We're meeting for lunch at a restaurant in Canary Wharf, where many of the major global banks are located. He is a man in his late 40s, inconspicuously dressed, and in possession of a firm handshake. He orders a Coke, and then a pasta dish he will dig in with great relish. In his volunteer email he said he was with a software firm (working in investment banking). When asked for a job description, he simply says he is a "quant".
"My parents discovered that I was of a mathematical bent aged three when I was apparently lining up my toys in order of size and then colour. I was one of these terrible, precocious kids who did their mathematics O-level aged 12. After a long academic career I ended up doing theoretical physics for my PhD, and spent a couple of years at Cern in Geneva. Many people I know from back then are still at universities, doing research and climbing the slippery slope to professorships and fellowships. They work the same astonishing long hours as I do, yet get paid a fraction and, from a purely scientific perspective, get to do some really, really interesting science. I often say (only half jokingly) that I "sold my soul" – I make a little over £200,000 a year, including my bonus.
"I am in a world of data, and I build all sorts of models for banks. For instance, one that helps a bank decide whom to lend a mortgage to. You have all this data about the person who is applying, and then the model works out the risk of lending to that person. You look at both the probability of this happening, and at the size of the loss in such an event.
"No model is perfect and before the current crisis banks might allow for one in five decisions in residential mortgages to be potentially wrong, for example. In the crisis many got burned and the true extent of the losses that will be suffered by the banks will only become clear in perhaps five or 10 years' time, and now they might want the failure rate to be only one in 25. So you need to adjust the acceptance level of the model (same model, same predictors – irrespective of the economics) – but a risk that you might take on the margins before, you wouldn't take now – hence the relative closure of the residential mortgage market.
"By the way, you should always have people doing random checks on the computer models decisions. One should never leave a self-learning algorithm to make decisions on its own. This isn't a worry about the algorithm becoming a character from The Terminator – a model simply needs constant testing; in fact, this testing goes on throughout the model's life.
"This week I am working on a specific fraud detection model. The key element here is the sheer volume of data. When you pay something with a plastic card – either credit or debit – a whole range of data is created. In statistical terms each of these data items is considered an 'observation', for instance, who you paid, when and from where, whether the transaction was swiped through a machine or was a chip and pin transaction (and whether you got the pin right!), the amount, the currency, the exchange rate, the type of business the merchant is in…
"The model I'm doing this week, I've got perhaps 30m individual card transactions and multiple observations per transaction like this, and I'm building a neural network to identify fraudulent transactions. The key here is to ensure that it's a self-learning algorithm which has the advantage of reducing the false positive rate – a false positive is when a model suggest that a transaction is suspicious, you check it and it proves not to be.
"Maybe the best way to describe a neural network is say you buy your wife a £30 bouquet of flowers every Friday at 1pm in London. Now if you do the same thing next Friday, the model will begin to understand that this is normal behaviour for the card. If you then suddenly try to withdraw €3,000 from an ATM two minutes later in Mozambique, the model will generate an alert which an agent in a bank will then investigate and contact the cardholder. The cardholder is protected from anxiety, the bank is protected from loss and – assuming that the banks fraud team are on the ball – the person who has fraudulently obtained the card details might find themselves coming to the attention of the police. Everyone's happy.
"From a development perspective how it works in hyper-simplified terms, is that I have a subset of data points from confirmed fraud cases and build this as a portfolio model: from this, the neural net finds patterns in individual cards and builds a model that will itself learn to construct an ever more sophisticated pattern of clients' behaviours with each activity the cardholder takes.
"The most difficult thing in this work: making sure the client knows what the hell they want; you'd be surprised how often banks don't really understand what they want from their data to the nth degree: for example, they might say 'I want to reduce default' – well great, but how do you define default and what portfolio do you want to use for this, are there any exceptions from the modelling sample?
"We live in a world of enormous amounts of data. The number of approvals for payments that a major global bank processes on any given day can run into tens of millions. Do we still have an overview? For the non-financial data streams, it's mainly controlled 'by exception', as the expression goes. What this means is that the data streams themselves remain essentially unseen, but there are all sorts of built-in checks that throw out exceptions, which generate alerts in case of something unusual.
"In their business processes banks pre-define what legitimate data is, and these pass through the streams unhindered (such as a field in which the value will always be a number: in the event that it's a letter then that's thrown out). There is a risk that data seen as legitimate may prove not to be so, true but there's an enormous amount of work being done in the financial sector to prevent such a thing, though and so I wouldn't recommend being too paranoid about it!
"Also, there have been massive redundancies recently. Usually when people are fired, in the unlikely event a scam had been in preparation, it would fall apart, or whistleblowers come forward. That this hasn't happened suggests that the models are OK. That's not to say that I'm giving anyone a hospital pass – and I'm certainly not creating a hostage to fortune.
"I said I was a quant, derived from the word 'quantitative'. We're the people of a definite mathematical bent, and if you're looking for a warrior-like analogy, we are perhaps the "armourers" of the financial industry, or, let me think … Traders are the warriors of our world; they go out and fight. I think of them as 'egos on legs'. Sharp suits, looking very smart… We quants are the trader's brain. It's our model that defines not only the risks the trader can take, the model also calculates how much risk he is taking with his particular trades at any given moment and we also predict future movements in valuation, pricing and the like.
"Philosophically, mathematics is a common language by which to describe the world. It's a language I love, almost too much for my own good. If I look out my window and I see three boats coming down a river, I am going to calculate without thinking about it how they'll avoid collision, which will pass who, when and where. In traffic I drive my partner insane by doing manoeuvres that she finds terrifying, because she hasn't made the calculations about other cars' speed and direction that I have. In airports I study check-in or passport lines for a while, dividing each into subsets of people, analysing them for the presence of children, lots of baggage, single or travelling together … Only then will I chose which line to stand in. Most of the time (>90%), I'm right about the queue I choose which gives me a quiet, albeit, somewhat geeky, smile of self-satisfaction.
"I have been in banking for over 20 years, and for several years I was with one of the major international investment banks. I discovered that I am just not enough of an arsehole to make it there. Why the top people at investment banks are like that? Well you have a thousand vice-presidents vying for 10 managing director posts. What do you think will happen? People will do anything to get ahead, back-biting, back-stabbing, the whole nine yards. For those of us who find life surrounded by other people difficult enough as it is, the requirement to network is hellish.
"Is there a point about banks I would like to see more broadly discussed in public? Well, about the current crisis. First, banks were falling over themselves to lend and I don't remember much complaint or criticism from the people the banks were lending to? This was not just the banks' fault, so if someone were to think 'hypocrisy' about the attitude of the public then – yeah, I'll buy this.
"Furthermore, I think it's a mistake that after the bigger items of the crisis that nobody was sent to prison. That would have sent a signal, especially if some of the more overwhelming acts of avarice (ABN takeover by RBS, for example) were punished not only by the failure of the bank but by there being criminal charges levelled against the CEO and chairman of the organisations concerned.
"You ask why there hasn't been more of an internal revolt, after so many people inside finance who had nothing to do with the causes of the crisis got burned? I'd say there is a great degree of coalescence. People may whine and complain but they get things off their chests and have now morphed into this amorphous blob. It's a group that sticks together. Call it fatalistic. It can be over any moment so I am going to try and grab what I can.
"It can be dangerous, having a mind for data. If you want an example, see the film A Beautiful Mind about Dr John Forbes Nash, Jr, the former 'ghost of Fine Hall' at Princeton. For my part, I'm nowhere near as smart as Nash, but I find it hard to switch off. A few years ago I found myself taking advantage of psychiatric services; as a consequence of which, in my case, I'm taking a large dose of medication on a daily basis to try to stabilise myself and I'm conscious that I sacrificed a marriage on the altar of work. It's particularly hard for me, with mild Asperger's, which in my case wasn't picked up until adulthood. Suffice to say, it's very tempting to just stay in the world where everything can be understood in mathematical language.
"Not sure though that I'd voluntarily swap IQ points for EQ – even though I'm certain that I'm going to end up as one of the single old blokes that you might occasionally come across – nice, big house in the country, lots of dogs, materially comfortable and yet utterly alone and mad as a fish.
Later, when asked to elaborate on that final point, he responds via email:
"I've long been aware of the prospect (with some 'tongue-in-cheek') of becoming mad as a fish, and the attractiveness of the current imbalance between EQ and IQ is that I know that my biggest, deepest fear is failure. With the current imbalance, I know that the risk of failure is reduced to its current level: eg, small but still real. That fear of failure drives me and means that I know I'm giving up anything approaching EQ in pursuit of avoidance of failure."
Comments
1 December 2011 2:55PM
My mate Dwanye never did Maths at school, come to think of it he never went to school much at all, but he could cut you a quart. of 'coocha' when he was 9, by twiddling his hand around in his left pocket without weighing it. Now thats precocious!
1 December 2011 2:58PM
Interesting article. Just out of curiosity, is it possible to model a systemic crisis based on who you lend to, policy initiatives, political scenario et al assuming similar actions by every similar bank? If it is possible, that's the kind of model major goverments and/or central bank should invest in.
1 December 2011 3:05PM
Macho bollocks. Anyway, am I alone in finding this article desperately sad? Whilst his erstwhile colleagues at Cern are busy making discoveries that could fundamentally change our perception of the world around us, this guy is taking medication and worrying about dying unloved, mad and alone.
1 December 2011 3:05PM
This comment was removed by a moderator because it didn't abide by our community standards. Replies may also be deleted. For more detail see our FAQs.
1 December 2011 3:09PM
Disappointing. No discussion of how the assumptions of risk models are checked and what happens when they're being broken, such as with the CDOs during the unprecedented housing bubble. He did work at a major investment bank for a few years, he should have some insight into this.
1 December 2011 3:12PM
Speaking as someone who works in the quantitative finance field, (although I wouldn’t go so far as to describe myself as an actual quant), I have to say that people probably only hear about mathematical financial models when they go wrong. In fact, they have become pretty much normalized in business, an everyday such models are used for the pricing and risk assessment millions if not billion of financial transactions, without anything going awry. However, when they do go wrong, they can go spectacularly wrong, just as they did in the last credit crisis.
Why is this? Two reasons I can see, either a) the quants who developed the model knew of the flaws on their models, but were happy to keep quiet as long as the 6 figure salaries and fat bonuses were rolling in, or b) they genuinely believe that this time, their models had everything covered, all eventualities were accounted for and that nothing could go wrong.
So basically the reasons were either a) greed or b) hubris, which goes to show that either the brightest are subject to the same fatal human flaws as the rest of us – something which, paradoxically, I find quite comforting.
1 December 2011 3:17PM
The investment banks rely, for a large part, on the ratings agencies to assess risk. Unfortunately, getting a job at a ratings agency isn't seen as sexy (or as well paid) as working for an investment bank. The fact that working for a ratings agency should be seen as one of the most responsible, important positions in the industry doesn't seem to compute and leads to them being pretty useless at their job (hence triple A rated toxic debt).
1 December 2011 3:19PM
You’ll probably find that the Cern guys have the same problems, as they tend to be caused by being a geek, rather than by being a quant. And anyway, as I implied before, mathematical financial model are probably beneficial to the economy as a whole, so people’s lives are being improved as a result – and that’s nothing to be ashamed of.
1 December 2011 3:20PM
You think that working out a way to protect people from fraud without blocking their credit cards every 5 minutes is a waste of time?
In a world where we are producing ever more data, making sense of it is vital.
Clearly, there's a bit of geek in me as I found this fascinating.
1 December 2011 3:23PM
Brother bankers... there's strength in numbers!
1 December 2011 3:23PM
Fair enough, but he does admit that he's in it for the cash and work ruined his marriage, which seems to be a bit of a recurrent theme in these blogs.
1 December 2011 3:24PM
Good interesting article.
Good to see a geek being paid well for doing geeky work.
1 December 2011 3:27PM
Brilliant, fascinating, guy after my own heart ... more like these, please.
1 December 2011 3:29PM
To ensure that "nothing could go wrong" in the field of risk assessment surely you would have to eliminate risk itself. I do think the tools can be misused to engender a false sense of security though. I once saw a Monte Carlo simulation being run on a risk model to produce probability outputs to 30 decimal places of precision. What's the point?
I thought this was one of the more fascinating contributions to this series - thanks to whoever provided it and good luck to you.
1 December 2011 3:34PM
To be fair to him, what he is doing is actually very useful. From a certain perspective, it is more useful than what is going on at Cern, because what he does impacts countless people very directly.
There is something about this attitude that smacks of "holier-than-thou". He seems to enjoy what he does, so who are you to say he should have done something else?
On the other hand, one wonders whether it was him or people like him who screwed up pre-crisis - it was lending to people who should have been denied that caused the problems, and so you would assume the algorithms used to assess lending were wrong...
1 December 2011 3:38PM
Not quite - the banks rely on the ratings agencies to put a kite mark on things that they sell, for the "benefit" of investors. The better the rating, the better for the bank. And as the bank pays, guess what the ratings agencies tend to do? Give them lurvely AAA ratings.
The banks would never be so thick as to invest their own money on the basis of the credit ratings. What kind of dumbarse would do that?
1 December 2011 3:40PM
I'm usually first to stand up for someone's choice of career on these blogs. In this instance, though, he really doesn't seem very happy. As PatDavers has pointed out, that's probably more to do with an underlying condition than his employment, so point taken.
1 December 2011 3:44PM
It is an understanding.
If you model a world to be understood via mathematics, the real world has to conform to the model, inverting what is important.So when you state: 'mathematics is a common language by which to describe the world', you are not accurately describing that numbers are relational abstractions. Wittgenstein's view would be that the 'axioms of Mathematics' as a whole are a set of rules we have found that covers all the rules of all the 'games' under the name 'mathematics'.
1 December 2011 3:47PM
He has mild Aspergers. Many very intelligent geeky types do. It takes 2 people to ruin a marriage.
I find most people who work are only in it for the cash. The people who love their jobs are the truly lucky ones.
I have met a similar person to this, but he is penniless because he will not "sell his soul to the devil". His brilliant mind is wasted and he is unemployed. He just does not get on with people.
1 December 2011 3:56PM
While most are the quiet type, not every quant has Asperger, or worries about QE. Likewise, not everyone is a wannabe mathematician or physics researcher, many come from from economics/stats studies and happen to be genuinely interested in what they do.
1 December 2011 4:00PM
Maths isn't physics it doesn't always translate to the real world.
In my first day of Physics my maths teach said that in maths if you drop a pencil it never hits the ground because it has to fall half the distance left to fall before it falls the whole distance and when it falls that half distance it has another half the distance to travel before it falls to the floor, repeat to infinity never reaching the table as it has another half distance to fall but it does hit the table. Also three thirds isn't 1 it's 0.99999999999.....infinite
Mathematical modelling has been shown to be complete nonsense, but was simply an excuse to confuse people with something they didn't understand so they could scam more money and lie about the risk.
1 December 2011 4:00PM
Will the subject be responding to this blog? He mentions a model for assessing mortgage lending. Was this created during the subprime housing bubble with brokers who rendered underwriting standards a joke? If so, did he have any power to denounce the models as garbage?
1 December 2011 4:08PM
Everything can't be "understood" in mathematical language. Maths can be used to model and approximate reality - sometimes quite accurately - but it is not reality.
Models contain assumptions and approximations always differ from the actual results by some amount.
I have had a bit of experience of dealing with Quants (brilliant people who I wouldn't trust to run a Whelk Stall) and the big problem is trying to discover the unspoken assumtpions built into the model. For example, until it happened no-one thought Lehmann Brothers would go bust. Trades with them were done on the basis that they would always exist as a credit-worthy entity.
1 December 2011 4:14PM
0.9 recurring is one.
1 December 2011 4:14PM
those models are being examined rather carefully (and have been for few years now). With benefit of hind sight it's quite interesting looking back day by day (starting from let's say 2003) and going through the data inputs into the model and comparing the results with what is known today. Infact most banks have significantly expanded their risk teams, and in some cases such as RBS, it seems the pre-crash risk teams were always overriden by business/trading, and lacked any say (as well as talent).
This guy does not work in investment banking, credit card fraud (what he is modelling) is business of retail banks.
1 December 2011 4:15PM
You're wrong here. Your mistake is that you're assuming a third is 0.333333 recurring, rather than the more precise definition of a third as 1/3. If you definite correctly, then by definition, three thirds is 1.
1 December 2011 4:17PM
As a Quant specialist myself, I don't feel anything like i've sold my soul to anybody.
We live in an increasingly complex world where prices of everything fluctuates every minute (or even every second). Part of a highly developed and civilsed society is the ability to hedge risk appropriately and transfer them to people who know better.
1 December 2011 4:17PM
What utter tosh.
Actuaries have been successfully calculating risks for insurance companies for at least 200 years.
1 December 2011 4:19PM
thetrashheap
You had a terrible maths teacher if (s)he left you perpetually confused about Xeno's paradox or 0.999... = 1.
1 December 2011 4:20PM
*Zeno's. Whoops!
1 December 2011 4:23PM
The global financial system was so over leveraged in 2007-08, that all models of risk broke down.
The Lehman crisis was not the cause of the disaster, but just one of the many triggers that could bring down the unstable system of the most rapid growth of credit ever seen in the world from 2001-07.
1 December 2011 4:26PM
Most insurance companies are a scam that never want to pay out anything and always find a way to wriggle out. I had to go to the ombudsman to get my insurance company to pay out £2000, which is less than 3 years premiums. Despite claiming, in 10 years they made over £3500 from me.
1 December 2011 4:30PM
I like these finance 'interview/articles'. Almost feel sorry for this guy. One day he's going to wake up and be like this:
http://movieclips.com/usLo-the-bourne-identity-movie-why-would-i-know-that/
1 December 2011 4:30PM
Anecdotal evidence?
I have had mostly good experiences with Travel insurance, Health insurance and car insurance in my time.....
The service has been excellent with only the odd blip.
1 December 2011 4:42PM
Fascinating piece, the best thing on this site today.
This guy does not mention that the money he is earning makes him happy, and does not appear to particularly value it. Surely he would be much happier and more fulfilled back at CERN?
1 December 2011 4:49PM
Yeah, maybe (I would like to think so), but he also highlights his deep fear of failure. Perhaps in CERN he found himself surrounded by lots of other equally highly intelligent scientists; and thus personal success, if measured against the Jones's, was harder to achieve.
1 December 2011 4:56PM
It appears we do not live in a highly developed and civilised society. It seems the quants create risk models that are picked up by traders who then ignore those who know better when the assumptions of those models are broken.
1 December 2011 5:01PM
You aren't necessarily meant to get back what you put in. That's how insurance works! The whole point is that, for most people, they pay more to insurance companies than they get back. For some people, if something goes wrong, they get a lot more back than they ever paid it (to cover a loss that they suffered and the others didn't). It is a pooling of risk. Or, if you like, a transfer of risk to the insurance company in return for a fixed payment. It is the nature of risk that the bad event may never occur, but you have the safety of knowing that you won't suffer a big loss without being made good.
Assuming they do pay up when they're supposed to, of course. But that's a separate question.
1 December 2011 5:09PM
Nobody could really argue with having a sound 'maths'base to banking plans.
Unfortunately some greedy bastards are not happy to stay within parameters worked out by 'quants' and others such as our friend in this piece.
The problem is that some bag of shite will eventually think it a good idea to mix up a financial instrument to include 125% loans to dodgy creditors and flog the resultant package as AAA*.
1 December 2011 5:14PM
After the creativeness of the accounting I think you'll find you're living in the remote past.
1 December 2011 5:35PM
I have to say that my experience of insurance companies over the years has been pretty good too.
When I have claimed on travel, house or car insurance - it's always been paid in full and pretty quickly. A few years ago my mother-in-law's car was written off in an accident. The value they offered her for it was higher than I would have advised her to accept! Simialrly. my son's small car was written off after an accident in last year's snow; again, I though the valuation was fair.
1 December 2011 6:17PM
That's called 'good luck' in the real world.
You can't buld a model to price risk, and mathematicans know that. I used to do 'risk' modelling for industry. It wasn't called that then, but it would be now.
1 December 2011 6:21PM
Jesus wept. You have all this 'unreliable' data about a person you have never even see, let alone talked to. You don't even check that they are employed.
Building societies used to do this at little cost. The number one piece of data was the contents of the savings account.
The next was local knowledge.
1 December 2011 6:27PM
Wrong way round.
If you think one in five will go wrong, then you need a lot of capital to back that up and you lend cautiously.
If you think only one in 25 would go wrong (and they actually believed only 1 in 33 would go wrong) then you are asking for trouble when 1 in10 go wrong.
You should check with reality and the people in the banks had no connection to reality.
1 December 2011 6:31PM
Which bank is asking for this? The Bank of Toytown?
Yes.
1 December 2011 6:38PM
No they haven't. The risk was always calculated to be on the winning side. They never had to caculate the real risk.
If they had calculated risk succesasfully, insurance companies wouldn't be surprised that we are living longer.
1 December 2011 6:44PM
Who would have thought ? Someone who works in the world of finance that I have no desire to do anything unpleasant to. The lad needs to stop chewing himself up, find something he actually likes doing. If I had the smarts I would be camped out at CERN for a 100 + hours a week. Security would have to beat me with sticks to get me to go home.
Sleep well in the (£200.000 a year doing something that kills me so that when the time comes I can move out to the country, hide in a corner & collect my own urine in jars - bliss) fire.
1 December 2011 6:48PM
Sounds better than my bank, which has in the past decided that ordering foreign currency from its home delivery currency service, and then a week later using a bank card in the country where that very currency is legal tender was highly suspicious, rather than evidence that I had, say, gone on holiday. They even tried to defend this when I challenged them on the phone, saying it was "a pattern commonly associated with fraud".
1 December 2011 6:48PM
And you seem overawed by it. It's only a lot of simple data that needs a bit of computing power to analyse in a fairly simple way.
There's all kinds of fraud you can't possible detect.
Buying flowers in London and withdrawing money in Mozambique minutes later is not difficult to detect.
Buying flowers in London and withdrawing money in Mozambique could be fraud or not fraud. It could even have a 'reasonable' chance of being fraud, but all the analysis in the world isn't going to help. People don't behave in patterns.
1 December 2011 6:55PM
Your bank was probably using a model like the one being developed.
If you'd ordered a lot of currency and were then using a bank card in a country noted fror card theft, then they were probably right to check.
My credit card company rang to check my last four 'normal' transactions a few weeks ago.
They cancelled the card when my wife ordered flowers from a US florist, shortlly after we'd been there. A bit cross about that, but that's what I expect models to do.