Agent-based modeling and simulation is characterized by emergent complex behaviors based on the interactions of simple autonomous agents. We have been conducting research on agent-based models of financial systems, especially as a method of assessing regulatory approaches and policies.
The bond market agent-based model implements a somewhat stylized investor ecology, with participants trading a limited universe of bonds through dealers who provide transaction immediacy on a principal basis using a request-for-quote (RFQ) protocol. In selecting the set of agents, we aim to model representative corporate bond investor heterogeneity. While there are multiple ways to segment the investor base, we guided our selection of buy-side agents by the nature of their liabilities (leveraged versus non-leveraged, presence of inflows/outflows) and their investment mandate (passive versus active, long only versus long/short). As as result, the market model includes three buy-side agent classes representative of a mutual fund, an insurance company and a hedge fund.
The mutual fund acts as a real money investor who aims to replicate the performance of a defined benchmark which includes the full universe of bonds available in the initial model. Any dividends or capital gains distributions are assumed to be re-invested. As noted, the fund is long only and does not leverage positions. The fund also maintains a dynamic cash balance as a buffer against investor redemptions (limiting forced sales) and to minimize transaction costs by parking cash until sizable orders can be made.
The insurance company agent implements a long-term value investor with a liability drive investment strategy. This agent manages an investment portfolio across equity and fixed income markets and changes allocation between markets depending on overall market conditions. The insurance company is a "long only" investor with additional constraints limiting the concentration of risk in any specific bond. In the initial model, leveraged positions cannot be established (another real money investor) and we further assume there are no external inflows or outflows in the form of premiums or claims.
Trading activity for the insurance company results from changes in portfolio allocation between equity and fixed income markets. Macro allocation decisions are driven by a number of variables, including equity market volatility as well as the current level and slope of the yield curve. Time series of equity volatility measures such as the VIX are loaded into the model, so different time periods can be used to anchor the simulations in realistic business cycles.
The hedge fund agent acts as a short-term tactical trader who follows a relative value trading strategy. As such the hedge fund maintains both long and short positions and makes active use of leverage. In the real world, fixed income relative value hedge funds have historically been among the most leveraged market participants.
The hedge fund agent is not subject to external inflows (basically a closed end fund) or redemptions (assume investor lock up); its trading capacity is constrained only by the availability of secured financing (leverage) from broker-dealer agents. We assume the hedge fund finances all positions on margin through prime-brokerage style arrangements with some of the dealers. Broker-dealers limit leverage using security-specific haircuts that can be dynamically adjusted depending on market conditions.
All margining is assumed to occur on an overnight basis. At the start of each trading day (a tick in the agent-based simulation), margin requirements are calculated based on current market prices and security-specific haircuts, as set by the broker-dealer agents. The difference between margin requirements and current wealth determines the trading capacity. If the new margin requirements exceed current wealth, the hedge fund is forced to liquidate positions (deleverage) to meet margin calls. Any excess wealth is free to be invested.
Dealers respond to requests for quote (RFQ) from the buy-side agents and trade with clients on a principal basis (there is no inter-dealer market in the initial model). Asset owners must trade with the dealer offering the lowest price. Dealers can maintain both long and short positions. Dealer behavior is limited through regulatory constraints and market discipline, the latter expressed through a constraint on value-at-risk (VaR) relative to capital.
Abstract: In this extended abstract, we present some next steps in our work using agent-based modeling and simulation to better understand crisis dynamics in financial markets. In particular, the focus is on the US corporate bond market, which has increased dramatically in size and exhibits some signs of systemic risk. One specific risk is related to the growing role of open-end vehicles (such as mutual funds) that can exhibit run-like behaviors as investors seek to redeem shares under conditions of stress. Our goal is investigate the potential of agent-based modeling and simulation as a tool for evaluating regulatory policies aimed at reducing market risk. Some possible regulatory policies include cash-to-asset ratios, alternate lending facilities, investor lock ups or gates, and swing pricing mechanisms. Preliminary simulation results are presented to highlight the promise of agent-based approaches for evaluating potential regulatory interventions. This work is certainly aligned with the WITS 2019 theme of ``Markets for Policy Making and Sustainability.'' As part of the demonstration track, live simulations using the agent-based model prototype will let workshop attendees participate in designing and running experiments.
Abstract: In this paper, we introduce a small-scale heterogeneous agent-based model of the US corporate bond market. The model includes a stylized ecology of investors that trade a set of bonds through dealers. Using the model, we simulate market dynamics that emerge from agent behaviors in response to basic exogenous factors (such as interest rate shocks) and the introduction of regulatory policies and constraints. A first experiment focuses on the liquidity transformation provided by mutual funds and investigates the conditions under which redemption-driven bond sales may trigger market instability. We simulate the effects of increasing mutual fund market shares in the presence of market-wide repricing of risk (in the form of a 100 basis point increase in the expected returns). The simulations highlight robust-yet-fragile aspects of the growing liquidity transformation provided by mutual funds, with an inflection point beyond which redemption-driven negative feedback loops trigger market instability.
The simulations include an economic shock (delivered at tick 50) in the form of a 100 basis point rise in the interest rate. The corresponding price drops across all five bonds are easily (see the figures below). The small but noticeable drops in prices for the next 10 to 20 ticks are due to the redemption-driven feedback loop. After the market stabilizes, more normal trading activity returns and the bonds settle at new price levels. A similar situation exists at a 25% market share with a more pronounced feedback loop and recovery.
At a 35% mutual fund market share, the interest rate hike first causes the expected price drops. However, after that the simulations take a different path. After these shock-induced drops, the prices fall of a cliff. For the next 50 clicks the redemption-driven feedback loop causes a spiral of decreasing prices. In fact the concave curve means that the price drops accelerate with correspondingly dramatic wealth destruction. The price drops are more pronounced from the outset, but reach an inflection point followed by precipitous price drops until all the dealer capacity is consumed (and the market flatlines). In these simulations, the redemption-driven price drops dwarf the initial interest rate shock effects. The feedback loop causes price drops in the 35% to 44% range, as compared to the shock-induced drops of roughly 1% to 14%.
The most recent journal paper is: Berndt, D.J.; Boogers, D.; Chakraborty, S.; McCart, J. Using Agent-Based Modeling to Assess Liquidity Mismatch in Open-End Bond Funds. Systems 2017, 5, 54. Please see the links below.
We gratefully acknowledge the support of the National Science Foundation (NSF Award 1445403) and the Office of Financial Research (OFR) in providing funding for this research.
