Arrow–Debreu Securities

The atoms of payoff space

Think of the future as a finite set of possible states — say, three states tomorrow: boom, normal, recession. An Arrow–Debreu security for the boom state is a contract that pays exactly $1 if tomorrow is a boom and $0 otherwise. Its companion securities pay $1 in normal only, and $1 in recession only. Three states, three primitive securities. Their prices today — denote them q_boom, q_normal, q_recession — are called state prices.

Now consider any other tradable asset, like a stock with payoffs of $120 in boom, $100 in normal, and $80 in recession. By no-arbitrage, the stock's price today must equal the cost of the replicating portfolio: 120 A-D-boom plus 100 A-D-normal plus 80 A-D-recession. That is,

P_stock  =  120 · q_boom  +  100 · q_normal  +  80 · q_recession

The same formula prices anything in the market — call options, put options, bonds, futures — by changing only the payoff vector. The state prices are the universal coordinates. The set of A-D securities is to asset pricing what the standard basis {e₁, e₂, e₃} is to linear algebra: every vector is a unique linear combination.

State price = probability × stochastic discount factor

The state price q_s decomposes into two pieces:

q_s  =  π_s  ·  M_s

π_s  =  physical (real-world) probability of state s
M_s  =  stochastic discount factor (SDF) in state s
       =  β · u'(c_s) / u'(c_0)
       (intertemporal marginal rate of substitution)

The first factor is what we mean by "how likely is this state?" The second is "how much do I value $1 if it arrives in this state?" — the marginal utility of consumption in state s, relative to today, discounted at the time-preference rate β. States that arrive when you're already rich (high c_s, low u'(c_s)) have low M, so $1 there is worth less than $1 in a recession state when consumption is scarce. That is why insurance — payoffs concentrated in bad states — is valuable: the SDF puts more weight on bad-state dollars.

Sum the formula P = E[M · payoff] across all states and you have the central pricing equation of modern finance:

P_today  =  E[M · x]  =  Σ π_s · M_s · x_s  =  Σ q_s · x_s

Every asset-pricing model — CAPM, consumption-CAPM, Black-Scholes, multifactor models — is just a different specification of M. The Arrow–Debreu framework is the structure they all sit inside.

Worked example: pricing a call option from state prices

Suppose the stock above costs $90 today. The risk-free rate is 5%, so a zero-coupon bond paying $1 next period costs $1/1.05 ≈ $0.952. We have two equations and three unknowns (q_boom, q_normal, q_recession), so we need one more security to pin them down. Suppose a put option that pays max(100 − S, 0) trades at $4.25 — payoffs (0, 0, 20). We can now solve the linear system:

  0.952  =      q_boom  +     q_normal  +     q_recession    (bond)
 90.000  =  120·q_boom  + 100·q_normal  +  80·q_recession    (stock)
  4.250  =    0·q_boom  +   0·q_normal  +  20·q_recession    (put)

The third equation gives q_recession = 4.25/20 = 0.2125 directly. Substituting back leaves a 2×2 system that solves to q_boom ≈ 0.405 and q_normal ≈ 0.335. Notice the state prices are not equal to physical probabilities — they are skewed toward the bad state by the SDF.

Now price a call option with strike $100 — payoffs (20, 0, 0). The price is just 20 · q_boom = 20 · 0.405 = $8.10. No model assumption beyond linearity and no-arbitrage. Every option, every structured product, every derivative on this stock prices the same way.

Complete versus incomplete markets

If, from the assets we can trade, we can synthesize every Arrow–Debreu security on the state space, the market is complete. Every contingent payoff has a unique no-arbitrage price; risk can be perfectly transferred between agents; the welfare theorems give Pareto-optimal allocations. The fundamental theorem of asset pricing splits into two halves:

• No-arbitrage is equivalent to the existence of strictly positive state prices q_s > 0 (and equivalently a strictly positive SDF M_s > 0).
• Completeness is equivalent to the state prices being unique — a market without complete spanning admits multiple positive state-price vectors, each giving a different no-arbitrage price for the unspannable claims.

Real markets are spectacularly incomplete. Labor-income shocks are individual-specific and untradable. Catastrophic events (asteroid, supervolcano) have no insurance market. Long-dated nominal cash flows beyond 30 years have no traded zero-coupon bond. Health-status outcomes are partially insurable but with adverse-selection-driven gaps. The growth of options markets, prediction markets, and bespoke derivatives is, in effect, a long secular project to fill in A-D coordinates the natural economy left empty.

Arrow–Debreu vs neighbouring frameworks

Concept	State space	Pricing rule	What it adds beyond A-D
Arrow–Debreu (1953/1954)	Finite discrete states	P = Σ q_s · x_s	The base layer; everything else is a special case.
SDF representation	Discrete or continuous	P = E[M · x]	Folds π into M; lifts cleanly to continuous time and continuous state.
Risk-neutral pricing	Discrete or continuous	P = E*[x] / (1 + r_f)	Absorbs M into a change of measure; π*_s = q_s / Σ q_s.
CAPM (Sharpe 1964)	Continuous returns	E[R_i] − r_f = β_i (E[R_m] − r_f)	Assumes mean-variance utility; M is linear in market return.
Consumption-CAPM (Breeden 1979)	Continuous	E[R_i] − r_f = β_c,i · λ_c	M = β · u'(c)/u'(c_0); links asset prices to macro consumption.
Black-Scholes (1973)	Continuous Brownian	Closed-form option formula	Specific SDF (lognormal returns, constant rate, constant σ); a special case of A-D in continuous state.
Black–Litterman, factor models	Finite or continuous	E[R] from market priors	Practical estimation; still nest inside the A-D pricing rule.

Where A-D securities show up in real markets

• Binary options. A digital option that pays $1 if a condition holds at expiry (e.g. S > K) and $0 otherwise is the cleanest direct approximation. CBOE binary options on the S&P 500 trade actively, and many FX dealers offer one-touch binaries.
• Butterfly spreads. Long one call at strike K−h, short two calls at K, long one call at K+h. In the limit as h → 0, the payoff converges to a delta function at K — an A-D security on terminal stock price. Breeden and Litzenberger (1978) showed that the second derivative of the call-pricing function with respect to strike is the risk-neutral density.
• Prediction markets. Polymarket and Kalshi contracts pay $1 if an event occurs and $0 otherwise — literal A-D securities on real-world states. Election markets, weather contracts, and inflation-prediction contracts trade explicit state prices.
• Treasury STRIPS. Zero-coupon government bonds isolating $1 at one specific future date — a discrete-time A-D security on the trivial single-state economy at each maturity.
• Catastrophe bonds. Pay full principal unless a defined event (hurricane, earthquake) triggers; a payoff that approximates an indicator on a specific natural-disaster state.
• Variance swaps and volatility futures. Their payoff depends on the integral of the squared returns — implicitly a continuum of A-D securities across the strike spectrum, formalised by the Carr–Madan replication formula (1998).

Significance: why the abstraction earned two Nobels

Before Arrow's 1953 paper, "risk" in asset pricing was modelled crudely — usually as mean-variance trade-offs against a single aggregate uncertainty. Arrow's reformulation in terms of state-contingent commodities did three things simultaneously. First, it embedded risk into the same Walrasian general-equilibrium framework that priced ordinary goods, allowing the welfare theorems to carry over. Second, it gave a clean criterion for when markets fail to allocate risk efficiently — incompleteness — and pointed toward concrete remedies, namely the development of derivative markets. Third, it provided the mathematical scaffolding on which Black-Scholes (1973), Lucas (1978), Breeden (1979), and the entire SDF revolution of the 1980s and 1990s would be built. Cochrane's textbook (Asset Pricing, 2001) opens with "P = E[Mx]" as the one equation of finance, and that equation is the A-D pricing rule with π folded in.

Common pitfalls and confusions

• Confusing state prices with probabilities. The state prices q_s do not sum to 1 — they sum to 1/(1 + r_f), the price of the riskless bond. Probabilities sum to 1. Multiplying through by (1 + r_f) gives the risk-neutral probabilities π*_s, which do sum to 1 — but those are not physical probabilities either.
• Assuming completeness. Most asset-pricing courses begin with complete-market intuition and then "extend" to incomplete settings. Real markets are incomplete by default; the right question is what specific securities span which subsets of risk, not whether the entire space is spanned.
• Reading the SDF as a forecast. M_s is not a prediction of consumption — it is a pricing kernel that encodes investors' marginal utility. A model where M is uncorrelated with consumption is observationally indistinguishable from one where investors are risk-neutral; that is precisely the empirical "equity premium puzzle."
• Mixing up discrete and continuous frameworks. In a continuous-state model, there are no atomic A-D securities — only an A-D density q(s). The pricing rule is an integral, not a sum, and Breeden–Litzenberger recovers q(s) from the second derivative of call prices in strike.
• Forgetting time. A-D securities are typically indexed by (state, date). A claim paying $1 in boom-three-years-from-now is different from a claim paying $1 in boom-tomorrow, and the corresponding state prices live in a richer space.
• Treating prediction markets as physical probabilities. A 30% Polymarket price on an election outcome is a state price, not a physical probability. It equals π · M / (1 + r_f). For neutral SDFs the discrepancy is small, but for politically polarised outcomes with concentrated participation, it can be substantial.