JMAB 2.0: A Critical Update for the Macroeconomic Agent-based Model JMAB
This paper identifies critical shortcomings in the influential JMAB macroeconomic agent-based model, proposes specific corrections to address misspecifications and convergence issues, and introduces an updated JMAB 2.0 version that yields more plausible and stable outputs for robust policy analysis.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Economists have long struggled to build models that can predict how the economy will behave when things go wrong. For decades, the standard tools were like maps of a calm ocean, assuming that markets always find a balance and that people act with perfect logic. But when the financial crisis hit in 2008, these maps failed to show the storm. In response, a new approach emerged that treats the economy not as a single, smooth machine, but as a bustling crowd of individual people and companies, each making their own decisions based on limited information. This method, known as agent-based modeling, simulates how these individual choices interact to create the big picture of booms and busts. One of the most famous attempts to build such a model was a computer program called JMAB, which was designed to be a flexible tool for testing banking rules and understanding how credit flows through society. It was built to be a reliable benchmark, a standard against which other researchers could measure their own work.
However, a closer look at this popular program revealed that its internal gears were grinding against each other. Simon Hess, an economist at the University of Rostock, discovered that while the original model could produce some realistic-looking patterns, it was fundamentally flawed in ways that made its long-term predictions unreliable. The model was intended to be a stable laboratory for testing policies, but instead, it was prone to crashing or spiraling into unrealistic behavior. For instance, the way the model's simulated companies guessed their future sales was incompatible with how they set their prices, causing their profit margins to drift upward uncontrollably until the simulation broke. Similarly, the rules governing how banks decided who to lend money to ignored the actual cost of funding those loans, leading to credit decisions that no real bank would make. These errors meant that the model was generating results that looked like economic facts but were actually just artifacts of its own broken code.
To fix this, Hess did not just patch the holes; he rebuilt the engine, creating an updated version called JMAB 2.0. The core of the problem lay in how the simulated agents learned from their past. In the original model, companies used a simple averaging method to guess their future sales, but this method failed when sales were uneven, causing companies to consistently underestimate their success and raise their prices too high. In the new version, the companies use a smarter, rule-based approach that corrects their mistakes as they go, keeping their expectations aligned with reality. The model also overhauled how firms decide to invest in new equipment. Instead of a feedback loop that caused investment to swing wildly between zero and infinity, the new system uses a steady, step-by-step adjustment that keeps the economy stable over time.
The banking sector received the most significant overhaul, as the original rules for lending were dangerously disconnected from reality. In the old model, banks would approve loans even when the interest they earned was less than the cost of the money they borrowed, a situation that would bankrupt any real financial institution. The new model forces banks to calculate the true cost of their funds before approving a loan, ensuring they only lend when it makes financial sense. It also introduced a market where banks can lend excess cash to one another, a feature that helps them manage their liquidity more efficiently and prevents the interest rates on deposits from behaving erratically. Furthermore, the way the model handles the failure of a company or a bank was corrected to ensure that when an entity collapses, its assets are distributed fairly and its debts are cleared, rather than leaving it stuck in a loop of bankruptcy.
When the updated model was run, the results were markedly different. The chaotic volatility that plagued the original version disappeared, replaced by a more stable and plausible economic rhythm. The banking sector no longer showed implausible gaps in credit or erratic interest rates. Instead, the model produced a steady state that could be maintained over long periods, making it a much more trustworthy tool for policymakers. The study demonstrates that for these complex computer simulations to be useful, every internal mechanism must be logically consistent with the others. By correcting the misspecifications in how agents form expectations and how banks manage risk, the new version provides a solid foundation for understanding how financial regulations might play out in the real world, free from the distortions of the original code.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.