Policy Research Working Paper 9627 Social Transfer Multipliers in Developed and Emerging Countries The Role of Hand-to-Mouth Consumers Jessica Bracco Luciana Galeano Pedro Juarros Daniel Riera-Crichton Guillermo Vuletin Latin America and the Caribbean Region Office of the Chief Economist April 2021 Policy Research Working Paper 9627 Abstract This paper estimates the macroeconomic effects of social is 23 percent in developed economies versus 60 percent in transfer payments to individuals for a sample of 23 devel- Latin American countries. This evidence is interpreted with oped and Latin American countries. The findings show that a two-agent New Keynesian model. The findings show that the social transfer multiplier is 0.3 in developed countries, the difference in the share of hand-to-mouth consumers but 0.9 in Latin American economies. The paper studies explains 80 to 90 percent of the difference in the estimated the role of hand-to-mouth consumers, who have no access social transfer multipliers. The paper also documents that to financial markets and a high marginal propensity to con- the share of hand-to-mouth individuals in emerging coun- sume, as a first order factor to explain the heterogeneity in tries is in general 47 percent which suggests that a larger the size of social transfer multipliers. Using survey-based social transfer multiplier may be expected for this type of data from the Global Findex dataset, the paper finds that economy. the average share of the population living hand-to-mouth 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://www.worldbank.org/prwp. The authors may be contacted at gvuletin@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 Social transfer multipliers in developed and emerging countries: The role of hand-to-mouth consumers Jessica Bracco Luciana Galeano Universidad Nacional de La Plata University of Michigan Pedro Juarros Daniel Riera-Crichton Georgetown University World Bank Guillermo Vuletiny World Bank April 13, 2021 Abstract This paper estimates the macroeconomic e¤ects of social transfer payments to individuals for a sample of 23 developed and Latin American countries. The …ndings show that the social transfer multiplier is 0.3 in developed countries, but 0.9 in Latin American economies. The paper studies the role of hand-to-mouth consumers, who have no access to …nancial markets and a high marginal propensity to consume, as a …rst order factor to explain the heterogeneity in the size of social transfer multipliers. Using survey-based data from the Global Findex dataset, the paper …nds that the average share of the population living hand-to-mouth is 23 percent in developed economies versus 60 percent in Latin American countries. This evidence is interpreted with a two-agent New Keynesian model. The …ndings show that the di¤erence in the share of hand-to-mouth consumers explains 80 to 90 percent of the di¤erence in the estimated social transfer multipliers. The paper also documents that the share of hand-to-mouth individuals in emerging countries is in general 47 percent which suggests that a larger social transfer multiplier may be expected for this type of economy. Keywords: …scal transfer, social transfer, …scal policy, …scal stimulus, government spending, multiplier, hand-to-mouth, emerging markets. JEL Classi…cation: E62, E63. We would like to thank Martin Ardanaz, Leopoldo Avellan, Yan Carrière-Swallow, Fernando Castaño, Manuela Cerimelo, Antonio David, Guillermo Falcone, Roberta Gatti, Margaret Grosh, Chadi Habib, Ale- jandro Izquierdo, Norman Loayza, Nora Lustig, Jorge Neyro, Ha Nguyen, Steven Pennings, Carola Pessino, Emilio Pineda, Francisco Pizzi, Andrew Powell, Martin Rama, Diego Rojas, Oscar Valencia, and Lucila Venturi for helpful discussions. We would also like to thank seminar participants at the World Bank, the Inter-American Development Bank, and Universidad Nacional de La Plata. We also thank Ivana Benzaquen, Jose Andree Camarena, and Irene Ezran for excellent research assistance. The views expressed are those of the authors and do not necessarily represent the views of the World Bank, its Executive Board, or its management. y Corresponding author. Email: gvuletin@worldbank.org. Website: www.guillermovuletin.com 1 Introduction The macroeconomic e¤ects of …scal policy regained great interest since the Global Financial Crisis. This relevance has been recently redoubled as governments around the world evaluate the use of alternative …scal instruments at their disposal to cope with the COVID-19 pandemic. A category of spending that has been receiving particularly growing attention is social transfers. Social transfers comprise both ongoing social protection programs and emergency policy responses. Ongoing social protection programs refer to the disbursement of government funds to individuals who meet certain eligibility criteria. The main categories include unemployment bene…ts, family programs, and pensions. Overall, social transfers have been increasing continuously over time mainly due to the impact of aging populations on pensions as well as a consequence of expanding family programs targeting the poor and most vulnerable households (especially in the emerging world). In fact, on average, social transfers currently account for more than 50 percent of primary government spending in developed countries and about 40 percent in emerging markets (Galeano et al., 2021). In terms of emergency social transfers, for example, some current policy responses to COVID-19 have proven to be among the largest in history.1 In spite of this growing relevance, little is known about the e¤ect that a $1 change in social transfers has on the aggregate level of GDP –the so-called social transfer multiplier (STM, hereafter).2 What is the current state of the literature on the size of the STM and the mechanism behind it? There are a handful of empirical studies for developed countries and no evidence of any kind for emerging markets. Existing empirical studies vary among several critical dimensions including the identi…cation strategy, scope of the social transfer metric, econometric methodology, and coun- try(ies) analyzed. Therefore, it is no surprise that there is little agreement in terms of the size of the STM. Estimates of STMs range between zero and one (Gechert, 2015; Alesina et al., 2017; Parraga-Rodriguez, 2018; Gechert et al., 2020). Interestingly, among the few existing papers, there is a strong consensus that social transfer shocks a¤ect output mainly through consumption rather than through investment (Romer and Romer, 2016; Alesina et al., 2017; Parraga-Rodriguez, 2018; Gechert et al., 2020; Pennings, 2020). This empirical fact points out that the primary mechanism behind the social transfer shock occurs through the government allocation of funds to agents with a high marginal propensity to consume, rather than through supply-side channels. On the theoretical front, the literature has mainly relied on di¤erent variants of closed-economy two-agent New Keynesian (TANK) models which prompt consumption to play a dominant role in 1 For example, the …rst phase of the Coronavirus Aid, Relief, and Economic Security (CARES) Act in the United States included one-time tax rebates to individuals, expanded unemployment bene…ts, and the Supplemental Nutrition Assistance Program (SNAP) social transfers that represented about 586 billion dollars, or 2.8 percent of the GDP of 2019. This social transfer amount is similar to that spent on the American Recovery and Reinvestment Act (ARRA) of 2009 (Oh and Reis, 2012). 2 Throughout this paper, and unless noted otherwise, we therefore reserve the term STM to refer to the aggregate (as opposed to the local/regional) e¤ect of a change in social transfers. 2 response to a social transfer shock (Monacelli and Perotti, 2011; Coenen et al., 2012; Giambattista and Pennings, 2017; Mehrotra, 2018). The distinctive and essential elements of this family of models are the existence of two types of agents that di¤er in their access to …nancial markets coupled with a …scal authority capable of redistributing funds between these two groups of individuals.3 While unconstrained Ricardian agents have access to …nancial markets and are, thus, able to smooth consumption, constrained hand-to-mouth (HtM, hereafter) individuals consume their entire income in each period and, consequently, have a higher marginal propensity to consume. The government collects lump-sum taxes from the Ricardian agents to pay for government purchases and social transfers to individuals. Naturally, if social transfers only reached Ricardian agents or if there were no HtM individuals, the STM would be equal to zero. The key to deliver a positive STM relies on social transfers actually reaching HtM individuals (i.e., the social transfer shock needs to redistribute funds from low- to high-marginal-propensity-to-consume agents) along with the existence of HtM agents (who help propagate the e¤ect of the initial social transfer shock). Moreover, TANK models deliver STMs that are larger the higher the share of HtM agents in the population (which increases the average marginal propensity to consume of the economy) and the higher the share of social transfers reaching HtM agents (which increases the redistribution of funds from Ricardian to HtM agents). Among other relevant features, these models also allow monetary policy to be more or less accommodative (the zero lower bound being the extreme case of the former) and …scal policy to vary the degree of persistence of government spending shocks. When studying the key determinants a¤ecting the size of STMs, existing papers have focused on the importance of the persistence of the social transfer shock (Coenen et al., 2012; Romer and Romer, 2016; Alesina et al., 2017; Gechert et al., 2020; Pennings, 2020) and on the magnifying role of the more accommodative monetary policy on the STM (Coenen et al., 2012; Romer and Romer, 2016; Giambattista and Pennings, 2017; Mehrotra, 2018). Despite the above-mentioned progress, to the best of our knowledge, no study has focused on how the heterogeneity in the share of HtM agents and in the share of social transfers reaching HtM agents a¤ect the size of STMs. Existing quantitative papers estimating STMs for the United States (Giambattista and Pennings, 2017; Pennings, 2020) discipline the share of HtM agents to one-third based on Kaplan et al. (2014). Kaplan et al. (2014) relied on household surveys for the United States and eight other developed countries to measure the share of HtM agents based on liquid net worth, …nding modest heterogeneity among developed countries.4 Notably, no paper uses data to discipline the share of social transfers reaching HtM agents. It is generally assumed that social transfers solely reach HtM agents, which is referred to as perfect social transfers targeting. However, we document that this 3 The in‡ uential paper by Oh and Reis (2012) was the …rst to call attention to the need to have heterogeneous agents coupled with redistribution arguments to deliver a positive STM. 4 Kaplan et al. (2014) …nd that the shares of HtM households (expressed in percentage terms) are as follows: Australia (19), Canada (30), France (21), Germany (32), Italy (24), Spain (20), United Kingdom (33), and United States (31). 3 assumption does not hold, even in developed countries. Since both the share of HtM agents and the share of social transfers actually reaching them are …rst-order aspects of the mechanism embedded behind the e¤ect of a social transfer shock, we address the limitations of the existing literature by considering their heterogeneity in a more global sample (considering developed and emerging economies alike). For this purpose, we focus on a sample of six Latin American countries and 17 developed economies. We decided to include emerging market economies, such as these six Latin American countries, because they o¤er an ideal laboratory to further understand the implications played by the allocation of social transfers and, especially, by HtM individuals with no access to …nancial markets. In particular, we contribute to the growing literature on the size and acting mechanism of STMs on three speci…c fronts: We provide new empirical evidence on the size of STMs for a range of countries based on a consistent methodology. In particular, we use the well-known Blanchard and Perotti (2002) identi…cation strategy that imposes timing restrictions by assuming that while government spending changes are allowed to contemporaneously a¤ect economic activity within the quar- ter, it takes the government at least one quarter to respond to developments in the state of the economy. Although the identi…cation of exogenous shocks continues to be a serious challenge for most empirical studies, we make signi…cant e¤orts to ameliorate endogeneity concerns and further guarantee the unanticipated within-the-quarter nature of the social transfer shock in the context of a local projections approach (Jorda, 2005; Stock and Watson, 2007). In partic- ular, we instrument cumulative changes in social transfers at each time horizon t + h using the residual at time t of a regression of changes in social transfers (excluding the unemployment insurance spending component) on the lags of a long list of macroeconomic variables including the changes of social transfers, GDP, total primary spending, …scal revenues, and central bank interest rates. We exclude the unemployment insurance spending component from the social transfer metric used as instrument because of its inherent automatic and rapid response to developments in the state of the economy (McKay and Reis, 2016; Di Maggio and Kermani, 2016; Galeano et al., 2021). We also perform several other exercises to reduce concerns about omitted variable bias. We …nd that the size of the STM is three times larger in Latin Amer- ican countries than in developed economies. While the STM is on impact 0.3 in developed countries, it is 0.9 in Latin American economies. In line with existing empirical papers based on data for developed countries, both samples also show that the e¤ect on output is mainly driven by private consumption whereas private investment remains largely unchanged. We also …nd that while the macroeconomic impact of social transfers is important in the short- and medium-term, it tends to weaken in the long-term. This is the …rst paper to provide evidence on the size and mechanism behind the aggregate STM for emerging markets. 4 We provide novel stylized facts on the share of HtM agents and the share of social trans- fers reaching them using survey-based dataset from Global Findex. We identify HtM agents in‡uenced by the work of Lusardi et al. (2011) as those who are not able to cope with a …nancial shock.5 The recipients of social transfers are identi…ed directly from Global Findex. Two …ndings stand out from several interesting ones.6 First, the share of HtM agents is, on average, 23 percent in the sample of developed countries and 60 percent in the Latin American sample.7 Second, the share of social transfers reaching HtM agents is, on average, 25 percent in the developed sample and 65 percent in the sample of Latin American countries. That is to say, the share of social transfers reaching HtM agents is mostly a re‡ection of the share of HtM agents, rather than the result of a particular social transfers targeting ability. This result suggests that, at least when including all types of social transfers, the usual assumption that social transfers only reach HtM agents (known as perfect targeting) does not seem to hold. We also show that these empirical regularities equally hold for a sample of 99 developed and emerging countries. A calibrated standard two-agent New Keynesian (TANK) model is able to match key STM regularities identi…ed empirically. Moreover, we show that about 80-90 percent of the large di¤erence in the size of STMs between Latin American and developed countries is explained by the higher share of …nancially constrained individuals who live HtM in the Latin American sample relative to the developed countries. This evidence on the quantitative e¤ect of HtM arguments, coupled with the strong evidence supporting a larger share of HtM agents in emerg- ing markets, suggests that a bigger STM may be expected for emerging market economies in general. The rest of the paper proceeds as follows. Section 2 shows the empirical estimates of STMs for both the developed and the Latin American samples. Section 3 presents the TANK model. Section 4 provides evidence that the share of HtM agents and the share of social transfers actually reaching them in emerging markets are much larger than in developed economies. Section 5 shows that a 5 As discussed in Kaplan et al. (2014), while the identi…cation of HtM agents is di¤erent from those studies based on the agents’liquid wealth, the empirical evidence is notably similar in terms of the shares of population living HtM. This robust …nding is also present in our study when relying on Global Findex. For example, for the United States, Kaplan et al. (2014) …nd a share of HtM ranging between 25 and 40, with their preferred estimate being one-third. Ours, using Global Findex, is 27 percent. Section 4 provides extensive evidence of these similarities. 6 In light of the heated debate in developed countries about the so-called wealthy HtM individuals (i.e., those with signi…cant amounts of illiquid assets like real estate properties, yet high marginal propensity to consume out of transitory changes in income), we …nd evidence that wealthy HtM individuals also exist in emerging markets. Yet, the HtM phenomenon in emerging markets is largely driven by poor HtM individuals. 7 This evidence is also in line with well-established, more macro/aggregate, evidence that emerging market economies have less …nancial depth, intermediation, and development than their developed counterparts (e.g., King and Levine, 1993; Beck et al., 2000; Beck et al., 2007). More recent micro-based evidence from individual spending data and local-based experiments also support, to di¤erent degrees of explicitness, the relevance of …nancial fragility on the marginal propensity to consume and on the size of local STMs (e.g., Haushofer and Shapiro, 2016; Egger et al., 2019; Pennings, 2020; Gerard et al., 2020). 5 standard calibrated TANK model is able to match key empirical regularities shown in Section 2 in terms of the size, main macroeconomic variables involved, and temporal pro…le of the STM for both the developed and Latin American samples. Section 6 o¤ers some …nal thoughts. 2 Empirical evidence on the Social Transfer Multipliers Social transfers comprise both ongoing social protection programs and emergency policy responses.8 Ongoing social protection programs refer to the disbursement of government funds to individuals who meet certain eligibility criteria. The main categories include unemployment bene…ts (transfers to unemployed individuals), family programs (transfers essentially to the poor and most vulnera- ble households), and pensions (mainly transfers to individuals after retirement). Social transfers represent a large component of government spending comprising more than 50 percent of primary government spending in developed countries and about 40 percent in emerging markets. By and large, the largest category of social transfers spending is pensions (accounting, on average, for more than 75 percent of social transfers), followed by family programs (about 20 percent, on average). Interestingly, in contrast to the emerging world’s long history of social protection in terms of social security and the most recent wave of family programs, unemployment insurance programs are rare in emerging countries (they simply do not exist or, if they do, they have negligible coverage). In terms of their short-term ‡uctuating nature, unemployment insurance spending is, by design, coun- tercyclical (even at the quarterly frequency!).9 On the other hand, pensions and family programs spending tend to be more rigid and slow-moving in nature and are mainly driven by demographic and structural deep-rooted social problems, respectively.10 This section estimates STMs using unbalanced quarterly data for the period 1960:Q1 through 2019:Q4 for six Latin American countries (Argentina, Brazil, Chile, Colombia, Mexico, and Peru) and 17 developed countries (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ire- land, Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, United Kingdom, and United States).11 2.1 Methodology and identi…cation strategy We use the well-known Blanchard and Perotti (2002) identi…cation strategy that imposes timing restrictions by assuming that while government spending changes are allowed to contemporaneously a¤ect economic activity (i.e., within the quarter), it takes the government at least one quarter to 8 This paragraph signi…cantly relies on Galeano et al. (2021). See their paper for a detailed discussion of social transfers basic trends, composition, rigidity, and business cyclical aspects of social transfers. 9 In fact, the contemporaneous correlation between quarterly changes of real GDP and real unemployment insurance spending is -0.23 (statistically signi…cant). Pennings (2020) also provide some similar evidence for the United States. 10 In fact, the contemporaneous correlation between quarterly changes of real GDP and real pensions and family programs spending is 0.09 (statistically not signi…cant). 11 See Appendix 1 for description of data de…nitions, sources, and time coverage for each country. 6 respond to developments in the state of the economy (e.g., Fatás and Mihov (2003), Galí et al. (2007), Corsetti et al. (2012), Ilzetzki et al. (2013), Huidrom et al. (2020)). We estimate the e¤ect of innovations of social transfer shocks on economic growth using the single-equation approach proposed by Jorda (2005) and Stock and Watson (2007), which is based on linear local projections (LP). The use of LP provides several advantages over the traditional structural vector autoregressive (SVAR) methodology. Speci…cally, LP (i) can be estimated by single-regression techniques, (ii) are more robust to potential misspeci…cations, and (iii) can easily accommodate non-linear speci…cations that may be impractical in a multivariate SVAR context (a feature that proves to be crucial in this paper).12 While the identi…cation of exogenous shocks is a challenge for most empirical studies, we make signi…cant e¤orts to ameliorate endogeneity concerns and further guarantee the unantici- pated within-the-quarter nature of the social transfer shock in the context of a local projections approach (Jorda, 2005; Stock and Watson, 2007). Our empirical strategy follows a three-step pro- cedure. First, we estimate the unanticipated social transfer shocks (excluding the unemployment insurance spending component) using the residuals of the following regressions: wo ui PL PL wo ui PL PL STi;t = l=1 & l Yi;t l + l=1 l STi;t l + l=1 l Gi;t l + l=1 l Ri;t l + PL wo ui + l=1 l inti;t l + "ST i;t ; (1) for each country i, where Xi;t = (Xi;t Xi;t 1 )=Yi;t 1 and X being real social transfers excluding the unemployment insurance spending component (ST wo ui ), real GDP (Y ), real total primary spending (G), real …scal revenues (R), or central bank interest rates (int).13 We exclude the unemployment insurance spending component from the social transfer metric used as instrument because of its inherent automatic and rapid response to developments in the state of the economy (McKay and Reis, 2016; Di Maggio and Kermani, 2016; Galeano et al., 2021).14 Second, once we identify the unanticipated social transfer shocks excluding the unemployment wo ui insurance spending component "ST , we use a local projections approach to …nd a cumulative impulse response function (IRF) representing the STM at di¤erent time horizons. This methodology allows us to directly project the behavioral reaction of GDP to the unanticipated social transfer 12 We should note that Jorda’ s LP method does not consistently dominate the standard SVAR method for calculating impulse responses of endogenous variables with contemporaneous e¤ects. Since Jorda’ s LP does not impose any restrictions linking the impulse responses at h and h + 1, estimates can display an erratic behavior due to the loss of e¢ ciency. Additionally, the impulse responses sometimes display oscillations at longer horizons (Ramey, 2016). For these reasons, and to err on the safe side, we report estimates up to 8 quarters after social transfer shocks. 13 The number of lags L = 4 is determined through log-likelihood ratio information criterion tests (see Hamilton (1994), pages 295-296). The same lag structure was recommended when using the Akaike Information Criterion (AIC) or the Hannan-Quinn information criterion (HQC). 14 Interestingly, given the relatively small share of unemployment insurance spending on total social transfers we cannot reject the null hypothesis that the contemporaneous correlation between quarterly changes of real GDP and real social transfers excluding (or including) unemployment insurance spending is 0.09 (0.06); in both cases statistically not signi…cant. 7 shocks by computing estimates of the h-step-ahead cumulative average treatment e¤ect on the GDP variable.15 A common problem in the literature is matching the typical de…nition of a …scal multiplier (i.e., total output change at step h divided by total expenditure change at step h) with the estimated steps of an IRF originated from an initial spending shock. To bypass this problem, we follow Ramey and Zubairy (2018) and use a two-step instrumental variable procedure. In the …rst stage, we …nd the predicted cumulative social transfers at each step h from the unanticipated social transfers shock excluding the unemployment insurance spending component in time t from the following regression: ST wo ui Ph 1 ST wo ui PL PL STi;t+h = h "i;t + l=0 flh "i;t+h l + l=1 kl Yi;t l + l=1 ml Gi;t l + PL PL + l=1 l Ri;t l + l=1 #l inti;t l + ci;h + qt;h + i;t;h ; (2) where and ci;h and qt;h represent country and time …xed e¤ects, respectively. In the second stage, we use the predicted values ST \ i;t+h from the previous expression to instrument the cumulative social transfers in a regression against GDP: P 1 PL Yi;t+h = IV 1h \ ST i;t+h + IV 2h \ ST i;t+h LAC + h ST l=0 wlh "i;t+h l + l=1 l Yi;t l + PL PL PL + l=1 l Gi;t l + l=1 l Ri;t l + l=1 l inti;t l + ci;h + qt;h + i;t;h ; (3) where Xi;t+h = (Xi;h Xi;t 1 )=Yi;t 1 . We use Teulings and Zubanov (2014) bias correction by including social transfer shocks nested between times t and h as controls. LAC is a dummy variable equal to one if the country belongs to the Latin American sample and zero otherwise. Estimators IV IV 1h and 2h in speci…cation (3) represent the impulse response at each step h. By construction, each step in speci…cation (3) also gives us the cumulative multiplier de…ned as the accumulated output divided by the accumulated spending in social transfers. Given our non-linear approach, IV IV the STM for a Latin American country would be determined by 1h + 2h , while for a developed IV economy, by 1h . Standard errors are drawn from a two-way cluster-robust covariance matrix. Even though, in principle, one may argue that …scal policy is more ad-hoc than other gov- ernmental policies (e.g., monetary policy), we also perform several additional exercises to reduce concerns about omitted variable bias. The results of these exercises are not reported for the sake of brevity and, especially, because our …ndings remain virtually unchanged.16 We winsorize the sample (rounding outliers to at 1 percent and 99 percent cuto¤). We include other control variables such as country speci…c commodity prices, exchange rate regimes, and the level of debt (as percent 15 This methodology provides a ‡ exible alternative to VAR approaches. As described by Jorda (2005), linear projec- tions can be estimated by simple regression techniques (IV in our case) and they are more robust to misspeci…cation errors. Nonetheless, Ramey (2016) points out some limitations— e.g., impulse responses sometimes display oscilla- tions at longer horizons (particularly after 16 quarters). Since, in this study, we are interested in the short- and medium-horizon e¤ects of social transfers on output, we can safely disregard these drawbacks. 16 These exercises are available upon request to the authors. 8 of trend GDP). We also drop one country at a time and one year at a time. 2.2 Empirical …ndings Based on the empirical strategy discussed in the previous subsection, Panels A and B in Figure 1 show the size of STMs in developed (blue color) and Latin American (red color) countries, respec- tively.17 Looking at our results, the size of the STM is much larger in Latin American countries than in developed economies, particularly in the short- and medium-term. While the STM in developed countries is 0.3 (t=2.0) on impact, it is 0.9 (t=3.8), in Latin American economies.18 In both sets of countries, the peak is reached after one quarter, coming to 0.5 (t=1.5) and 1.1 (t=2.6) in the developed and Latin American sets, respectively. After the …rst quarter, the output e¤ects tend to decrease. In other words, while the temporal pro…le of the STM is similar in both set of countries, the size of the STM in Latin American countries is about three times larger during the short- and medium-term. The novel evidence of such large STM on a set of emerging markets –Latin Amer- ican countries in this case– contrasts sharply with the more modest size of the STM obtained in developed economies. It is worth noting that the size of the STM for developed countries is similar to previous country-speci…c and panel-data-based empirical …ndings (e.g., Gechert, 2015; Romer and Romer, 2016; Alesina et al, 2017; Pennings, 2020). As it has been common in this literature, we now analyze which are the main macroeconomic variables involved behind the …ndings of Panels A and B in Figure 1. Panels C and D in Figure 1 show multiplier estimates, now measuring the e¤ect of social transfers on private consumption. Panels E and F show similar multiplier estimates focusing on the impact on private investment.19 The empirical …ndings are quite clear: the output e¤ect observed in Panels A and B are driven by the response of consumption (Panels C and D) as opposed to the statistically insigni…cant response of investment (Panels E and F). This is true for both developed and Latin American countries. While there is no previous empirical evidence for emerging markets in regard to the relative response in consumption and investment to a social transfers shock, the response estimated for developed countries matches that of previous studies. Lastly, as the shock persistence has played a role in the literature in explaining the heterogeneity of the size of the multipliers (Coenen et al, 2012; Romer and Romer, 2016; Alesina et al., 2017; Gechert et al., 2020; Pennings, 2020), we estimate autoregressive social transfers growth regressions 17 Recall that the STM measures the e¤ect of a $1 change in social transfers on the level of GDP. For example, a STM of 0.7 indicates that an increase in social transfers of $1 increases GDP by $0.7. 18 It is worth noting that if one had not excluded the unemployment insurance spending component from speci- …cations (2) and (3), the estimated STM would have been slightly lower, especially for developed countries (STM would have been 0.25 instead of 0.3). This upward bias, especially for the of group of countries with a relatively large presence of unemployment insurance mechanisms, would have wrongfully indicated that the impact of social transfers has less of a positive impact on the economy because of the automatic (and countercyclical!) within the quarter nature of this shock absorber program. 19 These multipliers were estimated the same way as described above for the output multiplier– just changing the dependent variable. 9 with a time trend for each country. We …nd a median autoregressive point estimate of the shock persistence ( ) of 0.86 for the developed countries and 0.84 for Latin American countries. More- over, we cannot reject the null hypothesis that the estimated shock persistence for these groups of countries are statistically the same. 3 Model We examine the e¤ects of social transfers on output in a closed-economy two-agent New Keynesian (TANK) model with two types of agents that di¤er in their access to …nancial markets and, therefore, have di¤erent marginal propensity to consume. Subsections 3.1 to 3.5 present the model. We follow the Giambattista and Pennings (2017) version of the TANK model very closely. Subsection 3.6 discusses the link between the share of HtM agents and the share of social transfers reaching them which, in turn, helps determine the degree of social transfers targeting. Lastly, Subsection 3.7 shows the STM analytical result of a simpli…ed version of the model which highlights the most salient determinants a¤ecting the size of the STM, including the role played by the share of HtM agents and the share of social transfers reaching them. 3.1 The households’problem There are two types of households: a Ricardian household, and a HtM household, with population shares (1 ) and , respectively. 3.1.1 The Ricardian household’s problem There is a unit mass of individuals, i 2 [0; 1], within the Ricardian household (agent 1). These individuals supply di¤erentiated labor inputs to intermediate-goods producers and only di¤er in their ability to change their nominal wage each period (as wages are sticky à la Calvo). The Ricardian household has access to buy/sell non-contingent bonds in the …nancial markets (which allows its members to smooth consumption over time) and owns the capital in the economy (which is rented by its members to intermediate-goods …rms). Consumption is equalized across individuals within the household due to the existence of Arrow securities (i.e., markets are complete within the household). Each individual at the Ricardian household solves the following problem: P 1 t L1;t (i)1+' M ax E0 ln (c1;t ) ; (4) c1;t ;bt ;It ;L1;t (i) t=0 1+' subject to the following budget, capital accumulation, and labor demand constraints: c1;t + It + bt = w1;t (i)L1;t (i) + M P Kt Kt 1 + (Rt 1 = t ) bt 1 + t T ax1;t + (1 )T rt ; (5) 10 Kt = (1 )Kt 1 + [1 S (It =It 1 )] It ; (6) "w L1;t (i) = W1;t (i)=W1;t L1;t ; (7) where is the intertemporal discount factor, ct is real consumption, Lt (i) is desired labor hours, ' 1 is the Frisch elasticity of the labor supply, It is investment, bt is real debt, real wages are de…ned as wt (i) = Wt (i)=Pt , M P Kt is the real gross rate of return on capital, Rt 1= t is the real interest rate, t are pro…ts from retailers, "w is the sticky wage elasticity, T ax1;t is the lump-sum tax paid to the government, T rt are the lump-sum social transfers received from the government by both households, 2 [0; 1] captures the share of social transfers reaching the HtM household, and, consequently, (1 ) identi…es the share of social transfers “leaking out”to the Ricardian household. If < 1, then part of the lump-sum taxes paid by the Ricardian household comes back, also in a lump-sum manner, in the form of social transfers. It is also assumed that the capital adjustment cost follows the expression S (It =It 1 ), where S (1) = S 0 (1) = 0 and S 00 (1) > 0 in steady-state. 3.1.2 s problem The HtM household’ There is a unit mass of individuals, i 2 [0; 1], within the HtM household (agent 2). Unlike the Ricardian household, the HtM household is …nancially constrained (i.e., it cannot lend or borrow) and, consequently, it consumes its entire income each period. Each individual at the HtM household solves the following problem: P 1 t L2;t (i)1+' M ax E0 ln (c2;t ) ; (8) c2;t ;L2;t (i) t=0 1+' subject to the following budget and labor demand constraints: c2;t = w2;t (i)L2;t (i) + T rt ; (9) "w L2;t (i) = W2;t (i)=W2;t L2;t : (10) where, as previously explained, 2 [0; 1] captures the share of social transfers reaching the HtM household. On one extreme, when = 1, all social transfers reach HtM agents and there are no social transfers “leaking out” to the Ricardian individuals. On the other extreme, when = 0, no social transfers reach HtM individuals and all social transfers “leak out” to the Ricardian agents. 3.2 The …rms’problem and sticky prices There is a continuum unit measure of competitive intermediate-goods producers that rent capital from the Ricardian household and hire di¤erentiated labor inputs from both Ricardian and HtM 1 households, aggregating labor through a Cobb-Douglas production function Yt = Kt Lt , where 11 Lt = L1 20 Retailers convert intermediate goods into …nal goods. Final output, Y f , is 1;t L2;t . t produced by a continuum unit of retailers, l, who buy di¤erentiated intermediate goods, Yt , at price P int in a competitive market, di¤erentiate it at no cost, and sell a variety of …nal output, Yl;t , at R1 1 R1 1 1 price Pl;t . Aggregate …nal output and prices are Ytf = ( 0 Yl;t dl) 1 and Pt = ( 0 Pl;t dl) 1 , where > 1 is the elasticity of substitution among varieties. Each retailer takes into account that it may not be able to change their price with probability p when choosing the optimal price. Then prices are sticky à la Calvo and retailers face a downward sloping demand curve for their variety. This optimization problem leads to a standard New Keynesian Phillips curve ^ t = Et ^ t+1 X^t, where a hat (^) denotes a log deviation from steady-state. Variables and X = = ( 1) represent the in‡ation rate and average mark-up. The parameter = (1 p )(1 p )= p captures the slope of the Phillips curve that determines the responsiveness of in‡ation and output to demand shocks. The higher the price stickiness (i.e., the larger p) the more …rms are not able to change prices to the desired level and, therefore, demand shocks generate a larger response on output. When prices are ‡exible (i.e., p = 0 or ! 1), demand shocks only a¤ect prices. The relative price of intermediate goods is de…ned as the inverse of the retailer’s average markup, Ptint =Pt = 1=Xt . Then, the marginal product of labor and capital in terms of intermediate goods must be divided by the markup to generate the real marginal product.21 Aggregate real wages are then given by w1t = (1 )(1 ) (1=Xt ) (Yt =L1;t ) and w2t = (1 ) (1=Xt ) (Yt =L2;t ). In steady state, the HtM household receives a share of labor income, due to the Cobb-Douglas speci…cation. In order to simplify the steady state, it is assumed that the HtM household receives a transfer of capital income and retailer’s pro…ts and pays a share of government spending. Therefore, the HtM household accounts for a share of total consumption. 3.3 Sticky wages Workers are able to set their wages at a steady-state markup above their marginal rate of substitution w = w= ( w 1), which implies that individuals have market power in their labor supply decisions. The labor supply of Ricardian and HtM households are composites of di¤erentiated labor inputs: R1 R1 L1;t = 0 L1;t (i)di and L2;t = 0 L2;t (i)di , respectively. Each individual of the Ricardian and HtM households can reset their nominal wage with constant probability 1 w in each period. Hence, the (nominal) wage decision of a HtM member i at time t = 0 is to choose W2;0 (i) to maximize (8) subject to (10) and other constraints, taking a lower adjusted discount factor 2 = w which incorporates the fact that W2;0 (i) can be reset in the future. The problem is analogous for the 20 Giambatista and Pennings (2017) show in their Online Appendix 2.1 that when labor across Ricardian and HtM households are perfect substitutes and also coupled with the existence of a labor union, as in Galí et al (2007), the exact same allocations and multipliers are achieved as in the Cobb-Douglas speci…cation. See Giambatista and Pennings (2017) for further details of this equivalence. 21 Deviations of Ytf from Yt are second order in the neighborhood of the steady state, and so in the …rst-order approximation we have Ytf = Yt . 12 Ricardian individuals. This optimization problem leads to a New Keynesian wage Phillips curve ^w w k;t = Et ^ k;t+1 ^w i;t , where = (1 w )(1 w )= w (1 + ' ) is the slope of the wage Phillips curve that determines the reaction of wage in‡ation (and then labor income) to demand shocks, k = 1; 2. 3.4 Monetary policy The Central Bank follows (in a log-linearized form) a Taylor rule with interest rate smoothing ( R) that reacts to deviations of in‡ation ( ) and output ( Y) from steady state: ^t = R ^ R Rt 1 + (1 R )( ^t + ^ Y Yt ); (11) ^t = lnRt where R lnRss is the log-deviation of the gross nominal interest rate from steady state. 3.5 Fiscal policy The government runs a balanced budget such that unproductive government purchases and social transfers are …nanced through lump-sum taxes on the Ricardian agents each period: T ax1;t = T rt + Gt : (12) As taxes are levied on the unconstrained agents, the timing of the taxes does not a¤ect the multiplier due to Ricardian equivalence. The paths for T^ ^ t (where a hat (^) for taxes, social rt and G transfers, and government purchases denotes a deviation from steady-state as a share of GDP) are exogenous and follow an AR(1) process T ^rt+1 = T ^ t+1 = G ^rt + eT r;t+1 and G ^ t + eG;t+1 where eT r;t+1 and eG;t+1 are a zero-mean i.i.d shock and captures the persistence of the spending shocks. The model is closed by the standard aggregate resource constraint Yt = c1;t + c2;t + It + Gt . 3.6 Social transfers targeting A key aspect of the model is that the marginal propensity to consume of HtM agents (who are …nancially constrained and cannot lend or borrow) is larger than that of Ricardian agents (who have access to …nancial markets). This di¤erence makes the …scal allocation of social transfers between these two types of agents of special relevance and it is captured in the model by the parameter 2 [0; 1]. The larger the , the larger the share of social transfers reaching the HtM agents (see equation 9) and the lower the “leak out” to the Ricardian agents (see equation 5). Interestingly, the parameter does not di¤erentiate whether the social transfers actually reach HtM agents as the result of (i) a special e¤ort by the …scal authority (either discretionary and/or in terms of social transfer design) to particularly target those HtM agents as opposed to Ricardian agents or, rather, (ii) as the result of the mere existence and representation of HtM individuals in the population (e.g., in an economy largely populated by HtM agents it should be, in principle, quite 13 easy to reach them extensively even without any targeted e¤ort). For this purpose, we decompose into two components. One component aims at measuring the social transfers targeting (ST T , hereafter) e¤ort, which is de…ned as the ability to reach HtM agents beyond their share of the population . That is to say, ST T . A second component captures the share of HtM agents in the population (captured by parameter ). For example, a high value of (e.g., = 0:8) could re‡ect a high prevalence of HtM agents (e.g., = 0:8) coupled with zero targeting (i.e., ST T = 0) or, on the other hand, be the result of a low prevalence of HtM agents (e.g., = 0:2) along with an important targeting e¤ort in identifying and reaching HtM agents (e.g., ST T = 0:6). Moreover, social transfers targeting could, in principle, be “misdirected” showing a negative impact in terms of its targeting (i.e., ST T < 0) when > . As discussed in the introduction, the current literature generally assumes that social transfers solely reach HtM agents (i.e., = 1), which is referred to as perfect social transfers targeting (i.e., ST T = 1 ). While through the lens of the TANK model, understanding the driving forces behind (either the …scal e¤ort in targeting or the result of a larger prevalence of HtM agents) does not seem to be of particular interest, this distinction is relevant both from a conceptual and …scal policy point of view. While HtM agents’ prevalence, , is a structural parameter considered to be outside of the …scal policy scope and more related to the degree of economic and …nancial development as well as economic policies in the …nancial sector, social transfers targeting (ST T ) is a key aspect of …scal policy. Moreover, while a large driven by a large (the extreme case being = and ST T = 0), shows that social transfers are able to reach HtM agents “e¤ortlessly”as a byproduct of having too many HtM individuals, a large driven by large e¤orts in social transfers targeting (i.e., ST T 0) indicates that a certain targeting e¤ort is able to explicitly identify and reach those agents that are actually …nancially constrained. In fact, Section 4 shows cross-country evidence supporting that, when considering all types of social transfers, the share of social transfers reaching the HtM agents, , is, by and large, re‡ecting the prevalence of HtM agents, , and less so social transfers targeting, ST T . 3.7 Analytical STM from a simpli…ed model In this subsection, we show an analytical expression for the STM considering a simpli…ed version of the model which makes the STM constant over time:22 1 dY (' + 1) ST M = (1 )+ >0 (13) dT r (1 ) 22 In particular, following Giambattista and Pennings (2017), it is assumed ‡ exible wages ( ! 1), no capital ( ! 0), no steady state government spending (Gss = 0), a simpli…ed Taylor rule where nominal interest rates only respond to contemporaneous in‡ ation ( R = 0 and Y = 0), a wage subsidy in steady-state equal to sss = Xss 1, that the Taylor principle holds ( > 1), and that the HtM share is not too high ( < (2 + ') 1 ). 14 where = [(1 )( ) = (1 )] [ (1 ) = ] > 0. From STM expression (13) it is straight- forward to show that: (i) dST M=d < 0 (a larger makes Ricardian agents cut their consumption more as the real interest rate goes up), (ii) dST M=d' 1 > 0 (due to larger labor supply e¤ects), (iii) dST M=d < 0 (because a lower reduces the need of Ricardian agents to cut their consump- tion due to lower labor supply e¤ects), (iv) dST M=d p > 0 (demand shocks have larger e¤ects on output rather than on prices), (v) dST M=d > 0 (mainly because a larger share of HtM agents, , increase the average marginal propensity to consume of the economy), (vi) dST M=d > 0 (as social transfers particularly reach those HtM agents with high marginal propensity to consume).23 It is important to note that while having a larger share of HtM agents, , or a bigger share of social transfers reaching the HtM agents, , increases the size of the STM, these increases are in and of itself associated with lower and higher levels of well-being, respectively. Why? Because a larger points to more individuals not being able to smooth consumption, whereas a bigger depicts more social transfers being allocated to constrained individuals. In other words, while having a larger ampli…es the e¤ect of a social transfer shock, it is not something that, naturally, should be commended as desirable. 4 The share of HtM individuals and the share of social transfer reaching them in data Given the nature of our contribution, it is essential to be able to measure the share of HtM individ- uals, , as well as the share of social transfers reaching them, . As discussed in the introduction, no paper disciplines the share of social transfers reaching HtM agents, , by data. It is generally assumed that social transfers solely reach HtM agents (i.e., = 1), which is referred to as perfect social transfers targeting. Regarding the share of HtM agents, , quantitative papers estimating STMs for the United States (Giambattista and Pennings, 2017; Pennings, 2020) discipline the share of HtM agents to one-third based on Kaplan et al. (2014). There have been two main strategies to identify HtM agents in the United States and other developed countries. A …rst group of studies relies on survey data on household portfolios, with an emphasis on savings, net worth, and the liquidity of assets (e.g., Kaplan and Violante, 2014; Kaplan et al., 2014; Aguiar et al. 2020). In their seminal paper, Kaplan et al. (2014) …nd that between 25 and 40 percent of United States households are HtM, with their preferred estimate being one-third. Of all HtM households, about one-third are poor HtM (who hold little or no liquid wealth and no illiquid wealth) and two-thirds are wealthy HtM (who also hold little or no liquid wealth but have signi…cant amounts of illiquid assets on their balance sheets, led by real estate properties). 23 It is worth noting that the empirical evidence obtained from properly adding interaction terms to empirical speci…cation (3) with respect to country-varying variables , , and (see Sections 4 and 5 for measurement details) delivers the model-based sign estimates. Results are not shown for brevity. 15 Just like the poor HtM households, wealthy HtM households have a large marginal propensity to consume out of small transitory income shocks (Kreiner et al. 2012; Broda and Parker, 2014; Jappelli and Pistaferri, 2014). Kaplan et al. (2014) also measure the shares of HtM households (expressed in percentage terms) for seven other developed countries …nding modest heterogeneity among developed countries: Australia (19), Canada (30), France (21), Germany (32), Italy (24), Spain (20), and United Kingdom (33). A second group of studies notes that holding assets is not the only means to cope with an unexpected …nancial shock. For example, individuals can also have access to credit or depend on the help of family and friends, among many other ways and margins to cope with a …nancial shock. These studies rely on surveys based on self-assessed measures of one’s capacity to deal with …nancial s own assets, capacity to borrow, shocks, regardless of whether the source of funds is the respondent’ a network of family and friends, or something else (e.g., Johnson and Widdows, 1985; Worthington, 2004; Lusardi et al., 2011; Vandone et al., 2011; Brunetti et al., 2016). For example, in an in‡uential paper, Lusardi et al. (2011) conducted a survey for a total of 9,148 individuals between the ages of 18 and 65 in the United States (where 2,148 individuals were surveyed) and other seven developed countries (with about 1,000 respondents per country) between June and September of 2009. Their essential strategy to identify …nancially fragile households relied on the following question: “How con…dent are you that you could come up with $2,000 if an unexpected need arose within the next month?” The $2,000 …gure is chosen because it is of the same order of magnitude as the cost of an unanticipated major car repair, a large co-payment on a medical expense, a legal expense, or a home repair. For the survey conducted in the United States, 22 percent of households responded with certainty that they would not be able to come up with the funds and some other 28 percent reported that they would probably not be able to come up with those funds. 4.1 Database used We use individual-level data from the World Bank’s Global Findex which is a comprehensive data- base on how individuals 15 years old and above save, borrow, make payments, and manage …nancial risk. The data are collected in partnership with Gallup through nationally-representative surveys of more than 150,000 adults in over 140 economies (i.e., covers about 1,000 individuals per country as in Lusardi et al. 2011). Global Findex has been published every three years since 2011. We use its latest survey for the year 2017 because it allows one to identify all types of social transfers. Previous surveys of 2011 and 2014 did not include public pensions (this is of particular importance for our paper because the empirical evidence shown in Section 2 includes all types of social transfers). The country samples were designed to be nationally representative and were subsequently weighted to re‡ect each country’s population. Given the focus on emerging markets and developed countries, we exclude from our descriptive analysis what the IMF classi…es as “low-income developing countries.” In total, we end up with 99 countries (including those 23 countries of our developed and Latin 16 American sample). In particular, the Global Findex database includes two sets of questions to each surveyed individual which are essential for our purposes of measuring and : Question FIN24 asks: “Now, imagine that you have an emergency and you need to pay [1/20 of GNI per capita in local currency]. Is it possible or not possible that you could come up with [1/20 of GNI per capita in local currency] within the next month?”The possible answers were “yes” or “no”.24 We use this question, like in the second group of studies which rely on self-assessed measures of capacity to deal with …nancial shocks, to calculate the share of HtM individuals, . Questions FIN37 and FIN38 ask if “[i]n the past 12 months, have you, personally, received any …nancial support from the government? This money could include payments for educational or medical expenses, unemployment bene…ts, subsidy payments, or any kind of social bene…ts. Please do not include wages or any payments related to work” and “[i]n the past 12 months, have you, personally, received a pension from the government, military, or public sector?”, respectively. The possible answers were “yes”or “no”. We use these questions to identify the share of social transfers reaching HtM individuals, . 4.2 Share of HtM individuals is more than twice as large in emerging markets Figure 2 shows the share of HtM individuals, , in each country. Red bars depict emerging markets while blue bars indicate developed countries. For ease of reading, countries with labels correspond to those used in our sample of six Latin American and 17 developed ones. The visual impression is striking: a majority of blue bars lie to the left of the …gure and the majority of red bars lie to the right (indicating a larger share of HtM individuals in emerging countries). In fact, the average share of HtM individuals is twice as large in emerging countries as in developed countries (47.5 percent vs. 23.8 percent, with a statistically signi…cant di¤erence). For our sample of six Latin American countries, the average share of HtM individuals is even larger, reaching 60 percent.25 While using a di¤erent period of coverage, measure, and identi…cation strategy, this evidence seems to match fairly well that of country-speci…c studies. For example, for Mexico, we …nd that 72 percent of individuals are HtM and Cugat (2019) …nds that for years 2016-2018 about 58 percent of Mexican households have no access to formal …nancial markets. For the United States, we …nd that 27 percent of individuals are HtM and Kaplan et al. (2014) …nd that for the period 1989-2010 between 25 and 40 percent of American households are HtM, with their preferred estimate being one-third. For Korea, we …nd that 20 percent of individuals are HtM and Song (2019) …nds that 24 In the case of the United States, this 1/20 of GNI per capita in local currency for the year 2017 was equivalent to 2,380 dollars which is very similar to Lusardi et al. (2011). 25 It is worth noting that the Global Findex survey of 2014 would reach virtually identical results regarding the share of HtM individuals, , with emerging (developed) countries showing slightly lower (higher) values of , of about 2 percentage points. 17 for the period 2012-2017 between 25 and 30 percent of households are HtM. For Japan, we …nd that 18 percent of individuals are HtM and Hara et al. (2016) …nd, following Kaplan et al.’s (2014) approach, that the share of HtM is about 13 percent for the years 1989-2009. Appendix 2 provides further evidence regarding key characteristics about the nature and main characteristics of HtM and non-HtM. First, it shows evidence that while a vast majority of non- HtM individuals in the developed world would rely on savings to cope with a …nancial shock, the top source of funding of individuals living in emerging markets would be aid from family, relatives, and friends. Furthermore, in both set of countries, few individuals would rely on selling assets for providing support, so illiquid wealth (such as real estate properties) does not seem to be important for dealing with a …nancial shock. Second, and using information on whether individuals have a standing property loan, there is strong evidence that wealthy HtM individuals are also present in emerging markets (and not solely in developed countries). Interestingly, yet not surprisingly, while wealthy HtM represent an important share of all HtM individuals in developed countries, it is less the case in emerging markets where the HtM phenomenon is largely driven by poor HtM individuals. Lastly, this appendix analyzes whether the capacity to deal with a …nancial shock varies across several plausible relevant economic and demographic, individual and household, characteristics. As in previous studies relying on data from developed countries, individuals living in households with higher income, greater educational attainment, male, and employed show a higher capacity to deal with a …nancial shock, both in developed and in emerging countries (with even similar predicting capacity values among these groups of countries). 4.3 Evidence on allocation of social transfers Figure 3 shows the share of social transfers reaching HtM individuals, . This is the …rst study showing evidence about this very important aspect of social transfer …scal policy.26 There is a much larger share of social transfers reaching HtM individuals in emerging countries. In fact, the average share of social transfers reaching HtM individuals is about twice as large in emerging countries as it is in developed countries (45.7 percent vs. 23.4 percent, with a statistically signi…cant di¤erence). For our sample of six Latin American countries, the average share of social transfers reaching HtM individuals is even larger, achieving 64.6 percent. Figure 4 shows the relationship between the share of HtM individuals, , (x-axis), and the share of social transfers reaching HtM individuals, , (y-axis). Most countries, both developed and emerging, lie very closely to the 45 degree line, implying that the higher the share of HtM individuals, the higher the share of social transfers that reach HtM individuals.27 Alternatively, Figure 5 shows 26 Naturally, there is a large related micro-literature focusing on how speci…c social programs can reach and impact vulnerable people based on their income levels, poverty status, and other socioeconomic/demographic aspects (e.g., Grosh, 1992; Gasparini et al., 2013; Cruces and Gasparini, 2012). 27 In fact, and based on a simple regression of against , we cannot reject the hypothesis that such a coe¢ cient equals statistically one for the whole sample of countries (as well as for industrial and emerging markets separately). 18 the relationship between the share of HtM individuals, , (x-axis), and social transfers targeting, ST T , (y-axis), depicting no systematic relationship between ST T (recall ST T ) and .28 ; 29 In other words, when considering the universe of all types of social transfers, countries’ability to reach HtM individuals seems to mainly re‡ect the prevalence of HtM individuals, , as opposed to a …scal targeting e¤ort aimed to reach HtM individuals beyond their population representation. Therefore, social transfers in emerging markets are able to reach those HtM individuals mainly because a large part of its population is in the HtM group as opposed to a particularly exceptional targeting of social transfers. The same lack of evidence of high-quality social transfers targeting holds also true in developed countries. 5 Quantitative results from the TANK model With the ability to measure and , we now proceed to solve the TANK model presented in Section 3 numerically in order to analyze the extent to which a standard calibrated TANK model is able to match key empirical regularities shown in Section 2 in terms of the size, main macroeconomic variables involved, and temporal pro…le of the STM for both the developed and the Latin American samples.30 Initially, in our benchmark calibration, and in order to direct our attention to the role played by and , we calibrate the rest of parameters based on values frequently used in the literature for developed countries. Later, we also allow other parameter values (other than and ) to be calibrated for Latin American countries and …nd, indeed, very similar results to our benchmark calibration due to the actual dominant role played by and . Some parameter calibration values deserve special attention. A key mechanism in the literature on …scal multipliers is the reaction of the nominal interest rate to in‡ation and output gaps by central banks (i.e., parameters and Y in the Taylor rule of equation 11). We follow Iacoviello (2005) in choosing a value of = 1:27, Y = 0:13, and R = 0:73. The persistence parameter = 0:86 is based on our own estimates shown in Section 2. The rest of parameter values are selected as follows: Frisch elasticity ' 1 = 1 (as in Christiano et al., 2005), discount rate = 0:995, Calvo probabilities = = 0:75 p w (from Barattieri et al., 2014), sticky wage elasticity "w = 20:50 (matching 5% unemployment rate), steady-state government purchases Gss = 0:20 (from World Development Indicators), steady-state capital adjustment cost S 00 = 1:5 (from Altig et al., 2011) as well as depreciation rate = 0:016, average mark-up X = 1:10, and capital share = 0:30 (in order to match K/Y from KLEMS). Measurement and sources of data regarding parameters and were discussed in detail in Section 4. In particular, we take the average and for each sample in the Latin American and developed 28 Based on a simple regression of ST T against , we cannot reject the hypothesis that such a coe¢ cient equals statistically zero for the whole sample of countries (as well as for developed and emerging markets separately). 29 Appendix 3 shows the share of HtM individuals ( ), the share of social transfers reaching HtM individuals ( ), and the social transfers targeting (ST T ) for each of the 99 developed and emerging countries included in our sample. 30 We would like to thank Steven Pennings for sharing the Dynare code from Giambattista and Pennings (2017). 19 groups (i.e., Developed = 0:23, Developed = 0:25, Latin America = 0:60, Latin America = 0:65).31 Figure 6 compares STMs, on impact, from the calibrated TANK model (dotted bars) with the empirical estimates shown previously in Section 2 (solid bars). Panel A shows the size of STMs both for the developed and Latin American countries. Panel B presents the di¤erence in STMs between these two groups. Panel A shows that the calibrated TANK model is largely able to account for the observed empirical evidence. The TANK model delivers much larger STMs for the Latin American sample than for the developed one. The quantitative STM delivered by the model on impact is 0.21 in developed countries and 0.92 in the Latin American sample. Notably, these results are well within the reported 90 percent statistical range associated with the empirical STM estimates which are associated with STM point estimates of 0.31 in developed countries and 0.90 in the Latin American sample.32 As observed in the empirical estimates, most of the macroeconomic e¤ect of the social transfer shock is driven by the reaction of private consumption as opposed to private investment. In other words, much like the …ndings of papers focused on developed countries, the e¤ect on output mainly occurs through consumption while private investment remains virtually unchanged. Panel B shows, now focusing on the di¤erence in size of STMs, that …ndings from Panel A imply that the model also accounts for the observed di¤erences between these two groups of countries. For example, the di¤erence in the e¤ect of social transfers on output based on the model is 0.71 (second dotted grey bar) which is well within the reported 90 percent statistical range associated with the empirical evidence of 0.58 (…rst solid grey bar). Four other aspects of the quantitative TANK model results are worth noting. First, the long- run multiplier is also about 4 times larger in Latin American countries (with a STM equal to 0.15) than in the developed sample (with a STM of 0.04).33 Second, as shown in Figure 7, the model is also able to match quite well the temporal pro…le depicted in the empirical estimates of Section 2. In fact, the rank-correlations observed in GDP, consumption, and investment between the quantitative results of the model and the empirical estimates range between 0.73 and 0.95 (in all cases statistically signi…cant). Panel A shows that the STM is, on impact, four times larger in Latin American countries than in developed economies. Panel B shows that private consumption increases in response to a positive social transfer shock (especially in Latin American countries) due to the redistribution of funds from low to high marginal propensity to consume households. Investment decreases in both developed and Latin American countries given that as in‡ation goes 31 It is also worth noting that the standard deviation of within each sample of countries is relatively low: 7 percent and 9 percent in Latin American and industrial countries, respectively. On the contrary, for the joint sample it is 17 percent. The equivalent percentages for are 7 and 10 for Latin American and industrial countries, respectively. On the contrary, for the joint sample it is 18 percent. 32 As a reference, if one had used the usual assumption in the literature that social transfers only reach HtM agents, known as perfect social transfers targeting (i.e., = 1), the implied quantitative STMs would have been much larger and disconnected with our empirical estimates. In particular, the quantitative STMs would have been 0.83 for the developed countries and 1.42 for the Latin American economies. 33 The long-run multiplier is de…ned as the discounted sum of changes in output divided by the discounted sum of changes in social transfers. 20 up, monetary policy reacts by increasing the real interest rate (i.e., > 1). In‡ation increases more in Latin American countries due to the greater change in aggregate demand (i.e., higher ) and, therefore, the larger the increase in the real interest rate, the larger the drop in investment. Third, we now analyze how much of the di¤erence in the e¤ect of social transfers on output depicted by the model (i.e., 0.71) re‡ects the prevalence of HtM individuals, , vis-à-vis the …scal targeting e¤ort aimed to reach HtM individuals beyond their population representation as depicted by the ST T (recall that ST T ). Recall that while the share of HtM agents, , is a structural parameter considered to be outside of the …scal policy scope and more related to the degree of economic and …nancial development as well as economic policies in the …nancial sector, social transfer targeting, ST T , is a key aspect of …scal policy, especially determined by . Naturally, since in the benchmark parametrization only and are allowed to vary, these two mechanisms jointly account, by design, for all the di¤erence in the e¤ect of social transfers on output depicted by the model. Also recall that, as discussed in Subsection 4.3, based on a large sample of countries, generally , which in turn implies that ST T 0. That is to say, when considering the universe of all types of social transfers, countries’ability to reach HtM individuals seems to mainly re‡ect the prevalence of HtM individuals, , as opposed to a …scal targeting e¤ort aimed to reach HtM individuals beyond their population representation (i.e., there is very little social transfer targeting). This global regularity does not escape to our developed and Latin American samples where Developed = 0:23, Latin America = 0:60, ST TDeveloped = 0:02 and ST TLatin America = 0:05. While Latin American countries have a much larger share of HtM individuals, , than developed countries, social transfers targeting, ST T , is very similar across both samples of countries.34 It should prove no surprise then that while the di¤erence in the share of HtM individuals, , explains 91 percent of the STM size di¤erence between developed and Latin American countries, the share of social transfers reaching them, ST T , explains 9 percent of such di¤erence.35 In other words, when considering all types of social transfers, Latin American countries depict large STMs mainly as a consequence of having, unfortunately, a large part of its population being HtM individuals as opposed to a particularly exceptional targeting of social transfers.36 Lastly, we now allow other parameter values (other than and ) to be calibrated for Latin American countries. In particular we follow De Mello and Moccero (2011) who …nd more accom- modative Taylor rules in Latin America (in particular, = 1:19, Y = 0:01, and R = 0:61). Other 34 It is also worth noting that the standard deviation of ST T within each sample of countries is relatively low: 2.9 percent and 2.2 percent in Latin American and developed countries, respectively. The same holds true when using the joint sample: 2.6 percent. 35 Using the baseline calibration for developed economies, the counterfactual multiplier when the economy has Latin America = 0:60 and ST TLatin America = 0:05 (i.e. Latin America = 0:65) is equal to ST M = 0:65. Therefore, it explains 91 percent of the di¤erence: (0.71 - 0.65)/0.71)*100 = 91. The remaining 9 percent di¤erence in the STM is explained by the fact that ST TLatin America = 0:05: 36 Appendix 4 shows on-impact STMs values using our benchmark calibration and combinations of and which are allowed to vary between 0 and 1. This exercise provides additional insights as to the relative importance of vis-à-vis driving large STMs. 21 parameter values are selected as follows for Latin American countries: discount rate = 0:988, sticky wage elasticity "w = 10:50 (matching 10% unemployment rate), steady-state government purchases Gss = 0:19 (from World Development Indicators), and capital share = 0:35 (in order to match K/Y from KLEMS). Since macroeconomic and in‡ation volatility is larger in emerging markets, the price-setting behavior in those countries is less rigid (see Barros et al., 2009, for Brazil). For this reason, we calibrate a higher frequency of price adjustments in developing countries (1.5 times a year relative to once a year in the developed world). This implies Calvo probabilities p = w = 0:625. Based on this alternative calibration exercise, the quantitative STM delivered by the model on impact is 0.84 in the Latin American sample, only 0.08 smaller than the benchmark calibration. Moreover, the prevalence of HtM individuals, , and the …scal targeting e¤ort aimed to reach HtM individuals beyond their population representation, ST T , help explains about 80 and 8 percent, respectively, of the overall di¤erence in the e¤ect of social transfers on output depicted by the model (i.e., 0.71). This implies that about 12 percent of the overall di¤erence in the e¤ect of social transfers on output depicted by the model (i.e., 0.71) is explained by other parameter di¤erences between developed and Latin American countries.37 To sum up, a standard calibrated TANK model is able to match key STM empirical regular- ities shown in Section 2 in terms of the crucial role of the higher share of …nancially constrained individuals who live HtM in the Latin American sample relative to the developed countries has in explaining the large di¤erence in the size of STMs previously estimated as well as regarding the main macroeconomic variables involved, and temporal pro…le for both the developed and Latin American samples. 6 Final thoughts Based on novel empirical evidence and quantitative results, we have shown the critical implications of having large shares of …nancially constrained individuals on the size of the STM. A larger share of individuals living hand-to-mouth causes social transfer shocks to easily reach individuals with a high marginal propensity to consume which, in turn, increases aggregate consumption and output. For this reason, the e¤ect on output is mainly driven by consumption while investment remains mostly unchanged. These …ndings coupled with the evidence supporting a larger share of HtM agents in emerging markets (beyond that of our six Latin American countries) suggest that a larger STM may be expected for emerging market economies in general. Two further re‡ections emerge from our analysis, especially when translating our …ndings into …scal policy action. First, given the large size that the STM can achieve especially in emerging markets, social transfers emerge as a natural …scal policy tool to help vulnerable families who are 37 In particular, 6 percent is due to di¤erences in monetary policy, 4 percent given by heterogeneity in nominal rigidities and 2 percent of the di¤erence by other factors. 22 …nancially constrained and at the same time help the economy to recover faster. In this sense, social transfers seem to provide an inclusive manner to deal with temporary and deep recessions, like during the COVID-19 pandemic. Second, because most of the e¤ect of a social transfer shock impacts the economy especially in the short- and medium-run (as opposed to having truly long-lasting e¤ects) and through private consumption (as opposed to via increasing the economy’s productive capacity and investment), this type of …scal policy tool is far from ideal to increase long-term growth and productivity. References Aguiar, M. A., Bils, M., Boar, C., 2020. Who are the hand-to-mouth? NBER Working Paper No. w26643. Alesina, A., Barbiero, O., Favero, C., Giavazzi, F., Paradisi, M., 2017. The e¤ects of …scal consolidations: Theory and evidence. NBER Working Paper No. w23385. Altig, D., Christiano, L. J., Eichenbaum, M., & Linde, J., 2011. Firm-speci…c capital, nominal rigidities and the business cycle. Review of Economic dynamics, Vol. 14, No. 2, pp. 225-247. Barattieri, A., Basu, S., & Gottschalk, P., 2014. Some evidence on the importance of sticky wages. American Economic Journal: Macroeconomics, Vol. 6, No. 1, pp. 70-101. Barros, R., Bonomo, M., Carvalho, C., Matos, S., 2009. Price setting in a variable macroeco- nomic environment: Evidence from Brazilian CPI. Mimeo, Getulio Vargas Foundation and Federal Reserve Bank of New York. Beck, T., Demirgüç-Kunt, A., Levine, R., 2007. Finance, inequality and the poor. Journal of Economic Growth, Vol. 12, No. 1, pp. 27-49. Beck, T., Levine, R., Loayza, N., 2000. Finance and the sources of growth. Journal of Financial Economics, Vol. 58, Nos. 1-2, pp. 261-300. Blanchard, O., Perotti, R., 2002. An empirical characterization of the dynamic e¤ects of changes in government spending and taxes on output. Quarterly Journal of Economics, Vol. 117, No. 4, pp. 1329-1368. Broda, C., Parker, J. A., 2014. The economic stimulus payments of 2008 and the aggregate demand for consumption. Journal of Monetary Economics, Vol. 68, pp. S20-S36. Brunetti, M., Giarda, E., Torricelli, C., 2016. Is …nancial fragility a matter of illiquidity? An appraisal for Italian households. Review of Income and Wealth, Vol. 62, No. 4, pp. 628-649. Christiano, L. J., Eichenbaum, M., Evans, C. L., 2005. Nominal rigidities and the dynamic e¤ects of a shock to monetary policy. Journal of Political Economy, Vol. 113, No. 1, pp. 1-45. Coenen, G., Erceg, C. J., Freedman, C., Furceri, D., Kumhof, M., Lalonde, R., Laxton, D., Linde, J., Mourougane, A., Muir, D., Mursula, S., de Resende, C., Roberts, J., Roeger, W., Snudden, S., Trabandt, M., in’ t Veld, J., 2012. E¤ects of …scal stimulus in structural models. American Economic Journal: Macroeconomics, Vol. 4, No. 1, pp. 22-68. Corsetti, G., Meier, A., Müller, G., 2012. Fiscal stimulus with spending reversals. The Review of Economics and Statistics, Vol. 94, No. 4, pp. 878-895 . Cruces, G., Gasparini, L., 2013. Políticas sociales para la reducción de la desigualdad y la po- breza en América Latina y el Caribe. Diagnóstico, propuesta y proyecciones en base a la experiencia reciente. CEDLAS Working Paper No. 142. Cugat, G., 2019. Emerging markets, household heterogeneity, and exchange rate policy. Mimeo, Washington, DC: International Monetary Fund. 23 De Mello, L., Moccero, D., 2011. Monetary policy and macroeconomic stability in Latin Amer- ica: The cases of Brazil, Chile, Colombia and Mexico. Journal of International Money and Finance, Vol. 30, No. 1, pp. 229-245. Di Maggio and Kermani, 2016. The importance of unemployment insurance as an automatic stabilizer. NBER Working Paper No. 22625. Egger, D., Haushofer, J., Miguel, E., Niehaus, P., Walker, M. W., 2019. General equilibrium e¤ects of cash transfers: experimental evidence from Kenya. NBER Working Paper No. w26600. Fatás, A., Mihov, I., 2003. The case for restricting …scal policy discretion. The Quarterly Journal of Economics, Vol. 118, No. 4, pp. 1419-1447. Galeano, L., Izquierdo, A., Puig, J., Vegh, C., Vuletin, G. Can automatic government spending be procyclical? Mimeo, World Bank and Inter-American Development Bank. Galí, J., López-Salido, J. D., Vallés, J., 2007. Understanding the e¤ects of government spending on consumption. Journal of the European Economic Association, Vol. 5, No. 1, pp. 227-270. Gasparini, L., Cicowiez, M., Escudero, W., 2012. Pobreza y desigualdad en la Argentina. Temas Grupo Editorial. Gechert, S., Paetz, C., Villanueva, P., 2020. The macroeconomic e¤ects of social security con- tributions and bene…ts. Journal of Monetary Economics (in press). Gechert, S., 2015. What …scal policy is most e¤ective? A meta-regression analysis. Oxford Economic Papers, Vol. 67, No. 3, pp. 553-580. Gerard, F., Naritomi, J., Silva, J., 2020. The e¤ects of cash transfers on formal labor markets: Evidence from Brazil. Mimeo, Washington, DC: World Bank. Giambattista, M., Pennings, S., 2017. When is the government transfer multiplier large? Euro- pean Economic Review, Vol. 100, pp. 525-543. Grosh, M., 1992. De los principios a la práctica: la focalización de programas sociales en América Latina. Programa de Estudios Regionales del Banco Mundial. Hamilton, J. D., 1994. Time series analysis, Princeton, NJ: Princeton University Press. Hara, R., Unayama, T., Weidner, J., 2016. The wealthy hand to mouth in Japan. Economics Letters, Vol. 141, pp. 52-54. Haushofer, J., Shapiro, J., 2016. The short-term impact of unconditional cash transfers to the poor: experimental evidence from Kenya. The Quarterly Journal of Economics, Vol. 131, No. 4, pp. 1973-2042. Huidrom, R., Kose, A., Lim, J., Ohnsorge, F., 2020. Why do …scal multipliers depend on …scal positions? Journal of Monetary Economics, Vol. 114, pp. 109-125. Iacoviello, M., 2005. House prices, borrowing constraints, and monetary policy in the business cycle. American Economic Review, Vol. 95, No. 3, pp. 739-764. Ilzetzki, E., Mendoza, E.,Vegh, C., 2013. How big (small?) are …scal multipliers? Journal of Monetary Economics, Vol. 60, No. 2, pp. 239-254. Jappelli, T., Pistaferri, L., 2014. Fiscal policy and MPC heterogeneity. American Economic Journal: Macroeconomics, Vol. 6, No. 4, pp. 107-136. Johnson, D., Widdows, R., 1985. Emergency fund levels of households. In: K. Schnittgrund (Ed.), Proceedings of the 31st Annual Conference of the American Council on Consumer Interests (pp. 235-241), Columbia, MO: American Council on Consumer Interests. Jorda, O., 2005. Estimation and inference of impulse responses by local projections. American Economic Review, Vol. 95, No. 1, pp. 161-182. Kaplan, G., Violante, G. L., 2014. A model of the consumption response to …scal stimulus payments. Econometrica, Vol. 82, No. 4, pp. 1199-1239. Kaplan, G., Violante, G. L., Weidner, J., 2014. The wealthy hand-to-mouth. NBER Working Paper No. w20073. 24 King, R., Levine, R., 1993. Finance and growth: Schumpeter might be right. The Quarterly Journal of Economics, Vol. 108, No. 3, pp. 717-737. Kreiner, C. T., Lassen, D. D., Leth-Petersen, S., 2012. Consumer responses to …scal stimulus policy and households’cost of liquidity. CEPR Discussion Papers 9161. Lusardi, A., Schneider, D. J., Tufano, P., 2011. Financially fragile households: Evidence and implications. NBER Working Paper No. w17072. McKay and Reis, 2016. The role of automatic stabilizers in the US business cycle. Econometrica, Vol. 84, No. 1, pp. 141-194. Mehrotra, N. R., 2018. Fiscal policy stabilization: Purchases or transfers? International Journal of Central Banking, Vol. 14, No. 2, pp. 1-50. Monacelli, T., Perotti, R., 2011. Redistribution and the multiplier. IMF Economic Review, Vol. 59, No.4, pp. 630-651. Oh, H., Reis, R., 2012. Targeted transfers and the …scal response to the great recession. Journal of Monetary Economics, Vol. 59, pp. S50-S64. Parraga-Rodriguez, S., 2018. The dynamic e¤ects of public expenditure shocks in the United States. Journal of Macroeconomics, Vol. 56, pp. 340-360. Pennings, S., 2020. Cross-region transfers in a monetary union: Evidence from the US and some implications. World Bank Policy Research Working Paper No. 9244. Ramey, V. A., 2016. Macroeconomic shocks and their propagation. NBER Working Paper No. w21978. Ramey, V. A., Zubairy, S., 2018. Government spending multipliers in good times and in bad: evidence from US historical data. Journal of Political Economy, Vol. 126, No. 2, pp. 850-901. Romer, C. D., Romer, D. H., 2016. Transfer payments and the macroeconomy: The e¤ects of social security bene…t increases, 1952-1991. American Economic Journal: Macroeconomics, Vol. 8, No. 4, pp. 1-42. Song, S. Y., 2019. Leverage, hand-to-mouth households, and heterogeneity of the marginal propensity to consume: Evidence from South Korea. Review of Economics of the Household, pp. 1-32. Stock, J., Watson, M., 2007. Why has U.S. in‡ ation become harder to forecast? Journal of Money, Credit, and Banking, Vol. 39, No. s1, pp. 1-33. Teulings, C. N., Zubanov, N., 2014. Is economic recovery a myth? Robust estimation of impulse responses. Journal of Applied Econometrics, Vol. 29, No. 3, pp. 497-514. Vandone, D., Bacchiocchi, E., Anderloni, L., 2011. Household …nancial vulnerability: An em- pirical analysis. Departmental Working Papers 2011-02, Department of Economics, Management and Quantitative Methods at Università degli Studi di Milano. Worthington, A. C., 2004. Emergency funds in Australian households: An empirical analysis of capacity and sources. Financial Counseling and Planning, Vol. 15, No. 1, pp. 21-30. 25 Figure 1. Social transfer multipliers:   Empirical estimation for Latin American and developed countries       Panel A. Effect on GDP   Panel B. Effect on GDP   in developed countries  in Latin American countries      Panel C. Effect on private consumption   Panel D. Effect on private consumption   in developed countries  in Latin American countries      Panel E. Effect on investment   Panel F. Effect on investment   in developed countries  in Latin American countries      Notes:  The  STM  measures  the  effect  of  a  $1  change  in  social  transfers  on  the  level  of  GDP  (Panels  A  and  B),  consumption  (Panels  C  and  D),  and  investment  (Panels  E  and  F).  Dark,  medium,  and  light  areas  show  standard errors at 68, 90, and 95 percent confidence intervals, respectively.       Figure 2. Country share of hand‐to‐mouth individuals   Notes: Authors’ calculation based on Global Findex database.            Figure 3. Country share of social transfers reaching hand‐to‐mouth individuals  Notes: Authors’ calculation based on Global Findex database.                   Figure 4. Country share of social transfers reaching hand‐to‐mouth   individuals vs. share of hand‐to‐mouth individuals    Notes: Authors’ calculation based on Global Findex database.                   Figure 5. Country social transfers targeting vs.   share of hand‐to‐mouth individuals    Notes: Authors’ calculation based on Global Findex database.                                       Figure 6. Social transfer multipliers:  Empirical estimation versus model quantitative results  Panel A. Size of social transfer multipliers    Panel B. Difference in size of social transfer multipliers    Notes: Standard errors depict 90 percent statistical significance error bands.          Figure 7. Social transfer multipliers: Model quantitative results      Panel A. Effect on GDP  Panel B. Effect on consumption and investment            Appendices Appendix 1. Data de…nitions and sources All series are quarterly and seasonally adjusted. If they are not seasonally adjusted from the original source, we apply a seasonal adjustment following the U.S. Census Bureau’ s X-13ARIMA-SEATSX- 13. All series are expressed in millions of national currency, in constant prices, de‡ ated using the implicit GDP de‡ ator derived from quarterly national accounts for each country.38 GDP, private consumption, private investment (gross …xed capital formation to be precise), and GDP de‡ ator sources: Argentina (Dirección Nacional de Cuentas Nacionales, INDEC), Brazil (Instituto Brasileiro de Geogra…a e Estatística, IBGE), Chile (Banco Central de Chile), Colombia (Departamento Administrativo Nacional de Estadística, DANE), Mexico (Instituto Nacional de Estadística, Geografía e Informática, INEGI), Peru (Instituto Nacional de Estadística e Informática, INEI, Banco Central de Reserva del Perú, BCRP), United States (Economic Research Division, Federal Reserve Bank of St. Louis), and European countries (Eurostat [Last update: 16.09.19. Extracted on: 17.09.19]). Social transfers sources: Argentina (Dirección Nacional de Política Fiscal y de Ingresos. Sub- secretaría de Programación Macroeconómica. Secretaría de Política Económica, Ministerio de Ha- cienda), Brazil (STN, IBGE, IPEA e BCB), Chile (Dirección de Presupuestos (DIPRES) / Banco Central de Chile), Colombia (Ministerio de Hacienda y Crédito Público), Mexico (Dirección General de Estadística de la Hacienda Pública. Unidad de Planeación Económica de la Hacienda Pública.), Peru (Government Finance Statistics (GFS) - IMF), United States (Economic Research Division, Federal Reserve Bank of St. Louis), and European countries (Eurostat [Last update: 19.07.19. Extracted on: 17.09.19]).39 Total government spending and revenues: Argentina (Ministerio de Economía en base a datos de la Secretaría de Hacienda), Brazil (STN, IBGE, IPEA e BCB), Chile (Dirección de Presupuestos (DIPRES) / Banco Central de Chile), Colombia (Dirección General de Política Macroeconómica, Ministerio de Hacienda y Crédito Público), Mexico (Dirección General de Estadística de la Hacienda Pública. Unidad de Planeación Económica de la Hacienda Pública.), Peru (Ministerio de Economía y Finanzas, Banco de la Nación), United States (Economic Research Division, Federal Reserve Bank of St. Louis), and European countries (Eurostat [Last update: 19.07.19. Extracted on: 17.09.19]). Central bank interest rates: Global Financial Data and central banks’websites. Sample period to the estimation of social transfer multipliers: Argentina (2004Q1-2019Q4), Austria (2001Q1-2019Q1), Belgium (1995Q1-2019Q1), Brazil (2010Q1-2019Q4), Chile (2005Q1- 2019Q4), Colombia (2000Q1-2018Q4), Denmark (1999Q1-2019Q1), Finland (1999Q1-2019Q1), France (1980Q1-2019Q1), Germany (2002Q1-2019Q1), Greece (1999Q1-2019Q1), Ireland (2002Q1-2019Q1), Italy (1999Q1-2019Q1), Luxembourg (2015Q1-2019Q1), Mexico (2007Q1-2019Q1), Netherlands (1999Q1-2019Q1), Norway (2002Q1-2019Q1), Peru (1995Q1-2018Q4), Portugal (1999Q1-2019Q1), Spain (1995Q1-2019Q1), Sweden (1995Q1-2019Q1), United Kingdom (1987Q1-2019Q1), and United States (1960Q1-2019Q4). 38 IMF (2017). Quarterly National Accounts Manual. Statistics Department, International Monetary Fund, 2017 Edition, Washington, D.C. 39 Since the series on unemployment insurance spending are not available at quarterly frequency, we proceed to interpolate it based on annual frequency data for each country. As a robustness check for this approach, and assuming that short-term ‡ uctuations in unemployment insurance spending are mainly driven by changes in the number of unemployed workers, we use such information on unemployment (which is available at quarterly frequency) to guide the weights used in the interpolation procedure. This alternative approach leaves our …ndings virtually unchanged. Appendix 2. Basic facts of HtM and non-HtM individuals On top of questions FIN24, FIN37, and FIN38 (already discussed in the main text, sub-section 4.1), in this Appendix, we also used the following questions from Global Findex Data to help characterize basic relevant facts of HtM and non-HtM individuals: Question FIN25 (which is included in the …nancial resilience section) asks the surveyed in- dividual in case of responding positive to FIN24 “What would be the main source of money that you would use to come up with [1/20 of GNI per capita in local currency] within the next month?” The possible answers being one of the following: “savings” , “family, relatives, or friends” or “money from working” , “borrowing from a bank, employer, or private lender” , “selling assets”, or “some other source” . We use this question to show that while savings is the most important (yet hardly the single) mechanism to cope with a …nancial shock in the developed world, it is more limited in emerging markets where “family, relatives, or friends” seem to be more relevant. We also show that selling assets is not very relevant either in devel- oped or emerging markets supporting that illiquid wealth does not seem to be very relevant in dealing with a …nancial shock. Question FIN19 (which is included in the borrowing section) asks the surveyed individual “[d]o you, by yourself or together with someone else, currently have a loan you took out from a bank or another type of formal …nancial institution to purchase a home, apartment, or land?” , respectively. The possible answers being “yes” or “no” . We use this question to provide an insight regarding the so-called wealthy HtM individuals identi…ed in previous studies for developed countries. In particular, we distinguish HtM individuals into those having a property loan (as a proxy of a particular type of relevant wealthy HtM individual) and those not having a property loan (as a proxy of the so-called poor HtM). Individual characteristic variables regarding income, educational attainment, age, sex, and labor status, in particular depending on whether the individual is part of the workforce and his/her employment status. 2.1. How relevant are individual savings to deal with a …nancial shock? Figure A.2.1 shows the main source of funds non-HtM individuals report would use to deal with a …nancial shock. In line with the literature available for developed countries which relies on liquid wealth (as opposed to total wealth which also includes illiquid assets, e.g. houses), individuals from this set of countries would mainly rely on savings (see blue bars in Panel A) and very little on selling assets (see blue bars in Panel E). In fact, the main reliance of non-HtM individuals on savings as a source to cope with a …nancial shock in developed economies is of 59.3 percent, distantly followed by funds originating from family, relatives, or friends (with an average of 16 percent) and money from working (with an average of 12.9 percent). On the other hand, selling assets as a main source of funding is supported only by 1.4 percent of non-HtM individuals. Notably, while still important, savings are much less relevant in emerging markets. In fact, the funds originating from family, relatives, or friends are, on average, 36.6 percent (i.e., more than twice of that depicted in developed countries), followed by savings and money from work (representing 28.3 and 24.2 percent, respectively). It is also worth noting that, in emerging markets, the share of non-HtM individuals asserting to the selling of assets to deal with a …nancial shock is also very low and virtually identical to that observed in the developed economies. On average, only 1.5 percent of non-HtM individuals in emerging markets would rely on selling assets which, in turn, also support that liquid wealth seems to be more relevant in dealing with a …nancial shock in this part of the world. 2.2. Wealthy HtM individuals: Does this kind also exist in emerging markets? We now explore the relevance of wealthy HtM arguments outside the developed world. Unfortu- nately, we do not have rich portfolio data. However, we can identify whether HtM individuals have a real estate property loan or not. While not ideal, this is very relevant information because, as discussed in great detail in Kaplan et al. (2014), housing wealth is an important source of wealth behind wealthy HtM and because between 50 and 70 percent of owner-occupied homes have a mort- gage in developed countries.40 For example, Kaplan et al. (2014) …nd that HtM households due to “only house wealth” is about 6 percent of all households in the United States and 2 percent in Germany. Panels A and B in Figure A.2.2 split the share of HtM individuals reported in Figure 2, into those who have a standing property loan and those who do not. Making a stretch in the use of the words, we refer to the share of wealthy HtM ( W ) and poor HtM ( P ). Panel A shows that, like in the developed world, wealthy HtM individuals are also present in emerging markets. On average, the share of wealthy HtM is 4.1 and 3.6 in developed and emerging markets, respectively.41 Interestingly, yet not surprising, Panel B in Figure A.2.2 shows that while wealthy HtM constitute a relevant share of the population in developed countries, it is less so in emerging markets where the HtM phenomenon is largely driven by poor HtM individuals. On average, the share of poor HtM is 17.1 and 39.1 in developed and emerging markets, respectively. 2.3. Main characteristics of HtM individuals We now analyze whether the capacity to deal with a …nancial shock varies across several plausible relevant economic and demographic individual and household characteristics available in Global Findex. Table A.2.1 shows the share of HtM individuals by household income quintile, education level, age, sex, and labor status, in particular depending on whether the individual is part of the workforce and his/her employment status. Table A.2.2 reports marginal e¤ects from probit regressions in which the dependent variable equals 1 if the individual is HtM and 0 if he/she is not HtM; both one-characteristic-at-a-time and all of them jointly. Both tables report these relevant data for all countries as well as for emerging and developed countries separately. The …ndings support that individuals with higher household income, greater educational attain- ment, male, and employed report a higher capacity to deal with a …nancial shock, both in developed and emerging countries (with even similar R2 values among these groups). The main discrepancy between developed and emerging countries is related to the relevance of age. While in emerging markets the older the individual, the lower their capacity to deal with a …nancial shock, in developed countries, respondents aged 15 to 34 report themselves as being more …nancially vulnerable. All in all, the relevance of this set of economic and demographic characteristics as well as their ability to predict HtM individuals is comparable to previous studies. These …ndings are broadly consistent with those of other studies focusing on the developed world (e.g., Lusardi et al., 2011). Moreover, if 40 See, for example, https://…vethirtyeight.com/features/how-many-homeowners-have-paid-o¤-their-mortgages/ and https://www.statista.com/statistics/957803/homeowners-with-and-without-an-outstanding-mortgage-in-eu-28- per-country/#:~:text=An%20average%2026.5%20percent%20of,like%20the%20Netherlands%20and%20Sweden.. 41 It is worth noting that our …ndings based on a not ideal measure are quite similar to those relying on more sophisticated analysis for industrial countries like in Kaplan et al. (2014). For example, Panel A reports that in Figure A.3.2 the share of HtM individuals with property loan is 4.6 percent in the United States and 1.5 percent in Germany. 28 one also included all possible interactions among these economic and demographic characteristics, the R2 would reach about one-third (regression results are not shown for the sake of brevity). Appendix 3. Country share of HtM individuals, share of social transfers reaching HtM individuals, and social transfers targeting Table A.3.1 shows the country share of HtM individuals, share of social transfers reaching HtM individuals, and social transfers targeting from Global Findex dataset (based on its latest survey for the year 2017). Appendix 4. Additional insigths regarding the relative importance of vis-à-vis driving large STMs Table A.4.1 shows on-impact STM values using our benchmark calibration and combinations of and which are allowed to vary between 0 and 1 in 0.05 intervals. The shaded cells indicate combinations of and such that ST T < 0. Recall that since ST T , then ST T < 0 implies that > . This exercise provides additional insights as to the relative importance of vis-a-vis driving large STMs. In particular, the results show that it is not feasible to obtain truly large on-impact STMs (e.g., ST M > 1) for low to moderate values of the share of HtM individuals, , (e.g., < 0:4). This implies that, in spite of important social transfers targeting e¤orts, even if was the maximum possible (i.e., = 1), in order to be able to achieve truly large on-impact STM, there exists the need to have large propagation forces via high average marginal propensity to consume, which can only be achieved by having a relatively large share of HtM individuals, . In the extreme, for a su¢ ciently high share of HtM individuals, , (e.g., > 0:8) it is possible to have truly large STMs even when is low enough such that ST T < 0. This quantitative importance of the share of HtM individuals, , provides another relevant dimension as to how to rationalize its practical importance in determining the size of STM in Latin America as well as in other emerging market economies with high prevalence of HtM individuals.     Figure A.2.1. Main source of money to deal with a financial shock. Percent of non‐hand‐to‐mouth individuals      Panel A. Savings  Panel B. Family, relatives, or friends      Panel C. Money from working  Panel D. Borrowing from a bank,   employer, or private lender      Panel E. Selling assets  Panel F. Some other source                                Figure A.2.2. Share of HtM individuals with and without property loans    Panel A. Share of HtM individuals with property loan     Panel B. Share of HtM individuals without property loan        Table A.2.1. Share of hand‐to‐mouth individuals, by economic    Table A.2.2. Probit regressions explaining HtM individual category   and demographic characteristics. Percent of respondents  with economic and demographic characteristics                                          Figure A.3.1. Country share of HtM individuals (α), share of social   transfers reaching HtM individuals (θ), and social transfers targeting (STT)    Notes: See Section 4 for definitions and details. Source: Global Findex 2017.             Table A.4.1. On‐impact STMs values using benchmark calibration and   combinations of α and θ which are allowed to vary between 0 and 1  Notes: Since the STM is not defined for α=0.00 and α=1.00, we selected α=0.0001 and α=0.9999, respectively. The shaded cells indicate combinations of α and θ such that STT<0. Recall that since STT ≡ θ‐α, then STT<0 implies that α > θ.