CROSS-REGIONS TRANSFERS IN A MONETARY UNION: EVIDENCE FROM THE US AND IMPLICATIONS FOR EUROPE STEVEN PENNINGS* Abstract. Transfers to individuals from the federal government vary greatly across states in the US, but are largely absent across countries in the European Monetary Union (EMU). Using exogenous variation in cross-state transfers as part of recent temporary stimulus packages and earlier permanent social security increases, I show that US states receiving larger transfers tend to have faster relative short-term growth in non-transfer income. The results are consistent with an open-economy New Keynesian model. Despite this, the model generates only a small reduction in output volatility if US-style countercyclical cross-region transfers are applied in Europe, especially when compared to the effects of locally-financed countercyclical transfers. JEL: E62, F45, F41

“Some economists...argue that [the] regional insurance scheme provided by the federal government is one of the key reasons why the system of fixed exchange rates within the United States has survived without major problems.” Sala-i-Martin & Sachs (1991) p20 “Indeed, America’s fiscal union is so good at absorbing shocks that it is often cited as a model for the more accident-prone euro zone.” (The Economist, Free Exchange, 28 November 2015)

1. Introduction Transfers to individuals from the federal government vary greatly across states in the US, but are largely absent across countries in the European Monetary Union (EMU). In the US, a $1 decrease in per capita income in a US state/region over time is associated with around a $0.20-0.40 net transfer from the federal government to the residents of that state/region, with the bulk of the Date: January 2016. First draft: August 2013. *Development Research Group, The World Bank. Email: [email protected] or [email protected]; Address: World Bank, 1818 H St NW, Washington DC 20433 USA. The views expressed here are the author’s, and do not reflect those of the World Bank, its Executive Directors, or the countries they represent. I would like to thank Virgiliu Midrigan, Mark Gertler, Bill Easterly, Emmanuel Farhi, Glenn Follette, Aart Kraay, Hyun Oh, seminar participants at NYU, the World Bank, Federal Reserve Board, the NY Fed, the Bank of England, Midwest Macro, UNSW, Monash, ANU, Georgetown and CIDE. 1

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

2

adjustment occurring through the tax system (Feyrer and Sacerdote 2013, Sala-i-Martin and Sachs 1991, Bayoumi and Masson 1995). In contrast, the European Commission’s budget is only about 1% of EU GDP and so the variation in comparable cross-country transfers within the EMU is negligible, almost by definition. Transfers between the US federal government and individuals represent the majority of US federal spending and the majority of recent stimulus packages (Oh and Reis 2012).1 Recent recessions in countries on the European periphery have been severe, perhaps made deeper by the inability of those countries to independently loosen monetary policy and devalue their exchange rates. While the experience of these countries has called into question the viability of the EMU, the US monetary union remains solid. Part of the difference might be due to fiscal policy. The United States has long been seen as an example of risk sharing across states — largely through cross-region transfers — as the opening quotation from Sala-i-Martin and Sachs (1991) more than 20 years ago indicates. This positive view of the US “fiscal union” continues to be influential among policymakers and in the popular press. But the validity of this hypothesis hinges on the effects of cross-region transfers on the regions receiving them. This paper investigates the causal effect of cross-region transfers on non-transfer income in the US data and uses the results to calibrate a New Keynesian (NK) model, which is then used to evaluate the effects of a US-style fiscal union in Europe. Given the relative absence of cross-country transfers in Europe, there is no way to estimate the effects of a potential fiscal union in Europe without using a model. Yet the effects of cross-region transfers will depend on the type of model used, so a careful analysis of empirical evidence from the US is needed to discipline that choice. The results suggest that cross-region transfers do in fact boost short-term non-transfer income in US states receiving them, with larger increases in non-transfer incomes for more persistent crossregion transfers. Those effects are consistent with a New Keynesian model featuring home bias in consumption and a share of non-Ricardian households. Despite the empirical evidence from the US, model simulations suggest that hypothetical countercyclical cross-country transfers in Europe 1

The Federal bailout of Texan banks during the Savings and Loans crisis of the late 1980s reached 20% of Texan GDP (Hill 1990). Transfers between federal and state governments are much smaller than those involving individuals. For example in FY2010, transfers to individuals from the federal government (Retirement/Disability Payments and other direct payments) were $1734bn, whereas grants to state and local governments were less than half the size ($680bn). Traditional government spending — salaries and procurement — accounted for about $870bn (Source: CFFR 2010)

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

3

provide only limited stabilization of regional shocks, especially compared to a counterfactual of selffunded countercyclical transfers by European countries themselves. This surprising result stems from fact that countries on the European periphery are simply too open, and recessions are not sufficiently persistent, for countercyclical cross-region transfers to have a large quantitative effect. The mechanisms driving the size of cross-region transfer multipliers are derived analytically in Section 4. Empirics Transfers to individuals are typically endogenous to state-level income growth (e.g. unemployment benefits), so the correlation between transfers and short term state performance is uninformative about the effect of the transfers. I solve this problem, I identify the causal effect of cross-region transfers on state income growth using permanent changes in US social security (aged pension) payments around 1970, and temporary transfers as part of recent stimulus packages (Section 2).2 As the policy change is at the aggregate level, the total size of the transfer cannot be driven by differential state growth rates. The share of the aggregate payment going to each state is based on largely preexisting demographic characteristics of states (similar to a Bartik instrument). For example, an increase in old-aged pension rates across the whole US leads to an increase in payments to Florida, where lots of retirees live, relative to states like Alaska (which is less popular with retirees). This cross-region transfer is clearly unrelated the strength of the business cycle in Florida vis-a-vis Alaska. In general, only the aggregate size of the transfer is available from official sources, and so I have to construct the cross-state allocation manually. I estimate the effect of transfers on state-level labor income (W × L) and quarterly state-level GDP, though the latter is unavailable for most of the sample. Figure 1 here An example of a specific temporary transfer is the 2008 Bush Economic stimulus package starting in 2008Q2, which paid out $300 to those who paid no taxes (but had some other income), and $600 to those with a tax liability. Figure 1 shows that states which received larger transfers (as a share of labor income) tended to grow faster in the quarter, with a “relative multiplier” (line slope) of about 0.3. These payments were a much larger share of income in poorer states like Mississippi 2

Wilcox (1989), Romer and Romer (2015), Parker et al (2013) and Johnson et al (2006) (among others) investigate the effect of these programs on aggregate consumption, but no-one (to my knowledge) has examined their effects on incomes at the state level.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

4

(MS) than richer states like Connecticut (CT). As the relative income distribution of Mississippi (MS) and Connecticut (CT) is effectively fixed quarter-to-quarter, it is unlikely that differences in the growth of labor income could be driving differences in transfers. To control for other factors which might be correlated with both the transfer and growth in labor income, I estimate a range of regression models which control for state-specific trends (state fixed effects) and aggregate variation (time fixed effects), among other things. For robustness, I also pool over a range of similar temporary transfers, such as those from the 2001 stimulus package. The estimated multipliers suggest that an extra dollar of temporary transfers to a state boosts contemporaneous labor income in that state by around an extra quarter-dollar (relative to states that did not receive the extra dollar of transfers). I estimate the effect of permanent cross-region transfers using a similar methodology applied to a range of ad-hoc social security rate increases from the 1950s through to the mid 1970s (partly in compensation for inflation). Often, monthly benefits would increase permanently by 10-20%. I find that states which received an extra $1 in social security payments increased their labor income by around $1-2 in the first quarter or two, depending on the specification. Model Relative transfer multipliers of this size can be rationalized by a multi-region New Keynesian model but are inconsistent with the baseline neoclassical model (see Sections 3 and 5). In the model, one of the regions is small (representing a US state or a European country), with the rest of the monetary union combined into the other region. Each region produces its own variety of a tradeable good, prices and wages are sticky, and unconstrained households trade a non-contingent bond. To generate empirically realistic transfer multipliers, the model also needs (i) some home bias in consumption to localize the effects of demand in the region receiving the transfer, and (ii) a fraction of the population must consume their income hand-to-mouth to prevent temporary transfers from being saved (I calibrate the fraction to about 1/3). In the short term, output is partially demand determined so when cross-region transfers boost demand for local goods, nontransfer income also increases (as in the data). In the neoclassical model and in the NK model in the long run, local goods become more expensive, expenditure shifts towards foreign goods and output falls as wealthier households choose to work less.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

5

Mechanisms To understand why a positive transfer multiplier in the US might not necessitate a European fiscal union, I derive an analytical expression for the cross-region transfer multiplier in a simplified model (Section 4).3 This multiplier can be decomposed into two parts: a self-funded transfer multiplier which is available to countries pursuing their independent countercylical policies and a second “pure” cross-region transfer multiplier which depends on the funds being received from abroad. For temporary transfers, the cross-region transfer multiplier is mostly driven by the former; for permanent transfers it is mostly driven by the latter. At the frequencies of European business cycles, the self-funded transfer multiplier makes up more than half of the cross-region transfer multiplier, which drastically reduces the stabilization gains of a fiscal union. The analytical expressions also show that the transfer multiplier is strongly increasing in the degree of home bias — a practical problem given European countries are typically quite open. Counterfactual of a European Fiscal Union How much shallower would European regional recessions be with a US-style fiscal union? Using the calibrated NK model consistent with US data, I produce estimates for a country experiencing a recession as deep and persistent as that in Ireland over 2009-12 (Section 6). Following Farhi et al (2014), I generate a regional recession using a large and persistent increase in borrowing costs which reduces the demand for home goods. Without cross-country transfers (as is the case currently), I match the relative fall in output in 2009 of 3.78% in the data, and also the approximate path of output over the next few years. With countercyclical transfers of the size they are in the US, the fall in output is about 8.5% smaller than under the EU fiscal system (i.e the fall in output is 3.46% rather than 3.78%) — quite a modest gain. As foreshadowed in the analytical results, more than half of these gains are available to a European country who is able to fund its own countercyclical transfers of a similar size. The quantitative gains in terms of the overall reduction in GDP volatility are very similar. The relatively small advantage of “pure” cross-region transfers stems from the fact that the households who respond differently to a transfer from Dublin vis-a-vis Brussels also save the majority of temporary transfers. Of course, a permanent transfer would be spent, but then it wouldn’t be countercyclical.4 3The

ability of countercyclical cross-region transfers to boost output in a region in recession depends on the size of the cross-region transfer multiplier. 4A caveat: this paper has little to say about other potential gains of a fiscal union, such as reduction of sovereign debt spreads.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

6

Related Literature Despite the size and policy importance of cross-region transfers, there has been little empirical research on whether transfers across regions actually stimulate non-transfer incomes in the regions receiving them. To my knowledge, this is the first paper that estimates the effect of transfers (to individuals) on incomes across regions in a monetary union.5 The existing empirical literature has largely focused on whether transfers are consumed, rather than whether transfers boost incomes as investigated there (eg Parker et al 2013, Johnson et al 2006, Wilcox 1989). Very little of this analysis is at a regional level. Another strand of the empirical literature focuses on the effects of other types of government expenditure across states (e.g. Nakamura and Steinsson 2014, Suarez and Wingender 2012), or the effect of spending by state and local governments (Clemens and Miron 2012, Shoag 2011, Chodorow-Reich et al 2012). Theoretical work on fiscal policy within a monetary union has generally been focused on government purchases rather than transfers (e.g. Gali and Monacelli 2008, Nakamura and Steinsson 2014, Farhi and Werning 2013), though both of the latter papers include a brief discussion of the financing of government purchases. Farhi and Werning (2014) examine optimal cross-region transfers in monetary union in response to productivity shocks, and find that the benefits of a fiscal union are large if asymmetric shocks are persistent or economies are quite closed. My calibration suggests that countries are in fact not sufficiently closed, and shocks not sufficiently persistent, for the benefits of a fiscal union to be sizable. Unlike Farhi and Werning (2014), my model includes hand-to-mouth households, which are important for matching the effect of transfers in the data.6

2. Empirical Evidence In this section, I estimate the causal effect of temporary transfers from the 2001 and 2008 stimulus packages (Section 2.1) and permanent social security transfers in the 1950s-1970s (Section 2.1) on growth in labor income and GDP. Identification Estimating the effect of transfers on economic activity is difficult because causality can run in both directions. Even if federal transfers stimulated non-transfer incomes in the regions receiving them, regions with low levels of growth for other reasons might attract higher 5Contemporaneous

work by Feler (2015) examines the medium-run effects of Bolsa Família transfers in Brazil. theoretical work on the effect of transfers on incomes has mostly been in closed economy settings (Oh and Reis 2012, Giambattista and Pennings 2015). 6Other

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

7

levels of federal transfers (such as unemployment benefits), biasing the estimates. My identification strategy addresses this problem by studying federal transfers that stem from aggregate policy changes which affect all states, such as temporary federal stimulus packages, and permanent increases in social security payments. As the policy change is at the aggregate level, by definition it cannot be driven by differences in conditions in one state vis-a-vis another. The key identification assumption is that the allocation of transfers across states is unrelated to other sources of contemporaneous income growth in that state. For the policies examined here, the allocation of spending is based on largely predetermined state characteristics, such as the number of social security recipients in each state, or the number of people in different parts of the income distribution. Conceptually, the levels of these variables will not change much from quarter to quarter, eliminating any reverse causality running from growth to transfers.7 In reality, identification is even stronger because policymakers use lagged information — such as from the previous year’s income tax reports — to allocate transfers across states. In general, I had to manually construct the cross-state allocation of many of these transfers using information on eligibility criteria. What I can and cannot identify This identification strategy is good at identifying the shortterm “relative multiplier” of transfers across states. However, it is not designed to address: (i) aggregate effect of the transfers (as the size of aggregate transfers may be endogenous — especially for the 2001 and 2008 stimulus measures) or (ii) the medium/long run effects of transfers across states. In the latter case, the longer the lags considered, the greater the chance of other correlated policy changes in the intervening period. Accordingly, I concentrate on the contemporaneous effect, and allow for up to a quarter lag, which means I am more likely to be picking up short-run demand-side effects rather than long-run supply-side effects. It also rules out most measures of state-level GDP before 2005, which are only available at an annual frequency.8 Dependent variable and specification In the main regressions, I investigate the effect of transfers on the growth rate of real quarterly labor income (equivalent to the wage bill) ∆W Li,t ≡ (W Li,t − W Li,t−1 )/W Li,t−1 which is available for the whole sample or quarterly GDP 7For

example, Online Appendix Figure 3 shows that across states, the size of the 1972Q4 increase in social security payments was almost exactly proportional to the 1970Q2 increase in social security payments. 8I cannot examine the effect on consumption which is unavailable at the state level (even prototype measures are annual). State-level quarterly seasonally adjusted data are not available for many alternative series such as hours worked, with heads-based employment only available since 1990.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

8

from 2005 ∆Y i,t ≡ (Y i,t − Yi,t−1 )/Yi,t−1 . All variables are deflated by the national Quarterly Personal Consumption Expenditures (PCE) Chain-type Price Index (GDP deflators or CPI are not available by state).9 Labor income data comes from the Bureau of Economic Analysis (BEA) under the official title of “Earnings by place of work” and mostly consist of Wage and Salary Disbursements (70%) but also includes non-wage payments by employers which contribute to labor costs such as employer contributions for pensions and social insurance (12%) and social insurance taxes (5%).10 It excludes income from transfers and is before income taxes. As state-level quarterly labor income can be extremely volatile, especially for small states, I drop outliers from all specifications.11 The measure of transfers is scaled by lagged labor income or GDP to retain the “multiplier” interpretation of coefficients ∆tri,t ≡ (tri,t − tri,t−1 )/W Li,t−1 or ∆tri,t ≡ (tri,t − tri,t−1 )/Y i,t−1 i.e. the dollar value of extra labor income or GDP produced by an extra dollar of transfers. All regressions include state fixed effects (µi ) and quarter fixed effects (γt ), which allow for a parsimonious specification while reducing potential omitted variable bias. The state effects control for all state-level trends (e.g. the faster growth rate of sun-belt states), and the quarter fixed effects control for the aggregate size of the transfer program, the US business cycle, monetary policy and any other aggregate shocks. The estimated equation is:

(2.1)

∆W Li,t or ∆Yi,t = β0 ∆tri,t + β1 ∆tri,t−1 + γt + µi + eit

2.1. Temporary Transfers. 2008 Economic Stimulus Act Around $100bn was transferred to households as one-time payments in 2008Q2-Q3 as part of the Bush administration’s Economic 9When

the BEA produces “real” estimates of state-level GDP it deflates using national prices for different industries, weighted using state-specific industry weights. These quasi-GDP deflators are difficult to map into the model in Section 3, so I use nominal variables and deflate them by the PCE. 10Earnings by place of work are sourced mostly from high-quality administrative records (BLS’s Quarterly Census of Employment and Wages (CEW) program), and make up around 3/4 of Personal Income before taxes. They also include partnership and sole trader income (12%) (all figures from 2011). 11Specifically, I drop observations where the quarterly growth rate of labor income growth is more than 3 SD from the mean (in either direction). For the 2000s sample used for temporary transfers, 3SD corresponds to quarterly growth (not annualized) (i) greater than 4.4% or less than -3.5% for labor income, (ii) greater than 4.8% or less than -4.2% for GDP growth. For the sample used for permanent transfers , the outliers are defined as quarterly growth in labor income (not annualized) greater than 6.9% or less than -5.3%. Outliers are around 1.5-1.75% of the sample. All data is used to calculate outliers; 2000-13, 2005-13 and 1948-2013 respectively. I also drop the District of Columbia from the estimation, given the large number of commuters across state boundaries.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

9

Stimulus Payments (ESP) (Parker et al 2013).12 There were two main components of the package: (i) $300 per capita payments made to those paying no net taxes but with at least $3000 in eligible income (around $30bn, which I call the low income rebate component) and (ii) $600 per capita payments made to those paying net taxes with a phaseout for those earning over $75,000 (around $70bn, which I call the middle-income tax refund component).13 Most of the empirical results turn out to be driven by the low income component. Parker et al (2013) exploited randomization in the timing that checks were sent out to find that (for the whole package) about 12-30% was spent on non-durables in the months they were received, or 50-90% including durables. I allocate the payments for the low-income component using the cross-state allocation from BEA (2009), which is primarily based on the geographic distribution of recipients of refundable Earned Income Tax Credits (EITCs). According to the BEA’s allocation, about 95% of the refundable 2008 low-income ESP were paid out in 2008Q2 (as against 85% of the whole package).14 Eligibility for the 2008 stimulus payment was made based on 2007 income as the level of 2008AY income is not known in 2008Q2. I allocate the middle income tax rebate payments across states using IRS data on 2007 income tax returns by state and income combined with the eligibility rules for the tax rebate (see the appendix for further details).15 2001 Stimulus payment In 2001Q3, the Bush administration transferred $38bn to households in a one-off payment as part of the Economic Growth and Tax Relief Reconciliation Act. Individuals paying net taxes mostly received $300 per capita, though unlike the 2008 stimulus there was no payment for those with no tax liability, and no phaseout for those on high incomes. I allocate the $38bn in transfers across states using IRS state-level data on individual tax returns for the 2000 tax year (see the Appendix for details). Exploiting randomization of payment dates, Johnson et 12The 2001 and 2008 stimulus payments are much cleaner examples of one-off stimulus payment than the components

of the 2009 American Recovery and Reinvestment Act (ARRA). Although the ARRA was larger ($787bn), only a very small share was spent on comparable one-time payments ($13bn for $250 payments to social security recipients in 2009Q2), and other programs like the Making Work Pay tax credits were spread across several years rather than being concentrated in one quarter. 13Households receiving either payment also received $300 for each eligible child, with eligibility rules similar to those for the Child Tax credit. The eligible income includes earned income and social security benefits (among other things). 14This is consistent with a high percentage of recipients receiving their ESP by direct deposit, which is faster than by check. IRS data show that around 80% of those with refundable EITCs filed electronically. Direct deposit payment were made in the first half of May 2008, leaving plenty of time for output to be affected in 2008Q2. 15Allocating these transers across states using IRS microdata on individual level tax returns delivers similar results, however microdata are not necessarily more accurate because around 50% of respondents in the microdata have their state identifier missing.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

10

al (2006) found that 20-40% of the payment was spend on non-durables in the months that it was received, with a higher MPC for the poor and credit constrained, and no response for durables. Scatter plot Figure 1 shows the relation between total transfers in 2008Q2 as a share of labor income on the x-axis, and the growth of labor income on the y-axis. States which received larger transfers (as a share of labor income) tended to grow faster contemporaneously, with a “relative multiplier” (line slope) of about 0.25.16 This estimate is similar to what I get in the regression analysis. The cross-state variation in transfers is striking: the 2008 ESP ranges from around 2.3% of quarterly labor income in Connecticut (CT), to 6.4% of quarterly labor income in Mississippi (MS). Two factors are at play here: first, the level of per capita labor income is much lower in MS than CT and so fixed dollar payments are mechanically more important. Second, much of the variation is driven by the low-income rebate (those paying no net taxes), which is focused towards poor states. Of the 4.1 percentage point gap in stimulus transfers in 2008Q2 between MS and CT, 3.4 percentage points are due to the cross-state allocation of the low-income payments. Regression Results Table 1 shows regressions of growth in labor income or GDP on the size of the transfer at the state level over 2001-08. Unlike the scatter plot, the regressions control for state and time effects, and impose that falls in transfers the following quarter (as the stimulus is withdrawn) reduce labor income or GDP growth. Pooling across all measures (Column 1) show that a $1 increase in temporary transfers to a state (relative to other states) increases labor income in that state (relative to others) by $0.24, significant at the 1% level. In Column 2, I break down these transfers into the three components discussed above. The results seem to be driven by the 2008 low-income rebate, with a relative multiplier of 0.33, significant at the 1% level. The effect of the 2001 transfer package has a similar multiplier, but has much larger standard errors. The 2008 mid-income tax rebate has little effect on labor income growth. The two final columns investigate the effect of transfers on quarterly state GDP growth, which is available since 2005. Nonetheless, the results are quite similar. The simplest specification (Column 3) suggest that the cross-state relative GDP multiplier is larger than the cross-state labor income multiplier, and is borderline significant at the 5% level. As before, results seem to be driven by the low-income rebate component, which is significant at the 1% level (Column 4). While the 16The

relationship is even stronger between cross-state transfers and labor income growth in 2001Q3 (not reported).

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

11

estimated coefficient is twice as large as before, the coefficient on tax refunds is more negative, perhaps suggesting some over-fitting. Table 1 here 2.2. Permanent Transfers. A sample of ad-hoc social security increases Before 1975, social security payments (largely the old age pension) were not indexed to inflation and were increased by act of Congress on an ad-hoc basis (Wilcox 1989). Unlike in the previous section, these were mostly permanent increases in transfers — a higher monthly stipend received by the elderly and their dependents — and so are more likely to be spent by unconstrained consumers. As the transfer increases varied in size and timing — for example, a 10% increase in stipends in June 1971, a 20% increase in October 1972 and no increase in 1973 — they would not be subsumed into seasonal factors. From 1975 onwards, social security payments were indexed to the CPI and generally adjusted annually, making them much more predictable and so are excluded from the analysis.17 The sample of social security payment increases at the aggregate level covers 1952-74 and is taken from Table 1 of Romer and Romer (2015). An earlier version of this paper used the aggregate permanent social security increases from Wilcox (1989), which covered 1965-74 or 1968-74.18 These increases over 1965-74 are much larger than those during 1952-64 (2.5 times larger on average; Table 6) so I also consider them separately in the analysis. Unsurprisingly, Wilcox’s and Romer and Romer’s sample of permanent social security changes is very similar for the period they overlap. Romer and Romer’s sample also includes a number of temporary social security increases, and I add these as controls in the regression. As Romer and Romer’s data is monthly whereas mine is quarterly, I spread the adjustments over two quarters if the increase in payments occurred mid-quarter.19 17Wilcox

(1989) examines the effects of these transfers over 1965-85, and finds an insignificant coefficient on consumption in the second half of the sample. Romer and Romer (2015) extend Wilcox’s sample back to 1952 and forward to 1991. If I extend my sample to 1952-91, the estimated permanent relative transfer multiplier is smaller (about 0.8) and only significant at the 10% level (t=1.8 ). 18Wilcox (1989) and Romer and Romer (2015) study the effects of aggregate changes in social security benefits, and find (among other things) that higher social security benefits are consumed in the first few months after receiving them. In contrast, this paper focuses on the effect of cross-state variation in these transfers on cross-state variation in labor income growth, with quarterly dummies subsuming all aggregate-level variation (i.e removing variation in output, consumption or monetary policy). 19For example, a $1 increase starting on 1 June will become a $0.33 increase in Q2 and a $0.66 increase in Q3.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

12

I allocate the increase in aggregate social security payments across states in proportion to that state’s share of social security payments a year before (see Appendix for details). This depends on the stock of retirees in each state, which is both slow-moving and largely exogenous. The largest permanent increases were in 1972Q4 (a 20% increase in benefits) of around 1.45% of quarterly labor income in West Virginia, Arkansas and Florida, which have a large number of retirees as a share of the population. In contrast, the increase in transfers to residents of Alaska was 0.2% of labor income in the same quarter. Table 2 here Results Table 2 shows that a permanent $1 increase in transfers to the residents of a state tends to increase that states’ relative labor income by $1-$2 in the first quarter or two. The first specification (Column 1) is the simplest: it includes no lags or controls, other than state and quarter fixed effects. The relative permanent transfer multiplier is 1.9, significant at the 1% level. Column 2 uses the same 1952-74 sample, but adds a control for temporary social security payments, an extra lag of transfers, and a lagged dependent variable. The contemporaneous estimate is almost identical, with some weak evidence (significant at the 10% level) of a further increase in labor income after a quarter. The estimated coefficient on temporary social security transfers is insignificant, but the standard errors are sufficiently wide that, despite the negative point estimate, I cannot reject that the temporary relative transfer multiplier is around 0.25 (as above).20 Column 3 uses the same specification, but over the 1965-74 sample when permanent increases in social security payments where larger (this was the sample used in some previous versions of this paper). During this period, the contemporaneous permanent transfer is insignificant, but the one-quarter lag is significant at the 5% level with a relative multiplier of 1.2. While 1.2 is smaller than 1.9 as estimated previously, the difference is only around one standard error. The final column returns to the full 1952-74 sample, but adds additional controls for (i) the rate of state population growth and (ii) the interaction between a state dummy and aggregate GDP growth.21 The results are 20I

don’t put much weight on these estimates (relative to those in Table 1) because they are imprecise: the standard errors are almost seven times as large. The imprecision stems from fact these temporary social security payments were much smaller and less variable than those in Table 1, and also occurred at the same time (and with the same cross-state allocation) as permanent social security increases (see Table 6 for descriptive statistics). 21The first variable helps control for interstate migration, which is not included in the model, and the second for a greater sensitivity of some states to aggregate GDP growth (for example, states with more cyclical industries).

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

13

similar to Column 2, but the estimated permanent transfer relative multiplier is smaller at 1.4 (though still significant). 2.3. Robustness Tests. The results presented above are fairly robust to a number of alternative specifications, briefly summarized here (see the online appendix for tables/figures and a more detailed discussion). • Extra controls. Estimated coefficients on pooled temporary transfer coefficients (Column 1 and Column 3 of Table 1), are fairly robust to adding extra controls for the rate of state population growth, a lag dependent variable and the interaction between a state dummy and aggregate GDP growth (See Online Appendix Table 1). Table 2 already includes these additional controls for permanent social security transfers. • Extra lags The contemporaneous coefficients estimated here are fairly robust to the addition of extra lags (see Online Appendix Table 1 and 4). Extra lags of temporary transfers are never significant, though they are sometimes significant for permanent transfers. • IV specification Instrumenting potentially endogenous BEA transfers with the constructed exogenous measures generally leads to significant coefficient estimates which are similar to (or slightly larger than) those reported above (Online Appendix Tables 1 and 4). First stage coefficients estimates are usually fairly close to unity and first stage F-stats are above 100, which explains the similar multipliers, and provides a cross-check of the measures constructed here. • Clustered Standard errors In extra results (not reported), I re-estimated the results in Table 1 and Table 2 clustering standard errors at the state level. Bertrand et al (2004) argues that this helps adjust for serial correlation in diff-in-diff studies. The significance of estimates in Table 1 are either unaffected, or are higher. For example, the coefficient on the pooled temporary transfers in the GDP regression (Model 3) now has a p-value of 0.02 (previously 0.05). For permanent transfers in Table 2, standard errors on key coefficients are actually smaller in Models (1), (2) and (4), but increase for Model (3) such that the first lag of permanent transfers is insignificant. • Cross-section If I estimate the cross-sectional relationship in Figure 1 in 2008Q3 when the stimulus is withdrawn, the estimated multiplier is slightly smaller (though it is significant

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

14

at the 10% level). As the transfer is negative in 2008Q3 for states like Mississippi, it suggests these states also have a larger fall in labor income in that quarter. For the 2001 stimulus and GDP, there is mixed evidence of a fall in non-transfer income as the stimulus is withdrawn (see Online Appendix Table 2) • Anticipation effects/Placebo tests There is no relation between payments in 2008Q2 and labor income growth in 2008Q1 (i.e. Figure 1 with the same X-axis, but income growth in the previous quarter (2008Q1)). This placebo test suggests the results are not driven by state-level trends and there is no evidence of announcement effects (the package was announced in February 2008). Similarly, an equivalent regression of labor income growth the quarter before the increase in permanent social security transfers yields an insignificant (and negative) coefficient. • Including outliers. If I add the outliers back into the sample with labor income as the dependent variable, coefficients on temporary transfers are almost unchanged in terms of size and significance (Table 1 Columns 1-2); the main coefficients on permanent transfers are usually slightly larger, and are significant at the 5% level (Table 2 Column 1-2). When I include outliers with GDP growth as a dependent variable, the coefficient on pooled temporary transfers is insignificant. This is entirely due to some outliers for Wyoming and Alaska, who received relatively small stimulus transfers, but in 2008Q2 grew at quarterly rates (not annualized) in excess of 8% and 5% respectively. Dropping these two small states (a combined population of a little over 1m people) yields coefficient estimates (including outliers) similar to those in Table 1 (Columns 3-4) which are significant at the 5% level. 3. Model I compare the effect of transfers on the economy in two benchmark models: (i) a multi-region New Keynesian model with sticky prices, sticky wages and a fraction of constrained households, (ii) a neoclassical model, which is just a special case as prices and wages become flexible, and the share of constrained households goes to zero. In this section I describe the model, in Section 4 I solve for the impact multiplier analytically in a simple version of each model. Section 5 presents quantitative results (including a comparison to the data), and Section 6 examines the ability of cross-region transfers to combat a regional recession in Europe, such as in Ireland in 2009-12.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

15

3.1. Model Description. Consider a monetary union consisting of a small region (home), like an individual US state or a small European country, and the rest of the monetary union combined (foreign, denoted with ∗). The small region has population n and the large region 1−n. Each region produces their own variety of good, named h (produced by home) or f (produced by foreign). Both goods are perfectly traded and are imperfect substitutes, with constant elasticity of substitution θT . The relative price of the two goods (the terms of trade) is st = Pf,t /Ph,t , and Pf,t and Ph,t are sticky in the Calvo sense. Good h has a weight of α in the utility function of the home consumer, with home bias in consumption if α > n. A fraction ω of households in each region consume their income hand-to-mouth (“HtM HHs”), with the remaining 1 − ω fraction being unconstrained and are able to trade a non-contingent bond.22 A prime (0 ) denotes the unconstrained HH’s variables, a double prime (00 ) the HtM HH’s variables. I solve the model by log-linearizing around the non-stochastic steady state where variables with a hat xˆ reflect log-deviations from steady-state values (except for fiscal variables, which are expressed as share-of-GDP deviation from steady state values).23 A list of log-linear equations is provided in the Online Appendix. 3.1.1. Unconstrained Household’s Problem (Home Region) ( l ∈ [0, n(1 − ω)], notation 0 ). Each unconstrained HH chooses aggregate consumption (c0t ), labor supply (L0t ) and real bond holdings (b0t ) (all defined in per capita terms) to maximize utility:

max{c0t ,b0t ,L0t } E0

∞ X

h i β t lnc0t − L01+ϕ /(1 + ϕ) t

t=0

subject to the budget constraint (written in nominal terms), where Pt is the consumption price index in the home region.

Pt c0t = Ph,t b0t − Rt−1 Ph,t−1 b0t−1 + Wt L0t + Ph,t, T rt0 + Ph,t, Π0t

Each unconstrained HH receives net transfers from the government (T rt0 ), receives profits from firms (Π0t ) and receives labor income Wt0 L0t . The problem of the unconstrained HH in the foreign 22When

st = 1, the home region consumes a share α of their own variety. An alternative assumption (which yields similar results) is that there is no home bias, but a non-traded goods sector with perfectly mobile labor across sectors. I assume (1 − ω)/ω ∈ Z+ . 23The steady state is zero-debt symmetric with equal per capita income (with s = 1). I solve using Dynare. t

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

16

region is analogous. When wages are Calvo-sticky (see Section 3.1.4), the HH supplies whatever labor is demanded at the sticky nominal wage.24

3.1.2. Hand-to-Mouth (constrained) Household’s Problem ( l ∈ (n(1 − ω), n], notation 00 ). Each HtM HH’s problem is similar, except that the household consumes its whole income hand-tomouth as in Gali et al (2007). Intratemporally, the household chooses aggregate consumption (c00t ) and labor supply (L00t ) to maximize utility:

max{c00t, ,L00t } E0

∞ X

h i β t lnc00t − L001+ϕ /(1 + ϕ) t

t=0

subject to the budget constraint (written in nominal terms), where Pt is the consumption price index in the home region.

Pt c00t = Wt L00t + Ph,t, T rt00 + Ph,t Π00t

The HtM household’s income consists of net transfers from the government T rt00 , labor income Wt L00t , and profits from firms Π00t . When wages are Calvo-sticky (see Section 3.1.4), the HtM HH supplies labor as demanded at the given sticky wage.

3.1.3. Goods demand and the aggregate resource constraint. Aggregate consumption in the home region (also per capita) is a constant elasticity of substitution (CES) index of varieties h and f produced by the two regions.

ct = (1 −

ω)c0t

+

ωc00t

  θ θT θT −1 θT −1 T −1 θT θT 1/θT 1/θT = α ch,t + (1 − α) cf,t

This results in the following standard demand equations for ch,t and cf,t and the consumer price index.25 24When wages are sticky,

each unconstrained HH is a member of one union, and HtM HHs supply labor to (1−ω)/ω ∈ + Z union(s) indexed by j. Wt L0t (l) is replaced by Wt (j)L0t (j) for unconstrained HHs, and Wt00 L00t is replaced by P 00 j Wt (j)Lt (j) for HtM HHs. This formulation allows each household’s consumption appear in the sticky wage Phillips curve (Equation 3.5). An alternative used in FRB-Sigma (Erceg et al 2006) and elsewhere is equivalent to setting ω = 0 in Equation 3.5, but generates almost identical multipliers. Each type of household trades arrow securities within its HH type to insure consumption against sticky wage shocks. 25Log linearizing, the consumption demand can be expressed in terms of the log terms of trade s ˆt = pˆf,t − pˆh,t : cˆh,t = cˆt + θT (1 − α)ˆ st and cˆf,t = cˆt − θT αˆ st . Note γ = 1 − n(1 − α)/(1 − n) is the weight on f goods in the foreign region’s utility function which sets per capita income equal in steady state across countries with Sss = 1.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

(3.1)

17

ch,t = α [Ph,t /Pt ]−θT ct

cf,t = (1 − α) [Pf,t /Pt ]−θT ct

h i1/(1−θT ) 1−θT 1−θT Pt = αPh,t + (1 − α)Pf,t Output of good h is consumed at home or abroad (an analogous condition for f ). gh,t are government purchases of home goods and c∗h,t is foreign consumption demand for the home good. gh,t can be either self-funded by increases in lump sum taxation on the unconstrained household in the home region, or funded by per-capita lump-sum taxation across the rest of the monetary union, though unless otherwise stated I set gh,t = 0 (ghSS = 0). The aggregate resource constraint (written in home per-capita terms, adjusting for different population sizes) is:

Yh,t = ch,t +

(1 − n) ∗ ch,t + gh,t n

3.1.4. Production, Sticky Wages and Sticky Prices. As is standard in New Keynesian models (eg Galí and Monacelli 2005), final output in each region is produced by a unit continuum of monop σ σX−1 ´ σX −1 X 1 . Each firm uses labor olistically competitive firms in each region Yh,t = 0 Yh,t (i) σX di ´1 as their only input such that for each firm i: Yh,t (i) = Lh,t (i) and Lh,t = 0 Lh,t (i).di. As the marginal product of labor is unity (i) the aggregate real marginal cost is the real product wage M Ch,t = wh,t = Wt /Ph,t , and (ii) the markup is the inverse of the real product wage Xt = 1/wh,t . ˆt. Taking a log-linear approximation around the steady state:26 mc ˆ h,t = wˆh,t = −X Firms face a downward-sloping demand curve for their variety, and must choose a nominal price, taking into consideration the Calvo probability θp that they may not be able to change their price each quarter in the future. As shown in Gali and Monacelli (2005) and elsewhere, this price setting problem leads to a standard New Keynesian Phillips curve (Equation 3.2), where π ˆh,t = lnPh,t − lnPh,t−1 is producer price inflation. 26The

steady state markup is X = σX /(σX − 1). Firms are identical apart from their ability to re-optimize prices each period. Deviations from these aggregates are second-order.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

18

π ˆh,t = βEt π ˆh,t+1 + κmc ˆ h,t

(3.2)

Wages are sticky as in Erceg, Henderson, and Levin (2000) and Galí (2008), generalized for the presence of HtM households. Lt is a CES composite of differentiated labor inputs supplied to a h i w
1 ´ (1−ω) 1− 1 1 w w −1 . The demand for each w dj continuum of “unions” indexed by j: Lt = 1−ω L (j) t 0 ´ (1−ω) 1 W t (j)1−w dj] 1−w . variety j is Lt (j) = (1 − ω)−1 (W t (j)/W t )−w Lt where W t = [(1 − ω)−1 0 Each union j sources labor from one unconstrained HH, and one HtM household.27 Labor is perfectly substitutable across household types, and so union j pays a common nominal wage Wt (j). ω The union enforces equal size-adjusted labor supply, such that L00 t (j) = L0 t (j). Combined 1−ω with the definition of output, in the log-linearized equilibrium:

ˆ 00 = L ˆ0 = L ˆ t = Yˆh,t L t t

(3.3)

Each union can reset its wage each period with probability 1 − θw , and in setting W t (j) it maximizes the size-adjusted utilities of its members. Given the wage-setting decisions by unions that do re-optimize, and the fact that unions that do not re-optimize must keep their nominal wages at last period’s value, there is an analogue of the Phillips curve (Equation 3.4). In particular, nominal wage inflation π ˆ w,t = logWt − logWt−1 will be a function of expected wage inflation tomorrow and the deviations of each type of household’s marginal rate of substitution from their steady state level.

(3.4)

π ˆ w,t = βEt π ˆ w,t+1 − λˆ µt

(3.5)

h

ˆ t + (1 − ω)ˆ µ ˆt = wˆh,t − (1 − α)ˆ st − ϕL c0t + ωˆ c00t | {z } real

27Because

cons

i

wage

there are (1 − ω) unions, but only ω ≤ (1 − ω) HtM HHs, each HtM household is a member of (1 − ω)/ω ∈ Z+ union(s). In steady state, each HtM HH supplies a fraction ω/(1 − ω) of it total per capita labor supply to each.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

where λ =

(1−θw )(1−θw β) , θw (1+ϕεw )

19

and wˆh,t = wˆh,t−1 + π ˆw ˆ h,t is the real product wage. The set up in t −π

the foreign region is analogous. Note that the equality of labor supply across types in Equation 3.3 weaken wealth effects on labor supply of HtM HHs receiving transfers: their desire to work less only affects their labor supply through the weighted average of marginal utilities in Equation 3.5.28 3.1.5. Monetary Policy & Interest Rates. Monetary policy is mostly irrelevant for the size of crossregion transfer multipliers because fiscal effects in one region are offset by changes in others, leaving aggregates largely unchanged. Nonetheless, I close the model by assuming aggregate nominal ˆ tAGG = φπ [nˆ πh,t + (1 − n)ˆ πf,t ].29 interest rates follow a Taylor rule R To ensure stationarity, I add a small debt-elastic interest spread ψ as in Schmitt-Grohe and Uribe (2003), which slowly moves the model back to steady state (where (1 − ω)ˆbYun,t is the deviation of debt

(3.6)

from

steady

state

as

share

of

GDP).

ˆt = R ˆ tAGG + ψ(1 − ω)ˆbYun,t + sp R ˆt

3.1.6. Relation between open economy relative multipliers and multipliers at the ZLB. Cross-region multipliers in a monetary union are completely different from multipliers at the Zero Lower Bound (ZLB) in a closed economy, despite the relatively fixed nominal interest rate in each case (Corsetti et al 2011, Farhi and Werning 2013, Nakamura and Steinsson 2014). The difference is that in the open economy, an increase in the price of home goods affects the terms of trade, which reduces competitiveness and demand for home goods. For temporary shocks, this means price changes are reversed in the long run, and hence long term real interest rates do not fall with higher short run inflation, mitigating the short run increase in consumption.

(3.7) 28Since

sˆt = π ˆf,t − π ˆh,t + sˆt−1

the union has market power they set the steady state wage at a markup εw /(εw − 1) above the household marginal rate of substitution. For small deviations from this, households are willing to provide extra labor to meet demand. The fact that each household only belongs to a finite number of unions means that the wage Phillips curve has the standard slope λ as in Erceg et al (2000) and Galí (2008), rather than the steeper slope of (1 − θw )(1 − θw β)/θw in Colciago (2011) and Schmitt-Grohe and Uribe (2005). 29R ˆ tAGG = lnRtAGG − lnRAGG is the log deviation of the nominal interest rate from its steady state level. nˆ πh,t + (1 − n)ˆ πf,t is monetary union-wide inflation.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

20

3.1.7. Fiscal Policy. Cross-region transfers (corresponding to a US fiscal system) consist of exogenous balanced-budget lump-sum transfers from households in the foreign region to households in the home region in proportion to their population shares (that is, the HtM household receives a fraction ω of transfers). Transfers are expressed as a deviation from steady state as share of Y

ˆ ), and follow a first order autoregressive process with persistence ρ. GDP in the home region (tr Because the home region is small, this is almost identical to results when lump-sum taxes fall on the whole monetary union. Self-funded transfers (corresponding to a EU fiscal system) consist of an equal-sized transfer to all residents of the home region funded by lump sum taxes on the unconstrained households in the home region (these can be thought of as the wealthier households, or those with a large mortgage interest tax deduction). Government purchases are wasteful purchases of home goods, which can either be crossregion (foreign) financed or self financed. If purchases are cross-region (US fiscal system), equal per-capita lump sum taxes are levied on the residents of the large foreign region.30 If purchases are self-financed, lump sum taxes are levied on home unconstrained households. The timing of tax payments doesn’t affect the results for either purchases or transfers (i.e. whether fiscal policy is balanced budget, or debt funded.) For self-financed transfers and purchases, this is an exact result because the home households paying the taxes are Ricardian. For cross-region transfers and purchases, it is an approximate result (exact as n → 0), because the home region is sufficiently small relative to the foreign region that the foreign economy is relatively unaffected by the taxes levied.

3.1.8. Measuring labor income and GDP using national prices. For the US, there are no statespecific GDP deflators, so instead I deflate nominal variables using the US CPI (in the model, US CPI and GDP deflator are identical). Because the small region and the monetary union as a whole consume goods in different proportions, the terms of trade sˆ appears in the expression for labor income or output deflated by the union-wide CPI. I label these measures “BEA” because its how they can be measured in the data.

30Given

the home region is small, results are almost identical if taxes are spread across the whole monetary union.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

21

In a sticky price & wage model, movements in wˆ and sˆ are small initially, so labor income and ˆ = Yˆh ). In the neoclassical model, GDP using national prices are very similar to real GDP (recall L because markups are constant, real product wages wˆh are also constant, and hence WˆLBEA = YˆBEA = Yˆ − (1 − n)ˆ s. For low values of armington elasticity θT , large movements in YˆBEA are possible in a neoclassical model due to big swings in the terms of trade.

(3.8)

ˆ − (1 − n)ˆ WˆLBEA = wˆ + L s

(3.9)

YˆBEA = Yˆ − (1 − n)ˆ s 4. Analytical impact multipliers in a simple model

In this section, I present some analytical expressions in a simplified version of the model to illustrate the drivers of the cross-region transfer multiplier (US fiscal system) in a New Keynesian (NK) model, and investigate how it compares to the self-funded transfer multiplier (EU fiscal system), the transfer multiplier in the neoclassical model and the purchases multiplier. The simple NK model involves three changes: (i) prices are fixed (θ → 1, κ → ∞), (ii) the home region becomes infinitesimally small n → 0,31 and (iii) there is no debt-elastic interest spread (ψ = 0). The simple version of neoclassical model only requires the final assumption. As policymakers are interested in the potential short-run stabilizing effects of the transfers, I focus on the impact transfer multiplier (denoted MT R ≡ ∂ Yˆ0 /∂tˆ r0Y ): the dollar-for-dollar change in output in the first quarter the transfer is received. 4.1. Analytical transfer multipliers in the New Keynesian Model. Quantitatively, impact multipliers in the simple model are very similar to those in the full model so long as the shocks are not too persistent (Table 4 in brackets). The reason is that in a sticky price model, firms are reluctant to change prices in response to temporary shocks. The other simplifying assumptions 31This

assumption is not strictly needed for characterizing the cross-region transfer multiplier (though it simplifies expressions). I need perfectly rigid prices to fix the lagged terms of trade sˆt−1 = 0. With sticky prices, sˆt−1 is an endogenous state variable, which means that output is not a constant multiple of the size of the transfer. With flexible prices sˆt can change instantaneously, making sˆt−1 unimportant.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

22

(n → 0, ψ = 0) have little quantitative impact. Proofs are in the Online Appendix, though follow from the lemma below. Because prices are fixed, the standard multiplier is equal to the BEA version using national prices. Lemma. Consider an AR(1) cross-region (US) or self-funded (EU) transfer shock with persistence h i ˆ ˆ ρ in the simple NK model. Then along the adjustment path following the shock, Yt − Y∞ also follows an AR(1) process with persistence ρ, where Yˆ∞ is the limit to which home output converges. Proposition 1. [Analytical New Keynesian Transfer Multipliers] In the simple NK model: A) the self-funded (EU-style) transfer impact multiplier is given by:

(4.1)

MSF,T R

  1−β αω 1− = 1 − αω 1 − βρ | {z } Relative M P C

B) the cross-region (US-style) transfer impact multiplier is given by:

(4.2)

MCR,T R = MSF,T R +

α 1−β 1 − α 1 − βρ | {z }

P ure CR multipler

Self-funded transfers (defined in Section 3.1.7) effectively transfer money from the unconstrained HHs to the constrained HtM HHs, with the transfer shrinking over time at rate ρ. As European countries can “self insure” using countercyclical self-funded transfers, the self-funded transfer multiplier is an important baseline against which I compare the size of cross-region transfer multipliers in Section 6. Proposition 1A (Equation 4.1) shows that there are two components to the self-funded transfer impact multiplier. First, there is the excess of marginal propensity to consume (MPC) out of transfers received by HtM HHs (unity) relative to MPC out of targeted transfers paid by unconstrained HHs (1 − β)/(1 − ρβ).32 This difference is maximized when transfers are perfectly front loaded (ρ = 0). Second, this term is weighted by αωT /(1 − αωP ), with ωT being the proportion of the transfer that accrues to HtM households (targeting), ωP the population 32The

unconstrained HHs reduce consumption by the present value of transfers paid. For a one-off transfer, this is the quarterly interest 1 − β. For permanent transfers (ρ → 1), consumption of the unconstrained HHs fall by unity.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

23

share of HtM HHs, and ω = ωT = ωP in the calibration in Equation 4.1 (untargeted transfers). α is the expenditure share on local goods. As such, αωT [1 − (1 − β)/(1 − βρ)] is the marginal propensity to consume out of temporary untargeted self-funded transfers. αωP is the marginal propensity to consume local goods from general temporary increases in income with 1/(1 − αωP ) being the traditional Keynesian Multiplier.33 The Cross-region transfers multiplier has two components (Proposition 1B, Equation 4.2). The first part is the self-funded transfer MT R,SF from above, which is only positive if transfers are temporary (ρ < 1) and there is a positive share of HtM HHs (ω > 0). The second component is what I call the pure cross-region transfer multiplier which applies even if those conditions are not met. Pure-cross region transfers boost output by increasing consumption demand by unconstrained HHs. As these HHs are Ricardian, the pure cross-region transfer is consumed because it never has to be repaid (unlike locally-financed transfers, or a loan). However, the unconstrained HHs only consume the permanent component of the transfer: (1 − βρ)−1 is the present value of the transfer payments and (1 − β) is the interest on that sum. If the transfer is permanent (ρ → 1), then all of the extra transfer income is consumed, though if it is very temporary (ρ → 0), then most of the transfer is saved. Corollary 1. In the simple NK model described above with home bias in consumption (α > 0) A. The cross-region transfers multiplier is always larger than the self-funded transfer multiplier (i.e. the pure cross-region multiplier is positive). B. The difference is only large if transfers are persistent (high ρ) or there is large home bias (high α). C. The share of the cross-region transfer impact multiplier explained by the self-financed transfer multiplier is decreasing in ρ Corollary 1 previews the main results of Section 6. Because European countries are too open (α too small) and recessions are not persistent enough (ρ too low), the short term stabilization gains from US-style countercyclical cross-region transfers are only slightly larger than those a European 33Equation

ˆ YSF,0 where αωP Yˆh,0 is demand 4.2 can be rearranged as Yˆh,0 = αωP Yˆh,0 +αωT [1 − (1 − β)/(1 − βρ)] tr |{z} | {z } Supply

generated by higher income.

Demand directly f rom transf ers

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

24

country could achieve for itself with a self-funded countercyclical transfer policy. With α = 0.56 and ρ = 0.935 as I assume in Section 6, home unconstrained households will save 87% of the cross-region transfer, and the “extra” impact multiplier on cross-region transfers is only 0.15 above what would be available with self-financed transfers. Moreover, the simple model assumes that prices are perfectly rigid, if not, the gains are even smaller. In terms of proportions, the share of the cross-region transfer multiplier explained by self-funded transfers decreases with ρ. With my parameters of ω = 1/3 and α = 0.56, almost all of the crossregion transfer multiplier is explained by the self-financed component when ρ = 0, almost none when transfers are permanent, and around 1/2 − 2/3 when transfers are as persistent as the Irish recession (ρ = 0.935). This suggests that unless cross-region transfers are very persistent, selffinanced transfers are almost as effective as cross-region transfers. Home bias The size of the pure cross region transfer multiplier is very sensitive to the degree of home bias (α) — this is the fraction of extra consumption spent on home goods without any change in prices (ˆ s = 0). Even if prices are fixed, if there is no home bias (α → 0) the crossregion transfer multiplier is zero. As the home economy becomes completely closed (α → 1), the transfer multiplier goes to infinity. As output is demand-determined, higher consumption only boosts demand if it is spent on domestically produced goods. To see why the multiplier has the form α/(1 − α), consider a perfectly persistent cross-region transfer (ρ → 1) which means that all the transfer is consumed by Ricardian HHs and MT R,SF = 0 (Figure 2 Panel A).34 The economy is originally at point A (terms of trade s = 1). The income expansion path has slope α/(1 − α) reflecting the fact that for every $1 of extra income, the household wants to spend $α on home goods and $(1 − α) on foreign goods. The home household receives a transfer from abroad of size AE, which shifts the budget constraint from BB to CC. But this is not an equilibrium, as there is now excess demand for home goods, and excess supply for foreign goods. As home output is demand-determined, output increases until the increase in consumption of the foreign good (= size of the transfer) is a fraction (1 − α) of the total increase in home income (tr + ∆y), at point G on budget constraint DD.35 The more closed the home region, the steeper the income expansion path, and the larger the transfer multiplier. 34I

thank Giancarlo Corsetti for suggesting this diagram, which is also similar to the one in some versions of Farhi and Werning (2014). 35Another way to see this is that ∆c =tr = (1 − α)(tr + ∆Y ), which is rearranged to ∆y = [α/(1 − α)] tr. f

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

25

Figure 2 here 4.2. Analytical transfer multipliers in the neoclassical model. The neoclassical model is simple enough to solve analytically with few additional assumptions. Proposition 2 suggests that the cross-region transfer multiplier is negative and often small. The household will only respond to the perpetuity value of the transfer (1 − β)/(1 − ρβ), and the extra income will make it want to work less due to the neoclassical wealth effect, reducing output by (1 + ϕ)−1 of the perpetuity value of the transfers (ϕ−1 is the Frisch elasticity of labor supply). The higher is ϕ−1 , the stronger are wealth effects, and the larger the fall in output. As the household saves the temporary part of the transfer, the fall in output is constant along the adjustment path. The self-funded transfer multiplier is trivially zero because there is only one type of household. Proposition 2. [Analytical Neoclassical Transfer Multipliers] In the neoclassical model (NC): A. The cross-region transfer multiplier is negative and given by:

NC MCR,T R = −

(4.3)

1 1−β <0 1 + ϕ 1 − ρβ

B. The cross-region multiplier using national prices (BEA) is of ambiguous sign and is given by:

(4.4)

NC MCR,T R (BEA)

  1 1 + αϕ 1−β = −1 ≶0 1 + ϕ (1 − α) [(θT − 1)(1 + α) + 1] 1 − ρβ

C. The self-funded transfer multiplier is trivially zero (as there is only one type of agent). The response to a permanent cross-region transfer is shown in panel B of Figure 2. The situation is initially the same as the case of fixed prices. However in the NC model, the economy responds to excess demand for home goods by increasing their relative price s0 = 1 → s1 < 1. The household also wants to spend a fraction of its extra income on leisure, which reduces labor hours and output.36 Combined, these factors cause the budget constraint to shift in and swivel to DD, the household produces less, and the household substitutes consumption towards foreign goods (at the lower relative prices), and consumes at point G. 36With

log preferences, the fall in output does not depend on the movement in the terms of trade, and hence does not depend on θT .

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

26

Even though output falls in response to a cross-region transfer in the neoclassical model, labor income or output at national prices (Equation 3.8-3.9) could rise or fall due to movements in the terms of trade (Equation 4.4). The higher the armington elasticity of substitution across goods θT , the more muted price movements, and the more negative the multiplier.37 For my default parametrization (α = 0.56, ϕ−1 = θT = 2), movements in labor income and output at national prices are close to zero even for permanent transfers, which is inconsistent with estimates in the data (see Section 5.3 for a detailed discussion). 4.3. Extension: Analytical Purchase multipliers (NK model). A number of papers have found large federally-funded government purchase multipliers at the regional level (Nakamura and Steinsson 2014, Suarez and Wingender 2012), but smaller self-financed government purchase multipliers (Clemens and Miron 2012). Could financing explain the difference? Proposition 3 shows that in the simple NK model, purchase impact multipliers are one unit higher than transfer multipliers financed in the same way (and with the same persistence). A corollary is that the difference between federally-financed and self-funded purchase multipliers is

the

pure

cross-region

transfer

multiplier

[α/(1 − α)] [(1 − β)/(1 − βρ)]. The pure CR transfer impact multiplier is around 0.17 with my default parameters (ρ = 0.935; α = 0.56), or 0.3 with Nakamura and Steinsson’s (ρ = 0.93 and α = 0.69) so even with fixed prices, the difference in financing cannot explain large federallyfinanced purchase multipliers. These results are consistent with Nakamura and Steinsson (2014), who show numerically (in a similar model) that the federally-financed multiplier is only marginally greater than the locally-financed multiplier.38 Proposition 3. [Analytical New Keynesian Purchase Multipliers] In the simple NK model, A. The cross-region purchases impact multiplier (MCR,G ) is equal to one plus the cross-region transfer multiplier (MCR,T R , defined above)

(4.5)

MCR,G = 1 + MCR,T R

α → 0, the multiplier is (1 + ϕ)−1 (θT−1 − 1)(1 − β)(1 − ρβ)−1 . and Steinsson’s (2014) multipliers are over two years, rather than on impact, so numbers are not directly comparable. The gaps between federally funded and locally financed impact multipliers are slightly smaller in the full NK model, and the difference falls over time as prices adjust.

37With

38Nakamura

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

27

B. The self-funded purchases impact multiplier (MSF,G ) is equal to one plus the self-financed transfer multiplier (MSF,T R defined above)

(4.6)

MSF,G = 1 + MSF,T R

C. The difference between cross-region and self-funded purchases multipliers is the pure crossregion transfer multiplier.

MCR,G − MSF,G =

α 1−β >0 1 − α 1 − βρ | {z }

P ure CR T rans M ult

Intuition As output is demand-determined, a $1 increase in gh increases home output by $1 today, and hence increases incomes by $1. This explains the “1” in Proposition 3A and B, though the size of the other term depends on what happens with this extra $1 of income. If g is foreignfinanced, no taxes need to be paid so incomes in the home region today increase by the full $1, which has the same macroeconomic effects as a $1 cross-region transfer (explaining MCR,T R ). If g is self-financed, the $1 income gains is partially offset against the perpetuity value of taxes paid, as in a self-funded transfer (explaining MSF,T R ). 4.4. Relation to theoretical literature. To my knowledge, the decomposition of cross-region transfer impact multipliers into “pure” and “self-funded” components is new, as well as all expressions involving hand-to-mouth households (ω > 0), or self-funded transfer multipliers. A number of other expressions are related to those in Farhi and Werning (2013, 2014) in special cases.39 Specifically, my expression for the cross-region transfer impact multiplier (Proposition 1B) matches that of Farhi and Werning (2013) Proposition 9 (the same as Farhi and Werning (2014) Proposition 17) if I set ω = 0 and they set t = 0. In the long run (with t → ∞), these expressions converge to my multiplier in the neoclassical model (Proposition 2A). Farhi and Werning’s (2014) foreign financed purchase multipliers (Proposition 8) are the same as in my Proposition 3A when ω = 0. 39Farhi

and Werning (2013) do very briefly discuss purchase multipliers in the open economy with hand-to-mouth households as an extension (their Proposition 13), but only consider complete markets for optimizing households (which undoes the effect of transfers), and they evaluate their expressions numerically. Farhi and Werning call ρN F A0 the annuity value of the cross-region transfer (where ρ is the discount rate), which with my set up (and timing assumptions) is (1 − β)/(1 − βρ)tr . They refer to home bias as 1 − α, whereas in this paper it is α.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

28

5. Quantitative results 5.1. Calibration . Most of the parameters such as price stickiness and the CES Home-Foreign elasticity come from Nakamura and Steinsson (2014) who have a similar multi-region model (see Table 3 for main parameters) or are relatively standard in the literature.40 Moreover, most of the parameters are appropriate for both a typical US state or a small European country (Ireland). • Ireland is about 2% of EU GDP, which is the same as the average US state, so I calibrate n = 0.02. • Kaplan et al (2014) argue that a third of households are hand-to-mouth in the US, UK and Germany, so I calibrate ω = 1/3.41 • I calibrate home bias in consumption α = 0.56, based on non-traded services as a share of US personal consumption expenditures. This implies export value added is around half of GDP, which is close to the value in Ireland.42 However, one important difference is the degree of wage stickiness, which is much higher in Europe than the US. In the US, nominal wages tend to be set on average once a year, so I set θw = 0.75 (as is standard in the literature). In Ireland, the average length of collective bargaining agreements is 3 years (Du Caju et al 2008), so I set θw = 0.912. High rates of wage stickiness are also needed to produce recessions as persistent as they are in the data (Section 6). Table 3 here 40I

choose the same Frisch elasticity of labor supply as estimated in Smets and Wouters (2007), and close to the average of the values used by Bernanke et al (1999) and Christiano et al (2005). It is also in the middle of the range of a number of macro studies (2.8) and micro studies (0.8) surveyed by Chetty et al (2011), but it is slightly higher than the value of unity used by Nakamura and Steinsson (2014). Other parameters are ψ = 0.00005 for the debt-elastic interest spread (Schmitt-Grohe and Uribe 2003; annual figure of 0.00074 converted to quarterly figure); β = 0.99 for the discount rate (Nakamura and Steinsson 2014), and CES elasticity across labor varieties in the New Keynesian model (εw = 21; Christiano et al 2005) 41In an earlier version of this paper, I defined households as HtM if they had less than $1000 in liquid assets in the 2010 Survey of Consumer Finances. 1/3 is similar to other values in the literature, such as 18-35% estimated by Kaplan and Violante (2014) for the US, 36% as estimated by Iacoviello (2005) and between 26% (Cogan et al 2010) and 50% (Campbell and Mankiw 1989). 42The level of home bias is lower than in Nakamura and Steinsson (2014) (α = 0.69) as they calibrate to a US statistical region, which is much larger than a US state. Midrigan and Philippon (2011) calibrate α = 0.75 for non-durable goods. As the effect of cross-region transfers increases with home bias, my calibration can be viewed as conservative. For the US, I exclude “Transportation” and “Financial Services and Insurance” from non-traded services as they could be purchased across state borders. Data from BEA Table 2.3.5U “Personal Consumption Expenditures by Major Type of Product and by Major Function” for 2014. For Ireland: Over 2005-09 Irish Exports to GDP averaged 84% of GDP. However in 2009 only 58% of these were domestic value added (OECD/WTO Trade in Value added indicators, May 2013). This suggests a export value added ratio of around 48% of GDP.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

29

5.2. Impulse responses in the full model. Whether counter cyclical transfers boost output depends on the class of model used to analyze them. As foreshadowed in the analytical results above, the New Keynesian model predicts a sizable increase in output initially (because prices and wages are sticky, and constrained households spend transfers), whereas the neoclassical model predicts a small fall in output (most of the transfer is saved, but some of the rest is “spent” on extra leisure). The size of impact multipliers are shown in Table 4, and are quite similar to those from the simple model so long as transfers are not too persistent. For one-off transfers, the multipliers on the self-funded transfers and cross-region transfers are almost identical to each other (and almost identical to the simple model). In contrast, when transfers are perfectly persistent, cross-region transfer multipliers almost triple, but self-funded transfer multipliers are zero. Transfer multipliers in the neoclassical model are close to zero for one-off transfers, and negative for more persistent transfers. The LHS of Figure 3 shows the response to a 1% of GDP cross-transfer shock with persistence ρ = 0.935 (similar to that of the Irish recession in Section 6) in both types of models. Output falls in the neoclassical model because prices of home goods rise, leading consumers to switch their expenditure away from home goods (Figure 3 RHS). Over time prices and wages also adjust in the NK model, leading output to fall. As such, the cross-region transfer multiplier in a NK model is only positive in the short term; long-term multipliers tend to be negative.43 How quickly output falls in the NK model depends on price and wage stickiness. If wages are as sticky as they are in Europe (green line), then the impact multiplier is slightly larger, but relative prices fall much more slowly, and the increase in output is much more persistent. Figure 3 here Table 4 here 5.3. Comparison of model and data. The response of a state economy to a transfer shock in the data over the first few quarters is consistent with the New Keynesian model, but inconsistent with the neoclassical model (Figure 4). Figure 4 is expressed in changes rather than levels, to be consistent with the empirical specification in Section 2.44 In a sense, this exercise can be thought 43With

our default calibration, present value multipliers (the present value of the change in output divided by the present value of the transfer) are negative for all but the most short lived transfer programs. 44In the model Y ˆt − Yˆt−1 = lnYt − lnYt−1 ≈ (Yt − Yt−1 )/Yt−1

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

30

of as estimating the slope in Figure 1, where the slope equals the difference in the growth rate of labor income or GDP across states relative to the difference in the size of the transfer received. The empirical estimates are taken from Columns 1 and 3 of Table 1, and Column 1 of Table 2.45 The first two panels report the response of labor income and GDP (respectively) to a 1% of GDP temporary transfer, such that the change in transfer income is 1% for the first quarter, -1% for the second quarter, and zero thereafter.46 One can see that the estimates from the New Keynesian model line up almost perfectly with those from the data (especially for labor income). In contrast, non-response to a transfer shock in the neoclassical model is outside the 95% confidence interval for labor income (borderline for GDP), and a long way from point estimates in either case. In the final panel is the response of labor income to a permanent transfer (a 1% GDP change in period 0, zero thereafter). While the response of labor income to a permanent transfer in the NK model is below the estimated effect, it is within the 95% confidence interval and so is statistically consistent with the data. In contrast, the small response in the neoclassical model is outside the confidence interval.47 Figure 4 here 6. Counterfactual: A US-Style Transfer Union in Europe? Given the troubles of the Euro area in recent years, many economists have been encouraging European countries towards greater fiscal integration. The United States has long been seen as an example of risk sharing across states — largely through cross-region transfers — as the opening quotation from Sala-i-Martin and Sachs (1991) more than 20 years ago indicates. But would Europe be better off with a US-style centralized fiscal system, in terms of its ability to smooth regional shocks? In this section, I model the effect of countercyclical cross-region transfers on output volatility in the context of a deep and persistent regional recession, similar in scale and length as the Irish 45

The model, coefficients in Figure 4 are taken from an impulse response function, but they are almost identical when estimated by running an univariate regression of output growth on transfers in simulated data. 46 In the data, I estimate a panel with state and quarter fixed effects, and report the impulse response function of the relative multiplier. In the model I report the impulse responses of home output to a cross-region transfer shock. These are comparable because the large region is absorbed almost entirely by the time fixed effects, and the output in the foreign region barely responds to the size of the transfer. 47 Although output and labor supply fall in the neoclassical model, the rise in the price of the home good more than offsets this, leading to a small increase in labor income measured at national prices.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

31

recession from 2009-12. As Europe doesn’t have countercyclical cross-country transfers, the effect of such a policy can only be evaluated in a model. The results are of course dependent on the type of model used; which is why we needed to inform this choice by comparing the model’s predictions to the empirical response to a transfer shock in the US in the previous section. In a linear model, the effect of countercyclical transfers in response to an unanticipated shock only depends on the time path of output driving the countercyclical fiscal response; exactly which shock causes the recession does not matter.48 As such, I abstract from real-world details like the fall in house prices or the solvency of banks and instead generate a recession with a large and persistent increase in borrowing costs, similar to that used by Farhi et al (2014) (a “spread shock”). This also means that the comparison to the Irish recession shouldn’t be taken too literally — it is only the depth and persistence of the recession that are relevant. The main result is somewhat surprising in light of the fact that cross-region transfers stimulate regional output in the US in the short term. Specifically, the ability of countercyclical crossregion transfers to smooth regional output shocks in an “ultra-Keynesian” model (consistent with the data) is quite modest, and more importantly, it has very little advantage over a policy of countercyclical self-funded transfers (countercyclical fiscal policy by the EU country itself).49 Simulating a deep and persistent regional recession in the European Periphery (“Ireland”) The model used to study countercyclical cross-region transfers in Europe is identical to the New Keynesian model for the US, except that wages are more sticky in Ireland than the US (see Section 5.1 and Table 3).50 Following Farhi et al (2014), I increase the spread on borrowing by home consumers, sp ˆ in Equation 3.6 by 0.005 (2% annualized) and with quarterly auto-correlation 0.95.51 The shock leads unconstrained households to save more, reduces consumption demand of the home good and sends the economy into recession. Countercyclical cross-region fiscal policy 48

That is, a linear model rules out interaction between variables, which are second order. If there were occasionally binding constraints or other non-linearities, the results might be different. 49 This finding is consistent with Bayoumi and Mason (1995), who argue that European countries have been able to compensate for a lack of cross-regional transfers by borrowing during recessions and saving during booms. This of course depends on the solvency of European governments in times of crisis, an evaluation which is beyond the scope of this paper. 50 This calibration is appropriate as Ireland is about as large and open as a typical US state. Results are also robust to assuming steady state Irish debt of about 100% of GDP. 51Farhi et al (2014) used a spread shock to generate a recession for Spain (with same persistence). My shock is around half the size because my model has a higher inter-temporal elasticity of substitution/lower risk aversion, which makes consumption more sensitive to changes in interest rates. The spread shock leads to a large improvement in the trade balance for Ireland, as observed in the data.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

32

aims to cushion regional shocks which reduce output in the region/country relative to the rest of the monetary union. As such I calibrate the model to the change in Irish GDP relative to that in the EMU, assuming no countercyclical transfers.52 The spread shock in the NK model does a good job of matching the size and persistence of the Irish recession. Specifically, the initial fall in GDP of 3.78% in the model is the same as the fall in GDP in Ireland relative to the EMU over 2009Q1-Q3 (Table 5 row 1) and the model approximately matches the time path of the recovery (Figure 5). Adding cross-region transfers The empirical evidence in the introduction suggests that in the US, net federal transfers are around 20-40% of the fall in the income of the state. As such, in the counterfactual simulations I assume that that the residents of “Ireland” receive a lump sum transfer from the rest of the monetary union of 30% of any fall in GDP (and have to pay 30% of any rise in GDP to the rest of the EMU). Table 5 Column C1 shows that this policy does reduce the size of the fall in GDP following the spread shock: instead of GDP falling by 3.78% it instead falls by 3.46% (i.e. the recession is 0.32ppt or 9% shallower). Figure 5 shows that the effect of the transfer is largest initially, but after a few years it is hard to distinguish output paths with and without cross-region transfers. Stabilization and the size of the cross-region transfer multiplier The ability of the cross-region transfer to smooth output depend on the size of the cross-region transfer multiplier, which is why the rest of this paper investigates the size of the multiplier in the NK model and in US data. The fall in GDP initially is 3.78%, which generates a cross-region transfer of about 1.1% of GDP (30% × 3.78%).53 The fall in output has a persistence of around 0.935, which from Table 4 suggests the cross-region transfers generated have an impact multiplier of around 1/3. As such, the countercyclical transfer boosts output by 1/3 × 1.1=0.37ppts of GDP, very close to the 0.32ppt fall in simulations in Table 5.54

52On

one hand the Irish budget deficit increased over 2008-09, but on the other hand the structural balance actually tightened (WEO database April 2015). Given the differences, I adopt a compromise position and assume that no countercyclical (or procyclical) transfers. 53The gains from stabilization are modest in part because of the modest size of countercyclical cross-region transfers. Larger countercyclical cross-region transfers — for example as part of regional bank bailouts — might have larger effects, but would also affect credit spreads, which make them harder to assess. 54The decomposition is not exact because along the IRF, GDP does not follow an AR(1) process exactly.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

33

Comparison with self-financed transfers An alternative policy available to members of the EMU is to conduct their own (self-funded) countercyclical transfers. As before, this policy involves a lump-sum transfer from unconstrained HHs to all agents in the economy, though the transfers from the unconstrained household to themselves net out. To make this comparable with countercyclical cross-region transfers, I assume that this self-funded transfer is 30% of any fall in GDP. Column C2 of Table 5 shows that the initial fall in output is only slightly larger with self-financed countercyclical transfers than with cross-region transfers (-3.57% vs -3.46%). Visually, it is difficult to distinguish the path of output in economies with self-financed vs cross-region transfers (Figure 5). Quantitatively, self-financed transfers provide around 2/3 of the benefits in terms of output smoothing as cross-region transfers. This was foreshadowed in Table 4, where the impact multiplier of self-financed transfers with ρ = 0.935 is around 2/3 that of cross-region transfers. Why the limited stabilization benefits of cross-region transfers? Proposition 1 suggests that countercyclical self-funded and cross-region transfers have similar smoothing properties because the pure cross-region transfer multiplier is small. In turn, this is because (i) countercyclical transfers are temporary (if they weren’t temporary they couldn’t be countercyclical) and (ii) European economies are reasonably open. The lack of persistence is important because only the unconstrained HHs respond differently to transfers from Dublin vis-a-vis Brussels (HtM HHs will spend both).55 Unconstrained households are Ricardian and save temporary transfers. With transfers of a similar persistence to those considered here (ρ = 0.935), the analysis following Corollary 1 suggests that the vast majority of the cross region transfer is saved. Extension: Output volatility A system of countercyclical cross-region transfers ideally reduces overall output volatility in regional economies in addition to smoothing out a single recession. To test this, I simulate the model for 1000 quarters with business cycles driven by a random sequence spread shocks (as above), with either no transfers, countercyclical cross-region transfers (a “fiscal union”), or countercyclical self-funded transfers. In the baseline case with no transfers, I match the standard deviation of quarterly Irish GDP growth over 1999-2015 of 1.9% (Table 5 Row 55This

is also true in the more general case where the home government borrows and then raises taxes in the future on both HtM and unconstrained households — HtM households don’t respond today to higher taxes tomorrow. Results are similar in an earlier version of the paper, when transfers were debt financed and taxes on both households responded to the level of debt as in Leeper et al (2010).

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

34

2 Columns A-B).56 Because the 2009-12 recession in Figure 5 was unusually severe I reduce the standard deviation of the spread shock from 2% to 1% (annualized). I find that the percentage fall in output volatility with countercyclical cross-region transfers is almost exactly the same as in the single crisis event above. Specifically, countercyclical cross-region transfers reduce output volatility by around 9%, from 1.9% (with no transfers) to 1.76% (with countercyclical cross-region transfers). However, gains from self-funded countercyclical transfers are not a lot smaller: the SD of GDP growth falls by 6.2% from 1.9% (with no transfer) to 1.81% (with the self funded transfer). Figure 5 here Table 5 here Robustness The fact that pure cross-region transfers are saved by Ricardian HHs means that even in very closed economies, the cross-region transfer multiplier is not large enough to substantially weaken the depth of recessions on the European periphery. Take the extreme parametrization of α = 0.9, which implies impact multipliers (for permanent transfers) above four, and well above the 95% confidence intervals in the US data. With ρ = 0.935 and an initial countercyclical crossregion transfer of 1.1% GDP (as assumed above for the Irish recession), the cross-region transfer only boosts GDP by 0.6ppts above that with self-funded transfers — which does little to reduce the severity of the recession. Given that economies on the European periphery are quite open — the Irish export value added to GDP is around 50% (suggesting α is closer to 0.5) — large stabilization gains from cross-region transfers are even less plausible. With these parameters (ρ = 0.935, α = 0.56), the stabilization gains from optimal cross-region transfers are also small in Farhi and Werning (2014). It is worth noting that that this combination of shock persistence and other model features (such as wage stickiness) generates higher transfer multipliers than one might get with alternative specifications. For example, the 2009-12 Irish recession was much more persistent than most recessions (which increases the impact multiplier): the average peak-to-trough post-war US recession lasts

56The

SD of the difference between Irish GDP growth and EMU-12 GDP quarterly growth is almost identical at 1.8%. In the model I abstract from the effect of aggregate shocks on regional GDP volatility.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

35

less than a year (nber.org/cycles.html). Other calibrations with less sticky wages, for example, would also lower cross-region transfer multipliers.57

7. Conclusion The presence of countercyclical cross-region transfers in the US is a key difference between EMU and US fiscal systems. As the future of the European fiscal system is debated and the United States federal system is put forward as a potential model for the EMU, it is important to have a better estimate of the effects of cross-region transfers in the US and a deeper understanding of the extent to which cross-region transfers can help to smooth regional shocks. My results suggest that even though cross-region transfers boost short-run non-transfer income in the states receiving them in the US, the reduction in regional output volatility from implementing a countercyclical “transfer union” in Europe would be modest — and not much larger than the gains from locally funded countercyclical transfers. In closing, a qualification is necessary. This paper has focused on the ability of countercyclical cross-region transfers to boost non-transfer incomes in regions of a monetary union through standard demand channels. While this is an important component of a US-style fiscal union, it is not the only one. This paper has little to say on other merits of fiscal centralization, such as the effects of a “banking union” on financial markets or the ability of guarantees by federal authorities to reduce spreads in regional sovereign debt markets. These are interesting areas for future research. An online appendix with supplementary material is available at https://sites.google.com/site/stevenpennings/PenningsCrossRegionTransfersOnlineAppendix.pdf

8. Data Appendix Data sources and construction • Earnings by place of work (labor income), Personal Income and BEA transfer data (downloaded 30 Sept. 2015), BEA Regional Economic Accounts, Quarterly State Personal Income Tables SQ4 and SQ35. http://www.bea.gov/regional/downloadzip.cfm 57In

the neoclassical model, the impact transfer multiplier with this level of persistence is slightly negative and spread shock leads to an increase in GDP because lower consumption increases labor supply.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

36

• Quarterly State GDP (downloaded 10 Dec. 2015), BEA Regional Economic Accounts Quarterly Gross Domestic Product (GDP) by State, GDP, All industries. This is a new series. http://www.bea.gov/regional/downloadzip.cfm • All variables are deflated by the Quarterly Personal Consumption Expenditures Chain-type Price Index (PCE) from the BEA and St Louis FRED (downloaded 7 August 2013). • 2008 low-income rebate: aggregate payouts were $28.25bn in 2008Q2 and $1.35bn in 2008Q3 (not annualized). Aggregate payouts and cross-state allocation from BEA (2009). Also see BEA (2008) • 2008 mid-income tax refund. According to BEA (2008) the middle-income component was $49.75bn (not annualized) in 2008Q2. Parker et al (2013) reports $15bn in payouts in 2008Q3 in total, leaving around $13.65bn in the mid-income component once the refundable component has been subtracted. The $49.75bn and $13.65bn payments are allocated across states using IRS data on 2007 Income tax returns (Tax Year 2007: Historical Table 2 (SOI Bulletin)58), using eligibility rules for the tax rebate. The cross-state allocation of total transfers in 2008Q2 is almost identical to contemporaneous growth in “All other personal current transfer receipts” (Online Appendix Figure 2) as calculated by the BEA, suggesting quite an accurate cross-state allocation.59 • 2001 Stimulus Payment. Johnson et al (2006) report payout of $38bn (not annualized) in 2001Q3. This is allocated across states using IRS data on 2000 Income tax returns (Tax Year 2000: Historical Table 2 (SOI Bulletin)), using eligibility rules for the tax rebate.60 58https://www.irs.gov/uac/SOI-Tax-Stats-Historic-Table-2

This is the same source for 2000 tax returns used in calculating the cross-region allocation of the 2001 stimulus. 59This calculation involves a number of assumptions given the aggregated nature of the data and complexity of the eligibility rules. Payouts are calculated by the number of single and joint taxpayers in each income bracket in each state, after adjusting for the fraction of returns that don’t pay any income tax. The IRS SOI provide use fairly wide income brackets, so I use the Census Bureau’s Current Population Survey (Table HINC-06) to approximate the phase out of the payments for higher income earners within income brackets using the US-wide income distribution. My allocation adds to about $75bn, which is slightly larger than $63.4bn middle-income component of the package. To keep the aggregate size of the package constant, I allocate the $49.75bn in 2008Q2 and $13.65bn in 2008Q3 across states in proportion to each states share. Due to a lack of data on the number eligible children (for those with a tax liability) across states, I assume that the $300 child payments for those with tax liabilities are allocated across states in proportion to the other payments and so don’t affect the share of payments received in each state. Using IRS microdata on the allocation across states produces very similar results. 60Payouts are calculated by the number of single ($300), joint ($600) and head of household ($500) taxpayers in each income bracket in each state, after adjusting for the fraction of returns that don’t pay any income tax (and households receiving partial payments). Household head taxpayers are an approximation of single parent households, who were eligible for $500 payments. My cross-state allocation adds to around $41bn, which is slightly

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

37

• Social Security Payments. The aggregate size of social security payments are taken from Romer and Romer (2014) Table 1. They report monthly payments as a share of personal income at annual rates. This is converted into non-annualized payments in the quarter. These payments are then allocated across states using each state’s share of total social security payments one year beforehand. Lagged social Security payments by state (for cross state allocation) taken from BEA Quarterly State Personal Income Table SQ35, which provides a breakdown of Personal current transfer receipts by category (including social security payments).61 • State population estimates from the US Census Bureau. US Aggregate GDP (Chained 2009 Dollars) from the St Louis FRED (both downloaded 24 May 2015). • Ireland & EMU GDP (Section 6) from Eurostat (downloaded 11 August 2015). Chain linked volumes (2010), Seasonally Adjusted. Table 6 here

References [1] Barattieri A., S. Basu and P. Gottschalk (2014) “Some Evidence on the Importance of Sticky Wages”, American Economic Journal: Macroeconomics 6(1): 70-101 [2] Bayoumi T. and P. Masson (1995), “Fiscal flows in the United States and Canada: Lessons for monetary union in Europe”, European Economic Review 39: 253-274 [3] BEA (2008): State Personal Income (October 2008) http://www.bea.gov/scb/pdf/2008/10%20October/spi_text.pdf#esabox [4] BEA (2009): Regional Quarterly Report (Jan 2009) http://www.bea.gov/scb/pdf/2009/01%20January/0109_regqtrlyreport.pdf [5] Bertrand, M., E. Duflo and S Mullinathan (2004), “How Much should we trust Difference-in-Difference Estimates?” The Quarterly Journal of Economics, 119(1) 249-275 [6] Bernanke B., Gertler M. and S. Gilchrist (1999), “The Financial Accelerator in a Quantitative Business Cycle Framework,” Handbook of Macroeconomics 1, pp 1341-1393. [7] Campbell J. and G. Mankiw (1989), “Consumption, Income and Interest Rates: Reinterpreting the Time Series Evidence,” NBER Macroeconomics Annual 4, pp 185-246. larger than the $38bn of payouts according to official data. The final estimate is each state’s share × $38bn. Results are quite robust to alternative assumptions. 61The cross-state allocation is almost identical to using the total amount of monthly benefits in current-payment status from hand-entered social security bulletin tables.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

38

[8] Chetty, R., A. Guren, D. Manoli and A. Weber (2011), “Are Micro and Macro Labor Supply Elasticities Consistent? A Review of Evidence on the Intensive and Extensive Margins”, American Economic Review Papers and Proceedings 101(3), pp 471-75 [9] Chodorow-Reich, G., L. Feiveson, Z. Liscow, and W. Woolston. (2012). “Does State Fiscal Relief During Recessions Increase Employment? Evidence from the American Recovery and Reinvestment Act.” American Economic Journal: Economic Policy 4(3): 118-145 [10] Christiano L., Eichenbaum M. and C. Evans (2005), “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy” Journal of Political Economy, 113(1):1-45. [11] Clemens J. and S. Miron (2012), “Fiscal Policy Multipliers on Subnational Government Spending”. American Economic Journal: Economic Policy, 4(2): 46-68. [12] Cogan J., Cwik T., Taylor J. and V. Wieland (2009), “New Keynesian versus Old Keynesian Government Spending Multipliers,” Journal of Economic Dynamics and Control, 34(3), pp 281-295. [13] Colciago A. (2011), “Rule-of-Thumb Consumers Meet Sticky Wages,” Journal of Money, Credit, and Banking 43(2-3), pp 325-353. [14] Corsetti, G., K. Kuester, and G. Muller (2011): “Floats, Pegs and the Transmission of Fiscal Policy,” in Fiscal Policy and Macroeconomic Performance, ed. by L. F. Cespedes, and J. Gali, Santiago, Chile. Central Bank of Chile. [15] Du Caju, E. Gautier, D. Momferatou and M. Ward-Warmedinger (2008), “Institutional Features of Wage Bargaining in 23 European Countries, the US and Japan”, European Central Bank Working Paper No. 974 [16] Erceg C., Henderson D. and A. Levin (2000), “Optimal Monetary Policy with Staggered Wage and Price Contracts,” Journal of Monetary Economics, 46(2), pp 281-313. [17] Erceg C., L. Guerrieri, and C. Gust (2006) “SIGMA: A New Open Economy Model for Policy Analysis” International Finance Discussion Papers 2005-835 [18] Farhi E. and I. Werning (2013) “Fiscal Multipliers: Liquidity Traps and Currency Unions”, Working Paper (October 2013) [19] Farhi E. and I. Werning (2014) “Fiscal Unions”, Working Paper (July 2014) [20] Farhi E., Gopinath, G. and O. Itskhoki (2014), “Fiscal Devaluations”, Review of Economic Studies 81, 725–760 [21] Feler L. (2015) “Local multipliers and spillovers from cash-transfers to the poor”, Working Paper SAIS [22] Feyrer J. and B. Sacerdote (2013), “How Much Would US Style Fiscal Integration Buffer European Unemployment and Income Shocks? (A Comparative Empirical Analysis)”, American Economic Review: Papers & Proceedings 103(3): 125–128 [23] Galí J. (2008), Monetary Policy, Inflation, and the Business Cycle: An Introduction to the New Keynesian Framework, Princeton University Press, Princeton, USA

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

39

[24] Galí J. and M. Gertler and D. Lopez-Salido (2007), “Markups, Gaps and the welfare cost of business cycles”, The Review of Economics and Statistics 89(1): 44-59 [25] Galí J. and T. Moncaelli (2008), “Optimal monetary and fiscal policy in a currency union”, Journal of International Economics 76: 116–132 [26] Galí J.and T. Monacelli (2005) “Monetary Policy and Exchange Rate Volatility in a Small Open Economy” Review of Economics and Statistics 72(3): 707-734 [27] Giambattistia E. and S. Pennings (2015) “When is the government transfer multiplier large?” Working Paper (November 2015) [28] Hill, E. (1990), “The S&L Bailout: Some states gain, many more lose”, Challenge 33(3), pp. 37-45 [29] Iacoviello M. (2005), “House Prices, Borrowing Constraints, and Monetary Policy in the Business Cycle,” American Economic Review, 95(3), pp 739-764. [30] Johnson D., Parker J. and N. Souleles (2006), “Household Expenditure and the Income Tax Rebates of 2001,” American Economic Review, 96(5), pp 1589-1610. [31] Kaplan G. and G. Violante (2014) “A Model of the Consumption Response to Fiscal Stimulus Payments”, Econometrica, 82(4), pp 1199-1239 [32] Kaplan G., G. Violante and J. Weidner (2014) “The Wealthy Hand-to-Mouth”, Brookings Papers on Economic Activity, (Spring 2014), pp 77-138 [33] Leeper E., M. Plante and N. Traum, (2010), “Dynamics of fiscal financing in the United States”, Journal of Econometrics, 156: 304-321 [34] Midrigan V. and T. Philippon (2011) “Household Leverage and the Recession” Working Paper NYU [35] Nakamura E. and J. Steinsson (2014), Fiscal Stimulus in a Monetary Union: Evidence from U.S. Regions, American Economic Review 104(3): 753-92. [36] Oh H. and R. Reis (2012). “Targeted transfers and the fiscal response to the great recession”, Journal of Monetary Economics, 59(S): S50-S64. [37] Parker J., N. Souleles, D. Johnson, and R. McClelland, (2013) "Consumer Spending and the Economic Stimulus Payments of 2008", American Economic Review, 103(6), pp 2530-53 [38] Romer C. and D. Romer (2015) “Transfer Payments and the Macroeconomy: The Effects of Social Security Benefit Increases 1952-1991” Working Paper (October 2015) [39] Sala-i-Martin X. and J. Sachs (1991), “Fiscal Federalism and Optimal Currency Areas: Evidence for Europe from the United States”, NBER Working Paper No. 3855 [40] Schmitt-Grohe S. and M. Uribe (2003), “Closing Small Open Economy Models”, Journal of International Economics, 61: 163-185

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

40

[41] Schmitt-Grohe, S., and M. Uribe. (2005) “Optimal Fiscal and Monetary Policy in a Medium-Scale Macroeconomic Model.” In NBER Macroeconomics Annual, edited by Mark Gertler and Kenneth Rogoff, pp. 383–425. Cambridge, MA: MIT Press. [42] Shoag D. (2011), “The Impact of Government Spending Shocks: Evidence on the Multiplier from Public Pension Plan Returns” Working Paper, Harvard University [43] Smets, F., and R. Wouters (2007). "Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach", American Economic Review, 97(3): 586-606. [44] Suarez J. C. and P. Wingender (2012), “Estimating Local Fiscal Multipliers”, Working Paper UC Berkeley [45] Wilcox D. (1989), "Social Security Benefits, Consumption Expenditure, and the Life Cycle Hypothesis," Journal of Political Economy, 97(2): 288-304.

9. Tables

Table 1. Temporary transfer multiplier — labor income and GDP ;ϭͿWŽŽůĞĚ ;ϮͿŽŵƉŽŶĞŶƚƐ ;ϯͿWŽŽůĞĚ ;ϰͿŽŵƉŽŶĞŶƚƐ ĞƉsĂƌŝĂďůĞ͗ 'ƌŽǁƚŚZĂƚĞƐŽĨ>ĂďŽƌ/ŶĐŽŵĞ;ϮϬϬϭͲϬϴͿ 'W;ϮϬϬϱͲϬϴͿ WŽŽůĞĚdƌĂŶƐĨĞƌƐ Ϭ͘ϮϰΎΎΎ Ϭ͘ϯϲΎΔ ;Ϭ͘ϬϴͿ ;Ϭ͘ϭϵͿ ϮϬϬϴZĞďĂƚĞƐ Ϭ͘ϯϯΎΎΎ Ϭ͘ϳϭΎΎΎ ;ůŽǁŝŶĐŽŵĞͿ ;Ϭ͘ϬϵͿ ;Ϭ͘ϮϱͿ ϮϬϬϴdĂdžZĞĨƵŶĚƐ ͲϬ͘Ϭϲ ͲϬ͘ϰϵ ;ŵŝĚŝŶĐŽŵĞͿ ;Ϭ͘ϯϭͿ ;Ϭ͘ϱϵͿ ϮϬϬϭdƌĂŶƐĨĞƌƐ Ϭ͘Ϯϵ ;Ϭ͘ϮϱͿ ^ƚĂƚĞΘYƚƌ& z^ z^ z^ z^ KďƐĞƌǀĂƚŝŽŶƐ ϭ͕ϱϴϭ ϭ͕ϱϴϭ ϳϯϳ ϳϯϳ EŽƚĞƐ͗ZŽďƵƐƚƐƚĚĞƌƌŽƌƐŝŶƉĂƌĞŶƚŚĞƐĞƐΎΎΎƉфϬ͘Ϭϭ͕ΎΎƉфϬ͘Ϭϱ͕ΎƉфϬ͘ϭ͘YƵĂƌƚĞƌůLJ'WĚĂƚĂŽŶůLJĂǀĂŝůĂďůĞ ĨƌŽŵϮϬϬϱYϮ͘KƵƚůŝĞƌƐхϯ^ĨƌŽŵŵĞĂŶĚƌŽƉƉĞĚ͘ƐƚŝŵĂƚŝŶŐƋϭƵƐŝŶŐK>^͘ΔƉͲǀĂůƵĞсϬ͘Ϭϱ

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

Table 2. The effect of permanent SS transfers on labor income

Model: CC HH share ω Home Bias (α)

Table 3. Parameters New Neoclassical Source/Target Keynesian 1/3 Kaplan et al (2014) 0.56 Non-traded services as share cons. (US)

Country/Region Size (n) Frisch Labor Elasticity (ϕ−1 ) Calvo prob not change wage (θw ) Calvo prob not change price (θp ) CES H-F Elasticity (θT ) Note: see text for further details.

or export Value Added/GDP for Ireland Average Size of US State; Ireland/EU GDP Smets and Wouters (2007)

0.02 2 0.75 (US)

-

0.912 (EU) 0.75

-

Barattieri et al (2014)/ Christiano et al (2005) Du Caju et al (2008)# Nakamura and Steinsson (2014)

2 #Average Collective Agreement Length in Ireland is 3 years (Fig 7)

41

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

42

Table 4. Transfer Impact Multipliers

1. Temp. (ρ = 0) 2. EU Crisis** (ρ = 0.935) 3. Perm. (ρ = 1)

Full NK model [Simple NK Model] A. Cross-Region B. Self-Funded 0.23 [0.24] 0.23 [0.23]

Neoclassical Model C. Cross-Region* -0.01

0.31 [0.37]

0.20 [0.20]

-0.11

0.61 [1.27]

0 [0]

-0.67

*Self-funded transfers multipliers are trivially zero in the Neoclassical Model

**

Persistence of Irish recession with Irish Sticky Wages (θw = 0.912)

Table 5. Irish Data and Countercyclical Transfers A. Data*

B. Model** C. No Transfers C1. (1) Crisis Fall in GDP -3.78% -3.78% Diff. Rel. No Transfer: (2) SD(∆lnGDP ) 1.9% 1.9% Diff. Rel. No Transfer:

Model with 30% GDP Countercyclical Transfers Cross-region Transfers C2. Self-funded Transfers -3.46% -3.57% 0.32ppt (8.5%) 0.21ppt (5.6%) 1.76% 1.81% 0.16ppt (8.3%) 0.11ppt (5.7%)

* Crisis: GDP in 2009Q3 relative to 2009Q1 for Ireland (relative to the EMU as a whole). ** Driven by a annualized spread shock of 2% (Row 1) or 1% (Row 2). GDP Data from Eurostat.

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

Table 6. Descriptive Statistics

43

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

44

10. Figures

2008Q2 Economic Stimulus Payments Real Growth in labor income, 2008Q2 −.03 −.02 −.01 0 .01 .02

OK

MA

CT

WV

LA

TX

KS CO WY PA MO AK UT NE AR CAMD KY NM VA WI WA SC DE HI IA OR OH NC IL AL ND ME ID MI GA MT INTNFL NH AZ NJ MN VT

MS

RI SD NY NV

.02

.03 .04 .05 .06 Transfer stimulus payments 2008Q2, share of labor income

.07

Line: growth=−0.02+0.3*payments(t=2.04) Source: BEA

Figure 1. Growth in Labor Income and transfer stimulus payments 2008

͘&ůĞdžŝďůĞƉƌŝĐĞƐ͗ĞĨĨĞĐƚŽĨƉĞƌŵĂŶĞŶƚĐƌŽƐƐͲƌĞŐŝŽŶƚƌĂŶƐĨĞƌ;ŶїϬͿ ůŶ;ĐŚͿ

/ŶĐŽŵĞĞdžƉĂŶƐŝŽŶ ƉĂƚŚǁŝƚŚƐϬсϭ

 īĞĐƚŽĨљƐ;ƐϭсƉĨͬƉŚфƐϬͿ z



/ŶĐŽŵĞĞdžƉĂŶƐŝŽŶ ƉĂƚŚǁŝƚŚƐϭфƐϬ

ȴz  y



ƚƌ



ͲƐϬсͲϭ D

'





ͲƐϭ

ȴD

Figure 2. Cross-region transfers with fixed prices (LHS) and flexible prices (RHS)

 ůŶ;ĐĨͿ

CROSS-REGIONS TRANSFERS IN A MONETARY UNION

Figure 3. Impulse response (levels) in full model to a shock of same persistence as Irish recession (ρ = 0.935)

45

CROSS-REGIONS TRANSFERS IN A MONETARY UNION ȴt>;ͿΎ͗ϭй'WZWĞƌŵĂŶĞŶƚdƌĂŶƐĨĞƌ

ȴ'W;ͿΎ͗ϭй'WZdĞŵƉ͘dƌĂŶƐĨĞƌ Ϭ͘ϴ

46

ϯ͘ϱ ĂƚĂнϵϱй/;ĚŽƚƚĞĚͿ ϯ͘Ϭ

Ϭ͘ϲ

EĞǁ<ĞLJŶĞƐŝĂŶDŽĚĞů

Ϯ͘ϱ

EĞŽůĂƐƐŝĐĂůDŽĚĞů

Ϭ͘ϰ Ϯ͘Ϭ

Ϭ͘Ϯ

ϭ͘ϱ ϭ͘Ϭ

Ϭ Ͳϭ

Ϭ

ϭ

Ϯ Ϭ͘ϱ

ͲϬ͘Ϯ

Ϭ͘Ϭ Ͳϭ

ͲϬ͘ϰ

Ϭ

ϭ

Ϯ

ͲϬ͘ϱ

ͲϬ͘ϲ ͲϬ͘ϴ

ĂƚĂнϵϱй/;ĚŽƚƚĞĚͿ EĞǁ<ĞLJŶĞƐŝĂŶDŽĚĞů EĞŽĐůĂƐƐŝĐĂůDŽĚĞů

Ͳϭ͘Ϭ Ͳϭ͘ϱ

YƵĂƌƚĞƌƐ

EŽƚĞƐ͗ ŽƚƚĞĚ>ŝŶĞƐ͗ϵϱ/͘ĂƚĂͲ dĂďůĞϭŽůƵŵŶϯ͘Ύ'WŝŶEĂƚŝŽŶĂůWƌŝĐĞƐ

YƵĂƌƚĞƌƐ

EŽƚĞƐ͗ ŽƚƚĞĚ>ŝŶĞƐ͗ϵϱ/͘ĂƚĂͲ dĂďůĞϮŽůƵŵŶϭ͘Ύ>ĂďŽƌ/ŶĐŽŵĞŝŶEĂƚŝŽŶĂůWƌŝĐĞƐ

Figure 4. Comparison of IRF in model and data

Figure 5. Spread Shock (application to Ireland 2009-13)