bank’s purchasing power parity (PPP) conversion factor divided by the country’s
nominal exchange rate with respect to the U.S. dollar.16 The latter measures, relative to the U.S.,
the cost of the bundle of goods and services that makes up the gross domestic product of a
country.17

        We regress both the real exchange rate and the country rating against net domestic
saving, where the latter equals the current account and is expressed as the difference between
12
   Dynamic ordinary least squares (DOLS) has been shown to provide efficient estimators of long-run
relationships in panel data with co-integrated variables (Stock and Watson, 1993).
13
  Thus, the residual (uncontrolled for) component of real exchange rates is more likely to reflect demand side
than supply side effects.
14
  Safe haven effects are measured by interacting market volatility (the VIX) with a safe haven dummy that equals
one for the United States, Switzerland and Japan (and zero for all other countries).
15
  For example, the investment-to-GDP ratio is likely to be underestimated if nominal variables are used in times
of significant terms of trade gains because the GDP deflator rises faster than the investment deflator.
16
  Rodrik (2008) uses a similar index albeit from a different database (Penn World Tables instead of the World
Bank’s WDI). Note that because the exchange rate is expressed in terms of dollars, an appreciation is reflected in
an increase in e.
17
   Although the PPP conversion factor series was recently updated (after this project was initiated), we continued
to use the former version (with data up to 2012) because we detected in the updated version some counterintuitive
adjustments for some key Latin American countries.


                                                       13
investment and gross domestic saving (all as shares of GDP), with the coefficient of investment
constrained to be the opposite of that of saving.18 Consistent with other studies (e.g., Loayza et
al, 1999), saving is derived by subtracting consumption from national income, which is equal
to GDP plus net factor income and net unrequited transfers.

         (b) Full sample results

        Consider first the results of the OLS structural-form regressions for the full sample (i.e.,
not broken down by country income level), as synthesized in Table 1 (see Appendix C for the
full regression results). Tables 2 and 3 complement the saving regressions for this larger sample
with some OLS regressions for the smaller sample where a breakdown of saving by sector is
available. Overall results indicate a good fit. Except for  which is about twice as large under
DOLS than under OLS, all other elasticities are quite similar. We will therefore only discuss
the OLS results.

        Starting with the ES channel, the growth elasticity of saving, , is positive and
significant (Table 1). The strength of the ES channel comes mainly from the public and (non-
financial) corporate sectors (Table 2). There is however a remarkable difference between these
two sectors when their saving performance is compared across countries (Table 3). The share
of (non-financial) corporate saving (as well as that of foreign saving) in total domestic saving
rises in countries that save less than their peers (hence where savings are relatively scarce),
while that of the public sector declines. This confirms the key role corporate saving plays as
the most buoyant component of domestic saving. On the one hand, corporate saving boosts
aggregate saving during growth spurts, hence accounting for the bulk of the ES channel
response capacity. On the other hand, it fills up the saving gaps when saving is scarce. By
contrast, in the case of the public sector, when it rains it pours: its share of total saving is higher
the more a country saves relative to its peers.

         Turning now to the ER channel and the real exchange rate regressions, consistent with
the first leg of this channel, the real exchange rate appreciates as the current account worsens
(e>0). However, e loses significance for economies experiencing current account surpluses,
indicating that nonlinearities also affect the ER channel. To the extent that current account
surpluses are more likely to occur under slack conditions, this is consistent with a weakening
link between demand and prices as the economy drifts away from its productive frontier. There
is also an important multiplier effect operating through S, the saving’s response to the real
exchange rate. In addition, as postulated under the ER channel, investment rises (I < 0) and
growth accelerates (g < 0) as the exchange rate depreciates.

        Moving on to the IR channel and the sovereign risk rating regressions, as postulated,
ratings improve as the country’s current account of the balance of payments strengthens (r<0).
However, there are strong nonlinearities, as r loses significance (and switches sign) for
economies with current account surpluses. Thus, as expected, the IR channel becomes
operative only where debt sustainability becomes an issue. Also consistent with priors under
the IR channel, the real exchange rate appreciates (e>0), investment increases (I >0), and
growth rises (g>0) as country risk ratings improve. Yet, the condition e+er >0 is always



18
   Because investment and saving are expressed in logs, the difference between saving and investment (net saving)
is approximated by the difference of their logarithms.


                                                      14
satisfied, which indicates that, on average, the ER channel statistically dominates the IR
channel. Thus, on average, an increase in saving depreciates the real exchange rate.

            (c) Results by country income level

       Consider now the results by income level (Table 1). Middle-income countries (World
Bank definition) generally fit the model better.19 The responsiveness of saving to growth (the
ES channel) is higher in the middle-income countries than in the low-income countries and
comparable to that in the high-income countries. At the same time, only low and middle-
income countries show a strong ER association between real exchange rate appreciations and
increases in the current account deficit. The ER links between the real exchange rate and
investment, or the real exchange rate and growth are similarly stronger in the middle-income
countries, and so is the IR link between current account deficits and country ratings.

        Such heterogeneity at either end of the per capita income scale is not too surprising.
Higher-income countries should become less susceptible to ER- or IR-type macroeconomic
patterns as the frictions limiting cross-border and cross-sector factor mobility become eroded
away with economic development and global integration. In other words, in richer countries,
the substitutability between foreign and domestic saving is likely to be greater. Similarly, at
the lower end of the per capita income distribution, the relevance of the ER and IR channels
should also be limited, this time due to insufficient scale effects. In particular, it is reasonable
to expect the positive growth externalities associated with the development of the tradable
sector to start appearing only beyond some minimum threshold of economic size and
development. Similarly, IR-type debt and crisis dynamics are likely to require a level of
development of capital markets that is unlikely to be present in the lowest-income countries.

            (d) Impact analysis

        While the structural-form regressions support all three channels, they yield a modest
estimate of their impact. A ten percentage points of GDP domestic saving mobilization effort
(the effort needed to bring Latin American saving rates broadly at par with those of emerging
Asia) would raise yearly per capita income growth by only 0.3 percentage point in countries
with current account deficits (Table 4). About two-thirds of this impact comes from the ER
channel and only one-third from the IR channel. Furthermore, consistent with the existence of
strong asymmetries and nonlinearities, impacts are even smaller in the case of current account
surpluses. The ES channel provides a significant but limited contribution to growth. From each
dollar of additional investment, about 10 cents would be self-financed by the induced increase
in savings caused by the higher growth. This falls far short of the ES-on-steroids level.

        The structural-form regression estimates have severe limitations, however. They
pinpoint correlations but not directions of causality, and do not account for cross-equation
interrelations and simultaneity. To circumvent such limitations we run reduced-form,
instrumental variable-based estimates, based on the following set of instruments:20

               Domestic saving: old dependency ratio
               Real exchange rate: commercial openness

19
     This is consistent with results reported in the literature (see for example Rodrik, 2008).
20
  To avoid spurious correlations, we first regress the instruments against GDP per capita and country size, and
use the residuals, instead of the raw data.


                                                           15
               Country rating: enabling environment
               Investment: inflation crisis dummy
               Per capita income growth: population growth

        Arguably, all five instruments meet the exogeneity requirement. Demographics (the old
dependency ratio) determines saving but not the other way around. Commercial openness
affects the real exchange rate but it is hard to argue that the real exchange rate affects total
trade (even if it clearly affects the current account, hence the composition of trade between
imports and exports).21 The enabling environment (which reflects deep historical trends and
fundamental institutions) clearly matters for the country rating but it is hard to see a strong
reverse causality in the medium-term. Inflation crises (i.e., high inflation episodes) have
substantial effects on investor behavior and are obviously endogenous to the macroeconomic
environment (monetary and fiscal policy), but the reverse link going from investment to
inflation is tenuous at best. And although population growth is a function of the level of per
capita income, it is improbable that population growth would move significantly with the rate
of growth of per capita income.

         Reflecting high endogeneity in the system, only four elasticities {, e, g, } can be
retrieved, all of which are substantially above those obtained through OLS estimates (see
Appendix D). 22 The sub-set {s, I, g, e} has p values high enough to suggest non-zero
elasticities while the sub-set {I, r} does not. Thus, setting I = r = 0, adopting for {s, I, g,
e} the previously obtained structural-form OLS values, and retaining for the other four
elasticities the values indicated above yields a more significant growth acceleration (0.85
percentage points per year for a ten percentage points increase in the saving to GDP ratio).
Although the ES channel now makes a more significant contribution to growth, it still falls
short of 100 percent. Thus, the evidence continues to lean against an ES-on-steroids view.

       For current account deficits,  and e are no longer identifiable. Maintaining the values
obtained for current account surpluses but using for r its OLS value yields a modestly higher
impact (1.14 percentage points of higher growth for a 10 percentage points increase in the
saving to GDP ratio). However, since OLS estimates yielded a much higher e (about three
times larger) for current account deficits than for surpluses, this procedure is likely to
underestimate the true value of e, hence the potential impact of a saving boost in the context
of current account deficits. The same can be said for r, as the structural estimates actually
yielded a wrong-signed estimate for current account surpluses. Using plausibly higher (albeit
admittedly somewhat speculative) values for e and r yields a much higher range of impacts.23

5.         Exploring the macro patterns of the saving-growth channels

       Based on the above estimates, this section uses the benchmarking methodology
described earlier to contrast the macroeconomic patterns induced by the ER and IR channels.

21
     See IMF (2013) for a similar interpretation.
22
  The IV estimated value of 0.07 for  is very close to its DOLS estimated value (0.064). Also, the estimated
value of 5.15 for  is consistent with the range of values for key parameters (capital share, investment to GDP
ratio, and capital-output ratios) reported for middle-income countries (see Appendix A).
23
   Using a twice as large value for e as that estimated for current account surpluses (0.86) and adopting for r its
OLS value for middle-income countries (-0.18) would raise the growth rate (for the same 10 percentage points
increase in the saving-to-GDP ratio) by about 2.7 percentage points for countries with current account deficits.


                                                        16
It shows that these patterns are consistent with priors. The section concludes with a brief look
at Latin America’s recent macro history, which illustrates how the ER and IR patterns have
interacted over time. Since we are only interested in the relative positions of countries but not
in causality or policy impacts, we use the results of the more complete structural form
regression system.

       (a) Cross-country patterns

         Figures 2-6 reveal the starkly contrasting features of the ER and IR channels. Figure 2,
which interacts the real exchange rate gaps with the growth gaps, essentially reproduces the
findings of Rodrik (2008), whereby countries with more competitive real exchange rate growth
faster, in line with the second leg of the ER channel. In turn, Figure 3, which interacts the
saving gap with the real exchange rate gap, illustrates the first leg of the ER channel. Countries
that save more have more competitive real exchange rates. In addition, Figure 4 shows that
countries that have more competitive exchange rates are more export oriented (they have a
higher ratio of external to domestic demand). Thus, the central tendency of the entire sample
bears the ER signature. Yet, as is apparent in Figures 5 and 6, there are also IR forces clearly
at work. Countries that save more have higher ratings (Figure 5) and stronger current account
balances (Figure 6).

        The footprints of the ER and IR channels can be best identified in Figure 7, where we
plot the median growth gaps, rating gaps, and investment gaps for all observations located in
each of the quadrants of the same saving-exchange rate map that was displayed in Figure 3. To
help distinguish the ex-ante (pre-crisis) from ex-post (post-crisis) modes of the IR channel, we
also plot the median inflation rates in each of the quadrants. The results confirm the existence
of very distinct ER-IR patterns aligning along each of the diagonals in the map:

          In accordance with the ER pattern, the points in the top left quadrant (the
           undervalued-high savers) are associated with high growth and high investment (i.e.,
           positive gaps); instead, the observations in the bottom right quadrant (the
           overvalued-low savers) are associated with low growth and low investment (i.e.,
           negative gaps). Remarkably, there is not much difference between the two groups
           as regard their sovereign ratings, which suggests that, as long as the low saving and
           overvalued exchange rates do not raise fiscal and balance of payments sustainability
           issues, the country ratings remain unaffected.

          In accordance with the ex-ante mode of the IR pattern, the points in the bottom left
           quadrant (the undervalued-low savers) are associated with low sovereign ratings
           (i.e., negative gaps), while those in the upper right quadrant (the overvalued-high
           savers) are associated with high ratings (i.e., positive gaps). The growth
           performance of the observations in the overvalued quadrants is substantially below
           benchmark while that of the observations in the bottom left quadrant (undervalued-
           low savers) is only slightly above benchmark. This suggests that while
           overvaluation typically undermines growth, a necessary condition for
           undervaluation not to undermine growth is macro sustainability (as signaled by
           positive sovereign rating gaps).

          In accordance with the ex-post pattern of the IR channel, inflation is the highest in
           the bottom left quadrant (undervalued-low savers). High inflation countries are



                                               17
               typically the ones under the spell of financial repression and flight capital, which is
               a trademark of macroeconomic and financial crises.
           (b) The case of Latin America

        We now use the benchmarking framework to locate and contrast the macroeconomic
paths followed over the last three decades by the Latin America and the Caribbean (LAC)
region, particularly by its larger, middle-income countries. 24 When countries are divided
according to their average saving and real exchange rate gaps over the period 1990-2012, it is
first worth noting that LAC is the only region deep inside the bottom left quadrants of both
Figures 3 and 5. On average, LAC has conformed to an IR pattern of low saving, low rating
and undervaluation. Moreover, the region saving rate was low not just because LAC’s growth
rate was low (i.e., only a reflection of the ES channel). As shown in Figure 8, LAC’s average
saving rate was persistently below benchmark, whether calculated on the basis of the
benchmark growth rate or the actual growth rate.

         Yet, the above averages hide substantial diversity, both across countries and over time.
For simplicity, let us consider only the larger middle-income countries in LAC (henceforth, the
LAC1 countries). Looking first across countries, four groups of LAC1 countries stand out
(Figure 9). The first group (Chile, Mexico Panama and Peru) occupies the top left quadrant.
These are countries with positive saving gaps and undervalued real exchange rates that conform
to what one would expect for high savers under the ER channel. On the polar opposite side of
Figure 9 (the bottom right quadrant) stands another group (Brazil, Costa Rica, and Uruguay)
that also conforms to the ER channel but on the low saving side. Exchange rates were
overvalued because saving was low. The four remaining countries can in turn be assembled
into two IR groups. The group in the bottom left quadrant (Colombia and Ecuador) conforms
to an ex ante IR pattern of low savings yet undervalued exchange rates. Instead, the group
sitting in the top left quadrant (Argentina and Venezuela) conforms to an ex post IR pattern of
high savings and appreciated exchange rates, where financial repression and exclusion from
world capital markets forced private saving upward.

        Let us now focus, with the help of Figure 10, on the two groups of LAC1 countries that
followed the ER pattern in Figure 9. The contrast between these two groups is obvious in term
of their saving gaps (Figure 10.d): the high savers significantly exceeded their benchmarks
throughout the entire period while the low savers fell short by a wide margin, especially during
the 1990s, the stabilization years. Consider now the real exchange rates gaps (Figure 10.a).
Both groups started from deeply under-valued currencies in the crisis-laden 1980s and
subsequently appreciated, but the appreciation was much more pronounced in the case of the
low savers, with the real exchange rate plateauing in the case of the high-savers while the low
savers continued to appreciate over the whole period. Such stark end-of-period difference is
difficult to explain based on IR-based pressures. Indeed, following an major initial dip
characteristic of the 1980s, sovereign ratings rose steadily and ended up over-performing for
both groups (Figure 10.b). The evidence therefore suggests it was the ER channel, rather than
the IR channel, that was mostly responsible for the distinct paths followed by the real exchange
rate in these two groups of countries. As a result, the low savers experienced much higher
current account deficits, and ended up paying a much heavier price in terms of lower
investment and growth rates (Figures 10 f, e and c, respectively).


24
     We only include LAC countries with per capita income above $5,000 a year.


                                                       18
        The data also suggest the presence of interaction and sequencing effects that reflect the
relative dominance of the ER and IR channels. This can best be appreciated in Figure 11, which
plots the path followed by the correlation between the saving and real exchange rate gaps
during 1981-2015.25 During the 1980s, the correlation was mostly positive and rising, pointing
toward a clear IR dominance under the pull of widespread balance of payments crises. Instead,
during the 1990s, the correlation trended downward as the region’s macroeconomic policies
improved, suggesting a switch from IR to ER-dominated dynamics. The downward trend
continued during the 2000s, as saving rates declined and real exchange rates appreciated further
under the pull of the commodities boom. Thus, by 2011-12, LAC1 countries aligned closely
along the second diagonal, as predicted by an ER-dominated pattern (Figure 12). Yet, a new
reversal of the trend, back in the IR direction, took place in 2012 after the commodities boom
peaked and real exchange rates started again to depreciate.

        The above patterns therefore suggest some systematic dynamic interaction between the
IR and ER channels, with one channel feeding the other. The decline in saving, real
appreciation and increase in foreign liabilities brought about by the ER channel sowed the
seeds for subsequent IR dynamics when the ER path became unsustainable. In turn, the real
depreciation and domestic saving adjustment brought about by the IR channel as economies
lost access to foreign saving set the ground for a new round of ER-driven appreciation.

        Importantly, however, domestic or external macroeconomic events (such as large terms
of trade shock associated with commodity cycle) triggered or accentuated these dynamics. This
in turn is consistent with the potentially persistent yet ultimately transient nature of the ER
channel. As argued in Section 3, a stable link between saving and the real exchange rate should
naturally emerge and persist in a stochastic world where production factors adjust sluggishly
and that is constantly buffeted by exogenous shocks to which countries must respond by
adjusting their saving rate. Thus, in the recent commodities cycle, Latin American countries
had to choose how much of the windfalls to save. As predicted by the ER channel, countries
saving more experienced more moderate appreciations and better growth.

6.      Concluding thoughts

       Despite the clear limitations of this paper (further work is needed to refine the reduced-
form estimates, which at present rely on only a limited subset of elasticities; and a dynamic
framework is clearly required to fully account for the complex stochastic interactions between
the IR and ER channels), its main messages and results are worthy of attention. They point
toward two complementary directions for policy.

        First, there is an important role for short-term oriented policy in avoiding saving
collapses (consumption exuberance) in countries undergoing regime changes or exposed to
external shocks, such as surges of capital inflows or booms in commodity prices. From an ER
perspective, a surge in capital inflows is either an external response to a domestic saving shock
or call for a domestic saving response to an external shock. Thus, capital controls on inflows
would arguably be a suitable response if the inflows are originated externally, whereas macro-
prudential policy would better suit a situation where the inflows are triggered by a domestic



25
 Yearly gaps were calculated for LAC1 countries until 2015 using an expanded database. Time fixed effects
were estimated using the residuals of the structural form regression for the real exchange rate.


                                                   19
saving shock. In the case of surges in commodity prices, countries should save as much as
possible of the windfall gains when the latter are temporary.26

        Second, precautionary saving mobilization efforts may be a justifiable insurance policy
in middle-income countries with structurally low domestic saving rates, hence with a tendency
to overheat more quickly and go more readily into balance of payments deficits as domestic
demand pressures build up. Reflecting the strong non-linearity of elasticities (hence impacts),
structural saving mobilization efforts can help countries stay clear from the region where the
marginal impacts of demand shocks on the real exchange rate and the risk premium are higher,
thereby reducing demand-induced macroeconomic instability. Thus, during upturns, by
reducing the real exchange rate appreciation resulting from demand pressures, saving
mobilization efforts can limit the potential damage associated with extreme ER-led trajectories.
During downturns, by improving the structural balance of payments, such efforts can provide
the policy space needed to moderate the adverse growth implications of abrupt reversals into
IR-led trajectories.27




26
  Because benchmarks and gaps are controlled for changes in terms of trade, spending (instead of saving) a
windfall gain derived from a favorable terms of trade shock worsens a country’s saving gap.
27
  Such a policy approach should be immune to fallacies of composition whereby saving efforts by individual
countries lower world demand and weaken world growth. That is because collectively inefficient world saving
gluts are unlikely to arise as long as countries’ policy objective is limited to avoiding large and chronic current
account deficits, rather than further raising already positive current accounts.


                                                       20
                                       References

Aghion, P., D. Comin, P. Howitt and I. Tecu (2009). “When Does Domestic Saving Matter for
      Economic Growth?”, Harvard Business School Working Paper 09-080.
Aizenman, J., B. Pinto and A. Radziwill (2004). “Sources for Financing Domestic Capital: Is
      Foreign Saving a Viable Option for Emerging Countries?”, NBER WP No. 10624.
Banerjee, A. and E. Duflo, (2005). “Growth Theory Through the Lens of Development
      Economics”, Massachusetts Institute of Technology
Benigno G., and L. Fornaro (2013). “The Financial Resource Curse”, CEP Discussion Paper
      No. 1217.
Berg A., and Y. Miao (2010). “The Real Exchange Rate and Growth Revisited: The
      Washington Consensus Strikes Back”, IMF WP 10/58.
Berg, A., J.D. Ostry, y J. Zettelmeyer (2008). “What makes growth sustained?” IMF WP 08/59.
Bergstrand, J. H., 1991. "Structural Determinants of Real Exchange Rates and National Price
        Levels: Some Empirical Evidence," American Economic Review, vol. 81(1), 325-34.
Brock P. and S. Turnovsky, 1993. “The Dependent Economy Model with Both Traded and
      Non-traded Capital Goods, NBER WP No. 4500.
Busso, M.; A. Fernández, and C. Tamayo (2015). “Productivity and Savings in Latin America”,
      Mimeo, Inter-American Development Bank.
Calvo, G. (1998). “Capital Flows and Capital Market Crises: The Simple Economics of Sudden
      Stops”, Journal of Applied Economics, Vol. 1, No. 1, pp. 35-54
Campbell, J. and J. Cochrane (1999). “Force of Habit: A Consumption-Based Explanation of
      Aggregate Stock Market Behavior”, Journal of Political Economy 107 (2), 207-251.
Carroll, C., and D. Weil (1993). “Saving and Growth: A Reinterpretation”, NBER WP No.
      4470.
Cerra, V. and S.C. Saxena (2008) “Growth Dynamics: the Myth of Economic Recovery,” The
      American Economic Review, Vol. 98, No. 1, pp. 439–457.
Corden, W.M. (1981). “The Exchange Rate, Monetary Policy and North Sea Oil: The
      Economic Theory of the Squeeze on Tradables”, Oxford Economic Papers 33, 23-46.
De Gregorio, J., A. Giovannini and H. Wolf (1993). “International Evidence on Tradables and
      Nontradable Inflation,”NBER Working Paper No. 4438.
Dornbusch, R. (1980). Open Economy Macroeconomics, Basic Books, New York.
Eaton, J., and M. Gersovitz (1981). “Debt with Potential Repudiation: Theoretical and
      Empirical Analysis”, Review of Economic Studies XLVIII, 289-309.
Ebeke, C. (2014). “Private Saving Accelerations”, IMF WP 14/223.
Eichengreen, B. (2008). “The Real Exchange Rate and Economic Growth”, Working Paper No.
      4, Commission on Growth and Development, World Bank.
Fazzari, S., G. Hubbard and B. Petersen (1988). “Financing Constraints and Corporate
      Investment”, Brooking Papers on Economic Activity I, 141-195.
Fogli, A. and F. Perri (2015). “Macroeconomic Volatility and External Imbalances”, NBER
      WP No. 20872.
Giles, J. and C. Williams (2000). “Export-Led Growth: A Survey of the Empirical Literature
      and Some Non-Causality Results”, University of Victoria.
Gourinchas, P.O., and H. Rey (2013). “External Adjustment, Global Imbalances and Valuation
      Effects”, Handbook in International Economics Vol. IV, Gopinath, Helpman and Rogoff
      eds.
Gourinchas, P.O., and O. Jeanne (2013). “Capital Flows to Developing Countries: The
      Allocation Puzzle”, Review of Economic Studies, 10-17
Guariglia, A., X. Lin and L. Song (2011). “Internal Finance and Growth: Microeconometric
      Evidence on Chinese Firms”, University of Notthingam Discussion Paper 09/11.


                                            21
Harrison, A., and A. Rodriguez-Clare (2009). “Trade, Foreign Investment and Industrial
      Policy”, MPRA Paper 15561.
Hausmann, R., L. Pritchett y D. Rodrik (2005). “Growth Accelerations”, Journal of Economic
        Growth 10 (4): 303-329.
Hsieh, C.T., and P. Klenow (2009). “Misallocation and Manufacturing TFP in China and India”,
      Quarterly Journal of Economics, 124(4), 1403-1448.
Itskhoki O. and B. Moll (2014). “Optimal Development Policies with Financial Frictions”,
      NBER WP 19994.
International Monetary Fund (2013). “The External Balance Assessment (EBA) Methodology”,
      IMF Working Paper 13/272.
Jeanne, O. and A. Korinek (2010). “Excessive Volatility in Capital Flows: A Pigouvian
      Taxation Approach”, American Economic Review 100(2), 403-07.
Johnson, S., J. Ostry and A. Subramanian (2007), “The Prospects for Sustained Growth in
      Africa: Benchmarking the Constraints”, IMF Working Paper 07/52
Kaldor, N. (1958). “Alternative Theories of Distribution”, Review of Economic Studies 23(2),
      83-100.
Korinek, A. and L. Serven (2010). “Undervaluation Through Foreign Reserve Accumulation”,
      World Bank Policy Research WP 5250.
Leland, Hayne E. (1968), “Saving and Uncertainty: The Precautionary Demand for Saving,”
      The Quarterly Journal of Economics, Vol. 82, No.3, pp. 465-473.
Levy Yeyati, E., F. Sturzenegger and P. A. Gluzmann (2013). “Fear of appreciation”, Journal
      of Development Economics 101, 233-247.
Lewis, W. A. (1954). “Economic Development with Unlimited Supplies of Labor”, The
      Manchester School 22(2), 139-191.
Loayza, N., K. Schmidt-Hebbel and L. Serven (2000). “What Drives Saving Across the
      World?”, The Review of Economics and Statistics 82(2), 165-181.
Mendoza, E. (2010). “Sudden Stops, Financial Crises and Leverage”, American Economic
      Review 100, 1941-1966.
Morshed, M. and S. Turnovsky (2004). “Sectoral adjustment costs and real exchange rate
      dynamics in a two-sector dependent economy”, Journal of International Economics 63,
      147-177.
Modigliani, F. (1986). “Life Cycle, Individual Thrift and the Wealth of Nations”, American
      Economic Review 76, 297-313.
Prasad, E., R. Rajan and A. Subramanian (2007). “Foreign Capital and Economic Growth”,
      Cornell University.
Reinhart, C. and K. Rogoff (2013). “Financial and Sovereign Debt Crises: Some Lessons
      Learned and Those Forgotten”, IMF WP 13/266.
Rodrik, D. (2000). “Saving Transitions”, World Bank Economic Review 14(3), 481-507.
Rodrik, D. (2008). “The Real Exchange Rate and Economic Growth”, Brookings Papers on
      Economic Activity, 39(2 (Fall)), 365-439.
Rogoff, K. (1992). “Traded Goods Consumption Smoothing and the Random Walk Behavior
      of the Real Exchange Rate”, NBER WP No. 4119.
Romer, P. (1986). “Increasing Returns and Long-Run Growth”, Journal of Political Economy
      94(5), 1002-37.
Rowthorn, R. (1982). “Demand, Real Wages and Economic Growth”, Studi Economici No. 18,
      3-54.
Schmidt-Hebbel, K. (1998). “Chile’s Takeoff: Facts, Challenges, Lessons”, Working Paper 34,
      Central Bank of Chile.
Stock, J., and M. Watson (1993). “A Simple Estimator of Cointegrating Vectors in Higher
      Order Integrated Systems”, Econometrica Vol. 61 No. 4 pp. 783-820.


                                            22
Turnovsky, S. (2000). “Growth in an Open Economy: Some Recent Developments”,
     Discussion Papers, University of Washington 0015.
Van der Ploeg, R. (2010). “Natural Resources: Curse or Blessing?” CESifo WP No. 3125.
World Bank (2006). “Where is the Wealth of Nations? Measuring Capital for the 21st Century”,
     World Bank, Washington, D.C.
Yang, D.T., J. Zhang and S. Zhou (2011). “Why Are Saving Rates so High in China?” NBER
     Working Paper No. 16771.
Zeldes, Stephen P. (1989). “Optimal Consumption with Stochastic Income: Deviations from
     Certainty Equivalence”, Quarterly Journal of Economics, Vol.104, pp. 275-298.




                                            23
TABLES AND FIGURES




        24
Table 1. Summary Regression Results




                25
Table 2. Sectorial Saving Regressions




                 26
Table 3. Sectorial Saving Shares vs. Domestic Saving Gap




                          27
Table 4. Impacts




       28
                         Figure 1. Channels Linking Saving and Growth




Note: blue arrows represent the real exchange rate (ER) channel; red arrows the real interest rate (IR) channel;
and green arrows the endogenous saving (ES) channel.




                                                       29
                                   Figure 2. Real Exchange Rate and Growth Gaps
                                                     1990-2012

                         1




                       0.5

                                                          HI
  Real Exchange Rate




                                                    SSA
                         0
                                                       LAC
                                                                       EAP
                                                               MNA
                                                                     ECA

                       -0.5




                        -1
                              -7      -5       -3     -1            1        3         5           7
                                                    GDP per capita Growth
♦ LAC 1990-2012 Average ●Other countries per Period ♦Other groups of countries 1990-2012 Average
Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the real exchange rate and GDP per capita growth gaps for the whole sample.
Each period is a three-year average. See Section 3 for the details of how the benchmarks are
calculated.




                                                          30
                              Figure 3. Saving and Real Exchange Rate Gaps
                                                1990-2012


                    2




                    1
  National Saving




                                              EAP
                                                    MNA          HI
                    0
                                             ECA
                                                           SSA
                                                     LAC



                    -1




                    -2
                         -1           -0.5             0                    0.5                       1
                                               Real Exchange Rate
♦ LAC 1990-2012 Average ●Other countries per Period ♦Other groups of countries 1990-2012 Average
Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the national saving and real exchange rate gaps for the whole sample. Each
period is a three-year average. See Section 3 for the details of how the benchmarks are calculated.




                                                     31
                         Figure 4. Real Exchange Rate Gap and Demand Composition
                                                 1990-2012

                         1




                       0.5
  Real Exchange Rate




                                                    LAC    SSA
                                                          MNA
                         0
                                                            EAP

                                                           ECA



                       -0.5




                        -1
                              -2   -1.5   -1   -0.5      0        0.5         1         1.5           2
                                                 Demand Composition
♦LAC 1990-2012 Average ●Other countries per Period ♦Other groups of countries 1990-2012 Average
Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the real exchange rate gaps and the demand composition for the middle
income countries in the sample (World Bank classification). Demand Composition is the ratio of
external demand (exports) to internal demand (consumption plus investment). Each period is a three-
year average. See Section 3 for the details of how the benchmarks are calculated.




                                                    32
                              Figure 5. Saving and Country Rating Gaps
                                              1990-2012


                    1


                                                                            EAP


                                                                  MNA
                    0                                       ECA
                                                                        SSA
                                                                                HI
  National Saving




                                                              LAC




                    -1




                    -2
                         -1             -0.5                            0                             0.5
                                                 Country Rating
♦ LAC 1990-2012 Average ●Other countries per Period ♦Other groups of countries 1990-2012 Average
Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the national saving and country rating gaps for the whole sample. Each
period is a three-year average. See Section 3 for the details of how the benchmarks are calculated.




                                                    33
                              Figure 6. Saving and Current Account Gaps
                                               1990-2012


                    2




                    1
  National Saving




                                            EAP
                                                        HI MNA
                    0                                               ECA
                                                   SSA
                                                          LAC



                    -1




                    -2
                         -1          -0.5                0                    0.5                     1
                                                  Current Account
♦LAC 1990-2012 Average ●Other countries per Period ♦Other groups of countries 1990-2012 Average
Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the national saving and current account gaps for the whole sample. Each
period is a three-year average. See Section 3 for the details of how the benchmarks are calculated.




                                                   34
Figure 7. Saving and Real Exchange Rate Gaps, Median Inflation and Median Rating,
                    Growth, and Investment Gaps by Quadrant
                                    1990-2012

    


                                            II                                                   I

                                                                 Over‐saving
                               0.2    0.8     1.0   5.7                             1.0
                                                                                                             2.8

                                                                                          ‐0.6       ‐0.04
         National Saving Gap




                                            Undervalued                               Overvalued
                                                                 Under‐saving




                                      0.2           6.8
                                                                                    0.4                      3.7
                                             ‐0.5
                               ‐0.7                                                       ‐0.7
                                                                                                     ‐1.3


                                            III                                                  IV


                                                          Real Exchange Rate Gap
                                       ●
                                  Country Rating           ●
                                                          Growth                ●
                                                                         Investment       ●Inflation
   Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
   Notes: This figure is based on Figure 3. It shows the median country rating gap, GDP per capita growth gap,
   investment gap and inflation rate for all the observations in each quadrant for the 1990-2012 period for the whole
   sample.




                                                                35
                           Figure 8. Saving in Latin America: Actual and Benchmarks
                                                    1990-2012

                      22
                      21
                      20
                      19
  in percent of GDP




                      18
                      17
                      16
                      15
                      14
                      13
                      12
                               1990-1992


                                           1993-1995


                                                           1996-1998


                                                                       1999-2001


                                                                                   2002-2004


                                                                                                   2005-2007


                                                                                                               2008-2010


                                                                                                                           2011-2012


                                 Actual                Observed Benchmark                      Equilibrium Benchmark
Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: Sample includes Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, El Salvador,
Guatemala, Mexico, Peru, and Uruguay. Equilibrium (observed) benchmarks are derived from the predicted
values of the saving regression based on the equilibrium (observed) values of the rate of growth and real
exchange rate. See Section 3 for more details on how the benchmarks are calculated.




                                                                         36
Figure 9. Average Saving and Real Exchange Rate Gaps in LAC1 Countries
                              1990–2012

                    0.5
                                  ER high savers                                                  IR high savers
                                                              CHL               VEN
                                                                         PAN
                                                                                                         ARG
                                                              PER     MEX
                      0
                                            ECU
  National Saving




                                              COL
                                                                                                      URY
                                                                                     BRA
                                    IR low savers

                    -0.5



                                                                                      CRI           ER low savers


                     -1
                           -0.4              -0.2                                0                               0.2
                                                     Real Exchange Rate

Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the average national saving and real exchange gaps for LAC1 countries for the 1990-
2012 period. See Section 3 for the details of how the benchmarks are calculated. The dashed line is the linear fit
for the per-period version of the whole sample for 1990-2012.




                                                         37
Figure 10. Macroeconomic Gaps for ER High-Saving and Low-Saving LAC1 Countries
                                   1981–2012




                                     38
  Figure 11. Correlation Between the Saving and Real Exchange Rate Gaps for
                       LAC1 Countries, 1981-2015

    1                                                                                                                    The
          The Descent into IR                                    The Road to ER                                          Return
  0.8                                                                                                                    of IR


  0.6

  0.4

  0.2

    0

 -0.2

 -0.4

 -0.6

 -0.8

   -1
        1981
               1983
                      1985
                             1987
                                    1989
                                           1991
                                                  1993
                                                         1995
                                                                1997
                                                                       1999
                                                                              2001
                                                                                     2003
                                                                                            2005
                                                                                                   2007
                                                                                                          2009
                                                                                                                 2011
                                                                                                                        2013
                                                                                                                               2015

Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: Sample includes Brazil, Chile, Colombia, Costa Rica, Ecuador, Mexico, Panama, Peru, and Uruguay.




                                                                39
          Figure 12. Saving and Real Exchange Rate Gaps in LAC1 Countries
                                      2011-2012

                    0.5                  PAN*

                                  ARG*

                                                                                     VEN

                                                      PER
                                   MEX          CHL
                      0
  National Saving




                                                      ECU
                                                                               CRI

                                                                     COL

                                                                                                              BRA
                    -0.5                                                                URY




                     -1
                           -0.4          -0.2                  0                      0.2                     0.4
                                                       Real Exchange Rate

Sources: Authors' calculations based on UN, WDI and Institutional Investor data.
Notes: This figure shows the national saving and real exchange gaps for LAC1 countries for the 2011-2012
period, except for the data identified with a star: Argentina is 2005-2007 and Panama is 2008-2010. See Section
3 for the details of how the benchmarks are calculated. The dashed line is the linear fit for the per-period
version of the whole sample for 1990-2012.




                                                            40
                                        APPENDIX A

                               The Saving and Growth Model


1.       The growth equation

        Consider a single sector Cobb-Douglas productive sector, such that per capita output,
y, equals:
                          
              Y    K 
         y      A 
              L    L

Or, in rates of growth:




Since:



It follows that:




        Assuming a constant cross-sample average capital output ratio and making productivity
growth a function of the level of economic development, the exchange rate gap and the rating
gap, leads to a reduced-form growth equation such as equation (10) in the text. Since the
investment ratio is expressed in logs, and defining the average investment to GDP percentage
ratio as i* this implies a value of  such that:

              i *
         
              K /Y

        Using for middle-income countries a capital share of 0.4, an average investment to GDP
ratio of 25, and an average capital-output ratio of 2 yields a  of 5.




                                             41
2.       Forward model solution

                                                                             1
       Through straightforward algebraic manipulation, the cells of ( I  Q ) in (18) can be
expressed as in Table A1:

                                    Table A1. Forward Model Solution
                             S               i                     E                   r                  g
        s                  1-Zs             uZs               3  2 Zs           3   2 Zs          s)
        i                   -Zi            1+uZi               i   2 Zi         4   2 Zi           i
        e                  -Ze              uZe              1  2 Ze             e   2 Ze           e
        r                  -Zr              uZr                 2 Zr             1   2 Zr             r
        g                  -Zg             uZg             1   2 Z g          1   2Zg           g

Where:

     1   g  i                                             Zg  A / 

      2   i   s  1                                    Z s  ( s   A ) / 

      3   s  1                                          Zi  ( i   i r ) / 

      1   g   g e   4                                  Ze   / 

      2  u 4   s e   ( g   e g )                   Zr   r / 

      3   1   s e                                       A  1   r 5

      4   i   i e                                          1   A   e ( i   s )

      5   g   i                                           '   A

         e   e r                                            u  1  


3.       Backward model solution using instrumental variables

        Separating the matrix of exogenous elasticities, R, into two components, one (V) related
to the controls used as instruments, Yt k and the other (V’) to the rest of the controls, equation
(18) in the text can be rewritten as:

                 ˆ k  ( I  Q )1[VY k  (V ' SR ')Y k ]
                 X                                                                               (A1)
                   t                 t                t


         Writing the reduced-form estimates as:


                                                       42
                                            ˆ k  AY k  BY k
                                            X                                                                                          (A2)
                                              t     t      t


                                It immediately follows that:

                                          ( I  Q )1V  A                                                                             (A3)

                             We include in V all significant non-diagonal structural-form instrumental coefficients
                    and solve (A3) using the significant reduced-form instrumental coefficients in A. For an
                    elasticity to be retrievable (i.e., to ensure that the elasticity measures the effect of the
                    endogenous variable rather than that of its instrument)), its reduced-form instrumental
                    coefficient should be significant and the number of significant reduced-form instrumental
                    coefficients in the matrix row corresponding to this elasticity should exceed the number of
                    significant structural-form instrumental coefficients by at least the number of elasticities to be
                    determined in that row (see Tables A2 and A3).28




                                    Table A2. Instrumental Coefficients, Structural and Reduced Forms

                                          SAV                             INV                           REER                           RAT                           GRO
                                    ν              a               ν               a              ν               a              ν              a               ν            a
r_OldDepend CA<0              -0.0186**       -0.0201**       -0.00480       0.00156         -0.00407       -0.00286        -0.00103       -0.000962      -0.0354        -0.0294

r_OldDepend CA>0              -0.0232**       -0.0301***      -0.000740      -8.26e-05       0.00303        0.00592         -0.00225       -0.00289       0.0388         0.0663*

r_Inflation CA<0              -0.106          0.0364          -0.207**       -0.118+         -0.194**       -0.287***       -0.164**       -0.240***      -0.327         -0.689+

r_Inflation CA>0              0.0596          0.224*          -0.249*        -0.143+         -0.0216        -0.127*         -0.0954        -0.0851        -0.836+        -1.478***

r_LTradOpen                   0.222***        0.288***        0.0780*        0.124***        -0.153***      -0.168***       0.0439         0.0517+        0.802**        1.119***

r_EnabEnv                     0.0820          0.0521          -0.0535        -0.00481        0.0837**       0.0701*         0.201***       0.167***       -0.0372        0.586**

r_PopGrowth                     -0.0442         -0.104**       -0.0141           -0.0145       0.0778*** 0.0773*** 0.0160                     -0.00250       -0.873***   -1.095***
*** p<0.01, ** p<0.05, * p<0.1, + p<0.2
Notes: The coefficients in the colums entitled 'ν' correspond to the structural forms, and those under the columns entitled 'a' correspond to the reduced forms.




                    28
                       To ensure that we do not unduly discard structural-form instrumental coefficients, we adopt a high (20%)
                    significance threshold.


                                                                                         43
              Table A3. Instrumental Coefficients and Reduced Form Elasticities29




          The following four retrievable elasticities are thereby obtained:


        asg   s aeg                                                  r
                                                                      ag  1ae
                                                                              r
            g                                                g 
            ag                                                           arr

                                                              e               1
                                                     
        [1   ei   (aii   ei ag
                                       i
                                         )]                 1   e  s (ai  as ) e  ae
                                                                          i    i         i




            i
           ae            1                                              
e                                                     
       aii  as
              i
                1   (aii  i as
                                 i
                                   )                         1   e ( s  i  1 )




29
   Since the current account equals investment minus saving, e and r can be retrieved from the investment side
or the saving side.


                                                      44
                                        Appendix B

                                   Statistical Appendix




                        Table B1. Country Group Composition



Region   Countries
 LAC 1   Argentina, Bahamas, Barbados, Brazil, Chile, Colombia, Costa Rica, Ecuador, Mexico,
         Panama, Peru, Trinidad, Uruguay, República Bolivariana de Venezuela.
LAC 2    Belize, Bolivia, Dominican Republic, El Salvador, Guatemala, Guyana, Honduras,
         Nicaragua, Paraguay.
 EAP     Bangladesh, Bhutan, Cambodia, China, Fiji, Hong Kong SAR, China, India, Indonesia,
         Korea, Rep., Malaysia, Pakistan, Papua New Guinea, Philippines, Sri Lanka, Thailand,
         Tonga, Vietnam.
 ECA     Albania, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech
         Republic, Estonia, Georgia, Greece, Hungary, Kazakhstan, Kyrgyz Republic, Latvia,
         Lithuania, Macedonia, Moldova, Mongolia, Romania, Slovenia, Tajikistan, Turkmenistan,
         Ukraine.
 High    Australia, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Iceland,
Income   Ireland, Israel, Italy, Japan, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland,
         United Kingdom, United States.
 MNA     Algeria, Iran: Islamic Rep., Jordan, Lebanon, Morocco, Syrian Arab Republic, Tunisia,
         Turkey.
 SSA     Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Chad, Côte d’ Ivoire,
         Equatorial Guinea, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Kenya,
         Lesotho, Madagascar, Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Niger,
         Rwanda, Senegal, South Africa, Sudan, Swaziland, Togo, Uganda, Zambia.




                                              45
                         Table B2. Data Definitions and Sources

Variable      Description                                                                          Source
LDomSavings   Domestic saving as a share of GDP. Expressed in logs . Domestic saving is gross      UN data and World Development
              national disposable income (GDP plus net factor income and net unrequited            Indicators
              transfers). Gross Savings/GDP from UN; net factor payments/GDP and net
              unrequited transfers/GDP from WDI.
LInvestment   Investment as a share of GDP. Expressed in logs.                                     UN data


LRating       Country risk rating. Expressed in logs.                                              Institutional Investor database


GDPpcGrowth   Per capita income growth rate.                                                       World Development Indicators

LExcRate      Ratio of the purchasing power parity conversion factor to the nominal exchange       World Development Indicators
              rate with respect to the US dollar. The series used is the one which was available
              before the mid 2014 update. Expressed in logs.

LInv-LSav     Current Account calculated as the difference between investment and domestic         UN data and World Development
              saving. Expressed in logs                                                            Indicators

LGDPpc        Per capita income. Expressed in logs.                                                World Development Indicators

OldDepend     Ratio of old in the work population to total work population.                        World Development Indicators

Population    Total population.                                                                    World Development Indicators
PopGrowth     Rate of population growth.                                                           World Development Indicators

FuelExp       Oil exports as a share of GDP. Fuel Exports from WDI with missing data filled        World Development Indicators
              through linear prediction using WB Wealth of Nations data.                           and Wealth of Nations
TradOpen      Ratio of imports plus exports to GDP. Expressed in logs.                             World Development Indicators

CapOpen       Capital openness index                                                               Chinn and Ito (2006)
Transf        Net unrequited transfers as a share of GDP.                                          World Development Indicators
LPubCons      Public consumption as a share of GDP. Expressed in logs.                             World Development Indicators
FiscBal       Fiscal balance as a share of GDP.                                                    The Economist Countries Profiles


EnabEnv       Simple average of the Corruption and Rule of Law Indexes.                            World Governance Indicators
Inflation     Inflation crisis dummy.                                                              Reinhart and Rogoff (2011)

ExtDebt       External debt crisis dummy.                                                          Reinhart and Rogoff (2011)

RiskAv        Risk Appetite-Safe Haven: calculated as the VIX times a safe haven dummy that is     VIX and S&P-500
              equal to one for the U.S., Japan and Switzerland, and zero for the rest of the
              countries. VIX data extrapolated backwards using the S&P-500 index.
ToT           Terms of trades expressed in logs. Missing data from WDI completed through           World Development Indicators
              smooth pasting with IFS data.                                                        and International Financial
                                                                                                   Statistics
FinCorp       Savings of the financial corporate sector divided by total domestic savings          UN National Accounts Main
                                                                                                   Aggregates database.
NonFinCorp    Savings of the non-financial corporate sector divided by total domestic savings      UN National Accounts Main
                                                                                                   Aggregates database.

                                                                                                   UN National Accounts Main
Households    Household savings divided by total domestic savings
                                                                                                   Aggregates database.
Government    Government savings divided by total domestic savings                                 UN National Accounts Main
                                                                                                   Aggregates database.
ForSavings    Savings of the foreign sector divided by total domestic savings                      UN National Accounts Main
                                                                                                   Aggregates database.




                                                      46
                        Table B3. Data Statistics

   Variable    Mean     Standard Deviation   Minimum    Maximum
 DomSaving     21.03          11.25           -16.60      80.96
 Investment    23.01            7.40            4.74      67.88
GDPpcGrowth     2.53            3.58           -8.00      36.30
   ExcRate     61.60           29.33           18.71     175.42
   GDPpc     11103.03        11162.58         272.22    48352.78
 OldDepend     12.08            7.34           4.30       38.27
 PopGrowth      1.49            1.12           -1.71       6.95
   FuelExp      5.13           13.45            0.00     150.64
     ToT        4.63            0.23            3.33       5.56
  TradOpen      0.78           0.46            0.09        4.47
  CapOpen       0.39           1.58            -1.86       2.44
    Transf      3.74            6.00           -3.52      44.87
   PubCons     15.11            5.56           2.89       38.01
    FiscBal    -2.33            3.97          -23.39      23.69
    PrivCre    53.36           48.85            1.25     276.25
    Rating     45.90           25.07            8.20      97.23
   FinCorp      1.22            6.17          -29.73      19.18
 NonFinCorp     1.69            1.16           -2.14       7.02
 Households    11.37            4.48           -0.18      27.81
 Government     6.31            7.05          -18.16      29.74
 ForSavings     1.22            6.17          -29.73      19.18
  Population 5.20e+07       1.60e+08         99697.67   1.35e+09
  EnabEnv      0. 13            1.02           -1.53       2.23
   Inflation    0.08           0.24              0           1
   ExtDebt      0.18            0.35              0          1
    RiskAv      0.49            3.11             0        26.48




                                     47
                                                    Appendix C

                                             Full Regression Results



                               Table C1. Structural Saving Regressions30

                              Full Sample         Full Sample         Low Income   Middle Income High Income
                                   (1)                 (2)                (3)           (4)          (5)
                                 OLS                DOLS                 OLS           OLS          OLS

GDPpcGrowth                     0.0251**            0.0636*            0.00960       0.0314**         0.0380**

LExcRate                        -0.324**           -0.343**            -1.016**      -0.247+           -0.0232

LGDPpc                          0.340***           0.356***            0.739***      0.170**            0.294

Fuel Exp                       0.0157***           0.0160***           0.00626      0.0163***          0.0210*

OldDepend                      -0.0214***         -0.0289***           -0.0418      -0.0361***        -0.000596

PopGrowth                        -0.0544           -0.0876*             0.227+       -0.117+           0.00225

FiscBal                        0.0201***           0.0211***           -0.0194       0.0183*           0.0104+

Transf                         -0.000544           -0.00291             0.0150       -0.00824          -0.0375

ToT                              -0.300*            -0.417**           -0.00339      -0.403**          -0.520**

Constant                          1.278             1.876+             -4.580**      3.614***           2.151

Observations                       780               1,804               117           471               192
R-squared                         0.336              0.354              0.270         0.278             0.424
r2_a                              0.319              0.333              0.127         0.248             0.360
*** p<0.01, ** p<0.05, * p<0.1, + p<0.2
Notes: Statistical significance is reported for country clustering.




30
  The negative sign of the terms of trade reflects the fact that each component of aggregate demand (GDP) is
deflated by its own deflator. As the current account is calculated based on “real” exports and imports, national
incomes do not reflect the gains and losses caused by terms of trade changes. Instead, saving falls as consumption
tends to rise following terms of trade gains.




                                                           48
                           Table C2. Structural Real Exchange Rate Regressions
   
                                      Full Sample        Full Sample   Low Income   Middle Income High Income
                                           (1)                (2)          (3)           (4)          (5)
                                         OLS               DOLS           OLS           OLS          OLS

LRInv-LRSav CA>0                         0.0414             0.0455      0.190***      0.0717+       -0.0402

LRInv-LRSav CA<0                       0.138***            0.130***     0.0609+       0.0845**      0.0728+

LRRating                               0.208***            0.185***      0.136        0.146***     0.241***

LRGDPpc                                 0.186***           0.191***      0.149+       0.145***     0.412***

LRToT                                    0.385              0.746*       0.221        0.681**       0.667*

LRTradOpen                            -0.0974***          -0.0846**     -0.0356        -0.0573      0.0651

RiskAv                                 0.00972+           0.00896+                                 0.0107**

Inflation                              -0.181***          -0.193***     -0.00594       -0.124*      0.00275

RCapOpen                              0.000402+          0.000595**    -0.00178**    0.000419+    -0.000531+

LRPubCons                               0.130***           0.128***     -0.0690        0.0639       0.353***

Constant                                 0.0471             -0.0559      -0.155       -0.177**      0.145+

Observations                              807                2,287        123           491           193
R-squared                                0.734               0.721       0.546         0.519         0.721
r2_a                                     0.727               0.714       0.462         0.499         0.688
*** p<0.01, ** p<0.05, * p<0.1, + p<0.2
Notes: Statistical significance is reported for country clustering.                                             
   




                                                               49
                             Table C3. Structural Investment Regressions

                             Full Sample        Full Sample           Low Income   Middle Income High Income
                                  (1)                (2)                  (3)           (4)          (5)
                                OLS               DOLS                   OLS           OLS          OLS

LExcRate                      -0.357***          -0.228***               0.133       -0.301***      0.0115

LRating                       0.263***            0.148**               0.290+       0.270***       0.154

LGDPpc                         0.00219             0.0196               0.0177        -0.0114      -0.0537

Inflation                     -0.218***           -0.158**             -0.555***     -0.180***     -0.00969

Constant                       1.997***           2.191***             2.332***      2.159***      2.819**

Observations                     813                2,877                128           492           193
R-squared                       0.195               0.171               0.473         0.181         0.184
r2_a                            0.181               0.159               0.408         0.157         0.120
*** p<0.01, ** p<0.05, * p<0.1, + p<0.2
Notes: Statistical significance is reported for country clustering.
                                                             



                                  Table C4. Structural Growth Regressions

                             Full Sample        Full Sample           Low Income   Middle Income High Income
                                  (1)                (2)                  (3)           (4)          (5)
                                OLS               DOLS                   OLS           OLS          OLS

LInvestment                    3.181***           2.184***              2.257*       3.379***      1.555**

LExcRate                      -1.846***          -1.324***              -0.180       -1.836***     -1.440+

LRating                        1.117***           2.082***              0.872        0.984**       2.018+

LGDPpc                        -0.671***          -0.821***              -0.833        -0.389       1.131+

PopGrowth                     -0.849***          -0.500***             -1.221**      -0.831***     0.0926

Constant                       -5.514**           -4.493**              1.990         -8.041*     -25.00***

Observations                     813                2,857                128           492           193
R-squared                       0.340               0.361               0.244         0.360         0.576
r2_a                            0.327               0.352               0.143         0.340         0.540
*** p<0.01, ** p<0.05, * p<0.1, + p<0.2
Notes: Statistical significance is reported for country clustering.




                                                            50
                          Table C5. Structural Country Rating Regressions

                              Full Sample         Full Sample         Low Income    Middle Income High Income
                                   (1)                 (2)                (3)            (4)          (5)
                                 OLS                DOLS                 OLS            OLS          OLS

LInv-LSav CA>0                   0.0733              0.0739            0.112***        0.0541       0.118+

LInv-LSav CA<0                 -0.102***           -0.0978**           -0.00433       -0.178***    -0.112***

LGDPpc                          0.192***           0.197***             0.159*        0.234***      0.182**

CapOpen                        0.0308***           0.0382***            0.0181        0.0283*      0.0675***

TradOpen                        -0.0382+            -0.0295             0.150         -0.0551+      -0.0625

LPubCons                        0.00782              0.0490             0.0871         0.0246       0.0588

FiscBal                         0.00393             0.00464            0.00374        0.00536      0.00412+

Inflation                       -0.118**            -0.129**            -0.168        -0.100+       -0.215*

ExtDebt                        -0.395***           -0.400***           -0.366***      -0.413***    -1.917***

EnabEnv                         0.193***           0.181***            0.162**        0.218***     0.0809**

Population                    5.70e-10***         4.92e-10***         2.20e-09***    5.61e-10***   4.75e-10+

Constant                        2.071***           1.975***            1.822**        1.743***      2.030**

Observations                       807               2,207               123            491           193
R-squared                         0.884              0.875              0.753          0.776         0.796
r2_a                              0.881              0.872              0.702          0.766         0.771
*** p<0.01, ** p<0.05, * p<0.1, + p<0.2
Notes: Statistical significance is reported for country clustering.




                                                             51