The models used by policymakers before the 2008 crisis and in use among various policy circles is Dynamic Stochastic General Equilibrium models or DSGE in short. These models are used by policymakers around the world to make conditional forecasts but the success rate of these models have been abysmal. These models could not predict recessions or fat tail events, but due to lack of alternative options it has not been completely replaced. It is only recently that Janet Yellen has emphasized that models with heterogenous agents are required to guide the economics profession1.
The DSGE model accommodates all markets in the economy over time along with shocks to the economy based on probability distributions. The agents used in these models are homogenous in every respect. The idea was that macroeconomics can be explained using the microeconomic foundations. The model represents a top down approach by assuming identical agent behaviors.
Our economy now behaves like a complex system and the top down approach is not able to explain real life events. While the DSGE models are being used less , the reductionist thinking that it relies on still permeates.
Another approach called agent based modelling simulates the behavior of each agent in the system to show how simple patterns can lead to the complex patterns we observe in real life.
This approach helps us clearly observe emergent behavior when heterogenous agents interact in the economy. The invisible hand operating becomes visible and its workings can be examined. These models have also been able to explain behaviors of Clustered Volatility and Fat Tails exhibited by Asset markets.
The bank of England designed an agent based model to capture the dynamics of corporate bond trading by open ended mutual funds. The model looked at how investors redeeming corporate bonds for mutual funds can lead to feedback loops.
Figure 1 : Capturing non-linear relationships: the feedback loop following a shock to fund’s expected loss rate
It showed that if interest rates were to rise , existing corporate bonds may become less attractive leading to redemptions. Investors selling first might get a good price , but fall in liquidity would cause bond prices to fall. The price fall in bonds would lead to more redemptions and so more bonds would be sold causing a further decline in their prices. This feedback loop of redemptions is similar to that of bank runs.
Figure 2 : Reproducing stylized facts: the distribution of daily log-price returns
The model endogenously produced the distribution of daily log price returns. The overall fit is a match to the data. It is time economists and policymakers make agent based models a tool to analyze nonlinear relationships in the global economy.