Anticipation and Consumption Chapter 2

Figure 1: ESP Spending Responses—Average Impacts (US) Note: This figure presents estimates of the weekly spending response 𝛾𝑘 (weeks −4 to −1) and the cumulative spending response Γ𝑘 (weeks 0 to 7) from Equation (2) for various samples. For comparison, the shaded box denotes the range of point estimates reported by Broda and Parker (2014). The “Near scheduled date” sample consists of households receiving direct deposits three days leading up to the scheduled payment date or the weekend after. The “On scheduled date” sample consists of households receiving direct deposits on the date specified in Appendix Table 3. The “All households” sample consists of all households receiving direct deposits. Standard errors reported in parentheses are adjusted for clustering at the household level and obtained from a block-bootstrap procedure with 100 replicates. Calculated based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. Figure 2: ESP Spending Responses by Timing of Payment (US) Note: The panel on the left presents estimates from Equation (4) of the four-week cumulative ESP spending response Γ𝑤 4 for households receiving EFTs in the first (Group 1), second (Group 2), and third (Group 3) week of May, respectively, and the 𝑝-value labeled 𝑝123 corresponds to the null hypothesis of equality across groups. The panel on the right displays the difference in spending between Group 1 and Group 3, along with a 95 percent confidence interval (black line) and 90 percent confidence interval (vertical endpoints). Liquidity is an indicator for reporting that the household has at least two months of income available in easily accessible funds. Financial plan is an indicator for reporting that the household has gathered together its financial information, reviewed it in detail, and formulated a financial plan for the long-term future. Savings habit is an indicator for reporting that household members would rather save more for the future than spend their money and enjoy it today. Standard errors reported in parentheses are adjusted for clustering at the household level and obtained from a block-bootstrap procedure with 100 replicates. Calculated based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. Figure 3: ESP Spending Responses by Timing of Payment—Alternative Specifications (US) Note: The panel on the left presents estimates from alternative specifications of Equation (4) of the four-week cumulative ESP spending response Γ𝑤 4 for households receiving EFTs in the first (Group 1), second (Group 2), and third (Group 3) week of May, respectively. Panel A considers alternative sets of characteristics in the first step of the estimation, Panel B considers alternative sets of comparison households, and Panel C considers different specifications of the treatment group. The 𝑝-value labeled 𝑝123 corresponds to the null hypothesis of equality across groups. The panel on the right displays the difference in spending between Group 1 and Group 3, along with a 95 percent confidence interval (black line) and 90 percent confidence interval (vertical endpoints). Standard errors reported in parentheses are adjusted for clustering at the household level and obtained from a block-bootstrap procedure with 100 replicates. Calculated based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. Figure 4: MPC by Timing of Payment—Fixing Calendar Week (US) Note: Each bar shows estimates of the weekly marginal propensity to consume, holding the calendar week fixed, for households receiving EFTs in the first (Group 1), second (Group 2), and third (Group 3) week of May, respectively, over a one-month period, along with a 95 percent confidence interval. The estimates come from Equation (2) in Appendix B.2, extended to include fixed effects for the calendar week. The 𝑝-value labeled 𝑝123 corresponds to the null hypothesis of equality across all three groups, and the 𝑝-value labeled 𝑝13 corresponds to the null hypothesis of equality between Group 1 and Group 3. Standard errors are adjusted for clustering at the household level and obtained from a block-bootstrap procedure with 100 replicates. Calculated based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. Figure 5: Impact of Shorter Wait for Cash Transfers (Kenya) Note: Each dot corresponds to an estimate of the treatment effect, 𝛽𝑘 , from Equation (5) and the associated 95 percent confidence interval. Each specification corresponds to a different definition of the treatment group (short waiting times) and the comparison group (long waiting times), with “cutoff” denoting the threshold for defining a short waiting time and “max” denoting the maximum number of days of waiting time in the comparison group. Savings is an indicator for reporting nonzero savings, and the remaining magnitudes are reported in 2012 USD PPP. Colors denote statistical significance at the 1 percent (orange), 5 percent (green), and 10 percent (blue) levels. Figure 6: Relationship between Waiting Times and Outcomes (Kenya) (a) Savings (b) Assets (c) Durables (d) Investment Note: Each figure depicts the relationship between waiting times and outcomes in the form of a binned scatterplot. The line shows the fit of a global second-order polynomial. See Section 4.2 for details on the outcomes. Figure 7: Impact of Shorter Wait for Cash Transfers—Heterogeneity (Kenya) (a) Savings (b) Assets (c) Durables (d) Investment Note: Each figure depicts estimates of the treatment effect, 𝛽𝑘 , from Equation (5) and the associated 95 percent confidence interval for various samples of households. See Section 4.3 for details on the samples. Figure 8: Mental Accounting Model—MPC as a Function of Magnitude and Duration Note: This figure plots estimates of the marginal propensity to consume (MPC) in response to a windfall of the specified magnitude as a function of duration based on Table 2. The MPC is calculated by dividing the model-implied total spending response by the windfall size. The $1,000 windfall corresponds to the 2008 U.S. stimulus payment (Section 3); the $404 windfall corresponds to the GiveDirectly cash transfer in Kenya (Section 4); the $176.50 windfall corresponds to the windfall from the experiment in Malawi (Section 5). The diamonds indicate the range of waiting times present in the data from each setting. See Appendix Figure 18 for an analogous figure plotting the model-implied total spending response, and see Appendix Figures 19 to 21 for a depiction of how well the model fits the data in each setting. Table 1: Impact of Delayed Windfalls on Savings (Malawi) NBS account Formal savings Informal savings Total financial assets In-kind savings Total savings 1-day delay 8-day delay 𝑝-value -11.33 (7.32) -0.47 (12.34) 5.56 (10.56) 8.84 (18.90) -0.65 (24.36) -3.29 (31.89) -1.18 (7.16) 12.27 (14.04) 17.97 (10.56) 32.50 (20.60) 137.95 (36.57) 159.39 (47.74) 0.2034 0.5998 0.2255 0.2716 0.0003 0.0005 Note: Each row presents estimates of Equation (6) with the outcome variable as a measure of savings. The sample consists of 474 households receiving MK 25,000 (USD 176.50 PPP) windfalls from the field experiment by Brune et al. (2017). The no-delay treatment consists of 156 households receiving payments via cash or direct deposit without delay. The delay treatments consist of 318 households that receive payments after a one-day delay (158 households) or after an eight-day delay (160 households). The first column presents the estimate of 𝛽1 (the causal impact of receiving the windfall with a one-day delay relative to receiving the windfall immediately), and the second column presents the estimate of 𝛽8 (the causal impact of receiving the windfall with an eight-day delay relative to receiving the windfall immediately). Formal savings consist of balances in NBS bank accounts (the bank that facilitated the experiment), other bank or microfinance institution accounts, and employee-based Savings and Credit Cooperatives (SACCOs). Informal savings consist of balances in Rotating Credit and Savings Associations (ROSCAs), village savings clubs (kalabu yosunga ndalama), cash that is not for living expenses kept at home or in a secret hiding place or given to someone else for safe keeping. In-kind savings consist of advance purchases of farm inputs, business inventory, and bags of maize. Total financial assets consist of formal and informal savings combined. Total savings consist of total financial assets combined with in-kind savings. All values are reported in USD PPP adjusted using the 2014 exchange rate 141.64 MK/USD. Standard errors are reported in parentheses. The data come from the survey questions displayed in Appendix Figure 15. Table 2: Mental Accounting Model—Estimates and Fit Scaling factor Parameter estimates 𝛼 (time) 𝛽 (magnitude) Waiting equivalent Predicted monthly NCP spending Group 1 (actual: 65.25) Group 2 (actual: 45.24) Group 3 (actual: 18.73) 3.33 1 5 7.3 10 0.5789 (0.0394) 0.9984 (0.0002) 340.70 0.4323 (0.0445) 0.9971 (0.0003) 293.16 0.6447 (0.0410) 0.9986 (0.0002) 315.41 0.7107 (0.0440) 0.9988 (0.0002) 274.16 0.7698 (0.0478) 0.9988 (0.0002) 225.53 64.82 38.20 22.06 50.36 21.28 9.11 63.03 41.90 27.12 58.57 43.42 31.26 53.15 43.03 33.83 Note: Each column presents estimates of the model defined by Equations (7) and (8) for a different scaling factor. The top panel shows estimates of the parameters from Equation (9), and the the waiting equivalent refers to the magnitude (in dollars) that would result in a decrease in the MPC of the same amount as one additional week of waiting (computed as log(𝛼)/ log(𝛽)). The bottom panel displays the excess NCP spending implied by the model (see the data in Figure 2 for comparison). Standard errors reported in parentheses are adjusted for clustering at the household level. Derived based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. Table 3: Mental Accounting Model—Estimates and Fit: Heterogeneity Liquidity Parameter estimates 𝛼 (time) 𝛽 (magnitude) Waiting equivalent Predicted NCP spending Group 1 (monthly) Group 2 (monthly) Group 3 (monthly) Planning Savings habit Yes (1) No (2) Yes (3) No (4) Yes (5) No (6) 0.6016 (0.0623) 0.9979 (0.0003) 247.48 0.5718 (0.0499) 0.9988 (0.0002) 449.53 0.6616 (0.0635) 0.9979 (0.0003) 191.94 0.5304 (0.0483) 0.9988 (0.0002) 520.76 0.6811 (0.0576) 0.9976 (0.0004) 158.72 0.5013 (0.0535) 0.9991 (0.0002) 746.23 49.22 29.42 17.74 84.04 50.41 28.87 54.39 35.14 23.40 76.56 42.19 22.42 47.38 32.10 21.68 93.03 48.34 24.43 Note: Each column presents estimates of the model defined by Equations (7) and (8) for a different subsample of households. The subsamples follow those presented in Figure 2: liquidity unconstrained (Column 1), liquidity constrained (Column 2), those that make financial plans (Column 3), those that do not make financial plans (Column 4), those that classify themselves as savers (Column 5), and those that classify themselves as spenders (Column 6). The top panel shows estimates of the parameters from Equation (9), and the waiting equivalent refers to the magnitude (in dollars) that would result in a decrease in the MPC of the same amount as one additional week of waiting (computed as log(𝛼)/ log(𝛽)). The bottom panel displays the excess NCP spending implied by the model (see the data in Figure 2 for comparison). Standard errors reported in parentheses are adjusted for clustering at the household level. Derived based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business. Table 4: Mental Accounting Model—One-Time Payment vs. Reduced Withholding Panel A: Survey data from Sahm, Shapiro and Slemrod (2012) One-time Payment Percent mostly spend Percent mostly save Percent mostly pay debt 19 27 53 Panel B: Methods to convert survey data to MPC estimates One-time Payment Shapiro and Slemrod (2003) Parker and Souleles (2019) Coibion, Gorodnichenko and Weber (2020) Panel C: Model prediction MPC Reduced Withholding 13 33 54 Reduced Withholding 0.29 0.35 0.44 0.22 0.32 0.41 One-time Payment Reduced Withholding 0.32 0.27 Note: The top panel reports data from the Thomson Reuters/University of Michigan Surveys of Consumers documented by Sahm, Shapiro and Slemrod (2012, Table 1). The 2008 survey (Column 1) asks respondents how the tax rebates were affecting their spending; the 2009 survey (Column 2) asks respondents how the 2009 reduction in withholding is affecting their spending. The middle panel applies three methods from the literature to convert the survey responses into a measure of the MPC. The bottom panel states the prediction of our model; see Section 6 for additional details. Derived based on data from Nielsen Consumer LLC and marketing databases provided through the NielsenIQ Datasets at the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business.