 So, welcome back everyone. I hope you had a good lunch time and made lots of connections and got to know your fellow researchers. So, let's continue with session number two. We have three more papers actually that are mortgage market related. So, I'm really happy that we have so much interest in so many research papers on that topic because it's really important for central banks because, yeah, well, the mortgage market is hugely important from a financial stability perspective. So, let me briefly introduce the three speakers that we have in our lightning session. So, lightning session, it's the first time I hear this term. So, lightning session means it's relatively short presentations, 15 minutes, and then we don't have a discussant and the Q&A, but we just move on. So, it's really the possibility to get in a number of papers still and I hope you can all appreciate that. And then please use again the coffee break to follow up with questions. So, the speakers are in this order. First of all, Stelios Giannolaques from Athens University and he is doing research and firm dynamics and macroeconomics and he has already been at the ECB actually a couple of years back as a trainee. And the second speaker will be Jaunios Carmelavichos from the IMF where he is financial sector expert and his research interests are around banking, financial stability and macro potential policies. And then last but not least, Claes Backmann who is working at the Department of Economics and Business Economics at Arhus University and will very soon join the SAFE here at Frankfurt University. So, you have actually both affiliations almost and he has a strong research portfolio related to mortgage markets. So, I will give the floor to you Stelios. You can go over there. Hello everyone. Thank you for being here. I know that it's difficult to present after the lunch but I will do my best. Let me say that this is a joint work with Eugene Tereanu and Marco Forleta from the ECB and with Marco Gross from the National Monetary Fund. The main theme we are addressing in this paper is how borrower based macro potential policy measures can improve the resilience of both households and banks. Let me start with a few words about the motivation of the study. During the last decade, borrow based measures have been increasingly implemented across the European Union. Of course, the eruption of the pandemic crisis has slowed down the implementation both within and across the European countries. However, in the light of the recovery from the recession, borrow based measures have been called again upon to address the buildup of vulnerabilities in real estate markets. And indeed many European countries started to tighten their borrow based measures in 2022. Moreover, in the context of the review of the European macro macro potential framework by the European Commission, both ECB and the European systemic risk board has underlined the need for a harmonized definition of lending standards. The risk board took further by suggesting the direct inclusion of borrow based measures in the Polish toolkit of the European Union. So what we do in this study is to contribute into the literature by developing a quantitative tool for assessing the role of borrow based measures. First, in enhancing the households resilience and with the term resilience, we mean lower probabilities of defaults and losses given default. A key advantage of our paper is that it takes into account the side effects of borrow based measures, which is nothing else than a reduction in the mortgage loans volumes. Second, our model can evaluate the role of borrow based borrow based measures in improving banks capital ratios through the improvement of their mortgage portfolios. Finally, we can use our model to analyze the policy heterogeneity of borrow based measures across the income and wealth distributions of households. Okay, I can skip this to save some time. Now, which are the key messages of our study? We find that the implementation of borrow based measures and specifically their joint implementation can not only improve their resilience of households. Second, we find that borrow based measures can improve capital ratios of banks through the improvement of their mortgage portfolios. The lower probabilities of defaults and losses given defaults of households improve the mortgage portfolios of banks and through them their capital ratios. Finally, we found that the resilience benefits of borrow based measures for households is more profound for low income and low wealth households. Let me move to the mechanics of our model. Our model has three modules, a micro economic module that help us to describe the micro economic condition of each country, a micro module that help us to quantify the improvement of household risk metrics after the implementation of borrow based measures and a bank e-back module that help us to see how borrow the implementation of borrow based measures can improve the capital ratios of banks. Let me start with the micro economic module is a six variable structural bar. You can see the six variables here. The most important of them is unemployment, house prices and credit to the private sector. The micro economic module, the structural bar model serves two goals. First, it helps us to stimulate a bunch of forward paths for these six micro economic variables. And second, by using sign restrictions, we can obtain the responses of these six variable to a credit shock. We will use this e-back responses of these variables to quantify the adverse effect of borrow based measures, which is nothing else than a reduction in mortgage loans. Let's move now to the micro economic module. It has two parts. The first part is an employment status simulator that give us a bunch of forward paths for the employment status of household members. We obtain these forward paths in a way to match the aggregate unemployment rate from the structural bar. The second part of the micro economic module is a household balancing simulator. This simulator takes as an input the outputs of the previous modules and quantify the probabilities of defaults and loss given defaults for each household. Also, how we work on it, we try to detect default scenarios and default events for households. The default criterion we use is negative financial assets. Now, the negative financial assets of households depend on macro economic conditions. Recall that we have a bunch of forward paths for micro economic variables and first a bunch of different macro economic scenarios. This means that we have a bunch of different default scenarios for households. Since we have many scenarios default scenarios for households, we can define the probability of default. Given that we can define losses given default. The last module of our model is the bank e-back module. What this module does, it attaches the improved probabilities of default and losses given default or the mortgage portfolios of banks. So, we have lower unexpected loan losses, mortgage loan losses for banks and lower risk weighted assets. This means that we have better capital ratios. Very briefly, let me say a few words about the transmission of borrow based measures in our model. It is a two-step mechanism. First, we both regularity limits on the three borrow based measures we are examining here. The loan to value ratio, the debt service to income ratio and the debt to income ratio. These regularity limits can be either individual or joint, a combination of the three borrow based measures. And we both these regularity limits under the baseline macro scenario as it formed by the structure of our model of the macro economic module. These regularity limits restrict new high-risk lending, reducing the number of default episodes for households and thus leading to lower probabilities of default and losses given default. However, as I said earlier, borrow based measures have an adverse negative effect, which is nothing else than a policy-induced reduction in mortgage loans, in mortgage loan volumes. Our models quantify this policy-induced reduction and scale it using the impulse responses from the macro module as I said in the previous slide. Now, the quantified side effect is fed back into the micro economic module of our model, adapting the macro economic conditions. The deterioration of the macro economic conditions affect the financial assets of households, leading to more to a higher number of default episodes. Since we have a higher number of default episodes, we have a higher probability of default and given that higher losses given default. Of course, as we will see in the next slide, this side effect of borrow based measures is not quantitatively sufficient to cancel out the resilience benefits from borrow based measures. Very briefly, if you've heard about the data, our model combines micro economic data from the household finance consumption survey with macro and bank related data from several sources. You can see some of them here. Okay, I will skip this to save some time. Let's move very fast to the results. This figure here summarized the baseline results of our model. The left figure illustrates the distribution of country specific probability of defaults. Let me say that we simulated our model for 19 European countries. So here you can see this 90, the simulated 19 probability, the distribution of this 19 probabilities of defaults. We run four scenarios. The first scenario, it is the green bar here, is a scenario with a no-police scenario without the implementation of borrow based measures. There are four other scenarios, one with the implementation of each borrow based measure separately and one, the left one, with the joint implementation of these measures. The right figure shows the distribution of the country specific losses given default. Two important points here. First, both income and collateral based macro conditional policies improve the resilience of households in terms of lower probabilities of defaults and losses given default. Second, the resilience benefits for households are much stronger when policy limits are imposed jointly. Let me move to the results from the bank eBank module. So the reduced probabilities of default and losses given default improve the mortgage portfolios of banks, leading to lower expected losses and lower risk weighted assets. And these two positive effects lead to higher capital ratios. The left figure here shows the change in capital ratios of the 19 countries, 19 European countries. It shows the change between the non-police scenario and the scenario with the joint application of borrow based measures. As we can see here, the joint implementation of the measures has led to an increase in one percent and point approximately. The other two figures decompose this positive effect between the eBank of the reduction in lower losses and the impact of the fall in a risk weighted assets. As we can see, the 80 percent of the increase in the capital ratios is due to the fall in a risk weighted assets, while only the 20 percent of this positive effect is due to the reduction in mortgage loan losses. Now let's move to the distributional effect of borrow based measures. The left figure here shows the reduction in probabilities of defaults under the in the scenario with the joint implementation of borrow based measures relative to the non-police scenario. We do this for households with high and low income, namely for households above and below the median of the income distribution of households. As we can see here and in the right figure, we have the distribution of the reductions in losses given default. As we can see here, the resilience benefits from the implementation of borrow based measures is more profound for households with low income and we have the same results even when we examine the wealth distribution. Very briefly, let me summarize the main results of this paper. Again, we find that borrow based measures can not only improve the resilience of households and especially where they are jointly applied. We also find that borrow based measures can support bank solvency ratios through their improvement of their mortgage portfolios. And finally, we find that the resilience benefits are most important for households with low income and low wealth. Thank you. Thanks very much, Celius. Also for superb timekeeping. So, Janius, first yours. Hello everyone. It's a pleasure to be here. Today, I would like to present you a joint paper with my colleague Mantas Dirma. The paper is titled Microassessment of Macro-Predential Borrowed Based Measures in Lithuania, which is basically sharing of our experience and calibrating the BBMs using micro data. The usual disclaimer applies. I'm going to provide you some background. Overview the methodology, which gives an assessment of BBMs and lay ground for conclusions. So Lithuania adopted macro-predential borrow based measures back in 2011 through the enactment of responsible lending regulations, which we call ASN as per Lithuanian acronym. And the package, as a whole, consists of five measures that include the LTV of 85%, DSTI of 40%. Stressed DSTI of 50%. You have a maturity limit of 30 years. And you have a newest piece of regulation, the LTV for secondary mortgages, which was issued back in 2022. And it stands at 70%. Mortgage loan is said to be secondary. If during its inception, the household has at least one other active housing loan. And the package was aimed primarily to boost resilience and then desirably can also act counter-cyclically. However, during the low-rate environment, we were facing a couple of issues. First off, there was a really hot housing market in Lithuania, one of the hottest in the European Union, which kind of culminated in a mortgage credit overflow measured at 15%. We had a house price overvaluation of 20%. So a natural question may arise, how is it that we have had macro-predential policy for a decade? And yet we are still facing these vulnerabilities. So it may be that the BBM framework, as a whole, is not effective in putting a backstop to the successive growth or either the BBM parameterization is not stringent enough. The other issue we were facing was the increasing prevalence of secondary mortgages, which we found out that due default more often, they often have higher DSCI ratios and surprisingly, they have high LTVs, which are concentrated around 80%, just below the 85% headline limit. And as a bonus to that, they had additional fuel to the already hot housing market. What we do in this paper, we try to address three questions about the efficacy of the BBM package in containing credit risk, about the limits, their parameterization, is it adequate, and what is the necessary regulation for secondary mortgages in terms of LTV. And to do so, we develop a complete credit risk framework that models expected credit lifetime losses, which use PD, LGD models that are all based on micro data. So let me turn briefly to the methodology. The data is from the household credit register, which is combined with household income data spanning from 2004 to 2020. And for credit risk assessment, we model loan level default offense and credit risk is defined as the expected credit loss that is equal to the PD times LGD, where the latter two parameters are modeled separately in a one year ahead framework, particularly PD is modeled as a simple logistic regression, where we regress household default dummies on household loan characteristics, macro variables, and importantly, BBM related variables that are measured at loan origination at the point when BBM limits can really control household leverage. And they enter the equation nonlinearly using a cube explain specification. What is more, the LGD is computed using a simple accounting rule that is based on loan parameters and amortization schedule features because we don't really observe loan by loan losses. I'm not going into the detail about the PD model results, but all coefficient signs are as expected. They are highly statistically significant, and the model's discriminatory power in terms of OROC statistic that is measured out of sample stands at 90%, which is pretty high. Instead of merely looking at one year PDs and one year LGDs, we want to evaluate a lifetime's worth of credit risk, credit risk for each loan. And to this end, we compute the expected lifetime credit loss that is based on each loan's amortization schedule and the evaluated PD LGD parameters that also vary throughout its life cycle. We will be using this methodology to assess BBMs. And the first question in hand is the efficacy of the BBM package as a whole. And to shed light on that issue, we are displaying these charts which show the distributional characteristics of factual LTV, DSTI, and maturities throughout 2005, 2020 for loans that originated in the period. So preceding the GFC, we see a little bit of risk-taking that elevated along with elevated LTVs, DSTIs, and maturities. Post-GFC period saw market self-correction, thus the introduction of ASN requirements wasn't really distortionary at the time. However, current BBM framework still ensures that the market mortgage risk parameters are constrained and don't really go over the top. We used our modeling framework to simulate historical lifetime PDs and lifetime expected credit loss rates that are measured at loan origination. And from what you can see, the credit risk now is much smaller compared to the pre-regulatory era. What is more, we did this counterfactual experiment where we saw that during the crisis of 2009, mortgage losses would have been 83 percent smaller had ASN regulation been present in the 2000s preceding the GFC. And lastly, we did this sensitivity test where we introduced a drop in household income and an increase in interest rate. And the test really reveals that households are now much more resilient to stress than in the pre-regulatory era. So let me turn to the adequacy of BBM limits and more particularly the parametrization. Together with my colleague Mantas, we were thinking about the options for tightening had authorities decided to lean against the wind during the low-rate era and a natural starting point to look at that is to look at the distributional characteristics of how binding the measures are. And we concluded that LTV is not really an option for binding tightening because it's already binding for too many people, especially young families who have often have good collateral. And the measure is quite stringent from EU perspective. Therefore, we shifted our attention to DSCI since any binding of a DSCI measure would be less impactful but also less distortionary for the market and more risk targeted. However, it's not clear what is the right DSCI cap from risk perspective. And to shed light on this matter, we computed our model-based marginal effects. That is, what is the marginal impact of the change in DSCI on the probability of default at each different level of DSCI? And we mapped this relationship which resulted in these beautiful bell-shaped curves which have a strong peak maximum, which is basically saying that the probability of default is growing at the maximum possible rate at this point of DSCI. So if you were the regulator, you wouldn't, definitely you wouldn't want to find yourself in such a situation that this fast growing PD would want to be to the left of that curve. However, if you take a look at the ASN requirements in Lithuania, they are already beyond this point to the right, suggesting that during the low-rate environment Lithuania's DSCI limits were on the loose end. Suppose the regulator decided to tighten this measure to correct this. How would households behave? How households would definitely try to lengthen their loan maturities to minimize their loan payments to keep up with the new DSCI requirement. However, the model really showed that longs with long maturities are more likely to default over their long lifespans. Thus, we thought that the DSCI cap should be set in conjunction with the maturity limit. I'm not going to describe into detail of this chart, but suppose there was a situation where the authorities tried to lean against the wind and close the 15% credit gap. They would be exploring multiple options of reducing DSCI and maturity limits. However, the model is basically saying that if you want to close the credit gap and at the same time minimize a lifetime credit risk, the first best option is to do a joint reduction in DSCI and maturity limits to counter the feature that households would try to shift their maturities, prolong their maturities as a result of reduction in DSCI limit. Going back to the issue of secondary mortgages, the model is really showing that secondary mortgages are more likely to default compared to an otherwise identical but single loan. This is an issue, but we can really fix that as a regulator because if the PD of a secondary loan is high, we can counterbalance with strict LTV limits so that the LGD is lower and the expected credit loss is pretty much balanced. However, there is a bigger problem in terms of secondary mortgages because mere issuance of them imposes a negative externality of heightened default risk for existing housing loans that are already out there in the portfolio and you can't really affect them because you can't issue exposed regulation. So to counter that, you would want to tighten the LTV limits even more so to counterbalance the first and the second effects that I have just described. So how to do so? We did this calibration exercise with Mantas where we tried to find a personalized secondary LTV limit for every mortgage that was issued in the history and how did we do it? We were looking for a secondary mortgage limit that would solve this equation and equalize the credit risk of a household having two mortgages with the very same household having one mortgage but with maximal attainable ASN limits. And as a result, we have this chart which basically says that all secondary mortgages should have a LTV limit that is strictly lower than the headline LTV limit of 85%. There is a negative relationship between first mortgages, current LTV and secondary mortgages calibrated LTV limit and the relationship is a bit of a kinked around the 70% first mortgage threshold basically saying that if a household has a first loan that is relatively unamortized or high LTV, then he or she cannot be issued a loan with a relatively high LTV. So that is what actually the Bank of Lithuania implemented back in 2022, imposing the 70% limit on secondary loans and it's also differentiated by the first loan's current LTV limit. So let me jump back to the conclusions and in terms by limiting mortgage DSTI, LTV and maturity parameters which all affect credit risk, the BBM toolkit can be effective in containing credit risk and as we saw from our stress test results borrowers are now more resilient to stress than in the pre-regulatory era. What is more during low rate period according to our non-linear model, the DSTI measures were on the loose end. So if you want to achieve credit reduction targets at the same time minimize lifetime risk the best combination would be the best move would be to combine a joint tightening of DSTI and maturity limits and as per secondary mortgages not only they are riskier but they also impose negative externalities on other loans thus they rightly are and should remain regulated by imposing a more stringent cap. So thank you very much for your attention. Thank you Garnius and then we have Klaas as a final speaker. Hello everyone thank you for for being here. Today we're going to talk about Swedish macro potential policy a policy known as the amortization requirement that basically mandated that mortgages with a LTV ratio about 50% has to be amortized that is repaid. So in practice this this reform banned interest only mortgages for for high LTV loans. So this paper is joined with Patrick Gurand who's at the Fed board and Peter Van Santen who's at the University of Kroningen and the usual disclaimer is applied because the pad is at the Fed board. Okay so to kind of motivate this we're going to note that one of the most important features of the mortgage contract is the repayment schedule so deciding how fast you are going to repay the debt. So the repayment schedule introduces mandatory mortgage repayments mandatory savings that is designed to build home equity over time. This is one of the largest savings plans in the world so there is some data from the United States and that Bernstein and Kudji's paper saying it's about 300 billion dollars a year going into mortgage amortization. If we take a policy perspective on this we can I'm sure you all know that there's an active debate in many countries over whether to prohibit interest only mortgages or to further regulate mortgage design. So financial innovation has meant that there are new products coming up that in a lot of respects tend to target the repayment plan. So in the United States the most famous example is the interest only mortgage that was very popular in the pre-financial crisis that was later removed with the Dodd-Frank de facto ban. In Europe a lot of regulators are looking at banning interest only mortgages or at least regulating this part of the mortgage contract. Now our research question here is going to be what's the impact of these kind of policies on household borrowing and we're going to be specifically looking at borrowing at origination. And what we're going to do in the paper is we're going to exploit this reform that you already saw on the first slide that eliminated interest only mortgages for borrowers with an LTV ratio above 50%. As you saw we're going to document substantial bunching below the threshold meaning that households are voluntarily lowering their loan-to-value ratios in order to achieve you know to get rid of amortization payments. Our results are that this is driven by wealthy households. We don't find any evidence of credit constraint there. In fact only 14% of the households who are placing themselves at this threshold are doing so because of credit constraints. And we don't find any other supply side factors that could explain this. Most probably there's no changes in the interest rate terms. So that's on the empirical side. We want to or we're interested a little bit in the mechanisms behind this. Why are wealthy households so keen on getting an interest only mortgage? And to do that we're going to put up a lifecycle model of well with long-term ratios. And we find that in the baseline model there is no bunching. Households have a number of ways to undo required amortization payments. That means that this kind of policy is just not costly for them. They don't respond in the way that we see in the data. So we put in some some more behavioral features in the contract in the preferences and we're going to argue that the bunching is driven by households experiencing what we call flow this utility amortization payments. Essentially you can think of this as they think that amortization payments are a cost, an interest rate. That's the pattern that's most consistent with our data. And the implication of this is that interest only mortgages may substantially increase aggregate household debt because households are not taking this it's the true cost of the mortgage into account. Okay a little background on the policy and the institutional setting. So you can think of Swedish mortgages as essentially a bank loan with no set maturity where you can pay the interest and amortization. Most mortgages are adjustable rates with no set maturity and that's the kind of bank loan part of it. Pre-reform and majority of contracts were interest only. So Swedish borrowers typically don't tend to pay back their loans. They tend to save in other assets but that's another discussion. Sweden like many countries experienced high debt levels and rising debt levels you know in the aftermath of the financial crisis and you know there was more attention to household debt following this. So the regulator decided that this was a threat to macroeconomic stability and they wanted to do something about it. Their policy was to essentially say that if you have high LTV mortgages you're going to have to pay back on the principle. And the idea was to kind of reduce debt over time force people to amortize their debt. The thing that we're going to be looking at is this 50 threshold where essentially interest only mortgages is appropriate. It's also important to note that after you cross this threshold you can turn off amortization payments again. So if you start at 51 you amortize down to 50, you can call the bank and they'll stop taking money from your account. Now the empirics is nice to present because it's essentially in graph form. The intuition behind this is that the design of the policy where I should say the amortization payment is the amortization requirement is it's 1% of the entire mortgage. So there's a big discontinuous jump right at the threshold. But the design of this requirement creates a tradeoff where if you are close to this threshold you can choose between having a larger down payment or lower amortization payments. And this is going to allow us to identify the response of borrowers to changes in this in payments. For example let's say that you want to borrow against the $500,000 house if you pick a down payment of $250,000 you have an LTV of 50% and you don't have to make any amortization payment. If you pick a down payment of $240,000 your LTV is 52% and you have to amortize $200. It's a very salient choice for households sitting at the bank deciding how much money am I going to put in. There's a lot of stuff on the empirics in the paper as you can imagine. I'll note that the policy had bite as in there are very few interest only mortgages just to the right of the threshold after the reform right. So the upper graphs there are 2013, 14, 15, pre-requirement about 60% of contracts 50, 60% of contracts were interest only. After the reform to the right of the threshold there are no interest only contracts. Empirically we see that we take that graph that we had on the very first slide and we essentially just collapse that down and then we can elicit some behavior by households. So we find that 8% of borrowers within this kind of region they bunch so they place themselves at the point where they minimize amortization payments. They reduced their loan-to-value ratio by 5%, 2.5 percentage points over 50 so 5%. We find a little missing mass which we're going to use later in the model part and we find that 86% of these borrowers do not face any binding credit constraints. So first of all they have 50% equity in their house right so they're fairly wealthy on that side. They also don't face any payment constraints or binding payment constraints. And we don't find any other ends for supply side factors that could explain this so most importantly no changes in the interest rates around this threshold. So we want to explain this or the referees want us to explain this. They're right so we should think a little bit more about this. It is a little bit of a puzzling result. So what we're going to do is we're going to construct a lifecycle model of consumption, housing and mortgages where households are going to get utility from consumption and housing. There's going to be heterogeneity and risk in the model coming from initial assets, initial income and income risk. There's no house price risk in this model. I think we could add that. I don't think it will change but no house price risk. Households are allowed to choose consumption, liquid assets, housing so there's a rental market as well and houses of different sizes and they're allowed to choose, they have long-term mortgage contracts. We're going to model the reform as a world where the mandatory minimum payments are interest only or the Swedish policy where you have to advertise 1% of the mortgage if you have an LTV ratio above 50. And you can also do cash out refinancing. The results in the baseline model is that there is no banishing below the threshold. So this is the LTV distribution from the model and then we just plot the fraction of errors within each bin. The reason this happens in the model is that there's no kink or notch in expected discounted utility. So what that means is or what you can see in the graph is that the baseline model is in blue. The interest only model is in orange, a little bit above. So it's above how those like to live in an interest only world but they have a number of different ways to undo required ambitious payments. They can do refinancing, they can do borrowing more, they can change the other savings. So this doesn't have a discontinuous impact at this threshold in the model. So what's going on? We're going to put in two behavioral wedges and household preferences. This is motivated by us finding nothing in the budget constraint that could explain this and no changes in interest rate. So if it's not in the budget constraint it has to be somewhere in preferences. And we're going to do two types of costs. One that is very local. So one of this utility that applies from borrowers turn off amortization. The easiest way to conceptualize this is a cost calling the bank. The second is this ongoing flow of this utility to amortization. You can think of this as some kind of psychic cost households viewing amortization payments as a cost. And there's some literature that talks about this. What this does is the one off cost is going to generate what's known as a notch in the bunching literature. So being right to the right of the threshold here is going to be have, is going to change, discontinuous change in the slope. So the change in the average rate. Importantly, what happens is that households are going to bunch, households are just to the right of the threshold are going to say, I don't really want to pay this cost. I'm going to move my borrowing a little bit to the left so that I avoid this one time cost. And then there's a region there that it's maybe difficult to see with these colors, but there's dominated region where no one wants to be. So there's an empty space there that's known as missing mass. The second one is going to generate a kink in expected utility. So that's a change in the slope. So everyone above the threshold who have to amortize have to pay this one, this cost. It's an ongoing cost. That changes about the slope of this line. And that's going to generate bunching, but everyone is going to kind of move down. So there's no missing mass. In the data, we find little missing mass above the threshold. So this kind of makes us think that flow this utility is more important. We also try to contact the banks and see what kind of cost there could be to refinancing. And they all kind of said that there's no cost to refinancing. It's enough to do a phone call or in the mobile app. The implication of this is that households, if this is true, households are going to choose a mortgage contract that comes with a higher lifetime cost, essentially because they're confused. And if you go the other way from this policy and you introduce interest-only mortgages, households are going to think that's really cheap. And so they're going to increase their borrowing by a lot, even though it's not actually cheap. So to conclude this very short talk, so maybe I don't need to reiterate these things again, but we find that households reduce borrowing to avoid amortization payments. I'll note that if you're thinking about this from a macro potential perspective, you're saying the amortization requirement reduced debt. Having households with a 50% loan-to-value issue who have a lot of income to support mortgage payments, maybe that's not the group that you were targeting. These are not high-risk borrowers, I think. And then we argue that most of the bunching comes from a king in preferences. And I'll note finally that overall impact on financial stability also depends on what you do with the money. So are you now more liquid because of this, with your savings behavior? So I don't think you can conclude that this improved financial stability, although in one dimension. But that's it. Thank you for listening. So thank you very much. That concludes session number two. We learned a lot about housing markets and how to apply macro potential policies to these markets. And I think we got a lot of evidence mainly based on national data, but that's just in the nature of this matter because we do have very different mortgage markets and also very different behavior, I think, of borrowers or households in general. And I think also data availability varies a lot. But I think we can still all learn from these papers and to see where we have similar problems of overheated housing markets and we see what types of borrowers measures may work and may not work and what are the effects and also the impact, the effect of the importance of different household characteristics. So I think that was very interesting. And I'd like to thank the three presenters again and everyone else in this session.