Haute Lumière

Commerce · II.02 · MMXXVI · daylight

La Bourse  /  Volume II  /  Nº II.02  /  Ten concept briefs

TEN CONCEPT BRIEFS · Chapter II.02 — Markets as Living Systems

One page each. A reader who reads only these ten pages has the chapter.

---

BRIEF 1 — What Arrow and Debreu Actually Proved

The idea. General equilibrium theory proved that under stated conditions a set of prices exists at which every market clears at once. It did not prove that such a price set is unique, and it did not prove that any real process finds one.

Three separate questions live inside the word "equilibrium":

Worked example. A GPS that proves a route from London to Rome exists has told you something true and nearly useless. You wanted the route, and whether your car can drive it. Existence proofs are the first of three questions, and most economic argument proceeds as though they were all three.

Why it matters. Almost every popular claim made on behalf of markets — that they find the efficient allocation, that they tend toward balance, that interference prevents convergence — is a claim about attainment, and attainment was never established. The theory is not being attacked here. It is being read at the level of precision it was written at.

You already know this because you have watched a plan that was provably possible fail to happen, and you did not conclude the plan was impossible. You concluded that possible and reachable are different properties.

---

BRIEF 2 — Sonnenschein–Mantel–Debreu

The idea. Take an economy of perfectly rational agents with well-behaved preferences. Add them up. Ask what the aggregate can look like. The answer, proved between 1972 and 1974, is: almost anything.

Exactly three properties survive aggregation:

1. Continuity
2. Walras's law — the total value of excess demand is zero
3. Homogeneity of degree zero — doubling all prices changes nothing

What does not survive: uniqueness, stability, and comparative statics. Any function with those three properties is the aggregate excess demand of some economy of impeccably orthodox maximisers.

Worked example. A room of a thousand people, each with perfectly consistent preferences over lunch, can produce a collective ranking that is cyclic, unstable, and reverses when an irrelevant option is added. Nothing is wrong with any individual. The aggregate is simply a different object from its parts.

Why it matters. The founding move of modern macroeconomics is to model the whole economy as one optimising agent. SMD says the population's rationality places almost no restriction on the aggregate's behaviour — so the representative agent is not a simplification, it is a substitution. Kirman (1992) worked the consequence out in detail.

You already know this because you have sat in a team where every member was individually reasonable and the team's decision was not, and nobody in the room could be blamed for it.

---

BRIEF 3 — Scarf's Orbit

The idea. Existence is not attainment, and here is the counterexample that settles it.

Three goods, three consumers. Each consumer owns one unit of one good and wants it in fixed proportion with one other — Leontief preferences, the most well-behaved case. Equilibrium exists at equal prices. Now run the textbook adjustment story: raise the price of whatever is in excess demand.

  z₁ = p₁/(p₁+p₂) + p₃/(p₃+p₁) − 1      and cyclically for z₂, z₃
  ṗ  = z(p)      conserves  p₁·p₂·p₃

Because the product of prices is conserved, the path is stuck on a closed orbit. Run from 25 percent off equilibrium for two hundred thousand steps: distance starts at 0.490, ends at 0.470, and the closest approach in the whole path is 0.457. It never arrives.

Worked example. A thermostat wired so that heat raises the setpoint. Every component works. The house never reaches temperature.

Why it matters. This is not a pathological corner. It is three consumers with textbook preferences, and it is why "the market will find it" is a statement requiring evidence rather than a statement of theory.

You already know this because you have been in a negotiation where every party responded rationally to the last offer and the thing simply circled.

---

BRIEF 4 — The Representative Firm Does Not Exist

The idea. US firm sizes follow a Zipf distribution with Pareto exponent α = 1.059, measured by Axtell (2001) on roughly 5.5 million firms.

At that exponent:

• the mean is finite (it requires α > 1 — just)
• the variance is infinite (it requires α > 2)

A distribution with infinite variance has no typical member. There is no scale around which firms cluster; the same shape repeats at every magnitude.

Worked example. Take average firm employment in your sector, then add one firm the size of the largest. Under a normal distribution the average barely moves. Under α ≈ 1, it moves a great deal, and it will move again next year. The average is not converging to anything, because there is nothing to converge to.

Why it matters. "The typical firm in our industry" is a phrase that feels like data and is not. Any policy, forecast or strategy built on a representative unit is built on a statistic the population does not possess. The right object is the distribution and its exponent — which is knowable, stable, and far more informative.

You already know this because you have never once been surprised to learn that a handful of firms account for most of the output in any sector you know well. You just were not told that this is a law with a number.

---

BRIEF 5 — The Cubic Tail

The idea. Financial returns are not normally distributed. Their tails follow a power law with exponent close to 3 — measured by Gabaix, Gopikrishnan, Plerou and Stanley (2003) across indices, individual stocks and markets.

Worked example. On 19 October 1987 the S&P 500 fell 20.47 percent in one day. Against a daily standard deviation of about 1.0 percent, that is a twenty-sigma move.

The age of the universe is 1.4 × 10¹⁰ years. The Gaussian model does not say the crash was unlikely; it says it was impossible, and it happened.

Why it matters. 161 years is a number an organisation can plan against — it means a treasurer with a forty-year career should expect to see roughly a quarter of one. Ten to the ninetieth is a number that removes the event from the risk register entirely. The choice of distribution is not a technicality. It decides whether the largest thing that can happen to you appears in your planning at all.

You already know this because you have lived through at least one once-in-a-century event, and probably three, and you noticed.

---

BRIEF 6 — Increasing Returns and Lock-In

The idea. When the returns to a technology rise with its adoption — networks, standards, platforms, skills — the winner is selected by history rather than by merit, and then defended by the returns.

Arthur (1989) modelled this as a Pólya urn: each adoption makes the next adoption of the same option more likely.

Worked example. Run 4,000 urns of 2,000 draws each, starting one ball to each of two technologies. The share held by technology A at the end, by decile:

  10.2%  9.8%  10.4%  10.4%  9.3%  10.2%  8.8%  10.6%  10.6%  9.8%

That is a uniform distribution. Every possible market share is equally likely, and every one is permanent. Mean 0.4999, variance 0.0840 — against 0.5 and 0.0833 for Uniform(0,1).

Why it matters. Under increasing returns, "the market chose it, therefore it was best" is an inference the mathematics does not license. The market chose it, therefore it was early. Diminishing returns give you a unique efficient equilibrium; increasing returns give you many, and the one you get is the one that got ahead first.

You already know this because you can name at least one dominant standard in your own field that everybody agrees is inferior to something that lost.

---

BRIEF 7 — Emergence from Adaptation

The idea. Put heterogeneous agents in a market, let them learn, and the market's characteristic behaviour appears without anyone having written it in.

The Santa Fe Artificial Stock Market (Arthur, Holland, LeBaron, Palmer and Tayler, 1997) gave each trader a bundle of forecasting rules and let a genetic algorithm select among them. Nobody coded fat tails, volatility clustering, bubbles or technical trading. All of them emerged, as a consequence of agents learning about a market their learning was changing.

Worked example. No individual bird in a flock has a rule about flock shape. Three local rules about neighbours produce the shape. Reading the shape and inferring a plan is a mistake about what kind of object you are looking at.

The honest boundary, and it is important. The published runs found regimes: at slow learning rates the market converged toward the homogeneous rational-expectations equilibrium; at faster ones it entered the complex regime with the realistic statistics (LeBaron, Arthur and Palmer, 1999). The neoclassical model is not refuted — it is located. It is the low-adaptation limit of a larger model.

Why it matters. This is the strongest form of the claim in this chapter: not that equilibrium theory is wrong, but that it is a special case, and that the special case requires agents to learn slowly.

You already know this because you have watched a market move on what traders thought other traders would think, and known that no fundamental had changed.

---

BRIEF 8 — The Exponents That Break the Metaphor

The idea. The economy is not an organism, and the evidence is a sign.

Biological metabolism scales sublinearly: Kleiber's law, B ∝ M^0.75, confirmed across twenty-seven orders of magnitude. Urban socioeconomic output scales superlinearly: β ≈ 1.15 for wages, GDP and patents (Bettencourt, Lobo, Helbing, Kühnert and West, 2007).

  β = 0.75   metabolism      per unit on doubling   0.841   (−15.9 %)
  β = 0.85   infrastructure  per unit on doubling   0.901   ( −9.9 %)
  β = 1.15   wages, GDP      per unit on doubling   1.110   (+11.0 %)

Worked example. Double an elephant and each gram works 16 percent slower. Double a city and each person produces 11 percent more. These are opposite behaviours, and no amount of metaphor reconciles them.

Why it matters. What survives the correction is the ecosystem analogy, not the organism one — many lineages, no shared objective, no target size, no designed death. Every useful inference in living-systems economics comes from the ecosystem; every sentimental one comes from the organism.

And the cost, from the same paper: superlinear growth on a finite resource reaches a singularity in finite time unless innovation resets the curve.

You already know this because you have never once expected a city to stop growing when it reached adulthood, and you do expect that of a body.

---

BRIEF 9 — May's Criterion

The idea. Diversity does not buy stability. In the only model where the claim has been made precisely, it buys the opposite.

May (1972) showed that a randomly assembled community of S species with connectance C and interaction strength σ is almost surely unstable when:

  σ · √(S·C)  >  1

Worked example. Haldane and May (2011) carried this straight into banking: a densely interconnected financial network amplifies a local failure rather than absorbing it. Connectivity is risk-sharing at low stress and contagion at high stress, and the switch between the two is the criterion above.

The companion result. Acemoglu, Carvalho, Ozdaglar and Tahbaz-Salehi (2012) showed that independent sectoral shocks decay as 1/√n only in a balanced network. With a heavy-tailed input-output structure, aggregate volatility at n = 1,000 sectors can be ten times what the law of large numbers predicts.

Why it matters. It disciplines the whole frame. The living-systems view does not imply that more diversity and more connection are always better — it implies you must measure S, C and σ and find out.

You already know this because you have seen a small supplier failure take down a line, in an organisation that thought it was diversified.

---

BRIEF 10 — The Window of Vitality

The idea. Efficiency and resilience are the same axis read from opposite ends, and the optimum is neither end.

Ulanowicz measures a, the relative ascendency — the share of a system's throughput that runs through organised, efficient, constrained channels. The remainder is reserve capacity: redundancy, slack, alternative pathways. Robustness is:

  R(a) = −a · ln a        maximised at  a = 1/e = 0.3679

Worked example. A firm with one supplier, zero inventory, one channel and one product has driven a toward 1, where robustness collapses. A firm that diversifies without limit pays coordination costs it cannot carry. Both failures are on the same curve.

The honest caveat, and it belongs in the brief. 1/e is a property of the function −a·ln a, not a measured constant of economies. No dataset has established that an economy at 0.368 outperforms one at 0.30. Use it as a shape, not a target.

Why it matters. It converts "we should be more resilient" — which loses every budget argument — into a position on a named axis, with a floor that can be written into a covenant and reviewed beside margin.

You already know this because you have watched an efficiency programme make an organisation faster and then make it brittle, and you could feel the turn before anyone could name it.

---

All figures in these briefs are computed in lib/verify/II_02.py and reproducible with python3 lib/verify.py II.02. Published constants are attributed at the line that prints them.