Policy simulation: How to build an economy-wide digital twin¹
Agent-based simulation models can be an extremely useful tool for policy makers, in particular for predictive analytics, scenario analyses, and plausibility testing. Lucidminds.ai is at the forefront of making this happen.
Agent-based simulations can be used as a decision support tool for social and economic decision making — to improve data-driven, evidence-based, sensible, and prudent decision making. These simulations can be used as an exploratory tool to explore various policy scenarios in different environments, before putting actual people or the economy at risk. This provides a powerful extension to the policy makers’ existing toolbox. In the end, it may also help to create a triple win for the private sector, the public sector, and the general public at large, by providing a risk-free way of testing policies before they are actually put into effect in the real-world.
Modelling a complex & interconnected world
In our increasingly complex and interconnected world, many societal challenges are caused by human action. Examples are climate change, global financial crises, and non-sustainable economic growth. These challenges may cause serious harm to our societies. To mitigate the adverse effects that human activities have on our environment is in the general interest of society and policy makers use a variety of decision support tools to tackle such issues. However, policy makers are also facing an increasingly difficult task to regulate markets and human actions, especially due to the global interconnectedness of our activities. A policy maker’s toolbox consists, among other things, of their own common sense and policy making experience. But it also consists of consulting with experts and the use of predictive models. At Lucidminds.ai we want to help improve these policy models by using agent-based modelling (ABM), and simulation.
The quality of decisions depends on the quality of the data and the decision making models.
Evidence-based decision making
Evidence-based decision making relies on data and models. The quality of the decisions, therefore, depends on the quality of these models. Unfortunately, at the start of the global financial crisis, policy makers could not foresee the occurrence of the crisis, precisely because their models did not allow for such severe downturns in economic activity. In September of 2009, Paul Krugman wrote in the New York Times that the macroeconomics of the past 30 years was “spectacularly useless at best, and positively harmful at worst.”
At the time of the financial crisis in 2008, the models in use by all of the key financial and economic regulators — such as the IMF, the World Bank, the Bank of England, the Financial Services Authority, and the European Central Bank — did not allow for the occurrence of such severe downturns in economic activity. Because their models did not include any mechanisms to consider the network of interactions or the individual behavior of the economic actors that make up our actual economies, policy makers could not see this crisis coming. Some even said that, when the crisis did occur, they were blinded by the models at their disposal, and that those models even led them astray in their decision making.
Agent-based modelling
Agent-based computational modelling has been developed over the last 20 years mainly in academia but it has already found its way into some policy circles such as Central Banks. Agent-based models consist of a population of agents that are implemented as software objects and can interact with each other. In this way, a complex system such as a bee hive, a traffic simulation, or an entire economy can be modelled and simulated.
We at Lucidminds.ai adopt agent-based computational modelling as a way to assist policy makers to explore the space of possibilities generated by the interaction of human actors and market mechanisms. Such models take full account of the heterogeneity of individual agents and the decentralized social interactions between those agents and their environment.
Modelling the network of social interactions makes agent-based models a most suitable tool for policy making.
Networks of interactions
In an agent-based model, we construct a system in which the network of social interactions produces the aggregate phenomena. This means that the aggregate outcome arises from the agent interactions without the need to model those aggregate phenomena directly. The agent behaviour is guided by behavioural rules that are modelled at the micro level and are directly observable. The aggregate phenomena resulting from the agent interactions and other model structures are grown from the bottom up, as it were. This makes agent-based models a most suitable tool for policy making. After all, policy makers try to resolve how people behave at the micro level and how this might affect the system at the macro level. The job of the policy maker is then to select the right policy instrument and the correct policy measure to achieve the policy objective, given the anticipated micro and macro effects.
Agent-based simulation models can be used as an exploratory tool for policy making.
Exploratory policy
Agent-based simulation models can be used as an exploratory tool for policy making by testing out various policy scenarios in a virtual environment. A simulation model can be used to simulate these scenarios and a user or analyst can explore the results. This serves as a virtual test-bed to test different policies before they are actually implemented in the real-world. There are many reasons why this is beneficial to policy makers, regulators, and practitioners. For example, the implementation of the policy in the real-world might be accompanied with large costs, or there might be uncertainty about the results of a policy if it were implemented directly in the real-world. The simulation model therefore plays a role very similar to that of a flight simulator or a wind tunnel experiment in engineering. It allows a policy maker to test out, and explore, the consequences of his or her actions in a virtual, digital twin of the system under study before these actions are actually applied in the real-world. It therefore provides for a risk-free way of testing our decisions, before putting any actual people or the economy on the line.
Smart, learning agents are a cornerstone of useful policy models.
Simulation technology
Due to the massive amounts of data that computer simulations can generate, we have to take into account big data considerations. This implies the models should be designed to be deployable on high-performance computing clusters (HPC systems), or run on cloud-based computing infrastructure.
Machine learning techniques and AI help us to develop agents that are able to learn dynamically inside a simulation model. Due to this ability to learn, our agents are able to adapt flexibly to their environment. Smart, learning agents are a cornerstone of useful policy models. Even if the policy changes, these smart agents will change their behaviour accordingly. This can give important insights to policy makers about the effects of various policy scenarios.
At Lucidminds.ai we have a team of ABM experts dedicated to the design of complex and large-scale — yet realistic — agent-based simulations. We employ the power of realistic simulations to gain insights into complex socio-economic systems. Drop us an email for any inquiry regarding the use of ABMs in business cases.
¹The title refers to remarks made by Andrew Haldane, Chief Economist at the Bank of England, in a speech held on 7 May 2019 at University of Sheffield, see p. 27: https://www.bankofengland.co.uk/-/media/boe/files/speech/2019/is-all-economics-local-speech-by-andy-haldane
About the author:
Sander van der Hoog is Head of Simulation at Lucidminds.ai.
Thanks to Bülent Özel and Oguzhan Yayla
