Macroprudential capital buffers in heterogeneous banking networks: insights from an ABM with liquidity crises

ABSTRACT

We study how the effectiveness of macroprudential capital buffers conditional to the systemic-risk assessment of banks responds to the degree of heterogeneity of the financial system. A multi-agent model is employed to build an artificial economy with households, firms, and banks where occasional liquidity crises emerge. The systemic importance of banks is captured by a score-based mechanism reflecting banks' characteristics in terms of size or interconnectedness. We compare three degrees of heterogeneity in the configuration of financial networks related to different banking concentrations in the loan market. The main findings suggest that: (i) reducing the heterogeneity of the banking network stabilizes the economy by itself; (ii) the identification criteria of systemic-important institutions are affected by the heterogeneity of networks; it is preferable applying systemic capital surcharges to the largest banks under high heterogeneity and targeting those most interconnected under low heterogeneity; (iii) the effectiveness of systemic capital buffers is preserved under high heterogeneity when a common asset holding contagion channel is added. However, simple measures based on risk-weighted assets capital ratios appear to be more effective in low heterogeneous systems. Thus, we argue that prudential regulation should account for the characteristics of the banking networks and tune macroprudential tools accordingly.

KEYWORDS:

• Agent-based model
• capital requirements
• capital buffers
• financial networks
• macroprudential policy
• systemic risk

JEL CLASSIFICATIONS:

• C63
• D85
• E44
• G01
• G21

Acknowledgments

The authors would like to thank Giampaolo Gabbi and Marco Bardoscia for helpful comments and suggestions, as well as participants of the Minsky at 100 conference in Milan, the 2020 ABM4Policy seminar series, the 69th AFSE meeting, the 27th Computing in Economics and Finance International Conference, the 2021 Deutsche Bundesbank Research Seminars, the 24.5th Workshop on Economics With Heterogeneous Interacting Agents. Financial support by the Hans-Böckler Foundation is gratefully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Gallup Poll Social Series Work, https://news.gallup.com/poll/266807/percentage-americans-owns-stock.aspx.

2 The model is stock-flow consistent, which means that by the aggregate balance sheet identity the negative net worth of the government is balanced by the positive net worth of the private sectors so that the aggregate net worth is zero. $\sum _{i\in N^H}n{w}_{i,t}^H+\sum _{j\in N^F}n{w}_{j,t}^F+\sum _{h\in N^B}n{w}_{h,t}^B+n{w}_t^G=0.$

3 Although the inclusion of further elements in banks' balance sheets, such as derivatives to hedge against risk, would increase the realism of the model, we choose to include the least number of elements to limit the overall complexity of the agent-based model.

4 The price of consumption goods is set by firms via a mark-up on the unitary cost of output, which includes the labor and credit cost. As the wage rate and the mark-up rule are equal across all firms, the cost of credit affects the chances of firms to sell the production in the competitive goods market. Thus, high-leveraged firms pay a greater rate on loans. Their final goods are comparatively more expansive and are subject to greater losses than those less leveraged. Therefore, the assessment of firms' default probability is simply expressed as a function of the leverage rate.

5 Despite a deposit guarantee scheme for firms and households is not implemented, by bankruptcy law (see “Recovery rates” in Section 2.3) depositors are the most guaranteed creditors.

6 To sell loans, banks first determine their liquidity need, then compute the fair value of their portfolio loan by loan. Next they determine $\mathrm{\Delta }q$ taking into account Equation (18(18) $p_{\tau }^i=p_{\tau -1}^i\left(1-\frac{\mathrm{\Delta }{q}_{b,\tau }^i}{q_t^i}\frac{1}{{ϵ}^i}\right),i=\{bills,loans\}$(18) ). Lastly, they choose which loans should be liquidated to reach their objective. The loans for sale are evaluated at their fair market value by discounting cash flows: $L_{bj}^{fv}=\frac{L_{bj}(1+M{r}^f)(1-{\rho }_j^f)}{(1+\underset{\_}{r}{)}^M}$where $L_{b,j}$ is the book value of the loan of bank b to firm j, M is the residual maturity, $r^f$ is the interest rate on the loan, ${\rho }^f$ is the default probability of firm j, and $\underset{\_}{r}$ is the risk-free rate. A similar but simpler process is put in place to sell bills. Since the maturity of bills is one period and default probability is zero, fair corresponds to the book value of bills.

7 We refer to the capital buffer for other (domestic) systemically important institutions (O-SII) absent any cross-jurisdictional activity of banks in our framework.

8 ‘The systemic risk buffer (SyRB) aims to address systemic risks of a long-term, non-cyclical nature that are not covered by the Capital Requirements Regulation’. (ESRB, https://www.esrb.europa.eu/national_policy/systemic/html/index.en.html). European financial authorities are free to define the SyRB as long as it does not interfere with any other capital requirements. This translates into different scopes and many ways to define the SyRB.

9 The similarity between the portfolios of the pair $(b,k)$ is $Jac{c}_{b,k}=\frac{\sum _{s=1}^{S}min({\mathrm{\Psi }}_{b,s},Ps{i}_{k,s})}{\sum _{s=1}^{S}max({\mathrm{\Psi }}_{b,s},Ps{i}_{k,s})}$, where S = 4 is the total number of assets and ${\mathrm{\Psi }}_{b,s}$ is the share of asset s in b's portfolio, such that $\sum _{s=1}^S{\mathrm{\Psi }}_{b,s}=1$. $Jacc\in [0,1]$, where the maximum similarity is achieved at 1 and the minimum at 0.

10 Since all median values are zero, the boxplot would not be informative.

11 An exception holds for losses on firms' loans to equity, especially under high and mid-levels of heterogeneity. The result, which is not necessarily true in reality, is a consequence of model design and due to increase leverage of firms, as explained above commenting Figure 16.

12 Data on bank concentration are available from Bankscope, Bureau van Dijk (BvD).

Additional information

Notes on contributors

Andrea Gurgone

Andrea Gurgone is a researcher at the Fondazione Eni Enrico Mattei (FEEM) and was a post-doctoral fellow at the Bamberg University from 2019 to 2021. He was awarded a Ph.D. in Economics from the Catholic University of Milan in 2017.

Giulia Iori

Giulia Iori received a Ph.D. in Theoretical Physics from the Sapienza University of Rome in 1993. She is Professor in Economics at the Department of Economics at City, University of London since 2005.