 Hi everyone, it's MJ and in this video I want to look at some of the challenges that we have to face when it comes to Modeling credit risk. I think the first one that we need to talk about is the lack of Data and this is different to let's say insurance. So with when it comes to say life insurance Very much the public and how many people died every year is like I say it's knowledge for for all It's accessible for all these life tables people can go in and see okay How many people actually died at each various age groups that data is is quite easy to to access However, when it comes to how many people defaulted on certain loans that information is very much seen by banks as Intellectual property they don't tend to share their experiences with each other So if you're a new company and you want to start a peer-to-peer lending platform Well, your first challenge you're gonna have is you're gonna have a lack of Data and now this is where things get interesting is because of the nature of credit risk because the frequency of it occurring is Low it means your severity data will be lacking even more And so you're gonna have a lot of challenges trying to fit a statistical distribution to your severity data for example Let's say I have a loan book of ten thousand that means I've made ten thousand loans to to people and Let's say there is a 1% default rate. That means we only have a hundred defaults So our severity data can only be built up on a sample of 100 which is Which means what is going to be the statistical significance of whatever our findings is and that is going to add to some of the complexity Of course, you can maybe use extreme value theory, which looks at extreme values specifically when you are lacking Data so there are attempts to do it But one of the big challenges is not only is there a lack of data This is further stressed by the fact that frequency is low and therefore to try and model Severeity is going to be very very tricky then to make it even more harder is that Different durations can't necessarily be used for the same thing So you can't even combine your data because sometimes the information is not going to be Homogenous you can think about it even with the private lending if you're lending money to a person over 40 who has a job has a degree and all you're lending money to a student who's got no proven record Those two pieces of information are quite different. You can't really join them to say, okay That's gonna tell me whether a 30 year old is gonna be able to pay their loans or not And we see this also in the bond market is One not all corporates have traded debt It's not like shares where every publicly listed company has a has a share that you can trade Not all of those publicly listed companies also have a corporate bond that you can trade And even those that do what's further complicated is that there's different Durations they could be one for a 10 year one for a 20 year So the data again, it's very very much Fragmented and this is going to just add to the complexity So the fact that these these these loans are not homogeneous and the fact that they can have different Durations is further going to put stress on the ability to to model So to try and get your hands on good quality data in the credit Industry is very much half of the problem But let's say you're focusing on bond durations or publicly traded debt And you have let's say a little bit of information that you're building your models on Sometimes it's not that easy because with bonds and publicly traded debt You can have these things called credit enhancements. So sometimes you can buy a bond That is guaranteed by another party and now your model has to say okay was the probability that the original party defaults and The guarantee isn't respected now sometimes and I don't know I remember reading the story It was absolutely crazy. It was happening in China apparently where these guys were buying junk bonds So bonds that had very very low credit rating. They were very very close to being defaulted So they're picking them up cents on the dollars. So buying absolute rubbish rubbish bonds And then what they did is they simply Created a fake document saying that a bank that had a triple a rating had Guaranteed these random junk bonds and other investors then said, oh well Well, since it's got that guarantee from that triple a bond I'm prepared to pay double the price that these people Paid for it. So they might have bought five cents on the dollar and then they sell it at ten cents on the dollar and There's a little bit of fraud. So these guarantees that are introduced into the market I mean that is a very crazy story But it is something to be aware of is that sometimes the guarantee could be false But even if it's not false, it still adds to complexity because now you need to model what this other thing is And sometimes there will be implicit guarantee So for for example, if the if a utility of a country's List bonds sometimes people assume that the government will Carry or cover those long those bonds Another good example is say the European Union for those Member states of the EU Some people believe that if they were to go bankrupt the EU central bank will bail them out Will provide a little bit of enhancement and this is why Greece which has a very low credit rating Still gets to enjoy a very low Coupon repayment because the market assumes that the central bank will will cover their back another thing to understand is Or thing that makes it difficult is you can also use derivatives to Modify your credit risk and you've got these things called credit default swaps and credit default obligations These are CDS's or CDOs But once again, they too can be abused. There's been crazy stories of a company that Let's say has got some bonds out It's you know needs to make some repayments another organization comes and they buy a credit default swap on that bond So that they're not they're not the person who Is owed that money, but they buy a credit default swap And they normally buy for quite a cheap premium and it's going to pay up big if that company defaults because the company's got a good rating This organization can then go to that company and say listen here We will offer you a loan on better terms If you if you default and so they just call up the CEO They say defaults on the bond and we will give you a bond at a better better rate That company then does it they then the organization that made the phone call can then go and cash in on their CDS and You can see this this gets abused and there's a lot of Legal risk and people can then be saying well, that's unfair What actually counts as a default risk is it simply a mispayment is it when the bond has to be liquid that you know There's there's a lot of complexity that comes around as soon as we start introducing credit enhancements Like I said With the changing correlation this can really really make any model that tries to incorporate diversification It's gonna put this under strain because you might create a model Which assumes that the market is gonna stay stable and their fixed relationships are gonna be held within assets, but You need to create that scenario where okay? Well, what happens if there's a recession and these correlations have to have to rise How exactly they rise and under what circumstances again? It's very very difficult to model that Credit modeling also becomes complicated where we have something called model mixing People will use different techniques for frequency and severity and then they have to try and join the two together so they might use This model to do frequency and then that model for severity and then they try and you know Multiply the two together or try doing it in a way And they could make a mistake so it introduces even model risk or sometimes if they're in different dimensions You might not be able to fit the data together Then the last one we're gonna be talking about it's not the least I mean if you had to do a lot of research, you'll see when you start doing credit modeling in reality You'll see that there's a lot of complications and challenges That's why a credit analyst is such a sought-after role because it is so difficult but the final one we're gonna be talking about in this video is the credit rating agencies and Sometimes what we'll see is they give conflicting information for a long time Moody rated South Africa as investment grade. We're a standard and poor and Fitch rated us as Non-investment grade now. What does your model depend on? Who's do you use? Do you use moody? Do you use S&P? Do you take an aggregate of all of them? What do you do? Then there's also potential things where businesses could game the system So if a business knows that moody looks at certain financial ratios and not other financial ratios then businesses can take you know counting Acrobatics in order to game the system to maintain their ratio Lehman Brothers with repo 105 is a classic example of how they were gaming the financial system Specifically around financial reports and accounting regulation. So it has happened before so go check up That Lehman Brothers case study if you want to know more of that And then of course there's also the conflict of interest which that movie the big short or the book by Michael Lewis He tried to point fingers at credit rating agencies saying since it's the businesses that pay for ratings And because there's more than one rating agency if one rating agency is going to give them a low rating then they simply go to the other agency and They say that there's a bit of a conflict of interest because the person who you're paying has to Grade your paper or grade your your debt and if you're unhappy with that you're simply going to go to somebody else So credit rating agencies have got an incentive to Underestimate credit risk of the organizations that they're grading in order to retain their business So these are some of the difficulties or complexities And it basically means that I mean I think today the big big investment houses don't really rely on credit rating agencies as much which is a bit of a problem because like I said with when we Look at the Jero Lando Turnbull model it very much looks at credit rating Agencies and the the mocks or the credit worthy in the status that they've given them But overall these are the challenges lack of data The information is fractured. It's not homogeneous. They are credit enhancements. There are some Weird and crazy stories of fraud and abuse Correlation is constantly changing Sometimes we have to do model mixing and our credit rating agencies who we used to depend on for a long time Aren't as perfect as we would like them to be but that is credit modeling challenges in a nutshell I'll see you guys in the next video