Back-of-Envelope Series: Estimating spending needs to meet SDG 2

**Disclaimer:** This is a working draft, i.e., research in progress, and, as such, published to elicit comments/advice/recommendations and to encourage debate. The views expressed in this article are solely mine.

Summary: I’m trying a new series called the “Back-of-Envelope” Policy Series, where I calculate (very) rough estimates of a policy issue that interests me. For this inaugural “Back-of-Envelope” note, I’m looking at costing estimates to meet “minimum hunger and nutrition needs” under specific scenarios.

This rough estimate is intended to contribute to efforts within the Sustainable Development Goals (SDGs) framework to end hunger, achieve food security and improved nutrition, and promote sustainable agriculture by 2030 (SDG 2) from a social protection perspective.

I analyzed three main scenarios using the cost and affordability of a healthy diet (CoAHD) from the Food and Agriculture Organization (FAO) and distributional data from the World Bank’s Poverty and Inequality Platform (PIP). I constructed different cash transfer targeting frameworks for each scenario and estimated the total cost. In scenario 1, I calculated the total amount needed to close the “gap” for individuals with average incomes below their CoAHD. Average spending needs are about 11.4 percent of GDP for LIDCs, 0.27 percent of GDP for EMEs, and 0.003 percent of GDP for AEs.

In a second scenario, I used a relative targeting framework with different amounts for the population’s first, second, and third income quintiles. The average spending needs were substantially higher, with about 17.57 percent of GDP for LIDCs, 3.16 percent for EMEs, and 0.81 percent for AEs. In the final scenario, I estimated an untargeted and universal approach (simply CoAHD multiplied by the population). This approach is extremely expensive, and the mean cost for LIDCs was well above 50% of GDP annually.

Overall, LIDC’s spending needs are extremely high compared to the size of its economy, highlighting the drastic external funding needed to achieve SDG 2 by 2030.

(Interested in a bespoke costing scenario, or have any questions? Reach out!)

Introduction

Despite substantial progress over the past decade, COVID-19 and the ensuing global food price crisis have threatened the momentum of SDG 2 (and, to a greater extent, many of the other SDGs). Moreover, the current higher interest rate and price environment make achieving the SDG 2 goal increasingly complex and financially demanding.

The Food and Agriculture Organization (FAO) and World Bank’s framework on the cost and affordability of a healthy diet (CoAHD) is a particularly useful metric for understanding the scale of the problem. The CoAHD framework considers the minimum cost required to procure a diet that fulfills both essential nutrient needs and dietary energy requirements for individuals and then provides estimates based on local food prices and availability.

Estimating accurate spending needs for targeted food assistance programs or other social protection programs typically involve in-depth distributional analysis using a country’s household consumption/income survey data. However, this is a “Back-of-Envelope” policy series!

For a rough estimate, I can utilize percentile-level distributional data from the World Bank’s Poverty And Inequality Platform (PIP) and the CoAHD data to estimate of the spending needs required to fully cover households’ costs (as a side note, a shout-out to the World Bank’s PIP team for such an amazing data source).

The rest of this note is structured as follows: Section I briefly reviews the data sources. Section II covers the methodology of this “Back-of-the-Envelope” targeted food assistance simulation, and Section III presents the results.

Section I: Data overview

The FAO’s cost and affordability of a healthy diet (CoAHD)

Healthy and nutritious diets are a core component for achieving SDG 2. The FAO considers a healthy diet composed of various nutritious and safe foods that provide dietary energy and nutrients in the amounts needed for a healthy diet, as defined in food-based dietary guidelines (FBDGs). For this note, I will not cover the detailed specifics of the FAO’s methodology for the CoAHD dataset. However, broadly speaking, the key components of the CoAHD include:

• Nutritional and Caloric Adequacy: ensures that the diet covers major macronutrients and micronutrients, alongside sufficient caloric intake to support health and activity levels. A healthy diet is based on a wide range of unprocessed or minimally processed foods, balanced across food groups, while it restricts the consumption of highly processed foods and drink products; it includes whole grains, legumes, nuts, an abundance and variety of fruits and vegetables, and can include moderate amounts of eggs, dairy, poultry and fish, and small amounts of red meat.

• Least Cost Estimation: identifies the lowest possible cost for acquiring a diet meeting these comprehensive nutritional standards, factoring in local food prices and availability.

• Dynamic and Context-Sensitive: uses observed retail food consumer prices to provide an operational measure of people’s access to locally available foods in the proportions needed for health.

• Cost in USD PPP per person per day: The main indicator is the total cost of a healthy diet, as well as the disaggregated costs of starchy staples, animal-source foods, legumes, nuts and seeds, vegetables, fruits, oils, and fats. These are estimated in each country.

Fig. 1 below shows the geographic coverage of the CoAHD dataset. Note the particular cost challenges in sub-Saharan Africa and South Asia, which face the double burden of hunger and malnutrition due to the prohibitively high cost of a diet that meets nutritional standards.

Figure 1

The World Bank’s Poverty and Inequality Platform (PIP)

The Poverty and Inequality Platform (PIP) is an interactive and web-based computational tool that offers individuals access to the World Bank’s estimates of poverty, inequality, and distributional data. PIP provides a comprehensive view of global, regional, and country-level poverty and inequality trends for more than 160 economies worldwide.

The specific PIP dataset used for my analysis is the percentile-level distributional data, which reports 100 points per country per survey year ranked from the smallest (percentile 1) to the largest (percentile 100) income or consumption. For each income percentile, the dataset reports (among other variables), the average daily per person income or consumption (avg_welfare) in 2017 PPP$, and the share of the population in the percentile (which might deviate slightly from 1% due to coarseness in the raw data) (pop_share).

In addition, the database reports the welfare measure (welfare_type) used in the survey data — income or consumption — and the region covered (reporting_level) — urban, rural, or national. Table 1 below provides an example of Angola in the 2000 survey. The poorest percentile of the population consumes, on (national) average, $0.19 a day. On the other hand, the richest percentile (100) consumes, on (national) average, $87.27 a day.

Table 1

If you’re having trouble thinking about this data, one way to think of percentile-level distributional data is if every country comprises roughly a hundred people!

There are clear caveats in using this percentile-level distributional data. First, conducting simulations on percentile-level data is not a substitute for using the full survey data. Moreover, percentile-level data should not be used to re-estimate poverty and inequality statistics. Estimates using the full survey data are available on the PIP website. For example, using the percentile dataset to estimate the Gini coefficient would result in a lower within-country inequality since it does not account for the inequality within each percentile. In-depth country analysis should always use the full survey data!

However, as stated in my introduction, my analysis is a rough “Back-of-Envelope” estimate focusing on global coverage. The analysis is also meant to be a conversational starting point, highlighting the large financing needs to achieve the SDGs. I also conducted a short feasibility analysis to examine how “rough” an estimate was. Using the percentile-level data, I was typically within 1–2 percentage points of poverty rate estimates based on the full-survey data. I considered that good enough for a “Back-of-Envelope” estimate.

Section II: Estimation Methodology for Spending Needs

Distributional analysis typically involves using household income survey data to understand how economic policies, such as cash transfer programs, affect different population segments. Additional considerations are examining how income is distributed across households and identifying how policies alter these distributions, potentially affecting poverty, inequality, and overall economic welfare.

I only focus on overall budgetary needs using a simplified means-tested cash transfer framework, where benefits are transferred only to percentiles below a certain threshold. I do not conduct additional distributional analysis on poverty and inequality metrics.

To estimate overall spending needs, I outline various scenarios that assume differing approaches to the eligibility requirements. However, several assumptions remain consistent across all scenarios.

Firstly, I assume 100% effective take-up and targeting. This means that the estimate assumes everyone eligible to receive cash does, and there is no leakage of funds. For example, suppose a spending scenario targets the poorest decile of the population as recipients of a cash transfer covering the cost of a healthy diet. In that case, the estimate assumes 100% perfect targeting and take-up. In reality, some eligible households do not apply for or benefit from a scheme due to administrative issues, etc.

Secondly, I do not consider dynamic effects. This is a “Back-of-Envelope” simple analysis, and I do not consider any long-term effects or changes in consumption preferences or labor market behavior.

Thirdly, the “transfer” amounts and eligibility does not consider additional characteristics such as, but not limited to, dependents, disability, or geographic location. The percentile-level distributional data do not contain information other than mean income/consumption per person per day,

Finally, I do not consider any existing schemes. The percentile-level distributional data provides net income measures and does not provide data on any existing transfer programs, subsidies, or support.

Scenario 1: CoAHD as the Poverty Line

In the first (and simplest) scenario, I set the cost of a healthy diet as the “poverty line” and calculate the difference (or gap) between it and the mean income/consumption for each percentile. For example, if the mean income per person per day for the poorest percentile was $1.00 and the cost of a healthy diet per person per day was $3.00, then the difference is $2.00 per person per day. The spending needed to close the gap for the poorest percentile would be $2.00 multiplied by the population share for that percentile.

I repeat this calculation for every percentile until the mean income per person per day is more than or equal to the cost of a healthy diet per person per day. In other words, I estimate spending needs by calculating the total value needed to close the difference between every person’s income and the cost of a healthy diet.

In Table 2, I show an example of the calculation for Colombia in 2016. The mean shortfall from CoAHD is simply the difference between CoAHD and the mean income of each percentile. The mean shortfall is then multiplied by the percentile share of the population (which, as mentioned in Section 1, isn’t quite 1% due to the peculiarities of the survey data) and summed across all percentiles. The total shortfall per day is multiplied by 365 and divided by GDP to calculate annual spending needs as a share of GDP.

Table 2

Scenario 2: CoAHD scheme with relative targeting

I constructed a targeting scheme in the second scenario using a relative targeting scheme. The first quintile (20 percentiles) receives a full amount of the CoAHD, the second quintile receives 2/3rds of the CoAHD, and the third quintile receives 1/3rd of the CoAHD. The richest two quintiles do not receive any benefits.

In Table 3, I show another example using Colombia. Note that the ”…” means the observations between. Percentiles 1 to 20 receive the CoAHD value of $2.84, percentiles 21 to 40 receive the CoAHD value of $1.89, and percentiles 41 to 60 receive the CoAHD value of $0.95. After calculating the allocated shares, I multiply by the share of the population in each percentile and sum the values.

Table 3

Scenario 3: Universal basic CoAHD income

In the final scenario, I estimate the spending needs for a universal basic CoAHD transfer. Effectively, every person in the population would receive a basic stipend to cover the cost of a healthy diet. This scenario is similar in concept to a Universal Basic Income (UBI), where the basic income is set as the cost of a healthy diet.

For obvious reasons, this scenario has the highest spending needs. Table 4 shows the basic calculation.

Table 4

Of course, there is an infinite combination of different targeting approaches and schemes to simulate. However, the example Scenarios are intended to reflect three levels of strictness and targeting, which offer a range of spending needs for comparison.

Section III: Results & Brief Discussion

Results scenario 1: CoAHD as the Poverty Line

As expected, the three scenarios provide widely varying estimates for spending needs, highlighting the importance of clearly defining the scope of meeting the SDG 2 goal. Figure 2 shows the spending needs to meet Scenario 1, where I set CoAHD as the absolute poverty line and calculated the total amount needed to close the gap. Average spending needs are about 11.4 percent of GDP for LIDCs, 0.27 percent of GDP for EMEs, and 0.003 percent of GDP for AEs. Spending needs are particularly high for LIDC compared to the size of their economy.

In Figure 3, I plot the spending needs of Scenario 1 with tax revenue to GDP to compare needs against the available budgetary resources. The issue is particularly stark for LIDCs. For example, Burundi’s estimated spending needs in Scenario 1 are 57% of GDP, compared to current tax revenue of 15.79% of GDP. For Burundi to meet SDG 2 (as defined in Scenario 1) by 2030, it is entirely impractical to expect attainment without substantial external donor support.

However, a more optimistic outlook for Scenario 1 is Cameroon. Cameroon has an estimated spending need of 2.93% of GDP, compared to the current tax revenue of 11.35%. While a 2.93% of GDP spending program is enormous, it is attainable within a medium to long-term development goal.

Moreover, the outlook for EMEs and AEs is substantially more positive from a budgetary perspective. In particular, it goes without saying that for AEs, meeting food-related needs is arguably a political choice and not a budgetary issue. Compared to the size of their budget, closing the CoAHD gap for households is within reach.

Results scenario 2: CoAHD scheme with relative targeting

Figure 4 shows the spending needs to meet Scenario 2, which I structured as a relative targeting approach. I find that average spending needs are substantially higher compared to Scenario 1, with about 17.57 percent of GDP for LIDCs, 3.16 percent of GDP for EMEs, and 0.81 percent of GDP for AEs.

The design choice with a relative targeting approach sometimes results in higher spending needs for countries with less absolute poverty but lower needs for countries with higher absolute poverty.

In Figure 5, I show that the relative targeting approach does have a higher cost as a share of GDP for EMEs and AEs. AEs would face spending needs between 0.25 to 1.25 percent of GDP, which is significantly higher than estimates for Scenario 1. Moreover, EMEs would face spending needs that are close to a third of their current tax revenue to GDP.

This suggests that the relative targeting approach of Scenario 2 would benefit from better targeting to reduce the estimated fiscal cost.

Results scenario 3: Universal basic CoAHD income

Finally, Figure 6 shows the spending needs under Scenario 3, or the universal basic income approach. I’ll only touch on this scenario briefly to say that an untargeted and universal approach is extremely expensive. For many LIDCs, the spending needs are well above 50% of GDP annually. Even for AEs, the costs are more than a percentage point of GDP in additional expenditure. Considering the opportunity costs, it’s difficult to justify expenditures that may replace needs in other sectors, such as electricity, old age care, or investment in infrastructure.

Conclusion

This “Back-of-Envelope” exercise highlights the dire need for poverty-related hunger and the substantial resources still required for countries to meet SDG2 by 2030. Specifically, LIDCs across all scenarios face spending needs substantially higher than the available budgetary resources. The SDG 2 for LIDCs will likely be entirely unachievable by 2030 unless they receive substantial external financing and support. Even in the most conservative Scenario 1 approach, the average spending needed for LIDCs was 11.4% of GDP compared to an average tax to GDP of 14.15% of GDP. The cost of meeting SDG 2 alone is approximately the size of their whole (domestic) budget.

The prognosis for Advanced Economies and Emerging Market Economies is far rosier. For AEs, the choice to achieve SDG 2 is largely a political one. For EMEs, a priority in their medium-term fiscal plans would push them closer to achieving parts of SDG 2.

I also want to emphasize my estimation approach from a social protection perspective. In that sense, I am purely (roughly) estimating what it would cost to ensure individuals have the minimum incomes to afford a CoAHD. Now, this perspective is only one of several sub-goals within SDG 2. Other sub-goals of SDG 2 focus on long-term agricultural investing needed to lower countries’ CoAHD or the preparations needed to integrate climate adaptation concepts into current agricultural practices. What do I mean by this? From a Theory of Change perspective, let’s imagine an example of additional public investment into a country’s food supply chain. This may reduce food loss from the farm to the table, which could translate into lowering a country’s CoAHD. A lower CoAHD also means that my estimates would be lower, reducing the estimated fiscal cost.

None of those additional perspectives are included in my estimates. Relying solely on a social protection approach without meeting the additional sub-goals of SDG 2 that contribute to long-term agricultural development may lead to foreign aid dependency.

Appendix

Bibliography

• Poverty headcount ratio at $2.15 a day (2017 PPP) (% of population) https://data.worldbank.org/indicator/SI.POV.DDAY
• https://openknowledge.fao.org/server/api/core/bitstreams/6ca1510c-9341-4d6a-b285-5f5e8743cc46/content/sofi-2022/cost-affordability-healthy-diet.html
• Poverty and Inequality Platform https://pip.worldbank.org/home

Misc.