230B: Public Economics Redistributing with In-Kind Transfers

Transcription

1 230B: Public Economics Redistributing with In-Kind Transfers Hilary Hoynes Berkeley 1

2 Thanks to others The slides I am using this term are the result of a collaboration across lots of other faculty teaching PE including: Emmanuel and Gabriel Raj Chetty Day Manoli John Friedman Nathan Hendren Owen Zidar Amy Finkelstein 3

3 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 2

4 In-Kind Transfers Health Insurance: Medicare, Medicaid Nutrition: food Stamps, School feeding programs, WIC Housing: public, vouchers Education Job Training 3

5 In-kind transfers are widespread and large

6 In-kind transfers 101 Basic Economics says cash dominates in-kind Cash superior in terms of recipient utility, since in-kind constrains recipient behavior So why ever have in-kind transfers? Costs of in-kind vs. cash: in kind may have higher administrative costs -e.g. public housing vs cash but see corruption / theft issues in developing countries Several rationales for why benefits of in-kind may exceed cash 5

7 Paternalism Potential rationales for in-kind transfers: Stronger: Individual consumption choices fail to maximize own utility Weaker: Agency problems within family - family doesn t maximize child well being Merit goods (Musgrave 1959) Want to encourage consumption of certain types of goods: Society cares about certain consumption goods for poor over and above effect on poor s utility (e.g. healthcare, food) Consumption externalities Interdependent preferences - my utility depends on your consumption (perhaps of some good) How distinguish empirically from merit goods? 6

8 Potential rationales for in-kind transfers: (cont) Commodity specific egalitarianism (Tobin 1970) Income inequality tolerated but want basic food, medical services, housing needs met Political economy (easier to sell this form of redistribution) perhaps because of paternalism, merit goods, consumption externalities and/or commodity-specific egalitarianism Market failures e.g. Insurance may be valued at more than cost and may not be provided by unregulated market (market failures like adverse selection) e.g. liquidity constraints may interference with e cient allocation of elementary school education (can t borrow against future human capital) 7

9 Potential rationales for in-kind transfers: (cont) Price / pecuniary effects (Coate et al 1994) cash transfers increase demand for normal goods increase price In kind transfers (some) similarly increase demand but also increase supply which lowers prices Because of supply effect, can be more effective potentially than cash transfer for a given government expenditure In-Kind and Targeting efficiency (Nichols and Zeckhauser 1982) - already discussed If demand for specific goods is correlated with unobserved characteristic, can transfer more efficiently (and lose some productive efficiency by inkind vs cash) 8

10 Price Effects: Empirical Evidence Cunha, De Giorgi and Jayachandran (2014) Price Effects of Cash Versus In-Kind Transfers Re-examine a 2003 RCT in rural Mexico that randomly assigned 200 villages to receive either boxes of food (trucked into the village), equivalently valued cash transfers, or no transfers Original purpose: study impacts on food consumption and malnutrition. Very nice example of re-purposing an empirical setting (we should do more of this!) Find evidence of pecuniary effect: food prices significantly lower under in-kind transfers compared to cash transfers Relative to control, in kind transfers reduced food prices by 4 percent, cash transfers had a positive but negligible effect on prices Price effects larger in remote villages (bigger supply side effect) 9

11 Economic Rationales for in-kind transfers: Little empirical evidence Better at screening than cash? I don t know of any empirical evidence! Re-Evaluate the cashout experiments? Pecuniary effects - supply side effect on local prices Cunha, De Giorgi and Jayachandran (2014) is the only paper I know Could one look at the building (or tearing down) of housing projects in the US?! Valued more or less than cash? health insurance - will look at this next week with 10

12 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 11

13 Supplemental Nutritional Assistance Program (SNAP), A Snapshot Previously known as Food Stamps In FY2017, SNAP served 42.2 million people in 21 million households at a cost of $68 billion dollars Average monthly benefit $258 per household, About $4.20 per person per day Central element of the U.S. social safety net and main government policy aimed at reducing food insecurity; available nationwide since 1975 Survived welfare reform in the 1990s intact but many current proposals for reforming the program: converting to a block grant, adding work requirements, etc.

14 Source: Hoynes and Schanzenbach (2015). SNAP is the biggest USDA program Share of spending by program, 2014

15 SNAP WIC Lunch Breakfast 1961: pilot 1975: permanent 1972: pilot 1974: permanent : pilot 1975: permanent $74.1B (2014) $6.2B $11.3B $3.7B Low-income households (universal) 46.5 M individuals/month (2014) Monthly benefit via EBT Low-income pregnant, postpartum women, infants <1, children <5 Low-income school children 8.26 M individuals 19.1M free 2.5M reduced-price Voucher for specific goods & quantities; Nutrition educ, screening Lunches conforming to latest Dietary Guidelines standards Low-income school children 10.4M free 1.0M reduced-price Breakfasts conform. to latest Dietary Guidelines standards Household benefits Individual benefits Individual benefits Individual benefits

16 SNAP WIC Lunch Breakfast Max = $511/month (3-person family), Avg = $133/pers./month $275/HH/month Gross inc <1.3*FPL; Net inc < 1.0*FPL assets <$2250; universal w/restrictions on able-bodied adults BRR: 0.3*net income Food package varies by need (infant formula) Gross income <1.85*FPL; At nutritional risk None: eligible for all or nothing Reimbursement rate (avg): $3.06/free; $2.66 reduced Subsidies: Free: Inc.<1.3*FPL RP: Inc.<1.85*FPL (categorical for SNAP recipients) Discontinuity at 1.3 & 1.85*FPL Reimbursement rate (avg): $1.93 free; $1.63 reduced (same as lunch) Discontinuity at 1.3 & 1.85*FPL Least in-kind: Voucher for dollar value Voucher for quantity (price insensitive) Most in-kind: Meal Most in-kind: Meal Elig. if on SNAP Elig. if on SNAP Elig. if on SNAP

17 SNAP Eligibility and Benefits Means tested: eligibility requires gross monthly income to be below 130 percent of poverty, net income below 100 percent poverty Benefits phased out as income increases B F = G 0.3 Net Income Benefits are vouchers that can be used at grocery stores and most households combine cash and SNAP to buy food Now distributed through debit cards Used to purchase most food items available in stores o Cannot be used for alcohol, cigarettes, vitamins, non-food grocery items or hot foods Federal program; no area variation and few reforms over time Federally funded program (only admin costs paid by states)

18 Take-up has varied a lot over time as well as varying across states

19 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 13

20 Why causal identification is difficult with SNAP Universal program (no ineligible groups) Federal program (little variation across states, localities) Little variation over time (few reforms) Negative selection: SNAP serves people when they need the program it is difficult to disentangle the (presumably positive) impact of SNAP from the (presumably negative) impact of the circumstances that made a family eligible for the program.

21 Strategies for causal identification Program rollout (Almond, Hoynes, Schanzenbach) common design used in lots of settings Use available policy variation across states/time (reduced form or IV) Leverage sharp time series temporal variation (e.g. expansion and subsequent reduction in benefits from federal stimulus) RCTs: Food Stamp cashout experiments in 1980s Comparisons of the same family pre- and post-snap takeup (Hastings and Shapiro) Use local price variation ( purchasing power of SNAP ) (Bronchetti, Christiansen and Hoynes)

22 War on Poverty: Other papers using rollout design Hoynes, Page and Stevens (WIC) M. Bailey (Title X family planning; impacts on fertility, mortality) M. Bailey & Goodman-Bacon (community health centers and elderly mortality) Almond, Chay & Greenstone (Civil rights and infant mortality) Finkelstein & McKnight (Medicare introduction) Cascio et al (Title I funding for schools) Ludwig & Miller (Head Start) 14

23 Results: The SNAP Benefit Cycle Despite documented benefits of SNAP, there is growing evidence that the benefits of the program fall over the monthly food stamp cycle (first identified by Shapiro 2005) Most benefits redeemed early in the month (Hastings and Washington 2010, Castner and Henke 2011, Smith et al 2015) Calorie intake declines by 10-25% over the month (Shapiro 2005) Admissions for hypoglycemia increase over the month (Seligman et al 2014) School disciplinary actions grow over the month (Gennetian et al 2015, Gassman-Pines & Bellows 2016) Unclear results for test scores (Gassman-Pines & Bellows 2015)

24

25 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 15

26 Food Stamps and Food/Nonfood consumption All Other Goods Region unattainable with food stamps B F With Food Stamps Without Food Stamps Food 18

27 Predictions for inframarginal recipient: Nonfood and food consumption increases (F 0 F 1 ) Out of pocket food spending decreases (F 0 F 2 ) Overall food consumption goes up by less than food stamp benefit 19

28 If desired food consumption is low relative to the food stamp benefit, then total food consumption increases more and out of pocket food costs decrease less. Support for food stamps as in-kind transfer (instead of cash) is partially derived from this prediction of a larger increase in food consumption. Only true for constrained families. 20

29 Evidence on SNAP and food consumption Early work largely based on comparisons of recipients to nonrecipients showed that SNAP lead to much more spending on food than cash (4-10 times as much!) 21 The few causal estimates show: (1) SNAP leads to more food spending, (2) most households are inframarginal, and (3) SNAP leads to increases in food spending that are similar to cash Hoynes and Schanzenbach AEJ Applied 2009 use the historical rollout of SNAP SNAP cash-out experiments also find little difference between SNAP and cash (Schanzenbach 2007) Difference-in-difference effects of the ARRA temporary increase in benefits (and then when it was removed) show similar effects of SNAP and cash.

30 Hastings and Shapiro WP 2017 Important contribution to this literature driven by novel data: Longitudinal data for large grocery retailer (across 5-states) tracked with loyalty cards a SNAP occasion =1 if SNAP is main payment (largest share of benefits) a month is assigned to be SNAP month if any SNAP occasion SNAP benefits = sum household spending on SNAP occasions Rhode Island administrative SNAP data , matched to quarterly UI wage data used only to look at transitions on and off SNAP 17

31 Hastings and Shapiro WP 2017 (cont) Research designs: (1) event study: transitions on and off SNAP (2) DD: ARRA changes in benefits (2) IV: using 6-month certification interval SNAP adoption = 6+ non-snap months followed by 6+ SNAP months 18

32 Unusual formulation of event time (omitted period): Each coefficient series is shifted by a constant so that the observation-count-weighted mean of the regression coefficients is equal to the sample mean of the corresponding dependent variable. Source: Hastings and Shapiro (2017 WP)

33 Source: Hastings and Shapiro (2017 WP)

34 Result: SNAP adoption $110 more food spending [Fig 4] SNAP adoption $200 more SNAP benefits MPCF = Much smaller response of non-snap spending Source: Hastings and Shapiro (2017 WP)

35 Diff-Diff using ARRA increase in benefits Source: Hastings and Shapiro (2017 WP)

36 2SLS y = change in SNAP eligible spending, endog var = change in SNAP benefits Instruments = 1 for adoption month (1), 1 for 6 month cycle (2), both in (3) Plus instrument for cash income using gas prices Source: Hastings and Shapiro (2017 WP)

37 Decline in shopping effort store brand coupons Interesting pattern of increase in coupons prior to adoption Source: Hastings and Shapiro (2017 WP)

38 Findings Hastings and Shapiro WP 2017 FINDINGS MPCF out of SNAP ranges from 0.5 to 0.6 across the three designs MPCF out of cash income is lower at about 0.10 (they use changes in gas prices to create a shock to income ) SNAP reduces store-brand share and coupon redemption Interpretation: Calibrated model shows that neoclassical model generates a MPCF of 0.15 Mental accounting: SNAP benefits are psychologically earmarked for spending on food Price misperception (Liebman & Zeckhauser): receipt of an inkind benefit may lead the household to misperceive the price of food. Only mental accounting model generates the MPCF as well as declining shopping effort 20

39 Hastings and Shapiro WP 2017 THOUGHTS Single store chain; changes when on and off SNAP (they bring evidence to this but unclear) Construct SNAP benefits from within store panel (??!!) Entry into SNAP is endogenous; intertemporal substitution (shock gets you on SNAP) use coping mechanisms to get you over SNAP hump 21

40 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 22

41 Results: SNAP and Food Insecurity Most compelling research designs show fairly consistent evidence that SNAP reduces food insecurity Comparisons of the same family pre- and post-snap takeup (Mabli et al 2013, Mabli and Ohls 2015) Variation in state implementation policies that generates differences in take-up across states over time (Mykerezi & Mills 2010; Ratcliffe et al Shaefer & Gutierrez 2013; Yen et al. 2008) Expansions in benefits from federal stimulus (Nord and Prell 2011)

42 Food Stamp Rollout and Infant Health Almond, Hoynes and Schanzenbach (2011) Use initial rollout of the Food Stamps to estimate the effects of the food stamps on infant health Incidence of low birth weight, average birthweight, infant mortality Mother is treated during pregnancy with varying access to food stamps depending on county and month-year of birth Why infant health? Health at birth is an important predictor of later life economic and health outcomes Data: registry data of birth, microdata on census of births 1968+, observe county of residence (and birth) No food stamp receipt on birth record; use program data to convert to TOT; examine how results vary across demographic groups and county characteristics 31

43 Historical Rollout of SNAP (Hoynes joint work with Almond and Schanzenbach) Use initial rollout of the Food Stamps, which took place across the approx. 3,200 U.S. counties over

44 Key legislative markers and population rollout of Food Stamps 33

45 Geographic Rollout by County Source: Hoynes and Schanzenbach, American Economic Journal: Applied Economics

46 Validity of Rollout Design Need to establish that the rollout is exogenous Hoynes and Schanzenbach 2009 show that Consistent with historical account, earlier county food stamp adoption occurs for counties poor, urban, north, less farming Yet these variables account for very little of the variation Not correlated with other war on poverty rollouts (Bailey & Duquette) Also: Need to establish that fertility doesn t change; if it does then harder to interpret the effect on birth outcomes (it could be due to change in composition of births) [we find fertility doesn t change] 35

47 Almond, Hoynes & Schanzenbach: Empirical Model Use variation across counties in difference-in-difference model: y FSP Z * t TP ct ct c t st 1 c60 2 ct ct Observations are at the county (c), time (t) level Identification comes from variation across counties over time in adoption of FSP (FSP ct ) Fixed effects for county, time and state*year (or county*linear time) We also control for possible confounders: 1960 county characteristics interacted with linear time (Z c60 ) Per capita annual county expenditures on other government transfer programs (TP ct ) Standard errors clustered on county 36

48 AHS RESTAT: Findings Access to food stamps improves infant health, reducing low birth weight and statistically insignificant effects on infant mortality (though signs show improvement) Magnitude: $1000 (2009$) in additional food stamp income (treatment-on-the-treated) reduces incidence of low birth weight 4% for whites and 2% for blacks Effects concentrated at the bottom of the birth weight distribution and in high poverty counties Statistically insig effects on neonatal natality (but sign of coef suggest improvement) 37

49 Impact of In Utero exposure to Food Stamps: Reduction in likelihood of birth weight below selected cutoffs Percent Impacts (Coefficient / Mean) Source; Almond, Hoynes and Schanzenbach, Review of Economics and Statistics 2011.

50 Other issues in food and nutrition: Intra-family considerations Who does the shopping may have more control How might offering meals at school affect allocation of food at home? There may be spillover effects of the targeted programs (WIC, NSLP, SBP) on nontargeted members of the family The firm side (much less work here) As a quantity voucher, WIC recipients are price insensitive. Firms (formula manufacturers, retail shops) have incentive to increase prices. School meal providers maximize profit

51 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 24

52 Housing Programs for Low Income Households 1. public housing 2. privately-owned, subsidized housing [supply side] reduced rents for a specified number of years in return for a below market interest rate loan Section 8 New Construction and Substantial Rehabilitation programs provided a direct rental subsidy to tenants Low Income Housing Tax Credit (LIHTC); started in 1986 the largest subsidy for the production of rental housing 3. Tenant based vouchers [demand side] Section 8 Existing Housing Program awards vouchers to low-income households to rent apartments on the private market.

53 Collinson et al, LOW-INCOME HOUSING POLICY

54 Means tested NOT an entitlement (waiting lists) Tenant pays 30% of the their income in rent (another phase out rate), federal government covers the difference Cost $40B/year

55 Public housing demolition Between 1993 and 2007, more than 150,000 units of public housing were demolished, or equal to 11% of the nation s total public housing stock at its height. These demolished units have been fairly geographically concentrated; 60% of them are located in just 33 cities. Led by concerns about concentration of poverty (in high rise public housing buildings) Chyn (2017 wp) Examines LR effects of Chicago demolitions using selective demolitions matched to good admin data Move to 21% lower poverty areas 3 years later Children who are displaced have 4 pp higher empl, 16% higher earnings, 14% fewer arrests in young adulthood

56 Source: Eric Chyn (2017 wp).

57 Source: Eric Chyn (2017 wp).

58 OUTLINE 1) Why redistribute using in-kind transfers? 2) US Food and Nutrition programs (Esp SNAP) 3) Empirical research on SNAP 3a) SNAP and consumption (Hastings and Shapiro 2017) 3b) SNAP and health (Almond, Hoynes and Schanzenbach 2011) 4) Public Housing in the U.S. 5) Chicago voucher study (Jacob and Ludwig) 26

59 EFFECTS OF HOUSING VOUCHERS The Chicago Lottery (Jacob and Ludwig) July 1997, voucher waiting list opened up for first time in 12 years 82,607 applications Randomly assigned to waiting list Notified top 35,000 and told they would be offered a spot within 3 years. By 2003, 18,110 offered vouchers Treatment = 18,110 offered vouchers Control = waiting list number above 35,000 [Those with waiting list number between 18,110 and 35,000 had an ambiguous status since they may have anticipated getting the voucher]

60 Labor supply incentives of section 8 Benefit formula: standard disincentive of welfare program (B=G-twh), income and substitution effect discourage work. G = fair market rent = 40 th percentile of rentals t = required to pay 30% of income towards rent BUT, if benefits of housing are complementary to work (e.g. located near employment, safer) then work may increase.

61 Particulars of a LOTTERY RCT Announce a new opportunity or program (Charter school openings, Head Start openings, Medicaid slots, housing vouchers) KEY: Oversubscription exists (number of those who demand the new opportunity exceeds the supply of slots) Randomization: is the OFFER of the slot ITT: intent to treat, compare those offered the slot to those not offered the slot (reduced form, program evaluation) TOT: treatment on the treated, effect of getting the treatment on your outcome (use offer of slot as the INSTRUMENT) Have to deal with issues of compliers (among those offered the slot, who takes it up?) and always takers (those who get the slot even if not offered the voucher) Standard checks for RCT: balance in pre-ra Xs between the T and C group

62 Jacob & Ludwig AER 2012 Data: Admin data on employment and earnings (state UI data), and welfare participation (AFDC/TANF, Food Stamps, Medicaid) Sample: limit to those <65, not disabled, living in private market housing at application 11,696 treatment, 30,662 control Balance check (Table 1): looks good Take-up: Half offered treatment take up

63 Empirical approach Reduced form: impact of T on outcome (separate dummy for before and after T is offered the voucher) IV: Use T as instrument for leasing section 8 voucher Event study

64

65

66 Overall results Significant reductions in labor supply, increase in welfare Little impact on neighborhood or environment Derived from CES utility function, calibrated with their results Income elasticity = Compensated Wage elasticity = 0.15 reasonable