Blog: Who benefits from the recent ease of access to mortgages: first-time buyers or big corporations?
Are we heading to another housing bubble? Why are housing prices in places like Amsterdam constantly rising? Who benefits from the ease of access to mortgages: the first time buyers or solely speculators?
Important questions such as these can be investigated and simulated via agent-based models (ABMs). At a relevant project led by some members of Lucidminds AI, we have designed and run ABM simulations to find out the underlining factors.
We have seen that unless the housing markets are regulated with the right incentives, an excess amount of mortgages can cause recurring financial crises and uneven wealth distributions. One wouldn’t need ABMs to confirm such a ‘fact’ that we are inclined to ignore or easily forget. However, what our novel approach provides is a controlled socio-economics laboratory to devise and test policies or individual incentives to prevent such catastrophic economic crises.
Introducing a mortgage-credit-rule
In our project, we propose a mortgage-credit-rule that can easily be implemented by banks or enforced by central banks and governments. Our proposed policy has a dual effect, at one hand it acknowledges and enables the positive impact of housing markets driving the growth in an economy, on the other hand, it leads more even mortgage distribution in the society and less likelihood of mortgage led defaults.
Simulations, or ABMs, in general, are powerful tools to develop insights and to conduct what-if-scenarios for policy-making. ABM simulations are very instrumental especially when there are insufficient data or when access to sensitive data may violate privacy rights.
The policy suggestions of our computational experiments from that project are very recently published in a peer-reviewed and science citation indexed journal.
A team within Lucidminds.ai is dedicated to designing complex and large scale yet realistic ABM simulations. We are able to create virtual worlds populated by autonomous and adaptive AI actors.
Simulating the economy
We simulated an economy that has generated the data behind the sample plots of this post. In the simulated economy, there are firms, banks, a central bank, and a government. These AI agents are interacting at malls, credit markets, financial markets, housing markets, etc. Firms are producing and make sales decisions; households are looking for jobs, making consumption decisions, investing in financial markets, and buying houses, etc.
The realistic nature of our ABMs is confirmed by the fact that they can reproduce the patterns of real economies; and more importantly, we are able to observe realistic variations.
As a first step, we configure a Genesis version, the initial version of a world, then, we create multiple instances of it and run them on supercomputers. In a way, we generate ‘parallel universes’ for the same initial socio-economic conditions and its population.
Data science with simulated economies
Next, we put our data science know-how and skill sets into practice in order to examine rich time series data generated by the thousands of simulation runs. Depending on the peculiarities of each universe, such as the government’s policies, citizens consumption choices, channels of wealth accumulation, etc., we can observe economies that thrive or economies that run into huge financial crises as well as uneven wealth distributions among their citizens.
The first figure in this post is one example of various ways one can test whether the model outputs are sound and valid. This particular figure confirms the similarities between our simulated virtual worlds and the real-world patterns as of the dynamics of business cycles. That is, by conducting advanced time series and other econometric analyses we compare our simulation results with analyses done by actual central banks such as ECB or Bank of England on real historical data sets.
More specifically, via this figure, we reveal cross-correlations between time series on GDP, mortgages, and loans to firms. We have observed that in our virtual economies growth/decline in mortgages in the system is leading the growth/decline in GDP which is followed by growth/decline in loans given to consumer good producer firms. The upper panel is the cross-correlation of GDP vs Mortgages while the lower panel is GDP vs Loans to firms. These patterns are in line with the patterns seen in actual economies in the world.
From insight to foresight
The second figure above is a comparison of different housing market regulatory policies. We particularly forecast the GDP growth given a combination of the two policies we suggest.
The first policy is denoted by the Greek symbol the beta. It is mainly similar to some mortgage credit assessment done by the actual bank agents when you ask for credit. They check your regular income with respect to your mortgage debt payment plans. They give you mortgage only if they believe that you can afford the monthly debt payments throughout the whole debt service period for the coming years. Sometimes, they adjust the monthly service payment amount and hence the duration according to your income level. Such durations can go up to 30 years.
Unless there is a strictly imposed government rule, how strict these rules are arbitrary and determined either by the bank’s own policy or by the personal decision/incentive of the dealer who is ‘taking care of you’. As we all learned from the great recession of the 2008 crisis, the looser the policies the higher the systemic economic risks.
However, our simulations have also pointed out a fact behind the government’s tendencies at promoting the construction sector and hence easier access to the mortgages. Banks by giving credits to households, they essentially create fiat money ‘out of thin air’, which pumps the demand for consumption and hence the growth in GDP figures followed by more investment decisions by consumption goods producers such as home appliances, cars, etc.
The problem is when such self-enhancing loops are left out of control they create temporary prosperity and booms followed by sudden catastrophic economic crises. We have designed a second policy to circumvent such a shortcoming. In the figure, this second policy is denoted by the Greek symbol the gamma. In more plain words, it additionally checks the current debt of a mortgage requesting entity, an individual or a corporation, before fulfilling the new request. The stricter the rule the more difficult for someone to acquire a new credit for a second house or an expensive new car etc. The figure summarizes a policy suggestion that promotes easy access to the first time buyers but makes it very difficult to get a second comparatively large credit. That is, to be eligible for a new mortgage, someone or some corporation should be free from existing debts or the total debt on the balance sheet should be low enough compared to the existing assets.
Wealth distribution and heterogeneity
The third figure in the post demonstrates the power of ABMs compared to other simulation approaches especially in the area of socio-economic systems. It is the ability to look into individual-level differences in simulated worlds. The upper panel in the figure compares a Gini index of mortgage distribution within different scenarios. This figure along with numerous other analyses that can be found within the paper points out that the policy combination we suggest promotes more even distribution on homeownership while assuring a sustainable growth at the aggregate level in the economy.
As a team of ABM experts at Lucidminds.ai, we employ the power of realistic simulations to gain insight into complex socio-economic systems such as the one we have mentioned in this post. Drop us an email for any inquiry regarding the use of ABMs in business cases.
The paper we have mentioned in this post has been very recently published in the Journal of Economic Interaction and Coordination. It can be downloaded from the link here. If the link doesn’t work for you, contact me directly via ResearchGate for a personal copy of it.
Bulent Ozel, Lucidminds, CEO
Acknowledgments: We’d like to thank Oguzhan Yayla, Susanne Weller, Dominik Heilig, and Hamza Zeytinoglu for their comments and contributions to the article.