MIT OpenCourseWareLecture 7: Stochastic Financial Networks
MIT OpenCourseWare
Introduction and course connections 0:00
Robert Townsend opens by linking this lecture on stochastic financial networks to the previous session and mentions a planned but missing third lecture that he may sketch at the end if time allows. He frames the topic as liquidity and the value of key players versus contagion dynamics, essentially asking whether interventions enhance or limit markets. His outline covers the economic environment, the definition of a stochastic financial network with centralized and fragmented market examples, ex-ante liquidity injections as buffers against shocks, identifying the most valuable recipient of such injections, empirical work using Thai village data, and finally financial centrality in relation to contagion and systemic risk.
Renaming financial centrality to liquidity value 3:00
Townsend explains that the term financial centrality, used in the paper and slides, has caused confusion because it is often associated with contagion-based network measures, which his measure is not equivalent to. He says the terminology will be changed to liquidity value of a player, though the slides for this lecture still use the older wording. He situates the work relative to three literatures: Darrell Duffie's search-friction models of over-the-counter markets, random-matching monetary models like those of Kiyotaki and Wright, and earlier partitioned trading setups used in the course, plus explicit random-participation models tied to the Federal Reserve's role in injecting liquidity, an idea traced back to Milton Friedman and reflected in a quote at the New York Fed about managing liquidity shortages rather than employment or price stability.
Defining liquidity value and market examples 9:01
Financial centrality, or liquidity value, is defined as the marginal social value of giving a small amount of extra purchasing power to an agent, conditional on that agent actually being able to trade with others, since liquidity given to someone in isolation is useless. Townsend shows a 2006 picture of the federal funds interbank market broken into half-hour intervals, where trading is thin at the start and end of the day and denser at midday, illustrating how markets vary over time. He notes that such pictures are no longer common because the federal funds market has largely collapsed, replaced by an active repo market where money market mutual funds lend to hedge funds and pension funds against collateral, and this repo market has become the main venue for Federal Reserve policy.
Risk sharing environment and shocks 11:03
The underlying model has a finite number of risk-averse agents with concave utility who maximize expected utility over random incomes drawn from a distribution, with an example given using constant absolute risk-averse exponential utility and a mean vector and variance-covariance matrix capturing income correlations. A second shock governs market participation, represented by a binary vector where zero means an agent is out of the market and stuck eating their own income, and one means they can share risk with others, subject to an aggregate resource constraint that holds for every possible combination of income and participation outcomes. The community objective is to maximize a weighted sum of expected utilities across these shocks, with consumption chosen to pool risk, so that ex-ante, agents agree that whoever ends up with high income transfers to whoever ends up with low income.
Stochastic market formation and fragmentation 15:03
Townsend illustrates how a network can generate random market participation through a host-and-invitation process, where one trader is randomly chosen as host and sends invitations to others, with adjacent nodes receiving them with probability q and more distant nodes with diminishing probability like q squared, reflecting something like technological latency or distance. He clarifies that the specific mechanism does not matter for the theory, only that participation ends up random. More generally, agents can be divided into fragmented clusters rather than simply in or out of a single centralized market, with a probability distribution over which partition of agents forms, and within each cluster agents are treated as fully interconnected for risk-sharing purposes, unlike the bilateral debt-based connections shown in the previous lecture. He notes that international markets today are becoming more fragmented due to trade embargoes and financial sequestering.
Formalizing liquidity value and its policy meaning 27:02
Liquidity value is formalized as the effect on the community objective function of marginally increasing one agent's income by an infinitesimal epsilon before market and income shocks are realized, akin to giving that agent a bit more of a liquid asset priced one-to-one with consumption goods. Townsend stresses this is not how the Federal Reserve actually operates, since the Fed intervenes through repo lending and borrowing facilities rather than targeted infinitesimal injections, but the exercise asks what a proactive monetary authority could do if it identified key bridging players in advance, given how fast financial shocks like a bank failure can unfold. He then extends this to a finite total amount of liquidity A that could be injected across multiple agents to maximize the social value function, showing there exist critical thresholds where, below a certain amount, all liquidity should go to just one most-valuable person before spreading to a second.
Shadow prices and the liquidity value formula 36:31
Using a Lagrangian approach, Townsend defines qs as the shadow price of the resource constraint in state s, reflecting how much the objective function would rise if that constraint were slightly relaxed, comparable to marginal utility pricing in a decentralized competitive equilibrium. The resulting proposition states that agent i's liquidity value equals the expected product of i's participation shock and these shadow prices, meaning the value is zero whenever the agent is out of the market and equal to the shadow price of the market they are in whenever they participate. He closes the section with a worked example where all agents share equal weights, a common utility function, and independent incomes with the same mean and variance, showing that the optimal risk-sharing solution simply gives each participating agent the average income of everyone currently in the market.
Prudence and the value of liquidity 40:01
Financial centrality, the expected value of giving liquidity to a trader, can be broken into pieces: the mean income in the economy, the marginal utility at that mean, and a term involving the third derivative of utility, called prudence. Prudence means that the more cautious a person is about future risk, the more they want to save, and this applies to common utility functions like power functions. Higher variance in income raises the value of a liquidity injection because there is more risk to pool, while adding more agents to the market has diminishing returns, since the pool is already close to the average.
Segmented markets and who deserves liquidity 43:02
In a segmented market, the same kind of formula applies, except you divide by the number of agents in the specific cluster where a person happens to be, and you take expectations over all possible clusters, since this is entirely a before-the-fact, or ex-ante, measure. Because market participation is a yes-or-no event, the formula reduces to the case where the person actually shows up, weighted by the probability they show up. The intuition is blunt: there is little point giving liquidity to someone who rarely appears in the market, so people who are reliably present are valued more.
Ex-ante commitment and injection timing 44:01
A student pressed on why an ex-ante measure matters if consumption and injections can respond to realized outcomes anyway. Robert Townsend clarified that the point of injections and consumption rules is to hedge both income risk and the risk of not being able to participate in the market, and that the location of the injection has to be fixed ex-ante, like a buffer stock set up before anything happens, rather than depending on which cluster ends up forming. The relevant shadow price used to decide who gets liquidity is an expected shadow price, conditioned on the draws of income and participation shocks, not the price after the fact.
Neighborhoods and network position 46:31
Extending the host model, an agent's neighborhood is the set of nodes directly connected to them, and centrality depends on the size of that neighborhood but also, somewhat counterintuitively, on the size of the neighborhoods of the other traders connected to the host. Comparing two network pictures, an agent whose neighbors are only lightly connected elsewhere turns out to be more valuable for liquidity purposes than one whose neighbors already have many other connections, because incoming players who bring their own trading partners add less marginal value once the market is already thick. Valued players are those present when the market is thin, when incomes are low, when risk is high, or when income shocks are positively correlated, since correlated shocks create aggregate risk that a large but IID market could otherwise diversify away.
Asset pricing and Nash bargaining interpretations 55:05
The liquidity value measure has two positive-theory counterparts. One treats it as the price of a personalized state-contingent bond that pays off only when a given agent is present in the market, following standard Arrow security pricing logic, though this requires participation to be exogenous rather than something the agent can choose. The other reframes it through Nash bargaining, where each agent's threat point is autarky, and the resulting risk-sharing formula for constant absolute risk aversion preferences gives each agent the mean income plus an intercept tied to their Pareto weight, which resembles a fixed effect in a risk-sharing regression.
Testing the theory on Thai village data 1:01:31
Running this regression on Thai data, the individual fixed effects line up with two measured quantities: the covariance of a person's participation with the number of other people present, and the covariance of their income variance with participation. Communities behave as if they implicitly reward steadfast moneylenders or gift givers, people who are reliably present, with higher average consumption, matching what the theory predicts even without an explicit insurance contract. Townsend notes he would like to test this on markets like New York or Switzerland but the data are confidential and it is unclear which variables would translate.
Contagion as disease versus ex-ante risk sharing 1:05:32
The book Contagion, Systemic Risk in Financial Networks treats financial crises like disease transmission, covering historical crises, bank balance sheets, and cascade mechanisms such as default contagion and fire sales, where thin markets force distressed sales that push prices down further. Its policy implication is to limit exposure of large traders ex-post so problems do not spread. Townsend contrasts this with the ex-ante approach of his own framework, which aims to keep markets thick and functioning to mitigate shocks before they happen, noting that no existing work seems to combine both approaches into a hybrid policy.
Repo markets, coordination, and regulation 1:09:35
In the US repo market, money market funds with excess liquidity lend to broker-dealers, who intermediate for hedge funds and pension funds needing short-term financing, using treasuries as collateral. Because money market funds cannot observe each other's arrangements, the market has multiple equilibria, echoing the coordination problem from the earlier lecture on circulating IOUs. Post-crisis Basel regulation now limits balance sheet size, including exposure from rehypothecation where dealers borrow and lend simultaneously, and this contributed to repo rates spiking to around 10 percent in 2023 while the Federal Reserve's policy rate was near 3 percent, prompting the Fed to step in as a liquidity-providing node through reverse repo. Townsend describes ongoing work with the US Treasury to decompose the repo network's flows into cycles and chains to identify sources and sinks of risk, while still accounting for the bilateral risk dealers take on even when it stays off their balance sheets.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
