Keywords

1 Future Challenges

Traditional dynamic stochastic general equilibrium (DSGE) models of the macroeconomy view fluctuations of GDP around its trend as the product of aggregate shocks drawn form a lognormal distribution. The investment function of these models produces the observed autocorrelation in error terms (GDP follows an AR process) while a number of alternative mechanisms have been proposed to account for the increasing secular trend. However, there are other features of the behavior of the aggregate output trend that these models cannot explain. For example, persistence, that is the tendency for output shocks in one period to be correlated with those in the period before, is usually captured by directly imposing such pattern on the structure of macroeconomic shocks (error terms are serially correlated). It is also well known that there is an asymmetry in the behavior of aggregate output between expansions and slumps, whereby the former are smooth and the latter are sharp. Acemoglu and Scott (1991) have demonstrated how the last two features, namely persistence and asymmetry, can be accounted by allowing heterogeneity among producers and some form of increasing returns at the firm level. Lastly, as Ascari et al. (2015) have shown DSGE models cannot explain the observation that the tails of the distribution of aggregate shocks are ‘fat’; that is are inconsistent with normal draws. This last observation has recently motivated researchers to consider alternative approaches, such as network theory (e.g. Acemoglu et al. 2012, 2017; Carvalho 2014), agent-based modeling (e.g. Ashraf et al. 2017) and also frameworks that are combine these two approaches (e.g. Battiston et al. 2007). In all these papers the amplification mechanism is produced within the production sector. The financial sector either is non-existent or when it exists, as in Ashraf et al. (2017), is not responsible for the excess volatility in the system.

In Chap. 20 of this volume, we have suggested an alternative mechanism according to which the excess volatility is created by network effects in the banking system. Support for our approach is offered by Schularick and Taylor (2012) who have empirically identified a historical link between the level aggregate credit in the economy and macroeconomic performance. The authors argue that aggregate credit can predict economic crises, especially, rare catastrophic events, like the Great Depression and the Great Recession. These where the two more severe macroeconomic crises of the last 100 years, and both were preceded by extreme events in the banking system. For example, in the period 1930–1933 approx. 10,000 banks suspended operations or failed. The corresponding losses from the more recent crisis have been well documented.

Our central feature of our model was the presence of ‘fire sales’ (see, Shleifer and Vishny 2011). This is a price externality (see, Caballero and Simsek 2013) imposed by failing banks on other banks. As failing banks sale their assets, the prices of these assets drop causing other banks to reevaluate their assets at these new lower prices potentially leading to their own failure. Our work thus far has suggested that this mechanism is a good candidate for accounting for macroeconomic fat tails. Our aim is to firmly establish that this is indeed the case but also within a more general model that potentially can account for the other features of aggregate output described above.

Thus, looking ahead we are planning a number of extensions of our benchmark model. The first two are direct extension of the benchmark model focusing on the banking sector. The following two will be addressing issues related to the impact of the financial system on the rest of the economy.

Network structure, fire sales and systemic risk. Our first task is to try to understand how the interplay between the distribution of idiosyncratic shocks and network structure generate the amplification mechanism. We can do that by producing estimates of network measures (e.g. average degree, centrality) for each period of our model and then by checking how these measures are correlated with the corresponding growth rates of aggregate credit at the end of the period and the aggregate shock (aggregate of idiosyncratic shocks) at the beginning of the period.

Core-periphery network structure, distribution of bank size and systemic risk. We can experiment with alternative network formation specifications where banks that perform better face a higher demand for loans from firms. Actual banking networks have a core-periphery structure where banks in the core are responsible for a higher proportion of lending activity. Within such a framework we should be able to address important policy issues related to institutions that are either ‘too big’ or ‘too connected’ to fail.

The above issues have to varying degrees already been addressed in the literature. What is novel in our approach, is the introduction of fire sales in the general model. This is important because we will be able to get a measure of systemic losses. In models without fire sales, as we have already shown in Chap. 5, the total losses are equal to the initial losses and thus these models can only address issues related to the total number of institutions being affected following a shock (more likely providing a minimum estimate as fire sales would more likely generate more failures) but not the impact of such failures on economic losses.

An agent-based model of the macroeconomy with a bank network. In its present form the model captures the dynamics of aggregate credit as generated by a dynamic interbank model. We plan to extend the model by introducing consumers (depositors) and producers (firms) in the model. By taking into account the saving decisions of agents and by explicitly introducing production technologies along with externalities we hope that our model will be able to account for (a) the empirical relationship between growth patterns and cyclical fluctuations, (b) the persistence in GDP growth movements, and (c) the asymmetry between booms and busts.

Inequality and Business Cycles. Recently, there has been a lot research on understanding global inequality trends (see, Piketty 2013). One issue that has been overlooked are the cyclical patterns of inequality. Our agent-based model will naturally produce inequality variations along the economic cycle and this can help us identify economic groups that are more likely to suffer from economic downturns.

As we argued in Chap. 5 the asymmetry between booms and busts can be naturally captured by fire sales. By introducing some externality at the firm level, following Acemoglu and Scott (1991), we can also explain persistence. For example, firms that succeed in one period are more likely to succeed in the following period. The additional advantage of such specification would be that potentially can account for the observed size distribution of firms (Gabaix 2011). Within such a generalized framework we could then compare alternative sources of macroeconomic stability (financial versus technological).