Odd Lots - The Creator of VaR Explains How Large Banks Measure The Risk Of Their Own Portfolios

Episode Date: June 25, 2018

Earlier this year, markets were spooked by blow-ups in a number of volatility-linked products. But dealing with volatility is the foundation of risk management on Wall Street and there's a particular model that's become pervasive among big investors and banks -- so-called Value-at-Risk (VaR) models seek to gauge how much a portfolio might gain or lose based on historic price movements. On this week's episode of the Odd Lots podcast, we speak to one of the original creators of VaR. Till Guldimann explains how he came up with the model while at JPMorgan, plus how it works, its limitations, how it can be gamed, and what he thinks of the volatility landscape now. See omnystudio.com/listener for privacy information.

You can get the news whenever you want it with Bloomberg News Now. I'm Amy Morris. And I'm Karen Moscow here to tell you about our new on-demand news report delivered right to your podcast feed. Bloomberg News Now is a short five-minute audio report on the day's top stories. Episodes are published throughout the day with the latest information and data to keep you informed. Yes, there are other products like this from a variety of news organizations. But they usually rerun their radio newscasts throughout the day. That's not what we do. We create customized episodes that can only be heard on Bloomberg NewsNow.

And we don't wait an hour to publish breaking news. When news breaks, we'll have an episode up in your podcast feed within minutes. So you're always getting the latest stories and developments. Get the reporting and the context from Bloomberg's 3,000 journalists and analysts we're all over the world. Listen to the latest from Bloomberg News Now on Apple, Spotify or anywhere you listen. And welcome to another episode of the Oddlots podcast.

I'm Tracy Allaway. And I'm Jill Wisenthall. So Joe, one thing we talk about quite a lot on this podcast, especially lately, has to be volatility. Yeah, absolutely. We didn't see much volatility in any markets in 2017. And it's picked up more in 2018, though not dramatically, but we certainly have seen some very interesting episodes across a range of markets, making things quite a bit more interesting. Yeah, I love that we have episodes when there's no volatility, and then we have episodes

when there's lots of volatility. But do we ever stop to ask about the backbone of volatility measuring or volatility models? No, we don't really. I mean, we talked a little bit. It's like we always sort of talk around it, don't we? Like we talked about how the people blew up when they were shorting the VIX and stuff like that. But it is true that to some extent, we feel like these measures of volatility are like handed down to us on stone tablets on the top of a mountain rather than something that people had to sort of come up with on their own. Yeah. So today we're going to be talking about the origins of a volatility model that is essentially the backbone of a lot of Wall Street risk management and a lot of the volatility modeling that we've seen in recent years, it's something

called value at risk. And I think you're already familiar with it. I have a vague idea of what it is, but I think it's something like if you have a big portfolio, you want to measure what is a sort of reasonable amount you might expect to lose on any given day over some time period to see how risky your portfolio is. That's pretty good, actually. That's impressive. So just to just to harden. Thanks, Tracy. I'm glad my very rudimentary definition was enough to impress you. No, but it's true. I mean, that's it. It's the amount of money that you might expect to lose at a given confidence level over a certain time period. So for instance, if you and I were running

odd lots capital, which we totally should do at some point, and we had a one-day value at risk of a million dollars, say, at the 95% confidence level, that would mean we would expect to lose more than a million dollars on one day out of 20 at a 95% confidence level. So this model, you know, it was invented in the 1990s by J.P. Morgan and it spread throughout all the banks. And it became the backbone, as I said, but it's also intensely controversial. And you see lots of criticisms of it. Nicholas Nassim Taleb is probably the most famous critic of the model. But I should say, today we're going to speak. with someone who not only invented value at risk, but can also explain what it is that the thing does

and what it is that it doesn't do. Great. Well, I really don't know much more about it beyond what I told you, so I am looking forward to learning more. So without further ado, let's bring on our guest for this episode. It is Till Goldman. Till, thank you so much for joining us. You're welcome. Glad to be with you.

So, Till, maybe just to begin with, you could walk us through your early, career history. You know, I mentioned that VAR Value at Risk was invented at J.P. Morgan, and you were obviously at the bank when you invented it. But how did you end up there? Different jobs in different locations. At the time, trading room in Hong Kong. And that was in charge of that. And from a numerical viewpoint, rather than from a making money viewpoint. And as a consequence, I didn't make that much money, but I had collected a lot of numbers. So they said, well, perhaps you're better off coming back to New York and apply the number skills and we make the money, which I accepted and I became head of asset liability management, which at the time was being in

charge of the balance sheet risk of the bag. And our trading was increasing all over and we then decided to use a new methodology to trading risks in addition to. And that's when all the numbers I had collected about foreign exchange became handy. I knew how to deal with large data sets and how to look at large numbers. And that's how we came across this value-at-risk system. Till, what years were these? Because obviously these days, it's unimaginable to think that there would ever be trading,

let alone in large-scale trading without a very significant quantitative or numerical bent. So when did this sort of transition start to happen? We started with looking at foreign exchange, which at the time we had about 15 trading rooms around the world and traded 20 currencies, probably five or six in volume. And we had limits around the world, which we set in terms of millions of dollars of dollar mark or millions of dollars of yen dollar or pound sterling you could take. And that was not very good thing to do because every time you invent, you want to give a new limit on a new currency, you had to set a new limit in amounts of outstandings. We decided we need a common measure of these limits and that measure was volatile. So how much of the value at risk model was influenced by the events of 1987 when you had, you know, the Black Shoals formula that may have contributed? to the 1987 stock market crash and really the first sort of systematic sell-off, I guess, in the market.

At that time, we were fairly well on the way of understanding how we wanted to do it, and we realized that a good part of the market wasn't, and we could explain why they weren't doing it the right way, because they were looking at normality of markets, which we knew was a reasonably good starting position to take, but not necessarily at the final. And so we could explain what happened in that instance, in that accident. And we stuck to our guns and said, well, in the absence of such extreme events,

we at least have something better than looking at. So can you explain, before we really get into the development of value at risk, walk us through a little bit more what risk management looked like in the old days before the sort of numerical approach? Because obviously risk management has been around for a long time. But how did people approach the concept before you started doing your work? I think you have to understand that risk management in the old banking days was mostly about credit risk. Can you lend somebody money and how probably is it that this person will give you the money back?

And then as the balance sheets of the banks got bigger and the interest rates started to move, a new risk came up and that was called the interest rate risk of banks. And that was if you were borrowing short term and lending long term, that is if you borrow overnight with a short term rate and lending for five years, you get squeezed when interest rates go up. And to measure that squeeze or that, asset liability management was developed in the 60s and 70s. Asset liability management was something which was long term, that is you developed your risks,

or you looked at the risks over years instead of over shorter periods. And then in the 70s and 80s, trading started to really take off. There wasn't much trading in banking before that. And the trading was concentrated in Fornix. And that risk was really an overnight risk, would take a position during the day and then some took these positions overnight. And overnight you had jumps in, tried to figure out how much could you lose overnight. And when these foreign exchange risks became substantial in the early 80s, we needed a measure

to the credit risk in addition to the balance sheet risks. So as you're developing this model, I mean, walk us through what it was, what it was, was exactly that was going into it. Like, what's the data? We know that it looks at historical data and what are the parameters that it also involves and how did you agree on those parameters? That you have a trader who has a limit of $10 million in dollar mark, or let's say, dollar pound. And you would simply ask yourself, if that trader had that position over the last year over the last two years or last three, what is the maximum amount he could have lost on that position? And when you looked at the history of these positions over, you know,

you came up with a bell curve, eventual gains and losses. And said, well, the best way to explain or to put a number of value on that beltility or standard deviation in mathematical terms. That was kind of, you look at the history and see how much you would have lost if you had in a past and then you say the past is a reasonably good representation of what could happen in the future. And if you lost no more than a million dollars for 95 of the 100 days, then that was a good measure of what you could potentially lose. Now you have to understand that at that time, the entire portfolio management theory in asset management, as well as options model, which gained Nobel prizes for their respective inventors,

were all based on the same concept, that is, measure volatility in terms of standard deviation in a normal distribution. So it was a reasonable assumption to make that same assumption that was used also in asset management and in the options world. So then the obvious question is, and the criticism that you hear now and that you've heard for a while is, okay, all these measures are based on sort of a normal world, but we get fat tails and the world doesn't often look like a bell curve. And so then the question is, what is the value of the model in light of what we know about sort of extreme events? Number one, you have to understand the context within which the traders and management of trading operations who are operating. And the context was, we now have a measure, a reasonably good measure of risk. Shouldn't we compensate our traders based on the profits they make in relation to the risk state? That was a very new concept.

And that was really very helpful because if you had one trader who was trading pork bellies and made a million dollars in profits, and another trader who was trading dollar mark and made a million dollars, which one was the better trader? The trader was the one which had less risk to the million traders he made. So you now had a benchmark to evaluate traders with the same profitability. and you increased the limits for the trader which had the amount of profits. It was fundamental in better managing. And that in touch, if you now put yourself in the position of a trade, if I'm getting paid with a bonus, if I make a million dollars with 10,

how about if I make my position so they don't look so risky? Motivated very to take position high tail risks. because height directly with the standard deviation volatility measurement. So there was a, and people didn't realize that. There was a bias created by this measurement of create positions which were not properly measured. It was number one perspective. You see it happens all the time. Whenever you measure something and you pay people based on that measurement,

then they try to game the system. So that measurement created the bias of creating more, the second basic, the interconnectivity of the markets. The markets in the old days dependent. Exchange rate didn't change much when Japanese interest rates went up and down. Or the interest rates in Uruguay didn't change much when the sterling interest rates. The financial markets started to together, the markets became. interdependent. And there's a standard theorem in engineering which more interdependent the market system is, the less stable it becomes. And stability is basically the antipotan normal.

So the more you get around the world, the more you made them interdependent, the more you created non-normality. So in my perspective, the two things, the interdependence and speed of markets and the gaming by the profit makers. So that's a really interesting argument. And I hadn't actually thought about how this basically gave rise to risk-adjusted performance for traders. But when value at risk is most heavily criticized is usually during the financial crisis or in the run-up to the financial crisis for failing to foresee big trading losses on. things like mortgage-backed securities and subprime bonds and stuff like that, is your argument that

because of the models, because they were so entrenched, traders were sort of clustering into things that didn't look risky, you know, things like AAA-rated portions of synthetic CDOs, but that actually were exposed to significant fat-tail risk, and also that the financial system was more interrelated and so you had correlations that the model wasn't necessarily capturing? Is that the criticism? Relations, it means that the correlations were not normal. The correlation is just a statistical measure. And again, there you assume normality in the correlation distribution.

But the more interdependent and the more non-normal depositions are, the less these standard measurement of volatility and correlations. I'm June Grasso, inviting you to join me for the Bloomberg Law podcast. Every weekday, we help you make sense of the legal stories that shape the nation and the world. Listen for complete analysis of the biggest court cases, the latest actions from Congress and regulators, and the legal moves driving the markets, from corporate law to constitutional law and from state courts to the Supreme Court. At Bloomberg Law, we go beyond the day's headlines.

We speak with top attorneys, judges, scholars, and policy experts to break down what the rulings really mean. We do this every weekday, then bring you the best conversations in our daily podcast. Search for Bloomberg Law on YouTube, Apple, Spotify, or anywhere else you listen. On the East Coast, listen as you start your day. And on the West Coast, catch up in the evening. That's the Bloomberg Law podcast with me, June Grosso. Subscribe today wherever you get your... podcast. So is the idea, let's say I were a foreign exchange trader and I was trading the euro and I was

trading the yen and the Korean wand and so forth. And you might look at the past 100 days or three years or five years of performance of each of those and come up with some sort of reasonable expectation that, okay, the wand falls this much and the yen falls this much and stuff like that. But the idea being that what happens now or what happens in intercorrelated markets is that they all move dramatically on the same day at the same time. And so they're not just sort of random and normal, but the performance of all the different positions in the books go to extremes at the same time. Is that sort of the idea of like how the model breaks down under more extreme correlations or more extreme interconnectedness? That's correctly of the system which makes these. You have to invent positions which don't look so risky.

We're taken. We're called long-dated forward options. That is, a trader said, well, dollar yen usually moves within 10%. How about if I make a contract, which says, I'll sell you an option that the yen doesn't move more than 50%. 50% over the next 10 years. And if you take that position, you make money every day, except when the shtes the fan. And that's when you lose really big. So this is a typical kind of position doesn't get picked up correctly.

And the traders were biased to take these positions because they got easier limits for those and could make more profits on a risk-adjusted basis. Why doesn't VAR pick up that risk? Because VAR is a, you know, it's a simple measurement which depends on the normality of the distribution. And if you didn't have that simple measure of, you couldn't do the math. It was simply, the math becomes away from normal, not feasible. So I have a related question. You walked us through how the traders would game the VAR.

system. How did value at risk actually fit into risk management at the banks? Like, how were the senior risk managers using it? How valuable did they find it as a risk management instrument? And what actually happened to people if they breached their var limits? I've never been able to get to the bottom of that. Everyone seems to have a different answer. Well, 15 report, which was every day at quarter past four in the afternoon, which would show the risk around the globe in all our trading positions. And with that report, we went to the chairman's office and said, well, here is your risk. And the chairman at that, yeah, that's like a nice number, but, you know, I don't really

believe so much into it. Perhaps it's not right. And he was rightfully skeptical. The number proved out to be a reasonably good assumption. Everybody became more. You could show the actual loss in the global positions of JP Morgan was only five in five out of a hundred days more than that number showed. So that was the verification of the calculations. Understanding that this is not the maximum you could lose.

it's the minimum you could lose in a bad situation. So in 95% of the time you would not lose more, but in the other 5% of time, you would actually lose more. So is part of the criticism of VAR a sort of mischaracterization of how it was used or perhaps present, in your view, a cartoonish view or a naivete that never actually really existed? I think that naivete didn't exist in Lottom, but there were many others. But then as the VAR became kind of a government regulated, a central bank regulated standard, of course, a lot of people started to use it without really understanding what the limits were. It's very much like

you drive on a highway, 10% or 20% above the speed limit, and you feel happy because you're faster. And then you see a big accident and you slow down. People, as they perceive it. And when somebody says 60 is the right number or var is the right number, then nothing bad happens much worse. And then when something happens, then they kind of rein themselves in and they go relax again over time.

So till in the aftermath of the financial crisis, we did see some banking regulators who tried to alter var models. They tried to make them more robust. You had things like stressed var, which was supposed to be better at measuring the fat tails in the probability distribution. You also had other models like expected shortfall. How useful do you think those changes actually are to value at risk? Starting point about them. Then you have to do stress modeling.

and you have to do simulation under extreme circumstances. So these are all very good further developments, particularly if you have a bias in the markets that runs against you. The bias are, as I mentioned before, the complexity and the trader motivation. So, Till, the other thing that we've seen happen since the financial crisis is that volatility trading has sort of become a thing in and of itself. or at least on a scale that we didn't necessarily see before 2008, 2009. Given your background in modeling volatility, what do you think about the explosion in volatility trading,

you know, retail investors buying and selling things like exchange traded products tied to the VIX? Is that inherently risky or does volatility not necessarily equal risk? No, I think like all other trading, particularly if it's done in retail, is done by an income poops who don't understand and the financial institutions are very happy to provide the casino environment because they can make money of it. Altility trading is like any other kind of trading. It's a little bit more complex and more. It has a better story to it.

It's just another way of speculation. It's a more sophisticated. So Till, clearly you're not still working at JP Morgan. When did you get out of finance and what are you up to now? Well, I got out of finance in the early 90s, mid-90s, when my at JP Morgan was stalling. And I didn't think that was right. So I moved out and came to Silicon Valley to work in a startup in financial technology. And I was very lucky that startup turned out very well.

me then to do other things like building a vineyard. That does sound like a pretty awesome turn of events there. Seen come a farmer in the first place. Are there any similarities between modeling risk at a large investment bank and growing grapes in California? An equal amount of both. We certainly am surprised how little we understand about how to grow good grapes and make good wine. It's all in art. I apply to it, the less certain I am.

I understand what's going on. You collect a lot of numbers, and you hope you get to some insight, and by the time you think you have some insight. I love that answer because it sort of blows up literally everything we think we know about the modern world, which is that all these old practices that people do, we could just perfect them more if we really apply some data or AI or machine learning, and that is your answer. sort of just undercuts the entire thing.

We industrialize the world. You know, we have to put numbers on things because you can only manage by numbers. And there is limits to how much you can, but in the wine industry, the world has become industrialized. The bulk of all the wine that's drunk. And a very small number of small producers getting older stories, but they don't, almost 95% of all the wine 10% and it was exactly the opposite 100 years ago. We'll have to do an odd lots episode on the changing wine market at some point, but I think we'll

have to leave it there for now. Till Goldemann, thank you so much for joining us. Really fascinating conversation. Thank you for having me. It was a pleasure. Oh, Till, wait, before you go, what's the name of your winery so people can look it up? It's called Chateau Heitzich Haiz, which comes from Swiss-German and means there. There is no chateau here. Perfect. Thanks, Till. So Joe, I found that conversation really, really interesting.

Again, I think it's great to actually go to one of the foundations of volatility as we currently understand it and talk about it today. But there were also things in there that I hadn't really thought about before, like the notion that because you had this model, you had a bunch of traders who essentially tried to game it by clustering into things that they didn't think the model would pick up. I thought that was fascinating. I had not thought about that either. And so now that's a whole new sort of avenue of thing.

I want to think about and explore. Also the idea that we look at VAR as this sort of regulatory measure, but it didn't start off that way. And so the idea that maybe at one point this was a thing that sophisticated people understood have limitations eventually becomes this thing that becomes a sort of, de facto measure of bank health inappropriately is a concept that also I had never really thought about before. And also, I think, sort of vindicates its usefulness, even with the well-known limitations that it has. Yeah, it kind of makes me think that the criticisms that you've seen

of value at risk, that they can't anticipate tail risks, you know, it's not really, it's not a great criticism of the model. Or what I mean is we shouldn't be criticizing the model. Maybe the thing we should be criticizing is the fact that we have these huge institutions that are so complex that you can't actually come up with any model that's able to accurately capture everything it is that they do and all the risks that that entails. Absolutely. And I think it's funny, you know, even though it was sort of a half-joking question maybe about the connection between wine growing and risk management, I do think there is a common threat of just humility. Like, yes, There's only so much we can know about things that will happen in the future. And that sort of commands us or requires us to not try to get too scientific about what could go wrong.

Yeah, absolutely. I'm going to go think about the usefulness of financial models and how much we actually know about finance over a glass of wine right now. And I am going to go back to my desk and do normal work. All right, fair enough. This has been another episode of the. Oddlots podcast. I'm Tracy Allaway. You can follow me on Twitter at Tracy Allaway. And I'm Joe Wisenthall. You can follow me on Twitter at the stalwart. And you should follow our producer on Twitter, Topher Foreheads. He's at Forrest T as well as the Bloomberg head of podcast, Francesca Levy, at Francesca Today. Thanks for listening. The news doesn't stop on the weekends. Context changes constantly. And now Bloomberg is the place to

stay on top of it all. Gura. Join us every Saturday and Sunday for the new Bloomberg this weekend. I'm Christina Rafini. We'll bring you the latest headlines, in-depth analysis, and big interviews, all the stories that hit home on your days off. And I'm Lisa Mateo. Watch and listen to Bloomberg this weekend for thoughtful, enlightening conversations about business, lifestyle, people, and culture. On Saturday mornings, we put the past week's events into context, examining what happened in the markets and the world. That on Sundays, we speak with journalists, columnists, and key political figures to prepare you for the week ahead.

Join us as soon as you wake up and bring us with you wherever your weekend plans take you. Watch us on Bloomberg Television. Listen on Bloomberg Radio, stream the show live on the Bloomberg business app, or listen to the podcast. That's Bloomberg this weekend. Saturdays and Sundays starting at 7 a.m. Eastern. Make us part of your weekend routine on Bloomberg Television, radio, and wherever you get your podcasts.