A Unified Framework for Monetary Theory and Policy Analysis

Time is discrete and each period is split into a decentralized “day” subperiod, with anonymous bilateral matching at arrival rate alpha and a double-coincidence probability delta versus a single-coincidence probability sigma, and a centralized, Walrasian “night” subperiod where all agents trade a general good (Sec. 2, pp. 3-4). Money is essential “for the same reason it is in the typical search-based model: since meetings in the day market are anonymous there is no scope for trading future promises, so exchange must be quid pro quo” (Sec. 2, p. 5). Special day-market goods and the general night-market good are both nonstorable, so the only way to carry value from day to night or across periods is the one perfectly divisible, storable object the authors call money (Sec. 2, pp. 4-5).

Because preferences over the centralized-market good are quasi-linear (U(X) - AH), the value of entering the centralized market with m dollars, W(m,s), is linear in m: W(m,s) = W(0,s) + phi*m, and the optimal money balance m_{+1} chosen there does not depend on how much money the agent brought in (Sec. 3, eq. 2-4, pp. 6-7). This “convenient result” means every agent leaves the centralized market with the same money holdings, so the distribution of money in the next decentralized market is degenerate – the feature that earlier search-theoretic monetary models (which allow unrestricted money holdings) could not obtain without numerical methods (Sec. 3, p. 7; Sec. 4, p. 18). The authors contrast this explicitly with Trejos and Wright (1995), who “cannot solve” the unrestricted-money-holdings version of their own model and so study only a two-point restriction, and with Molico (1999), who studies the unrestricted case only numerically (Sec. 4, p. 18).

In a double-coincidence meeting, Nash bargaining with threat points equal to continuation values yields the efficient quantity q* (where u'(q*) = c'(q*)) with no money changing hands, regardless of either agent’s money holdings (Appendix Lemma 1; Sec. 3, p. 7). In a single-coincidence meeting, generalized Nash bargaining with buyer bargaining power theta implies the buyer spends all of his money and receives a quantity q(m) < q* strictly below the efficient level whenever his real balances are below a threshold m*, and receives exactly q* once his balances reach or exceed that threshold (Sec. 3, eq. 6-8, p. 8). This underproduction relative to q* is the central inefficiency the rest of the paper’s policy analysis works around: real balances below the threshold constrain how much the buyer can “pay,” and so constrain how much the seller is willing to produce (Sec. 3, pp. 8-9).

Proposition 1 states that any monetary equilibrium has a degenerate money distribution, d = m, and q = q(m) < q*; a steady state q_s > 0 exists under a condition on the function e(q) = u'(q)/z'(q), and – given theta close to 1 or log-concave u' – that steady state is unique and increasing in beta, alpha, sigma, and theta (Sec. 3, Proposition 1, p. 15). Critically, the steady-state quantity converges to the efficient level q* as the discount factor beta approaches 1 if and only if full bargaining power theta = 1 (Sec. 3, Proposition 1, p. 15) – so even in the limit of patient agents, any buyer bargaining power below one leaves a residual inefficiency that monetary policy alone cannot undo.

When money growth is tau (so M_{t+1} = (1+tau)M_t), the Friedman rule tau = tau_F = beta - 1 (deflating at the rate of time preference) is always optimal, but it achieves the efficient outcome q_s = q* only if theta = 1; with theta < 1, q_s < q* persists even at the Friedman rule (Sec. 5, eq. 22, p. 22-23). The intuition is a holdup problem: “An agent who carries a dollar into next period is making an investment with cost phi… When he uses the money in the future he reaps the full return on his investment iff theta = 1; otherwise the seller ‘steals’ part of the surplus,’’ which lowers the incentive to hold real balances and hence lowers q (Sec. 5, p. 23). Because the Hosios condition for bargaining efficiency here requires theta = 1 (since all the match surplus in a single-coincidence meeting comes from the buyer’s side), theta < 1 means the welfare function V is kinked at the Friedman rule rather than flat, so the Envelope Theorem no longer applies and “a moderate inflation therefore has a bigger welfare cost when theta < 1” (Sec. 5, p. 23).

The authors fit a three-parameter vector Omega = (eta, B, sigma) – governing curvature of decentralized-market utility, night-market preference scale, and the single-coincidence meeting probability – to annual U.S. “money demand” data for 1900-2000, where L(i) = M/(PY) is fit against the nominal interest rate i, following Lucas (2000)’s methodology exactly (Sec. 6, pp. 28-29). Because Omega is not precisely identified from this data alone, the authors fix sigma = 0.5 for most experiments (which bounds the decentralized sector at roughly 10% of aggregate output at a 4% benchmark inflation rate) and separately calibrate the bargaining power theta to three values: theta = 1 (no holdup), theta = 0.5 (symmetric bargaining), and theta_mu = 0.343, chosen so the model’s implied aggregate markup matches an empirical target markup of 1.1 (Sec. 6, pp. 29-30).

With theta = 1, eliminating a 10% inflation is worth 1.4% of consumption (1 - Delta_0 = 0.014), which the authors describe as “basically consistent with the consensus view in the literature” and comparable to Lucas’s (2000) estimate of “just under 1%” (Sec. 6, Table 1, p. 31). With theta = 0.5, the same calculation rises to 3.2% of consumption, and with the markup-calibrated theta_mu = 0.343 it rises further to 4.6% – the authors’ headline claim that “it is not hard to generate welfare costs equivalent to 3% or 5% of consumption” depends specifically on calibrating this bargaining-power/holdup wedge, not merely on the search-theoretic structure alone (Sec. 6, Table 1, p. 31; Sec. 7, p. 35-36).

The authors note an alternative route to a degenerate money distribution due to Shi (1997), in which the decision-making unit is an infinite “family” whose members pool money each period so that, by a law of large numbers, every buyer within a family carries the same balance (Sec. 4, p. 18). The authors argue their centralized-market “trick” is preferable on three counts: it does not require the “unappealing,” to some, infinite-family structure; it avoids “serious technical complications” that arise when bargaining happens between individuals but accrues to a larger decision unit; and, unlike the family structure, it does not require assuming that individual agents act against their own momentary self-interest (reporting no customers to avoid production costs) purely because the household head says so (Sec. 4, pp. 18-19). The authors are careful to note that Shi’s family structure is “not” equivalent to a competitive insurance market, correcting what they describe as a common misreading (Sec. 4, p. 19).

The paper extends the deterministic framework to allow uncertainty, including match-specific preference/cost shocks and random lump-sum monetary transfers across agents (Erho = 0), and reports that with uncertainty the money constraint need not bind with probability one (Sec. 5, pp. 24 onward). The conclusion is explicit that the model’s tractability rests on the combination of quasi-linear preferences and a periodically-convening centralized market, and that the authors “think that we have only scratched the surface, and much more could be done in terms of applications and extensions in future research” (Sec. 7, p. 36) – the paper itself is framed as a tractable scaffold for later policy work rather than a final quantitative verdict.