Lecture notes

Day 2

Lectures from Chicago, Tuesday, July 28, 2026.

Chicago

Tuesday, July 28, 2026

3 lectures

Open the full notes (PDF)

Lecture 1

Building Startup Ecosystems in Developing Countries

A conversation with Samir Mayekar

Managing Director of the Polsky Center for Entrepreneurship and Innovation, University of Chicago.

The bottom line

A real innovation ecosystem is not one institution but a set of interlocking parts, capital, an industrial base, entrepreneurs, incubators, universities and national labs, government, and connectors, and no single piece is sufficient. For most cities and countries the binding constraints are early-stage capital and a mobile, risk-tolerant workforce. The task is to diagnose your genuine strengths, focus on one or two sectors where you can win, and build bridges to where the money already is, rather than trying to recreate Silicon Valley at home.

An ecosystem, not a single institution

Mayekar came to the university as a founder who had lived the full arc. With co-founders he turned a powder that makes lithium batteries last longer and charge faster into a company, winning a business-plan competition that delivered about $1.5 million in angel investment over a single weekend. The first plant went to Japan (with a Japanese partner) because the United States was barely present in that China-, Korea-, and Japan-dominated market; later factories were built in the Midwest. Over roughly ten years the company raised about $200 million, took the U.S. Army as its first scaled customer, onshored as the supply chain turned America-centric, and was sold to a private-equity firm this year.

That experience maps onto a simple framework (from the organization UIDP): a true ecosystem needs capital, an industrial base, entrepreneurs, companies, incubators and accelerators, government as a partner, research organizations such as national labs, and connectors. His own company drew on nearly every piece, a $75 million Series B led by a local fund of former Argonne investors; Midwest automakers as partners; a university incubator with lab space that also supplied student hires; a cross-disciplinary class where he met his co-founders; a small-business (SBIR) research grant as the first money in; and a local clean-energy competition. Without any one of these, the company might not have survived. The lesson: diagnose where your ecosystem's parts are strong and weak, and attack the weaknesses.

The binding constraints: capital and workforce

Chicago and the Midwest have real strengths, deep manufacturing muscle, a university-driven research engine, and the country's most diversified big-city economy, in which no single sector exceeds 15% of metro GDP. But the deepest weaknesses are early-stage capital and workforce fluidity. Chicago is a private-equity town; founders with a risky idea and no cash flow must raise in Silicon Valley, and since venture capitalists will not travel, the fix is to go to them, plant an office in San Francisco and send entrepreneurs there. The workforce constraint is cultural: only places like Kendall Square and Silicon Valley have the density of people who will join a startup knowing most fail and simply move to the next one.

Agglomeration and collaboration are the levers. Within four or five hours of Chicago sits, through the Big Ten research alliance, the largest research-and-engineering base in America. Mayekar's "Third Coast Foundry," a San Francisco hub built with eight universities, sits within five minutes of about $100 billion of venture capital; a recent demo day drew 250 investors managing $200 billion to 40 deep-tech startups. Government's highest-value role is a focused, patient bet, not picking companies: because the University of Chicago drew roughly half of federal quantum-research dollars, the governor placed a $500 million line item into a quantum park on a former steel-mill site, catalyzing a roughly $10 billion, largely private build-out and drawing DARPA and IBM, making Chicago the hub of the quantum economy. Such bets rest on years of research and must survive political cycles.

Evidence at a glance

Capital is concentrating in AI. Per venture-industry (NVCA) data, about 90% of U.S. venture dollars over the trailing 12 months went to AI, with round and fund sizes ballooning; Mayekar expects a correction in roughly 18 months.
University equity models vary sharply. U.S. universities typically take about 3–5% of a spinout (plus royalties), whereas a statutory 50% claim deters founders. The university's new $25 million fund writes $250,000 pre-seed checks into deep-tech spinouts, its Silicon Valley manager required to seat a partner in Chicago.
Seeding fund managers works. Illinois committed state dollars as 10–20% of emerging managers' funds, helping create 30–40 new Chicago venture capitalists, the most diverse batch in the country, roughly half of them women.
Access is the real barrier. About 60% of U.S. "unicorn" founders are first-generation immigrants; venture capitalists almost never fund cold outreach, warm intros through a small circle of schools dominate.

Policy implications

Focus, then connect. For most regions the fastest path is to pick one or two sectors with a genuine right to win, back professional early-stage investors, and bridge to global capital, not to clone Silicon Valley.

Seed professional investors, not companies. Governments and universities should not pick winners; put capital behind independent managers and accept that most bets fail.
Build bridges to capital. Send delegations and offices to where the money is, and cultivate a credible diaspora investor to anchor a country-focused bridge fund, as one did by helping U.S. firms enter Japan on condition they take its check.
Prize velocity over paperwork. Public funding is throttled by compliance built for the rare fraudster and by cycles too slow for fast-moving technology; celebrate second chances, since failure culture changes only when people in power stand publicly with founders who have failed.
Concentrate public money where early dollars are hardest. Science-based innovation (biotech, energy and materials, quantum) needs it; even early AI was seeded by public research.

Lecture 2

Fertility Decline and Economic Policy

A lecture by James Heckman

Henry Schultz Distinguished Service Professor of Economics and Public Policy and Director of the Center for the Economics of Human Development, University of Chicago.

The bottom line

The statistics that drive fertility policy (above all the total fertility rate) are point-in-time snapshots projected across a lifetime, and they systematically mislead. Fertility has fallen almost everywhere over the past two decades, driven far more by the changing status and autonomy of women than by any single national policy. But the economic alarm is largely misplaced: in steady state a smaller population barely changes per-capita income, and the real costs are transient.

Measure fertility before you legislate

The crucial first question is how fertility is measured. The total fertility rate is a period measure: it sums age-specific birth rates observed in a single year and projects them as if one woman lived through all of them, assuming a stationary world across a reproductive span of more than thirty years, an assumption that never holds. The meaningful measure is the cohort rate, the number of children women actually bear over their lives, but it requires waiting, so demographers approximate it with duration and hazard models and "tempo versus quantum" adjustments. Replacement is roughly a total fertility rate of 2.1.

This distinction is not academic; it determines whether a policy looks like a success. Hungary's very generous program (no income tax for anyone with three or more children) appeared to work on the misleading period measures, yet examined carefully it did not raise completed fertility. France's storied pro-natal policies tell the same story: what looked effective has faded, marking "the end of the French exception."

A global, converging decline

Fertility has fallen across almost every region over the last 20 to 25 years, and the striking feature is convergence. The declines are most dramatic in East Asia, including China; South Korea now sits around 0.72 (far below anything that could be called replacement) with the age at first birth near 32 and childlessness approaching 40% by age 35. India is below replacement, with urban fertility around 1.5 and rural around 2.1 as of 2024; rural India at replacement challenges the standard Western explanation built on the career opportunities of educated women. Iran fell rapidly after the Islamic Revolution; sub-Saharan Africa remains high; Europe is very low and alarmed.

The common thread is the changing status of women: more education, careers, and autonomy; postponement of the first birth; and rising childlessness (about 29% of Japanese women are childless at 40). Because the decline runs across every education level (not only among educated women) single-cause "opportunity cost" stories are too shallow. Culture transmits: Brazilian regions reached by soap operas portraying small, modern families saw fertility fall, a result later replicated in China. And across countries, fertility is higher where men take a larger share of childcare, near zero in Korea, where fertility is near zero. Notably, China's one-child policy is not the main cause: fertility kept falling after it was abolished and tracks a common East Asian trend.

Should we worry? Steady state versus transient

The economic case for alarm rests on confusing transient with steady-state effects. The transient burden is real, an older population strains pensions and public support during the demographic transition. But in steady state the penalty is small: adjusted-support-ratio calculations show per-capita consumption barely moving across total fertility rates from 1 to 3, with an "optimum" near 2.25. Heckman rejects the claim that productivity growth depends on the size of the population; productivity comes from ideas, research, education, and institutions, not headcount, and a recent paper finds no link between population and economic growth. Indeed, an older, higher-saving population raises the capital stock, so a shrinking population can actually raise output per worker and welfare.

Evidence at a glance

South Korea is the outlier. A national total fertility rate around 0.72, some regions near 0.32, childlessness close to 40% at age 35, and a first birth near age 32.
India below replacement. As of 2024, urban fertility around 1.5 and rural around 2.1, rural India at replacement, contrary to Western expectations.
U.S. childbearing has fallen at every education level. The share of women with four or more children fell from 38% to 18% (high school or less), 24% to 12% (some college), and 14% to 6% (college and above).
Steady-state economics is reassuring. Per-capita consumption ranges only about 0.49 to 0.55 across total fertility rates from 1 to 3; the burden is transient, not permanent, much as, historically, the Black Death raised survivors' incomes when labor fell but land was fixed.

Policy implications

Don't legislate off a snapshot. Fertility is a long-run, non-stationary process; policy built on point-in-time total fertility rates will misdiagnose both the problem and the success of any remedy.

Use cohort, not period, measures. Evaluate policy against completed cohort fertility and duration models; Hungary's apparent success dissolves under proper measurement.
Target the transition, not the steady state. Concentrate resources on the transient burden (pensions and elder support during the demographic transition) since long-run per-capita income is largely unaffected.
Do not expect pro-natal subsidies to reverse the trend. Generous cash and tax incentives have not durably raised fertility; the drivers are education, autonomy, and culture, and where policy does have leverage, fertility is higher where men share childcare.
Treat migration as a national, not global, offset. It can rebalance a shrinking national population but cannot change the world total, and migrant fertility converges downward.

Lecture 3

Actionable Evidence, Scalable Ideas: What Every Policymaker Should Ask

A lecture by John List

Director of the Becker Friedman Institute; Kenneth C. Griffin Distinguished Service Professor in Economics and the College, University of Chicago.

The bottom line

Before backing any policy, ask two questions: is the evidence actionable, does it establish cause, not correlation? (and is the idea scalable) will it survive outside the pilot? Most organizations get both wrong. Three "deception traps" come from broken comparisons you can design away; three more survive, and grow, when you scale. The frontier is a single field experiment that establishes cause, identifies who is affected and why, and tests whether it will generalize.

Two questions, and the literacy to answer them

Across firms as varied as United Airlines, Uber, Lyft, Meta, Walmart, and Anthropic, and governments from the United Kingdom to the White House to the Dominican Republic, every organization faces the same two questions daily: is the evidence actionable, and is the idea scalable? Almost all get it wrong. The root problem, List argues, is that schools do not teach causal literacy; teaching it would carry an enormous return, especially amid an AI-driven deluge of data and easy comparisons.

Six deception traps

The first three traps are broken comparisons that can be defeated by design. Selection appears when groups chose their own status: at Uber, the CEO noted that loyalty members spent about 400% more than non-members, but members were already higher spenders, and once that choice was accounted for the program was not incremental, in fact negative, because surplus flowed to inframarginal customers. Hidden third variables drive spurious links, as when ice-cream sales and drownings both rise with hot weather. Before-and-after comparisons where everyone gets the treatment mislead: Australia's plain-packaging rule coincided with smoking falling from 19.5% to 18.3%, but the full time series shows a pre-existing downward trend, the policy did not bend the curve.

The constructive principles follow: a natural experiment creates "statistical twins" when assignment is by luck or nature, and a field experiment creates them by controlling the assignment mechanism. Treating Uber's "bad trips" as good-as-random showed they cost about 6 to 9% of a rider's future revenue over 90 days; an apology recovered roughly 35% of that effect, but only when it carried a real cost, since words alone are cheap talk. The second three traps are correct comparisons that teach the wrong lesson, and they grow at scale. Under Goodhart's law, a third-grade math score predicts adult earnings, yet paying children to raise the score shifts the distribution without improving life outcomes, the mechanism (parents, teachers, schools, neighborhoods) is what matters, and AI benchmarks share the flaw. The data deluge guarantees spurious correlations amplified by confirmation bias, so "the data refiners are the new oil." Survivorship bias is the reflex to reinforce the bombers that returned where they were hit, ignoring those that never came back: you need the losers, not just the survivors.

From evidence to scale

Scaling adds two more failure modes: an irreplaceable chef will not scale where available ingredients can, and some inputs (great teachers) are hard to scale cheaply. The "voltage effect" is nearly a law, about 99.9% of ideas lose benefit-cost value at scale, so "move fast and break things" is art, not science. What transports across settings is models, not data: start from a simple model and keep generating data to find what is missing. At Walmart, 2022 price elasticities mislead today unless two situational features are pinned down, whether a price change crosses a competitor's price and an integer. At the Chicago Heights Early Childhood Center, the decisive feature was teacher quality, so the pre-kindergarten program was built to work with mediocre, good, and excellent teachers, not only excellent ones. For governments, because a deployed policy is hard to claw back, staggered rollouts plus automatic sunset clauses let you learn what works while everyone eventually receives it, and stop what does not.

Evidence at a glance

Apologies, done right. Uber bad trips cost about 6 to 9% of future revenue; a costly apology recovers roughly 35%, while cheap-talk apologies do not work.
Selection reverses conclusions. Once selection is corrected, Uber's loyalty program was not incremental (even negative) because surplus went to inframarginal spenders.
Before-and-after misleads. Australian smoking fell from 19.5% to 18.3%, but the pre-existing trend, not plain packaging, explains it.
Audiences differ. In charitable giving, women respond more to social pressure and men more to price effects, so the ask should differ.
Nudges pay. A Dominican Republic tax nudge raised about $185 million (roughly half a percent of GDP) in one tax season.

Policy implications

Design to learn. Treat every rollout as a chance to run a staggered, sunsettable experiment; the goal is not just a causal estimate but knowing who benefits, why, and whether it will scale.

Ask the two questions first. Is the evidence actionable, and is the idea scalable? Answer both before committing resources.
Interrogate the comparison. Diagnose selection, hidden variables, and before-and-after traps; prefer natural and field experiments that create statistical twins.
Distrust proxies, the deluge, and the survivors. A good predictor becomes a bad target once you optimize it; spurious correlations multiply with data and AI; and inference needs the losers, not just the survivors.
Expect voltage drop, and design to learn. Assume benefits shrink at scale, and identify the situational features (teacher quality, competitor prices) that determine whether a result transports; in government, use staggered, sunsettable rollouts so effects can be tested.