JFJ. Doyne Farmer

J. Doyne Farmer

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


J. Doyne Farmer has a total of 108 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Liquidity Spirals

    by Garbrand Wiersema, Esti Kemp, J. Doyne Farmer

    The financial crisis of 2007-2008 highlighted the risks that liquidity spirals pose to financial stability. We introduce a novel method for studying liquidity spirals and use this method to identify spirals before stock prices plummet and funding markets lock up. We show that liquidity spirals may be underestimated or completely overlooked when interactions between different types of contagion channels or institutions are ignored. We also find that financial stability is greatly affected by how institutions choose to respond to liquidity shocks, with some strategies yielding a “robust-yet-fragile" system. To demonstrate the method, we apply it to a highly granular data set on the South African banking sector and investment fund sector. We find that the risk of a liquidity spiral emerging increases when the pool of institutions' most liquid assets is reduced, while a liquidity injection by the central bank can dampen the spiral. We further show that a liquidity spiral may be due to the banking and fund sectors' collective dynamics, but can also be driven by an individual sector under some market conditions. The approach developed here canbe used to formulate interventions that specifically target the sector(s) causing the liquidity spiral.

  • Open Access

    Learning to Manage Investment Portfolios beyond Simple Utility Functions

    by Maarten Peter Scholl, Mahmoud Mahfouz, Anisoara Calinescu, J. Doyne Farmer


  • Open Access

    Quantitative agent-based models: a promising alternative for macroeconomics

    by J Doyne Farmer

    Agent-based models (ABMs) are dynamic computer simulations that abandon utility maximization and instead assume that agents are boundedly rational and make decisions using heuristics, myopic reasoning, and/or learning algorithms. Because ABMs do not need to compute optima they are more tractable, allowing a higher level of realism. Recent research has developed quantitative agent-based models that make time series predictions, modelling a specific economy at a specific point in time; some of these address questions that mainstream models cannot even ask, and some make predictions that are superior or equal to their mainstream equivalents. After explaining what ABMs are and how they are built in more detail, I review four examples of models from my own work for leverage cycles, the 2008 housing bubble, Covid, and a general-purpose micro-macro model. I conclude by discussing the advantages and disadvantages of agent-based models in comparison to standard models.


  • Open Access

    Measuring productivity dispersion: a parametric approach using the Lévy alpha-stable distribution

    by Jangho Yang, Torsten Heinrich, Julian Winkler, François Lafond, Pantelis Koutroumpis, J. Doyne Farmer

    It is well known that value added (VA) per worker is extremely heterogeneous among firms, but relatively little has been done to characterize this heterogeneity more precisely. Here, we show that the distribution of VA per worker exhibits heavy tails, a very large support, and consistently features a proportion of negative values, which prevents log transformation. We propose to model the distribution of VA per worker using the four-parameter Lévy stable distribution, a natural candidate deriving from the generalized central limit theorem, and we show that it is a better fit than key alternatives. Fitting a distribution allows us to capture dispersion through the tail exponent and scale parameters separately. We show that these parametric measures of dispersion can be useful to characterize the evolution of dispersion in recent years.


  • Open Access

    Discounting the Distant Future: What Do Historical Bond Prices Imply about the Long-Term Discount Rate?

    by J. Doyne Farmer, John Geanakoplos, Matteo G. Richiardi, Miquel Montero, Josep Perelló, Jaume Masoliver

    We present a thorough empirical study on real interest rates by also including risk aversion through the introduction of the market price of risk. From the viewpoint of complex systems science and its multidisciplinary approach, we use the theory of bond pricing to study the long-term discount rate to estimate the rate when taking historical US and UK data, and to further contribute to the discussion about the urgency of climate action in the context of environmental economics and stochastic methods. Century-long historical records of 3-month bonds, 10-year bonds, and inflation allow us to estimate real interest rates for the UK and the US. Real interest rates are negative about a third of the time and the real yield curves are inverted more than a third of the time, sometimes by substantial amounts. This rules out most of the standard bond-pricing models, which are designed for nominal rates that are assumed to be positive. We, therefore, use the Ornstein–Uhlenbeck model, which allows negative rates and gives a good match to inversions of the yield curve. We derive the discount function using the method of Fourier transforms and fit it to the historical data. The estimated long-term discount rate is 1.7% for the UK and 2.2% for the US. The value of 1.4% used by Stern is less than a standard deviation from our estimated long-run return rate for the UK, and less than two standard deviations of the estimated value for the US. All of this once more reinforces the need for immediate and substantial spending to combat climate change.


  • Open Access

    An Agent-Based Model of the Housing Market Bubble in Metropolitan Washington D.C.

    by Robert Axtell, J. Doyne Farmer, John D Geanakoplos, Peter Howitt, Ernesto Carrella, Benjamin Conlee, Jon Goldstein, Matthew Hendrey, Philip Kalikman, David Masad, et al.

    Several independent datasets concerning household behavior in the Washington, D.C. metropolitan area are combined to create an agent-based model of the recent housing market bubble and its aftermath. Comprehensive data on housing stock attributes, primarily from local government sources, are used as input to the model, as are administratively-complete data on household characteristics. Data covering all real estate transactions over the period 1997-2009 are used as targets for model output, including inventory levels and average days-on-market in addition to price statistics. Finally, a very large sample of mortgage service data (80-90 percent coverage of metro D.C.) serve as both input to the model (e.g., mortgage types, interest rates obtained) as well as target output (e.g., refinancing rates, foreclosure levels). The model consists of a large number of heterogeneous households who make rent or buy decisions, are matched to homes and commonly seek mortgages with which to purchase homes. These households have homogeneous rules of behavior but heterogeneous realized behavior since decisions depend on local household characteristics (e.g., size, composition, financials). This is a so-called agent-based computational model since each household and the banks originating mortgages are software agents while each home and each mortgage are soft- ware objects. The model is capable of running at full-scale with the metropolitan D.C. housing market, over 2 million households. Overall, we find that certain empirically- grounded household decision rules are capable of generating a home price bubble much like what was observed during this time period. The model does not get the absolute bubble level and the timing of its bursting exactly right but does a good job on certain market ’internals’ such as real estate sales, inventories and market tightness. Not in the model at present are fine-grained aspect of household decision-making (e.g., moving in advance of schools starting) and thus the model lacks certain well-known temporal phenomena like seasonality. Also, while the home purchase market is deeply represented in the model, few details of the rental market are present, a further weakness. These limitations and parameter sensitivities are described. Despite these flaws, we use the calibrated model to perform a few policy experiments. Our preliminary findings are that tighter interest rate policies would have done little to attenuate the price bubble, while limiting household leverage would have had a larger effect.


  • Open Access

    Gradient-Assisted Calibration for Financial Agent-Based Models

    by Joel Dyer, Arnau Quera-Bofarull, Ayush Chopra, J. Doyne Farmer, Anisoara Calinescu, Michael Wooldridge

    Agent-based modelling (ABMing) is a promising approach to modelling and reasoning about complex systems such as financial markets. However, the application of ABMs in practice is often impeded by the models’ complexity and the ensuing difficulty of performing parameter inference and optimisation tasks. This in turn has motivated efforts directed towards the construction of differentiable ABMs, enabled by recently developed effective auto-differentiation frameworks, as a strategy for addressing these challenges. In this paper, we discuss and present experiments that demonstrate how differentiable programming may be used to implement and calibrate heterogeneous ABMs in finance. We begin by considering in more detail the difficulties inherent in constructing gradients for discrete ABMs. Secondly, we illustrate solutions to these difficulties, by using a discrete agent-based market simulation model as a case study. Finally, we show through numerical experiments how our differentiable implementation of this discrete ABM enables the use of powerful tools from probabilistic machine learning and conditional generative modelling to perform robust parameter inferences and uncertainty quantification, in a simulation-efficient manner.


  • Open Access

    BlackBIRDS: Black-Box Inference foR Differentiable Simulators

    by Arnau Quera-Bofarull, Joel Dyer, Anisoara Calinescu, J. Doyne Farmer, Michael Wooldridge

    BlackBIRDS is a Python package consisting of generically applicable, black-box inference methods for differentiable simulation models. It facilitates both (a) the differentiable implementation of simulation models by providing a common object-oriented framework for their implementation in PyTorch (Paszke et al., 2019), and (b) the use of a variety of gradient-assisted inference procedures for these simulation models, allowing researchers to easily exploit the differentiable nature of their simulator in parameter estimation tasks. The package consists of both Bayesian and non-Bayesian inference methods, and relies on well-supported software libraries (e.g., normflows, Stimper et al., 2023) to provide this broad functionality.


  • Open Access

    Forecasting the propagation of pandemic shocks with a dynamic input-output model

    by Anton Pichler, Marco Pangallo, R. Maria del Rio-Chanona, François Lafond, J. Doyne Farmer