APPLICATIONS OF A DATA-DRIVEN�AGENT-BASED HOUSING MARKET MODEL�AT THE CENTRAL BANK OF HUNGARY

Microdata in Macromodels Workshop

29th April 2025, Vienna

András Borsos, Zsuzsanna Hosszú, Bence Mérő, Nikolett Vágó

1 |

OUTLINE

• Motivation
• Model description
• Datasets and generation of 1:1 mapping
• Roadmap of the development
• Applications
• Borrower-based macroprudential regulation
• Policy schemes supporting first-time home buyers

Disclaimer: The views in this presentation are those of the authors and do not necessarily reflect the views or positions of Magyar Nemzeti Bank.

2 |

OPTIMAL MACROPUDENTIAL POLICY

• Preventing the emergence of excessive financial risks in the economy
• In the context of the housing market:
• Loan-To-Value (LTV) requirement decreases loss given default
• Debt-Service-To-Income (DSTI) requirement decreases the probability of default
• The two measures also mitigate house price volatility

• Finding an optimal trade-off between stability and the costs of the regulation
• Minimizing efficiency loss in the economy
• Social, welfare and inequality consequences
• Housing stock quality, housing standards, energy efficiency

Policy objectives

3 |

MODEL DESCRIPTION

4 |

CONTRIBUTIONS TO THE LITERATURE

Geographic heterogeneity at the neighborhood level

Elaborated credit restructuring and foreclosure protocols

Construction �sector to model the �changes in the housing �stock (newly built flats, renovations)

Assessment of social and welfare loss due to the regulation

1:1 scale �mapping of all the �4 million Hungarian households using empirical data

Detailed credit market �with procyclical banking system

A high-resolution �housing market model, �which is suitable for complex macroprudential policy evaluation with interactions toward fiscal and monetary policies.

5 |

THE MODEL IN A NUTSHELL

6 |

THE MODEL IN A NUTSHELL 1/6 – FLATS

• We generate the housing stock using weights, such that it mimics the empirical distributions in several dimensions as closely as possible.
• Flats have three characteristics:
• Location
• 124 actual, interpretable neighbourhoods,
• we estimated a cardinal quality value to each.
• State
• A composition of several measures
• Each month a flat’s state depreciates,
• but renovation can increase it.
• Size
• Within each location, flats are grouped into buckets:
• Flats within a specific size and state interval constitute a bucket.

7 |

THE MODEL IN A NUTSHELL 2/6 – HOUSEHOLDS

• Households’ income is determined by
• their educational level
• stochastic labor market shocks,
• and by macroeconomic processes.
• Demography:
• We implemented birth and death of the agents according to empirical data.
• Households’ demand: Each period, some households may move
• If they can achieve an increase in their consumer surplus by selling their home and moving to a new one.
• If they have sufficient financing, they can make a bid for a preferred home.
• Households’ supply
• When they inherit a flat, they can sell it

8 |

THE MODEL IN A NUTSHELL 3/6 – INVESTORS

• Representative professional investor + HHs
• Demand is influenced by the obtainable capital gain, so their decisions are determined by
• price changes and
• vacancy rates and rental markups.
• Endogenous change in rental markups
• BUT the supply and demand sides can be temporarily detached 🡪 short term disequilibrium periods

9 |

THE MODEL IN A NUTSHELL 4/6 – CONSTRUCTION

• The construction sector
• is represented by a representative firm,
• which estimates demand for newly built flats heterogeneously for neighborhoods and flat categories.
• Construction process:
• The construction sector builds high quality flats.
• It needs land site to build, so it buys the flats with the lowest unit price for sale in the neighborhood where it wants to build.
• Construction takes 18 months, but the construction firm can sell the flats even before they are finished.
• Construction costs are proportional to the regional average salary (and higher than the renovation cost).

​

10 |

THE MODEL IN A NUTSHELL 5/6 – RENTAL MARKET

We distinguish between short-term (maximum one month) and long-term renting.

• Short term
• represents online market place platforms (e.g. Airbnb), which mostly serve the demand coming from tourism.
• Long term
• If a household does not have an own home, it can go to the rental market.

11 |

THE MODEL IN A NUTSHELL 6/6 – CREDIT MARKET

• There are housing mortgage loans, bridge loans and also consumer loans for renovation.
• There are fix and variable rate loans as well.

Bank behaviour

Default

Credit types

Constraints

• A household is eligible for a loan if: �(1) it meets the LTV and DSTI rules; �(2) its expected income covers the credit payments and a minimal consumption level; �(3) and it did not have a defaulting loan in the past five years.
• There can be only one mortgage on one flat

• The bank increases and reduces the credit supply procyclically.
• The bank determines the credit margins with a regression model estimated on actual empirical data.

• In case of non-performance, households first try �to reduce their consumption 🡪 the bank restructures the loan 🡪 finally the collateral will be liquidated.

12 |

DATASETS AND

GENERATION OF 1:1 MAPPING

13 |

1:1 SCALE MAPPING

Flats

Households

Loans

• All 700K housing loan contracts
• Central Credit Registry
• Start date, contracting value, maturity, principal outstanding, payment, interest rate, non-performing status, non-performing start date

• All of the 4 million Hungarian households
• Occupational classification, income, social welfare benefits, education, age, sex, place of living, etc.
• Central Administration of National Pension Insurance
• Demographic Yearbook

• cc. 200K flats (realtors)�+ all transactions (NTA)�+ aggregated statistics of �HCSO micro census 🡪 4M flats
• Reconstructing the housing stock �such that it mimics the agg. statistics
• 3 characteristics: neighborhood, size, �quality attributes.

14 |

EMPIRICAL DATA I

Households’ income, demographics – Pension Contribution Dataset

• Individuals’ characteristics: �Birth date, Sex, ID for merging with loan data
• Main variables:
• Individuals’ anonym ID,
• Gross income,
• Employment type,
• Weekly working hours,
• 4-digit ISCO code,
• Start and end date for income payment,
• Location (NUTS3).

Loan contracts – Central Credit Registry

• Unit of observation: �All home equity mortgage loan contract (at the end of 2017).
• Main variables:
• Contracting date, Loan amount, Maturity
• Principal outstanding, Current payment,
• Market value, Provision,
• County, Settlement type,
• Non-performing status

​

15 |

EMPIRICAL DATA II

Housing market transaction dataset

from the tax authority

• Unit of observation: �All housing market transactions (with tax burden).
• Time period: �2014-2017.
• Main variables:
• Price,
• Time of transaction,
• Size
• Location

Housing market mediator dataset

• Unit of observation: �Housing market transactions (cc. 200k)
• Time period: �2016-2017.
• Main variables:
• Price,
• Time of transaction,
• Size, room number,
• Real estate type, State, Year of building, Heating type, Floor, Lift, Orientation, Garage, Balcony,
• Location (holiday destination, settlement type, Distance from the capital/regional capital).

​

16 |

CALIBRATION

Average house prices

• Parameters are calibrated such that the dynamics of the observable variables match the empirical data:
• average regional prices,
• the number of transactions
• newly built housing stock
• We used data from 2018/19.

17 |

VALIDATION

• We tested whether the variables of the model which were not calibrated follow the empirical data.
• We used mainly lending market variables:
• Number of loan contracts,
• New credit flow,
• Distribution of loans based on income deciles, LTV and DSTI categories.
• But also some disaggregated housing market statistics:
• # of transactions at the regional level
• Average neighborhood quality of flats in transactions

18 |

ROADMAP OF DEVELOPMENT

„Pilot”: Housing market ABM development

Housing ABM Applications

Macroeconomic ABM development

• 1:1 empirical mapping
• High realism in agents’ behavior

• Construction cost shocks
• Family support policy
• Macroprudential policies
• Targeted FTB subsidies
• Optimal scaling
• ABM – SVAR hybrid model for macroeconomic feedback

• Reference model: Poledna (2023)
• 1:1 mapping with empirical data
• 6 stages of development
• Special focus on the financial system
• Integration with the housing ABM�

19 |

MACROPRUDENTIAL POLICY APPLICATIONS

20 |

MACROPRUDENTIAL APPLICATION SETUP

Official macro. numbers �until 2022Q1, central bank forecasts until 2024Q4:

• 2018-19: high growth rate, �low unemployment, inflation and interest rate
• 2020: COVID slowdown
• 2021: recovery in GDP, but growing unemployment, inflation and interest rate
• 2022-24: Very high inflation and interest rate environment

Versatile conditions:

• richer results
• higher validity

Macroeconomic environment

Policy �scenarios

Disaggregated�results

3-3 versions of LTV and DSTI:

• Either unchanged, or +/-10 percentage point change (3x3)

+1 „No limit” scenario:

• No regulatory rules, only credit history and consumption constraints

The results are always relative to the current official regulatory framework in Hungary:

• LTV: 80%
• DSTI: 50%

​

21 |

ASYMMETRIC DEVIATIONS FROM BASELINE, LTV SEEMS MORE EFFECTIVE

• Looser (or no) LTV 🡪�house price boom (especially in Budapest)
• „no limit” 🡪 bubble bursts endogenously
• Loose LTV 🡪 bubble bursts only in the crises
• Stricter LTV 🡪
• Does not decrease considerably the volatility of the house prices
• Effect is similar in the whole country
• DSTI
• only relevant when the LTV is looser

​

Figure 1: Changes in the house price index relative to the baseline scenario decomposed based on regions of the country

22 |

CREDIT AVAILABILITY (CA)

Credit availability (Kelly et al. 2018)

• What is the most restrictive constraint for a household to buy a fair, „justifiable” flat?
• „Justifiable”: average flat in the region and income decile for a given household

Owner occupiers

• They can mostly buy their „fair” flat.
• LTV/consumption constraints mainly below median income.

First-time buyers

• With LTV/DSTI 🡪 Barely anyone is eligible for credit. (Even in the 10th decile 50% is not eligible.)
• Without LTV/DSTI 🡪 First 3 deciles: not eligible for credit. Even in the 7th decile only 50% eligible.

​

Figure 5-6: Credit availability in the baseline (upper) and in the no limit scenario disaggregated based on income deciles and on FTB-OO categories, aggregated between 2018-2024.

23 |

POLICY SCHEMES SUPPORTING FIRST-TIME HOME BUYERS

24 |

DEMAND-SIDE PROGRAMS

• DSTI is tightened by 10 percentage points
• LTV is increased to 90% (from 80%)

2. Mortgage Guarantee + Higher LTV

• The state provides a guarantee up to 10% of the loan
• LTV is increased to 90% (from 80%)

1. DSTI Tigthening + Higher LTV

​

Higher LTV

• 10% of the flat’s price

4. Lump-sum Grant

• 2.5 million HUF (indexed by inflation)
• Same amount as the percentage subsidy for the average housing transaction.

3. Percentage Subsidy

​

Fiscal policy: financial support at the time of home purchase

Eligibility criteria:

• FTBs belonging to the 10th income decile are not eligible
• Flat’s value: under 30 M HUF (excluding ~50% in Bp, ~10% in the rest of the country).

​

25 |

DEMAND-SIDE PROGRAMS (2)

• Preferential loan up to 30% of the flat’s value.
• Only 2% down payment is needed.
• The FTB does not have to pay installment for the first 5 years and the loan does not interest in this period.
• After 5 years, the installment increases with the debtor's income.
• Interest rate: 3%.

6. Shared Equity

• The state owns 30% of the flat.
• Only 2% down payment is needed.
• For the first 5 years, the FTB does not have to pay anything.
• After 5 years it pays a rent: 3% per year of the flat’s purchase value.
• After 10 years, the FTB buys back the state’s share at market price.

5. Bridging mortgage facility

​

Fiscal policy: bridging schemes

Eligibility criteria:

• FTBs belonging to the 10th income decile are not eligible.

​

26 |

SUPPLY-SIDE FTB PROGRAMS

7. Rent-to-Buy

• The state buys the flat.
• The FTB must pay a monthly rent: 75% of the market rent
• The rent is considered as capital repayment (with 0% interest rate).
• The FTB is not allowed to move during the 10-year lease period (except: marriage).
• After 10 years, the FTB pays the remaining capital to become the owner of the property.
• Eligibility criteria:
• Only FTBs below median income.
• Upper limit of the flat’s price is the household's 10-year income.

8. Build-for-Sale

• The state builds new flats on state-owned sites,
• in a non-profit way by omitting both the land cost and the profit margin of sales.
• The annual number of new flats is fixed: 3000 units.
• The size of the flats is randomly selected from 2 buckets between 42-71 square meters with an even distribution.
• The territorial distribution is based on the: i) population, ii) the historical demand for newly built flats.
• Eligibility criteria:
• FTBs belonging to the 10th income decile are not eligible.

27 |

ADDITIONAL FTB TRANSACTIONS

Proportion of additional purchases and (forced) sales due non-performance within additional first home transactions compared to the baseline, based on the 2018 cohort

​

​

​

Main objective: facilitating access to housing for those who would otherwise not be able to finance a home purchase.

28 |

FISCAL COST OF THE STATE

Fiscal cost per subsidised FTB and per additional transaction

​

​

Note: Due to their marginal fiscal cost, the two higher-LTV schemes and Shared Equity are not shown in the figure.

The FTB programs under review imply fiscal and macroprudential costs as well as issues and risks related to the implementation of the programs that may entail additional costs.

29 |

CONCLUSION

• We have developed a 1:1 empirical representation of the Hungarian housing market
• The complexity of the model enables the investigation of various policies.
• The granularity of the model makes it possible to analyze the effects along multiple dimensions.
• Our long-term goal is to develop a 1:1 macro ABM for Hungary and merge the two models, enabling the complex evaluation of the major macroprudential instruments (and the interactions with monetary and fiscal policies).

30 |

REFERENCES

• GitHub: https://github.com/mnbabm/housing-market-abm-public.git
• Mérő, B. – Borsos, A. – Hosszú, Zs. – Oláh, Zs. – Vágó, N. (2023): A high-resolution, data-driven agent-based model of the housing market. Journal of Economic Dynamics and Control, 155, 104738
• Hosszú, Zs. – Borsos, A. – Mérő, B. – Vágó, N. (2025): The optimal choice of scaling in economic agent-based models. Journal of Economic Behavior & Organization, 232, 106928
• Hosszú, Zs. – Borsos, A. – Mérő, B. – Vágó, N. (2024): The impact of borrower-based macroprudential policies on financial stability, inequality, and housing stock with a high-resolution, data-driven model. Available at SSRN.

​

31 |

THANK YOU FOR YOUR ATTENTION

merob@mnb.hu

32 |