Storytelling in Economics

Why narratives matter in economics and how a result gets told well: an interactive contagion model, two exercises from the material, the session index, and the reading list.

ITAM · Master in Applied Economics · Spring 2025

What it is

The course has two halves.

One is Robert Shiller’s Narrative Economics, the anchor text and the academic side of the argument. Shiller’s claim is that economic stories spread through a population the way a disease spreads, and that they act as shocks to real outcomes while they do it: consumption, investment, panics. Shiller argues the claim should be studied quantitatively, and the course treats it as such.

The other half is the craft. An empirical result does not explain itself, and the order in which the evidence arrives decides how much of it a reader takes away. The sessions follow the arc of a research project, from the motivation to the figure, the paper and the presentation, through readings discussed in class and exercises on real data.

Both halves answer to the same test: tell one result to a technical audience, to someone making policy decisions, and to a reader with no training in economics.

How a narrative spreads

Narrative Economics takes its central mechanism from epidemiology. Kermack and McKendrick (1927) split a population into three groups: those who have not caught a disease, those who have it and pass it on, and those who have stopped transmitting. Shiller applies the same three groups to a story. Some people have not heard it, some are telling it, and some heard it and no longer repeat it.

Two rates move people between the groups. The contagion rate c sets how fast tellers pass the story to people who have not heard it. The forgetting rate r sets how fast tellers lose interest and stop. Their ratio is the basic reproduction number, R0 = c / r: the number of new tellers one teller produces while almost everyone is still susceptible. Above one the story becomes an epidemic. Below one it dies out.

sirSemilla = 0.001
sirDias = 200
sirPaso = 0.05

// Explicit Euler. S + I + R holds at 1 to within 1e-14 everywhere on the
// slider grid. Integrated to convergence, a step of 0.05 lands within 0.07 pp
// of the analytic final size, which for a seeded run is
// S_inf = S_0 exp(-R0 (1 - S_inf)) with S_0 = 1 - sirSemilla, not the
// S_0 = 1 form: at R0 = 1 the two differ by about 4 pp, so checking against
// the wrong one invents an integrator error that is not there.
// What the run does NOT do is reach that asymptote in every setting. Just
// above R0 = 1 the epidemic is still under way on day sirDias and the final
// state sits far from S_inf. That is the fixed window, not the step size, and
// refining the step does not move it. It is why the reading says "still
// rising" instead of naming a peak there, and why it always says "by day
// sirDias" rather than reporting a total.
// Days are recorded as k / porDia rather than k * sirPaso so the axis carries
// exact integers.
sirSimular = (c, r) => {
  const porDia = Math.round(1 / sirPaso);
  let S = 1 - sirSemilla, I = sirSemilla, R = 0;
  const filas = [{ dia: 0, S, I, R }];
  for (let k = 1; k <= sirDias * porDia; k++) {
    const nuevos = c * S * I * sirPaso;
    const salen = r * I * sirPaso;
    S -= nuevos;
    I += nuevos - salen;
    R += salen;
    if (k % porDia === 0) filas.push({ dia: k / porDia, S, I, R });
  }
  return filas;
}

sirEtiqueta = ({
  S: "Have not heard it",
  I: "Telling it",
  R: "Stopped telling"
})
sirDominio = [sirEtiqueta.S, sirEtiqueta.I, sirEtiqueta.R]
sirColores = ["var(--teal)", "var(--clay)", "var(--ink-soft)"]

sirLargo = traza => traza.flatMap(d => [
  { dia: d.dia, grupo: sirEtiqueta.S, valor: d.S },
  { dia: d.dia, grupo: sirEtiqueta.I, valor: d.I },
  { dia: d.dia, grupo: sirEtiqueta.R, valor: d.R }
])

sirResumen = (c, r) => {
  const traza = sirSimular(c, r);
  let pico = traza[0];
  for (const d of traza) if (d.I > pico.I) pico = d;
  const fin = traza[traza.length - 1];
  return { traza, R0: c / r, pico, alcance: 1 - fin.S, fin };
}

// The three series are labelled at the end of their own line, so two labels
// collide whenever two curves finish together: I and R both sit at zero when
// R0 < 1, and S and R meet when half the population has heard the story. This
// pushes them apart by a minimum gap in data units. It only ever moves labels
// up, and the top one can rise by at most two gaps, so it cannot leave the
// frame.
sirSeparar = (puntos, minimo) => {
  const orden = puntos.map(d => ({ ...d })).sort((a, b) => a.valor - b.valor);
  for (let i = 1; i < orden.length; i++) {
    if (orden[i].valor - orden[i - 1].valor < minimo) {
      orden[i].valor = orden[i - 1].valor + minimo;
    }
  }
  return orden;
}

sirPorciento = v => (v * 100).toFixed(1) + "%"

// The largest I in the trace is not always the peak of the curve. The window
// is fixed at sirDias, and for the six reachable settings just above R0 = 1
// (c/r of 0.11/0.10, 0.12/0.10, 0.12/0.11, 0.13/0.12, 0.14/0.13, 0.15/0.14)
// the story is still spreading when the window ends, so the maximum falls on
// the last day. Calling that a peak would name the edge of the frame as an
// event the chart does not show: the curve leaves the plot still climbing.
// Below R0 = 1 there is no peak either, only the seed decaying from day 0.
// So the reading reports a peak only where the curve has actually turned.
sirLectura = (c, r) => {
  const x = sirResumen(c, r);
  const partes = [
    `R0 = ${x.R0.toFixed(2)}`,
    x.R0 > 1 ? "it spreads" : "it dies out"
  ];
  if (x.R0 > 1) {
    partes.push(x.pico.dia === sirDias
      ? `still rising on day ${sirDias}`
      : `peak ${sirPorciento(x.pico.I)} on day ${x.pico.dia}`);
  }
  partes.push(`${sirPorciento(x.alcance)} have heard it by day ${sirDias}`);
  return html`<div class="data-meta">${partes.join(" · ")}</div>`;
}

// A display cell that is only a cell name renders Observable's inspector, so
// every visual output here is a function called from its own cell (CLAUDE.md
// section 13, bug 0).
sirGrafica = (c, r) => {
  const x = sirResumen(c, r);
  const largo = sirLargo(x.traza);
  const etiquetas = sirSeparar([
    { grupo: sirEtiqueta.S, valor: x.fin.S },
    { grupo: sirEtiqueta.I, valor: x.fin.I },
    { grupo: sirEtiqueta.R, valor: x.fin.R }
  ], 0.065).map(d => ({ ...d, dia: sirDias }));

  return Plot.plot({
    width: 680,
    height: 320,
    marginTop: 24,
    marginLeft: 46,
    marginBottom: 32,
    marginRight: 150,
    style: {
      background: "transparent",
      color: "var(--ink-soft)",
      fontFamily: "IBM Plex Sans, sans-serif",
      fontSize: "12px"
    },
    x: { label: null, domain: [0, sirDias] },
    y: { label: null, domain: [0, 1] },
    color: { domain: sirDominio, range: sirColores },
    marks: [
      Plot.gridY({ stroke: "var(--rule)" }),
      Plot.lineY(largo, {
        x: "dia", y: "valor", z: "grupo", stroke: "grupo",
        strokeWidth: 2, strokeLinejoin: "round", strokeLinecap: "round"
      }),
      Plot.text(etiquetas, {
        x: "dia", y: "valor", text: "grupo", fill: "grupo",
        textAnchor: "start", dx: 8, fontSize: 13, fontWeight: 500
      }),
      // The tip's data is the union of every series the chart draws, and its
      // title names the series it is describing (CLAUDE.md section 13).
      Plot.tip(largo, Plot.pointerX({
        x: "dia", y: "valor",
        title: d => `${d.grupo}\nday ${d.dia}\n${sirPorciento(d.valor)}`
      })),
      Plot.axisX({ stroke: "var(--rule)", tickSize: 0, ticks: 5 }),
      Plot.axisY({
        stroke: "var(--rule)", tickSize: 0, ticks: 5,
        tickFormat: d => (d * 100).toFixed(0) + "%"
      })
    ]
  });
}

viewof sirContagio = Inputs.range([0.10, 0.50], { label: "Contagion", step: 0.01, value: 0.30 })

viewof sirOlvido = Inputs.range([0.10, 0.50], { label: "Forgetting", step: 0.01, value: 0.15 })

sirLectura(sirContagio, sirOlvido)

sirGrafica(sirContagio, sirOlvido)

Simulation of the Kermack-McKendrick SIR model in the form Narrative Economics applies it to a story. The population is normalized to one and starts with one teller in a thousand. Both rates are per day, and the day is a convention of the exercise. Nothing here is estimated from data.

Two things follow from the shape of the middle curve.

The first is that it comes down on its own. Nobody has to refute the story. As it spreads, the group that has not heard it shrinks, and once that group falls below r / c each teller recruits less than one replacement. The peak arrives and the decline begins with the contagion rate exactly where it was. A narrative can disappear without ever having been contradicted.

The second is that contagion on its own settles nothing. A story dropped faster than it spreads goes nowhere. A story that spreads slowly against slower forgetting takes hold. The level of c sets how quickly the episode plays out; whether there is an episode at all is decided by c against r.

The model in three lines

Write S, I and R for the shares of the population that have not heard the story, are telling it, and have heard it and stopped, with S + I + R = 1. In each period c · S · I of the population starts telling it and r · I of the tellers stop. So S falls by c · S · I; I rises by c · S · I and falls by r · I; R rises by r · I.

I grows only while c · S > r. At the start almost nobody has heard the story, S is close to 1, and the condition reduces to c / r > 1. That ratio is the basic reproduction number R0: the number of new tellers one teller produces before stopping, with the population still susceptible. I reaches its maximum exactly when S has fallen to r / c, and declines from there with c unchanged.

The figure integrates those three lines forward in twenty steps per day, from a start of one teller in a thousand.

The original statement of the model is W. O. Kermack and A. G. McKendrick, “A Contribution to the Mathematical Theory of Epidemics”, Proceedings of the Royal Society A 115(772), 1927, 700-721.

The craft in two exercises

The craft half, in two exercises. The first is about what the eye resolves before it reads. The second takes one chart from a spreadsheet export to a finished figure, one decision at a time.

Preattentive attributes

digitosFilas = ["805391643725", "472680359168", "936158270436", "258073496153"]

// A display cell that is only a cell name renders Observable's inspector,
// so every visual output here is a function called from its own cell
// (CLAUDE.md section 13, bug 0).
gridDigitos = (activo) => html`<div class="digit-grid">${
  digitosFilas.map(fila => html`<div>${
    Array.from(fila, caracter =>
      activo && caracter === "3"
        ? html`<span class="digit-hit">3</span>`
        : html`<span>${caracter}</span>`)
  }</div>`)
}</div>`

viewof resaltar = Inputs.toggle({ label: "Highlight", value: false })

gridDigitos(resaltar)

There are seven. One change makes them visible before counting them.

The same chart, in six steps

indicadores = FileAttachment("data/labor-indicators.csv").csv({ typed: true })

etiquetaSerie = ({ TIL1: "Informality", TOSI1: "Informal sector" })

// The regimen column stores INEGI's own internal codes. The note under the
// chart promises the tooltip names the framework, so it gets a readable name
// rather than the code: "tradicional" is the regular ENOE, "enoen" the Nueva
// Edicion that ran while fieldwork recovered from the 2020 interruption, and
// "enoe_2023" the return to the standard ENOE on the 2023 framework.
etiquetaMarco = ({
  tradicional: "ENOE",
  enoen: "ENOEN",
  enoe_2023: "ENOE, 2023 framework"
})

serieDe = codigo => indicadores
  .filter(d => d.indicador === codigo)
  .map(d => ({
    serie: etiquetaSerie[codigo],
    fecha: new Date(d.anio, (d.trimestre - 1) * 3, 1),
    valor: d.valor,
    anio: d.anio,
    trimestre: d.trimestre,
    regimen: d.regimen
  }))
  .sort((a, b) => a.fecha - b.fecha)

// 2020-Q2 is genuinely absent (the ETOE gap). Without an explicit
// undefined point Plot bridges it with a straight diagonal, which reads a
// change of method as a gradual transition.
conHuecos = serie => {
  const resultado = [];
  for (let i = 0; i < serie.length; i++) {
    resultado.push(serie[i]);
    if (i < serie.length - 1) {
      const actual = serie[i], siguiente = serie[i + 1];
      const meses = (siguiente.fecha.getFullYear() - actual.fecha.getFullYear()) * 12
        + (siguiente.fecha.getMonth() - actual.fecha.getMonth());
      if (meses > 3) {
        resultado.push({
          ...actual,
          fecha: new Date(actual.fecha.getFullYear(), actual.fecha.getMonth() + 3, 1),
          valor: undefined
        });
      }
    }
  }
  return resultado;
}

seriesPaso = [serieDe("TIL1"), serieDe("TOSI1")]
puntosPaso = seriesPaso.flat()
lineasPaso = seriesPaso.flatMap(conHuecos)
finalesPaso = seriesPaso.map(s => s[s.length - 1])
dominioPaso = [etiquetaSerie.TIL1, etiquetaSerie.TOSI1]

textosPaso = [
  "Straight out of a spreadsheet. A border, gridlines, a marker on every quarter, a legend set apart from the lines, and the software's default colors.",
  "Chart border removed.",
  "Gridlines removed.",
  "Data markers removed.",
  "Axis labels cleaned up and sent to the background.",
  "Series labeled directly, legend dropped.",
  "Each label in its series' color. The chart is now in the style the data pages of this site use: their palette, their type, and one faint grid."
]

lineaPaso = (paso) => html`<p>${textosPaso[paso]}</p>`

graficaPasos = (paso) => {
  const sucio = paso <= 3;
  const casa = paso >= 6;
  const trazo = casa ? "var(--rule)" : "var(--ink-soft)";
  const marcas = [];

  if (paso <= 1) {
    marcas.push(Plot.gridY({ stroke: "var(--ink-soft)", strokeOpacity: 0.85, strokeWidth: 1 }));
  } else if (casa) {
    marcas.push(Plot.gridY({ stroke: "var(--rule)" }));
  }

  if (paso === 0) marcas.push(Plot.frame({ stroke: "var(--ink-soft)" }));

  marcas.push(Plot.lineY(lineasPaso, {
    x: "fecha", y: "valor", z: "serie", stroke: "serie",
    strokeWidth: casa ? 2 : 1.5,
    strokeLinejoin: "round", strokeLinecap: "round"
  }));

  if (paso <= 2) {
    marcas.push(Plot.dot(puntosPaso, {
      x: "fecha", y: "valor", z: "serie", fill: "serie", r: 2.8
    }));
  }

  if (paso >= 5) {
    marcas.push(Plot.text(finalesPaso, {
      x: "fecha", y: "valor", text: "serie",
      textAnchor: "start", dx: 10, fontSize: 13, fontWeight: 500,
      fill: casa ? "serie" : "var(--ink)"
    }));
  }

  if (casa) {
    marcas.push(Plot.dot(finalesPaso, {
      x: "fecha", y: "valor", z: "serie", fill: "serie",
      r: 3.5, stroke: "var(--plot-background)", strokeWidth: 1.5
    }));
    // The tip's data is the union of every series the chart draws, and its
    // title names the series it is describing (CLAUDE.md section 13).
    marcas.push(Plot.tip(puntosPaso, Plot.pointerX({
      x: "fecha", y: "valor",
      title: d => `${d.serie}\n${d.anio}-Q${d.trimestre} (${etiquetaMarco[d.regimen] ?? d.regimen})\n${d.valor}%`
    })));
  }

  marcas.push(Plot.axisX({
    stroke: trazo,
    tickSize: sucio ? 5 : 0,
    ticks: sucio ? 21 : 5,
    tickRotate: sucio ? -45 : 0,
    fontSize: sucio ? 10 : 12
  }));
  marcas.push(Plot.axisY({
    stroke: trazo,
    tickSize: sucio ? 5 : 0,
    ticks: sucio ? 14 : 5,
    tickFormat: sucio ? (d => d.toFixed(2)) : undefined
  }));

  return Plot.plot({
    width: 680,
    height: 340,
    marginTop: 30,
    marginLeft: sucio ? 58 : 44,
    marginBottom: sucio ? 58 : 30,
    marginRight: paso >= 5 ? 132 : 24,
    style: {
      background: "transparent",
      color: paso >= 4 ? "var(--ink-soft)" : "var(--ink)",
      fontFamily: casa ? "IBM Plex Sans, sans-serif" : "Arial, Helvetica, sans-serif",
      fontSize: "12px"
    },
    x: { label: null },
    y: { label: sucio ? "Percent (%)" : "%", domain: [0, 65] },
    color: casa
      ? { domain: dominioPaso, range: ["var(--teal)", "var(--clay)"] }
      : { domain: dominioPaso, scheme: "tableau10", legend: paso <= 4 },
    marks: marcas
  });
}

viewof pasoActivo = Inputs.radio([0, 1, 2, 3, 4, 5, 6], { label: "Step", value: 0 })

lineaPaso(pasoActivo)

graficaPasos(pasoActivo)

Labor informality (TIL1) and informal-sector employment (TOSI1), share of the employed population. ENOE, INEGI, original series, without seasonal adjustment. 2020-Q2 was not surveyed. The ENOE ran under three successive frameworks over this period, changing in 2020-Q3 and again in 2023-Q1. The line is drawn continuous across both, and the tooltip names the framework behind each point.

The two lines are worth reading for what they measure, not only for how they are drawn. Labor informality counts people who work without social security, wherever they work. Informal-sector employment counts people whose workplace is a business that never registered. Someone can be in one and not the other, and the distance between the lines is exactly that group: mostly salaried workers with no social security inside registered firms, along with paid domestic work and subsistence farming.

The gap is where the change happened. The series carries no seasonal adjustment, so the comparison runs second quarter against second quarter. Labor informality fell from 59.4% in 2005 to 55.1% in 2026, while informal-sector employment rose from 28.2% to 30.2%. The informal sector did not shrink. What narrowed was the gap, from 31.2 points to 24.9. The ground it gave up was unprotected work inside registered businesses, and a single “informality” number hides both movements at once, which is why the choice of series is already part of the story being told.

The sessions

Twelve sessions, each anchored in a paper and, where the session teaches one, a method: minimum wages and difference-in-differences, the China shock and shift-share instruments, folklore as a predictor of present-day economic choices, an election campaign that reactivates a historical episode from centuries earlier. Two of them, on storytelling with data, are where the exercises above come from. The course is taught in Spanish and the slides are in Spanish.

The twelve sessions, with slides

• 1. The importance of stories. Opening session. The origins of storytelling and the evidence that folklore predicts present-day values and economic choices. Readings: Shiller, Narrative Economics: How Stories Go Viral and Drive Major Economic Events (Princeton University Press, 2019); Michalopoulos and Xue, “Folklore” (Quarterly Journal of Economics, 2021); Stephens, Silbert and Hasson, “Speaker-listener neural coupling underlies successful communication” (2010). Slides (PDF, 0.48 MB)

• 2. Stories, statistics, and memory. Chapter 2 of Nexus, a review of randomization and Bayesian statistics, and an experiment on stories versus statistics. Method: randomization and Bayesian statistics. Readings: Harari, Nexus, chapter 2, “Stories: Unlimited Connections” (2024); Graeber, Roth and Zimmermann, “Stories, Statistics, and Memory” (Quarterly Journal of Economics, 2024). Slides (PDF, 0.20 MB)

• 3. The virality of narratives. Narrative contagion in Shiller, models as stories with structure, instrumental variables, and the Acemoglu and Albouy exchange. Method: instrumental variables. Readings: Shiller, Narrative Economics; Morgan, The World in the Model: How Economists Work and Think (2012); Acemoglu, Johnson and Robinson, “The Colonial Origins of Comparative Development: An Empirical Investigation” (American Economic Review, 2001), with Albouy’s Comment and the authors’ Reply (both American Economic Review, 2012); Angrist and Krueger (1991) as the worked instrument. Slides (PDF, 0.72 MB)

• 4. Minimum wages and difference-in-differences. Minimum wage models and the wage-price spiral narrative; difference-in-differences, with Card and Krueger and the Mexican case. Method: difference-in-differences. Readings: Card and Krueger, “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania” (American Economic Review, 1994) and Myth and Measurement: The New Economics of the Minimum Wage (1995); Calderón, Cortés, Pérez Pérez and Salcedo, “Disentangling the Effects of Large Minimum Wage and VAT Changes on Prices: Evidence from Mexico” (Labour Economics, 2023); Campos-Vázquez and Esquivel, “The Effect of Doubling the Minimum Wage and Decreasing Taxes on Inflation in Mexico” (Economics Letters, 2020); on the labor-market models behind the debate, Manning, Monopsony in Motion (2003), Diamond (1982), and Mortensen and Pissarides (1994); and on the estimator, de Chaisemartin and D’Haultfœuille (2020), Callaway and Sant’Anna (2021), Sun and Abraham (2021), and Roth, Sant’Anna, Bilinski and Poe (2023). Slides (PDF, 0.82 MB)

• 5. Regression discontinuity and behavioral economics. A regression discontinuity review, with an application to upper-secondary admission, and behavioral economics through the replication crisis. Method: regression discontinuity. Readings: Ortega Hesles, “Just Making the Admission Cut-Off” (2015); Kahneman and Tversky, “Prospect Theory: An Analysis of Decision under Risk” (Econometrica, 1979); Simmons, Nelson and Simonsohn, “False-Positive Psychology” (Psychological Science, 2011); Brodeur, Lé, Sangnier and Zylberberg, “Star Wars: The Empirics Strike Back” (American Economic Journal: Applied Economics, 2016); on the method, Thistlethwaite and Campbell (1960), Angrist and Lavy (1999), and Imbens and Lemieux (2008). Slides (PDF, 0.82 MB)

• 6. Job replacement and shift-share instruments. Narratives about jobs being replaced, and shift-share instruments applied to trade with China, robots, and artificial intelligence. Method: shift-share instruments. Readings: Autor, Dorn and Hanson, “The China Syndrome: Local Labor Market Effects of Import Competition in the United States” (American Economic Review, 2013); Acemoglu and Restrepo, “Robots and Jobs: Evidence from US Labor Markets” (Journal of Political Economy, 2020); Acemoglu, Autor, Hazell and Restrepo, “Artificial Intelligence and Jobs: Evidence from Online Vacancies” (Journal of Labor Economics, 2022); Borusyak, Hull and Jaravel, “A Practical Guide to Shift-Share Instruments” (Journal of Economic Perspectives, 2025); on measuring exposure, Felten, Raj and Seamans (2019), Webb (2020), and Brynjolfsson, Mitchell and Rock (2019). Slides (PDF, 1.22 MB)

• 8. Narratives and economic events: panic and confidence. Shiller covers two-way causality and the narratives of panic and confidence; Roos and Reccius define the collective economic narrative. Readings: Shiller, Narrative Economics; Roos and Reccius, “Narratives in economics” (Journal of Economic Surveys 38(2), 2024, 303-341); Cass and Shell, “Do Sunspots Matter?” (Journal of Political Economy, 1983); Vosoughi, Roy and Aral, “The spread of true and false news online” (Science, 2018). Slides (PDF, 0.32 MB)

• 9. Storytelling with data I: context, charts, and clutter. Covers Knaflic’s first three lessons: context, choosing the chart, and removing clutter. Closes with McCloskey on economic writing. Method: data visualization. Readings: Nussbaumer Knaflic, Storytelling with Data (2015), lessons 1 to 3; McCloskey, “Economical Writing” (Economic Inquiry 23(2), 1985, 187-222). Slides (PDF, 2.12 MB)

• 10. Storytelling with data II: attention, design, and story. Covers Knaflic’s last three lessons: attention, design, and narrative structure, worked through one example. Closes with McCloskey on rhetoric. Method: visual narrative. Readings: Nussbaumer Knaflic, Storytelling with Data, lessons 4 to 6; McCloskey, “The Rhetoric of Economics” (Journal of Economic Literature 21(2), 1983, 481-517); McKee, Story: Style, Structure, Substance, and the Principles of Screenwriting (1997); Vonnegut, “How to Write with Style” (1980). Slides (PDF, 1.94 MB)

• 11. Storytelling in public policy. The Jones and McBeth Narrative Policy Framework, applied to three development evaluations: school construction in Indonesia, deworming, and antimalarial bed nets. Method: randomized trials and difference-in-differences. Readings: Jones and McBeth, “A Narrative Policy Framework: Clear Enough to Be Wrong?” (Policy Studies Journal, 2010); Duflo, “Schooling and Labor Market Consequences of School Construction in Indonesia” (American Economic Review, 2001); Miguel and Kremer, “Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities” (Econometrica, 2004); Cohen and Dupas, “Free Distribution or Cost-Sharing? Evidence from a Randomized Malaria Prevention Experiment” (Quarterly Journal of Economics, 2010); on the framework and its critics, Stone, Policy Paradox, Roe, Narrative Policy Analysis (1994), and Schlaufer, Kuenzler, Jones and Shanahan (2022). Slides (PDF, 0.67 MB)

• 13. Storytelling in business. Narrative as brand, as investment pitch, and as internal culture. Two papers measure its effect on advertisements and on IPO prospectuses. Method: content coding and regression. Readings: Quesenberry and Coolsen, “What Makes a Super Bowl Ad Super? Five-Act Dramatic Form Affects Consumer Super Bowl Advertising Ratings” (Journal of Marketing Theory and Practice, 2014); Martens, Jennings and Jennings, “Do the Stories They Tell Get Them the Money They Want? The Role of Entrepreneurial Narratives in Resource Acquisition” (Academy of Management Journal, 2007); with Freytag’s five-act dramatic pyramid as the scaffolding. Slides (PDF, 0.48 MB)

• 14. Storytelling in political processes. Ideas, framing, and collective memory in politics. The central case measures how an election campaign reactivates a historical episode from centuries earlier. Method: difference-in-differences and spatial regression discontinuity. Readings: Ochsner and Roesel, “Activated History: The Case of the Turkish Sieges of Vienna” (American Economic Journal: Applied Economics, 2024); Mukand and Rodrik, “The Political Economy of Ideas” (NBER, 2018); Entman, “Framing: Toward Clarification of a Fractured Paradigm” (Journal of Communication, 1993); Halbwachs, Les cadres sociaux de la mémoire (1925); on historical persistence, Dell (2010, 2012) and Acemoglu, Johnson and Robinson (2001); Chetty et al., “Social Capital I: Measurement and Associations with Economic Mobility” (Nature, 2022). Slides (PDF, 0.50 MB)

Sessions 7 and 12 were given to student presentations of their paper proposals, so they have no slides.

How it is graded

• Class participation, with three slides per assigned reading (summary, strengths, and weaknesses), presented to the group by one student each session.
• Assignments: empirical exercises through the course.
• A five to ten minute presentation of the paper proposal to the group, with feedback from classmates.
• A two-page written proposal: the story to be told, the motivation, the elements that support it, and the plan of work.
• Final paper.

The syllabus allows AI tools for exploring ideas, drafting, and iterating, provided the submitted text is not copied from them and their claims are checked.

Readings

Books and articles the course draws on throughout, beyond the readings listed with each session.

• Robert J. Shiller. Narrative Economics: How Stories Go Viral and Drive Major Economic Events. Princeton University Press, 2019.
• Deirdre N. McCloskey. The Rhetoric of Economics. University of Wisconsin Press, 1985.
• Mary S. Morgan and Thomas A. Stapleford. “Narrative in Economics: A New Turn on the Past”. History of Political Economy 55(3), 2023, 395-421.
• Mary S. Morgan. The World in the Model: How Economists Work and Think. Cambridge University Press, 2012.
• Michael Roos and Matthias Reccius. “Narratives in Economics”. Journal of Economic Surveys 38(2), 2024, 303-341.
• Karla Borja and Suzanne Dieringer. “Telling My Story: Applying Storytelling to Complex Economic Data”. Eastern Economic Journal 49(3), 2023, 328-348.
• Edward E. Leamer. Macroeconomic Patterns and Stories: A Guide for MBAs. Springer, 2009.