RDP 2022-01: MARTIN Gets a Bank Account: Adding a Banking Sector to the RBA's Macroeconometric Model Appendix A: Literature Review

The motivation for this project is clear. We know that feedback between the banking sector and real economy can be an important source of shock amplification, so we want a modelling framework that incorporates at least some of these financial accelerator channels. Unfortunately, the best way to incorporate these channels is not obvious.

While the existence of financial accelerator mechanisms was known before the global financial crisis (Kiyotaki and Moore 1997; Bernanke et al 1999), it was this crisis that highlighted the failure of modern macroeconomics to fully appreciate their size and likelihood of occurring (Lindé et al 2016; Gertler and Gilchrist 2018). While progress has been made, the macro literature has advanced in a myriad of different directions; from adding linear mechanisms (Gertler and Kiyotaki 2010; Christiano, Motto and Rostagno 2014), to applying numerical solution methods to small macrofinancial models with simple nonlinearities (He and Krishnamurthy 2013; Brunnermeier and Sannikov 2014; Adrian and Duarte 2016), and using atheoretical statistical techniques to find highly nonlinear relationships between macro and financial variables (Adrian, Boyarchenko and Giannone 2019a, 2019b; Hartigan and Wright 2021). Moreover, very few of these strands of literature take financial intermediation seriously; common approaches either explicitly ignore banks, do not model them realistically, or implicitly assume their influence is captured in the reduced-form parameters of the model (see Jakab and Kumhof (2018) for a discussion). Unfortunately, a unified theory of macroeconomics and finance still eludes us.

This gap between theory and reality also exists within the central bank modelling universe. Macroeconometric models at central banks either do not include a financial sector, or include only linear financial accelerator mechanisms (Adrian 2020; Muellbauer 2020). At the same time, areas in charge of financial stability have constructed complex nonlinear stress testing frameworks (Burrows, Learmonth and McKeown 2012; Adrian, Morsink and Schumacher 2020; Correia et al 2020). When discussing the relationship between the Bank of England's macro modelling framework and their financial stability modelling, Hendry and Muellbauer (2018, p 311) state that:

It seems as though the linkage is almost entirely one way, from the ‘real economy’ to finance. The contradictory lesson from the global financial crisis apparently remains to be learned …

The most well-known of these central bank macroeconometric models, FRB/US, developed by the Federal Reserve Board (Brayton, Laubach and Reifschneider 2014) abstracts from both financial accelerator mechanisms and the financial sector. This abstraction is also a feature of the key semi-structural policy models of the Bank of Canada, LENS (Gervais and Gosselin 2014); European Central Bank, ECB-BASE (Angelini et al 2019); the Bank of Japan, Q-JEM (Hirakata et al 2019); and the Reserve Bank of New Zealand, NZSIM (Austin and Reid 2017).

The semi-structural model developed by the Norges Bank, the Small Macro Model (SMM) (Hammersland and Træe 2014), incorporates a linear financial accelerator mechanism in which lower asset prices reduce collateral values, thereby restricting investment and amplifying the asset price fall (akin to the Kiyotaki and Moore (1997) and Bernanke et al (1999) mechanisms). The Central Bank of Chile's model, MSEP (Arroyo Marioli, Becerra and Solorza 2021), incorporates a linear financial accelerator in which credit growth and risk directly affect GDP growth, while GDP growth directly affects credit growth and risk. Importantly, neither of these models explicitly incorporate a banking sector nor stress in the financial system.

De Nederlandsche Bank's semi-structural model, DELFI 2.0 (Berben, Kearney and Vermeulen 2018), includes a comprehensive banking sector that has many similarities with our framework. However, they differ from our approach by modelling the banking sector using a linear error correction framework (including for loan losses and capital shortfalls), and by estimating the relationships rather than calibrating from existing models. The Bank of Italy's Quarterly Model, BIQM (Miani et al 2012), includes a banking sector that, like our framework, feeds back into the real economy via lending rates. But like DELFI 2.0, BIQM uses a linear framework with estimated relationships.

Some central bank stress testing frameworks include an amplifying feedback to the real economy. The Bank of Japan uses an estimated medium-scale model with a macroeconomic sector, and a financial sector split into individual institutions; they show that the inclusion of feedback mechanisms more than doubles the effect of a shock to GDP (Kitamura et al 2014). The European Central Bank uses several models to incorporate amplifying mechanisms within their stress testing framework (Henry and Kok 2013). While the financial sector amplification mechanisms these frameworks incorporate are similar to our proposal, their focus on stress testing means they have a more detailed financial sector than our proposal but a less detailed macroeconomic framework. Therefore, as far as we can tell, what we are attempting has not been done before.