How market design can feed the poor

America's largest non-profit had a broken distribution system. University of Chicago economists fixed it.

For most of its history, Feeding America, the nation’s largest nonprofit, relied on a broken system to distribute its 220 million pounds of food per year. It ignored existing stocks and donations, flooding fully stocked food banks in Idaho with potatoes and warehouses in Alaska with five gallon buckets of pickles.

To fix how America fed its hungry would require the economics of market design. The new system increased food supply by 100 million pounds annually, equivalent to feeding an additional 60,000 people every day. It is one of the clearest successes of market design so far.

The failures of centralized food allocation

The old system, used since the late 1980s, created a ‘goal factor’ metric based on poverty rates and population size and then allocated food in direct proportion to this need. Food banks would receive calls or emails announcing they had been assigned a load, often with conditions like required pickup dates.

This centralized approach created systematic waste and misallocation. The system treated all pounds of food as identical. A pound of carrots received the same allocation priority as a pound of peanut butter despite vast differences in nutritional value, transportation costs, and demand. Potato chips were particularly problematic: they require ample space but offer poor nutrition.

The system operated in ignorance of what food banks needed. Food banks received an average of 20 percent of their food from Feeding America with the rest coming from local sources that headquarters knew nothing about. This created duplication and waste: a food bank might receive eggs from Feeding America the same week they had already received eggs from somewhere else.

The system also couldn’t account for the fact that some food banks had strong ties to local food sources while others had no alternatives, meaning that the ‘food rich’ banks received excess food while the ‘food poor’ ones often only received products they didn’t need.

Feeding America also had no visibility into food banks’ storage constraints or inventory levels, consigning lots of food to the trash. Sending cheese or produce to a food bank with a full fridge resulted in those products spoiling. Donors often gave food close to expiration dates, and the centralized system couldn’t quickly match time-sensitive donations with food banks that needed them. Most perversely, if food banks refused unsuitable loads it still counted against their need measure. This forced food banks to accept unwanted goods to maintain their ranking in the allocation queue, punishing waste avoidance. Despite these issues Feeding America could hardly say ‘no’ to a donation without risking the donor relationship altogether.

A choice system

In 2004, economist Canice Prendergast joined a task force of food bank veterans and University of Chicago academics to redesign Feeding America. Its purpose was to introduce a better, market-driven allocation system.

The internal politics of reform were complicated. Many directors feared that introducing market mechanisms would favor the privileged over the needy. One food bank director told Prendergast: ‘I am a socialist. That’s why I run a food bank. I don’t believe in markets. I’m not saying I won’t listen, but I am against this’.

The task force would spend over a year crafting a unique solution: a specialized currency called shares that went into an auction system that combined market efficiency and equitable distribution. To design the system they drew on the principles of market design, which Nobel Prize winning economist Al Roth called ‘the economist as engineer.’

Market design involves designing the incentive systems that govern what price is paid, like auction systems, or matching algorithms that allocate goods in two-sided markets in typically one-to-one matchings. The most widespread contribution of market design is auction theory, which has revolutionized auctions from Google’s online search ads to FCC spectrum auctions. Non-price matching algorithms have been used to help students get matched to better schools, allocate soldiers in the army, match medical school graduates to residency programs, and even improve organ donation.

Usually, price-based solutions are not employed in non-price contexts because ‘monopoly money’ is artificial. But in the case of Feeding America, a pure matching algorithm would not have worked because different goods have different values, and because the matches involve quantities.

Instead, the task force went with a currency known as ‘shares’. As far as I know, this is the only example of a successful ‘monopoly money’ auction system, let alone one at such scale. It was a spectacular success.

Anatomy of a food auction

In the Choice System, food banks receive shares in proportion to their goal factor – Feeding America’s measure of need – but once they have these shares, food banks can spend them however they choose exclusively through twice-daily online auctions. The system used a simple website interface that allowed food banks to become familiar with the system months before implementation, including training for food banks.

The Feeding America dashboard after the reform. Image credit: Prendergast 2016.

The auctions are sealed (bids are secret until there is a winner) and first-price (the winner pays what they bid). Food banks can search offerings by geographic constraints or food type, and easily access price histories for similar items, with 30-40 offerings posted at least two hours review time. Although a second-price auction is more efficient, it was rejected for the practical reason of being too confusing to the food banks. All shares spent during the day are redistributed at midnight using the same need-based formula, ensuring that high bids by wealthy food banks benefited all participants. Once the food bank directors got it, they became excited; this was crucial for driving acceptance.

Food banks could also access ‘credit’ to bid on expensive, high-quality items that exceeded their current share balance. Debts were repaid interest-free from future share allocations. This ensured smaller food banks could compete for premium products. Food banks could even bid negative amounts to accept hard-to-move products donors insisted on giving. Additional measures were introduced to help smaller food banks. Joint bidding allowed food banks to bid together to split truckloads that were too much for one organization but perfect for several. Less sophisticated food banks could delegate bidding to Feeding America staff, and a Fairness and Equity Committee could convene to review claims for additional shares due to unmeasured hardship.

Supply explosion

The Choice System’s impact was immediate and dramatic. Food supply increased by approximately 35 percent (100 million pounds) in the year around implementation. Within just the first seven months after launch, supply rose by 50 million pounds.

It also revealed how wrong the previous system’s assumptions were: auctions revealed that food banks were willing to trade 116 pounds of produce for a single pound of pasta.

Produce sold for just four percent of the average good’s price, while rice, pasta, and cereal cost roughly three times the median. The average price was 0.28 shares per pound, meaning food banks typically received about three pounds per share. But a quarter of loads sold for 0.3 shares or more, and eight percent of purchases involved food banks paying more than one share per pound.

Over 80 percent of the efficiency gains came from allowing food banks to choose to ‘buy’ lower quantities of higher quality food, which allowed food banks to correct skews in their other food sources. Compared to what they would have received under the old system, 30 percent of food banks chose to get more than twice as many pounds, while another 25 percent chose less than half as many pounds.

Donors contributed more knowing their food would be efficiently allocated, and Feeding America accepted previously rejected donations due to increased system liquidity. It also allowed food banks to sell their surplus to other food banks, adding another 12 million pounds annually. Under the old system, a food bank with ten percent greater perceived need received exactly ten percent more pounds of food. After the Choice System implementation, ten percent greater need resulted in only 4.3 percent more pounds, as food banks bought higher-quality food and chose not to spend all their shares. The system benefited smaller, food-poor organizations the most, as the behavior of food-rich organizations drove down prices for cheaper goods.

Perhaps the most telling indicator of the Choice System’s success was that the safeguards designed to protect smaller food banks fell into disuse. The Fairness and Equity Committee never convened and no food bank chose to permanently delegate bidding to Feeding America. And only 1.6 percent of winning bids used joint bidding.

Markets in everything

Some parts of market design never took off: despite being the first market design proposal, the Gale-Shapley algorithm, which was designed for university admissions, has still never been used in that setting. Even some of the success stories, like residency matching and kidney exchanges, took decades to get implemented.

Feeding America was not one of those cases.

I was at the conference in Chicago where Prendergast first showed these results. He said that the socialist board member who had opposed the system was now one of its strongest advocates. Today, Feeding America distributes over 5.2 billion meals per year.

Feeding America site. Image credit: Feeding America.

---

Evan Zimmerman is the cofounder of Edge and founder of Jonovo