Marc Hinterschweiger, Bank of England

​

Disclaimer: The views expressed here are those of the presenter and do not necessarily reflect those of the Bank of England or its policy committees.

Agent-based models for policy –

past, present, future

​

MacroABM 2nd Workshop

Vienna, 30 April 2025

Bank of England’s ABM journey

2

Experiences from three projects

3

2014

2019

2024

Experiences from three projects

4

2014

2019

2024

How did the journey start?

5

Developing a workhorse model for housing tools

Why ABM?

• Heterogeneity
• Interactions
• Non-linear dynamics
• Policies, involving threshold effects and/or targeting a certain segment of the market
• Distributional impacts of policies

Collaboration with Oxford/INET

Housing ABM: Timeline

6

2014

Gaps identified: study effects of housing tools on the mortgage and housing markets

​

Cross-Bank collaboration with Oxford/INET

2021

Agent-based model of the UK housing market

7

10,000 households

Heterogenous in age, income and savings

​

Agent-based model of the UK housing market

8

Owner-occupiers

BTL investors

Renters

10,000 households

Caps on LTI and LTV ratios,

and affordability tests

Central Bank

Agent-based model of the UK housing market

9

10,000 households

A mortgage lender

Central Bank

The model does not allow to assess the impact of policies on consumption, and the resilience of borrowers and lenders. As a result, the impact of policies on several risk indicators are used as proxy.

Caps on LTI and LTV ratios,

and affordability tests

Validation

10

the key housing and mortgage market statistics, i.e. house prices, mortgage approvals

Hypothetical experiments

11

The Central Bank imposes LTV and LTI limits for owner-occupier mortgages:

1. a hard LTV cap at 85 percent; or
2. a soft LTI cap – 15% above the limit of 3.35.

Hypothetical experiments

12

The Central Bank imposes LTV and LTI limits for owner-occupier mortgages:

1. a hard LTV cap at 85 percent; or
2. a soft LTI cap – 15% above the limit of 3.35.

Chosen to have the same level of bindingness as the LTV cap

Binding scenarios, i.e. 20% of borrowers are affected.

Experiments affect households differently

13

BTL investors benefit

FTBs are more negatively impacted than home-movers

FTBs are more constrained

Attractive investment opportunities & mortgage availability

Policies can spill over into the rental sector

14

BTL investors benefit

FTBs are more negatively impacted than home-movers

Demand for rental

Supply for rental

Policies can spill over into the rental sector

15

Demand for rental

Supply for rental

LTV cap

RENTS

Policies can spill over into the rental sector

16

Demand for rental

Supply for rental

LTI cap

RENTS

An LTI limit can also reduce LTV ratios

17

An LTI limit can also reduce LTV ratios

18

Average LTV ratio decreases from 68% to 65%.

Average LTV ratio decreases from 63% to 61%.

An LTI limit can also reduce high-LTV lending

19

An LTI limit can also reduce high-LTV lending

20

An LTI limit can also reduce high-LTV lending

21

Housing ABM: Other central banks

22

Experiences from three projects

23

2014

2019

2024

24

Building a macro ABM with a housing market

​

Macro ABM model: Popoyan et al. (2017) “Taming macroeconomic instability: Monetary and macro-prudential policy interactions in an agent-based model”

​

​

​

​

Housing ABM model: Carro et al. (2022) “Heterogeneous effects and spillovers of macroprudential policy in an agent-based model of the UK housing market”

​

25

• Main idea

26

Housing Model

Consumption

Income

Private Bank

Central Bank

Households

House Sale Market

House Rental Market

• Main idea

27

Housing Model

Macro Model

Consumption

Income

Private Bank

Central Bank

Households

House Sale Market

House Rental Market

• Main idea

28

Housing Model

Macro Model

Consumption

Income

Private Banks

Central Bank

Households

House Sale Market

House Rental Market

Overview: Agents and Markets

29

Bids & Offers

Bids & Offers

Offers

Bids

• Overview: Agents and Markets

30

Wages

Consumption Goods

Labour

Price

Bids & Offers

Bids & Offers

Offers

Bids

Overview: Agents and Markets

• Overview: Agents and Markets

31

Wages

Consumption Goods

Corporate

Credit Lines

Labour

Price

Mortgages

Bids & Offers

Bids & Offers

Offers

Bids

Deposits

Deposits

Overview: Agents and Markets

32

Wages

Consumption Goods

Corporate

Credit Lines

Labour

Price

Mortgages

Bids & Offers

Bids & Offers

Offers

Bids

Monetary

Policy

Prudential

Regulation

Reserves

& CB Loans

Deposits

Deposits

Overview: Agents and Markets

• Overview: Agents and Markets

33

Wages

Consumption Goods

Corporate

Credit Lines

Labour

Price

Mortgages

Bids & Offers

Bids & Offers

Offers

Bids

Monetary

Policy

Prudential

Regulation

Sales Tax

Reserves

& CB Loans

Deposits

Deposits

Bonds

Bond

Interest

Overview: Agents and Markets

• Housing Variables

34

Mean values

LTV

LTI

House-price-to-income

• Housing Variables

35

Housing tenure

Distribution of buy-to-let properties

Summary:

• Most averages are very well fit
• Some distributions are well captured, others deviate non-negligibly
• For most housing variables, fit is worse than in the standalone housing model

• Experiments

36

1. Increase in capital requirements affecting all lending
• Capital buffer increased 8% 🡪 15%
• Minimum capital requirement kept at 10%
• Thus, total capital requirement increases 18% 🡪 25%
2. Soft loan-to-income (LTI) limit affecting only owner-occupier morgages
• LTI limit set to 3, allowing for 10% of mortgages to go over it
3. Both regulations together

Capital stack

UK

Model

Demand for credit met in full subject to affordability tests

Credit is rationed

Bank is resolved

Hypothetical experiments

Experiments and findings

37

1. Tightening capital requirements leads to a sharp decrease in commercial (22%) and mortgage lending (28% for home-movers and 32% for BTL investors), and housing transactions (27%);
2. LTI policy leads to a sharp decrease in house prices (13%), and it spills into the BTL sector and also affects rental market;
3. LTI policy has distributional effects in the housing market; capital policy does not (figure above);
4. Both policies indicate a resource shift from housing to consumption and have a positive impact on real GDP (1% increase) and unemployment (0.5 pp decrease), while there is no material impact on inflation and the real interest rate.

For more, please see…

38

Experiences from three projects

39

2014

2019

2024

New features

• Detailed banking system
• Deposit insurance scheme (i.e., risk shifting from shareholders to depositors)
• Bank runs
• Contagion mechanism
• Equity market
• Resolution costs

​

• Physical capital

​

40

A forthcoming book chapter: The recent developments in the use of agent-based modelling at central banks

by Andras Borsos (Central Bank of Hungary), Adrian Carro (Banco de Espana),

Aldo Glielmo (Banca d’Italia), Jagoda Kaszowska-Mojsa (National Bank of Poland), Arzu Uluc

and Marc Hinterschweiger (Bank of England)

41

ABMs by central banks and related institutions

• Topics range from monetary policy, inflation dynamics, forecasting, credit cycles, housing market, financial markets, systemic risks (i.e., contagion channels), climate change to payment systems and CBDC.

​

• Why has there been an increasing interest in ABMs at policy institutions?
• Demand: Expansions in central banks’ remits and responsibilities; and emerging risks / new challenges.
• Supply: Methodological advancements; increased data availability and computational capacities.

42

ABMs have been published by 24 central banks and 7 other related policy institutions such as IMF, World Bank, BCBS.

ABMs by central banks and related institutions

43

Learnings from our experience over a decade

Obstacles to / issues with the more widespread use of ABMs in policy:

• Technical/modelling challenges
• Lack of coding skills
• IT systems
• Data limitations (need for micro data; privacy regulations…)
• Complexity

​

• Communication challenges
• with policymakers: (i) black box; (ii) “Could you add one more thing…?”
• with (mainstream) economists, i.e., unfamiliar with this approach / sceptical

​

​

44

Learnings from our experience over a decade

• Building a community

​

• Engaging in institutional capacity building
• Invest in building “cultural” acceptance (of ABMs and models in general)
• Get and preserve a slice of the resources pie
• Bring together people with the right skills / expertise
• Long-term projects – avoid key person risk, ensure effective data and storage management

​

• Working toward a (Kuhnian) paradigm shift?

​

​

​

​

45

ABM4Policy Group

• The goal of the ABM4Policy community is to exchange ideas and discuss open issues and challenges of using agent-based models to address policy relevant questions. Since 2019, ABM4Policy has been organizing research seminars and workshops with prominent scholars in the field.  

​

• Next workshop at ECB, Frankfurt, on 26-27 June: website.

​

• The organizing committee of ABM4Policy is: Eugen Tereanu, Radu Popa (ECB), Marco Bardoscia, Marc Hinterschweiger, Arzu Uluc (Bank of England), Adrian Carro (Bank of Spain), Mauro Napoletano (Université Côte d'Azur and OFCE), Lilit Popoyan (Queen Mary University of London), Andrea Roventini (Scuola Superiore Sant’Anna), Marco Gross (IMF).

​

46

Conclusion

47

Agent-based models have the potential to augment the toolkit available for policy-making at central banks.

​

​

How to overcome the barriers?