Nassim Taleb: Keynote Address, 2015 Fletcher Conference on Managing Political Risk

good afternoon everyone I hope you all enjoyed lunch and your respective workshops thank you to our workshop leaders for leaving those enlightening sessions I'd now like to introduce our moderator and discussion leader for keynote lecture Nadeem Sajadi who is the director of the Farris Center for eastern Mediterranean studies here at the Fletcher School of Law and Diplomacy mr. Shahada comes to us from London where he remains an Associate Fellow of the Middle East and North Africa program at Chatham House the Royal Institute of International Affairs in addition to his scholarly achievements mr. Shahaji brings more than 15 years of experience as an advisor to EU government European institutions and international organisations in drafting foreign policy and assistance strategies for the Middle East in North Africa and he is currently a senior member of st. Antony's College Oxford where he served as director of the Center for Lebanese studies from 1986 to 2005 as well as a member of the Executive Board of the centro de estudios state orientem medio de la fundación promoción Rocio de la cultura in Madrid thank you I practiced that thank you very much mr. shadi and he'll be introducing our keynote speaker okay Thank You Julia and wonderful Spanish pronunciation now it's a very risky business having to introduce the same polyp twice and to an audience that knows him so well and that's here so we all know that the theme has for his since spent 20 years as a derivative trader and in the in the city and then he repented and went to as I as I and went and started and became a philosopher and and a and a mathematician - - in a way to redeem himself so now his his books you all know them they've been translated into 36 languages but I also have developed a definition of political risk especially for for for this audience I think the definition of political risk is bringing Nassim Taleb to a roomful of economists political scientists and men with suits and ties I give you mercy here I'm actually even surprised to be invited never know so this is my lecture so I'm I just jotted down what I want to talk about and I realized so the first thing let's talk about this data data analytics data that kind of stuff okay and see and talk about the unpredictability by following the logical steps okay how many of you think that you can use data to predict you know systematically what's going on all right very good the are you aware of something the field called finance right how many correlation did we have in 2007 and finance before the crisis in your opinion the correlation matrix of finance to tell you how much data we had we had data and compute covariance matrix of the world how much how large do you think it was how big do you think sorry how much yeah how many correlations how many many okay keep going all right between half a billion and a billion correlations we have in finance data is not what we needed okay what we wanted and effectively I went and testified in Washington against spending more money on data like you know to get more confused by data and the Congress person actually the congressman he told me uh you know was a little bit of scorn mister Talib when you cross the street you don't look at stuff but unless exactly what I do when I cross the street I remove data because you have trillion pieces of information and all I'm interested in crossing the street I'm not interested in the eye colors of every driver or in the chemical composition of every item around I'm only interested in large items that are moving okay so he pretty much is called via negativa so the the and and the more data you put in of course unless you're going to see that anyway so let me first I'm gonna present the problem that they've had in economics for ages still not cured for reasons that I'll introduce called skin in the game like ask in the game in other words not the people paying for their mistakes and then we discuss the notion of fragility in place of predictive analytics because fragility is much more measurable and much more rigorous an approach than trying to just confuse people by making prediction that cause them to take more risk so let me start with the notion of fat tails okay you randomly select two people from the world and you end up they their total height is 4 meters and 28 centimeters that's in total okay for two people what is I don't know it would be in she's very tall okay the total what's the most likely combination you're gonna have one person is three meters the other one is 1 centimeter one meter and 20 no sorry okay so the two of them has to be the lot of deviation the two of them has to be two meters and 14 centimeters each that's very rare but any combination that's not this so give you the total would be even more of a rare event okay that's simple now you randomly select two people on planet Earth and they have a net worth of 42 million dollars okay what's the most likely combination you have between these two let's give me numbers most likely sorry most likely okay so it's gonna be 41 million nine hundred and ninety thousand and and 10k alright that's most likely combination all right this this explains the problem what I call fat tails now so can someone based on this define fat tails a rough sort of intuitive mathematical definition typically a good mathematical definition can be expressed with simple examples of course but can someone formalize it give me what is fat tail and then we can identify it and see the consequences of that yes you don't have higher other extreme outcomes you have lower odds of extreme outcomes all right for example if you're in a very very fat tail where the fewer rich people but those who are rich are very very rich so effectively the odds of deviations drop which is the biggest mistake in antenna fiber is it like when I was talking when I wrote a book called the Black Swan everybody said tala the same bats one are more common as they know they're not more common the more consequential it means they're less frequent but more consequential if you talk about inequality all right if one person has all the wealth that would maximum fat tails all right one person has all the wealth then you wouldn't say well there are more people in the tails there's only one person in the tails in fact the the so can someone give me the definition of fat tails and different one and another one no this is so technical and correct but technical something more fundamental actually that's helpful okay when you have a lot of data most of the properties are explained by the smallest amount of observations okay in other words we take how many days you've been working on Wall Street these are the variations for banks for example one day explain the bulk of the volatility you say maximum fat tails one day explains everything you say as if we take for example common examples in finance when you look at finance the world of finance almost everything is fat-tailed in the sense that small number of companies represent the bulk of the sales we have 800,000 approved drugs on the market of which hundred thousand are available worldwide to be purchased of these how many drugs represent the bulk of the sales very very small between five and ten of the profits even smaller take the book business you have a million manuscript on the market for novels in fiction most people who you know write books in America I don't know about Europe they usually call them they work for Starbucks you know so so that's a million million million authors of an English language of nonfiction novel of which 20,000 will be published and how many will represent half the sales how many books sorry good be numbers she said 10 she said five books are correct somewhere between 5 and 35 depending on if the lady from illiberal comes back or not all right forget the idea so this is pretty much so what's happening ok so fat tails is one a small number of observation cause the major effect on property okay something you learn in school this is my second lecture in this room I usually you know I come free to what you think I don't like her a lot so but when I was here I criticized something they teach in school called the law of large numbers which I told them automatically whenever you hear it used or you hear something called linear regression equated with all right and let me explain why because if you're sampling you need to be fast it doesn't work in a real world small number of observation deviations people would call them outliers determine all the properties if you're sampling wealth in America or in the world right you sample the first hundred million Indians all right not going to give you the average all right unless you hit on a few top people and if you know that the Bill Gates for example is wealthier than the bottom billion or something all right so you get the idea of fat deal so that's sort of what what I mean by fat tails and it has consequences all across they are domains like this one there are not fat tailed and that's the rule the rule is when you have large deviation okay the the the average and the maximum are close to each other okay and this one is when you have large deviation one observation typically will represent the bulk of you the maximum is what counts actually you can get most of the properties from the maximum or the second or the third or the first ten or first whatever but you can have a billion people you want to get the total wealth of the planet you get sampled one top one percent you get more than fifty percent okay don't tell the tax authorities because in fact numbers show about between 40 and 70 percent all right so it was it so you get the idea okay that's what I mean Dale and there's a super fat tail if you think violence small number of conflicts represent the bulk of people who died in in history for example okay and that's super fat deal you have two classes of thing death by knives and death by nuclear weapons or death by violence by big arms also sometimes on the contagion best buy knives can be like this okay now second point why the world has become is becoming more more fat-tailed something called I don't know if you've heard of it's called the globalization okay so you can imagine Google in the 18th century starting in a garage dorm and running the world okay it's not so small advantages winner if they call effects come from that and the Dacians with that that in ecology for example the numbers of diversity in nature is proportional to the the size of the island for exam or continent a large content will have many more species at an island but will have a lot more but they'll have a large content will have fewer per square meter so and you lose that diversity but and then of course now variables that we discover 2007 are becoming more and more fat tail and as we have fewer Christ but once it happened their deeper unpredictability that problem probably what's behind on predictability that you have nothing happening and then a big problem called the turkey problem and the turkey problem because the turkey is fed by a butcher for a thousand days and everyday confirms to the statistical Department of the turkey that the butcher loves turkeys with increased statistical confidence until that's giving minus 1/2 days I don't know what Massachusetts three days you have a big surprise for the turkey all right and here's revision of belief and then you can see the statistical machine didn't work that's pretty much what's happening on the fat tails we need a different machinery or at least you can tell yourself that the mean is not visible using conventional methods by sampling the means don't have the mean there's a book by a guy called Steven Pinker of a drop of violence and when we looked at data we realized two things one of one that he was wrong he didn't know how to compute the data and the second thing if anything is rise of violence but more concentrated so if take history is fewer and deeper that's pretty much fat tails and the conventional test statistic don't work for that okay now this is number one and this pretty much explain why if you're if you throw garbage after garbage of data it's not going to work but but but you can see extremes from data so if you can use data big data for extremes only things or for targeted thing you ask a question is he does he know anyone who has a beard without a mustache all right it's very good for terrorism but it's not good to predict the socio-economic events okay except then after the fact I mean it's like predicting catalyst all right you can't you know predict thing and then when you have a like a bridge collapsing people you know don't analyze in engineering these don't analyze the the colors of the last trucks that was honest that may have caused the collapse you look at how fragile the bridge is okay so this is so far my idea and here now let's talk about something skin in the game in moral hazard you've heard please call Wall Street or it's very generalized actually even outside Wall Street if you make a you know banks they like to make money steadily all right on their fat tails you don't see because I said a lot a lot of large numbers operates very slowly on the fat tails it's very very very slowly okay you need vastly more data to figure out if someone is really making money yet cooperation everybody they get a bonus something called a B if you make money for a year you get a bonus if you're right no so you make money you're one you get a bonus you're - you got another bonus you're three if you live in New York then suddenly all your jobs become funny alright make money three years in a row you get more and more money under management the bank gets bigger they expand okay and then we continue a lot of these and then comes a point when you have a turkey problem alright when hey you know what Thanksgiving one is to the equivalent for the bank 1982 banks lost more money than history banking and we didn't even have bonuses then at a time right more in a history of mice and the banking on one event 2007-2008 four point seven trillion before of course the taxpayer came to rescue them okay in an implicit way and of course here what people do is they write a letter saying the odds of this event so low that we can see it and surely there's a sentence surely is as much of a surprise to us as it was to you that was in 1998 when long-term capital management blew up and you saw it again in 2007-2008 with absolutely no linguistic evolution same same sentence alright now this I've generalized I'm generalizing this to the following any situation in which you had the upside without the downside you're invited to people statistical property we have hidden risks that blow up rarely and then using the Marco which Marco is all that people will give you here will tell you that is very safe when in fact it's highly risk and you cannot make these claims based on the structure of the portfolio so with time those who survive whether cooperation the minute cooperation go to the market they start developing this payoff visible profits and steady and hidden losses not borne by them I called that the Bob Rubin trade 120 million dollars collected from Citibank and when Citibank happened he said while I was a so unexpected he didn't return the hundred nineteen million he should return all right you know I keep one for for the drivers and stuff like that but no only twenty million okay so I call that I said you know John Gotti for example the Mafia they never made that money okay same thing happened recently after the crisis I don't have you heard JP Morgan the whale well same thing all these people are getting thirty million dollar bonuses on something that would you really was hiding risk and of course they lost it all and then they spun a story and people kept their previous bonuses no no clawback all right or minor clawback if it ever happened once in Swiss bank right so but this we can generalize to any professional ones that are more popular sorry don't have something I know you have your microphone now but feel free to interrupt if you're angry or something so know it okay so here we have a class of a thing and I call it no skin in the game if you have skin in a game at all times you don't have that problem okay this sort of explain this kind of setup explains the modernity modernity was people getting benefits from action and with the adverse effect not being affecting them because they're not visible they're delayed and they hit them later okay and you can generalize to a lot of situation in which say you're a bureaucrat you're going to do something that improves your year end job assessment but then you hide risk and then of course when thing blows up you see always unexpected okay and so you have an invitation to have steady thing rather than volatile and effectively based on this principle you're gonna see when you look at countries that anything anything that's very volatile when things are very volatile guess what the more stable one thing are steady they're very stable there's something people in finance are definitely I mean it's like a generalized fraud because if you look at a metric called Sharpe ratio which is average return divided by standardization of return the first thing the standard deviation return doesn't work under fat tails or at measuring risk the ones with highest the funds from Lehman Brothers that lemur the firm that went bust a Bear Stearns now remember the Bear Stearns the funds that the highest funds the ones that went bust were from Bear Stearns they have never lost money until they lost money pure turkey problem okay all right so now I've set up I've explained unpredictability I've explained fat tails explain why how people tend to position themselves on delayed blow ups if you have a job assessment no morals and no skin in the game hedge fund incidentally don't have that problem you know why why they're forced exactly he said it is that forced to a hedge fund manager is forced to eat his own cooking so they have usually fifty percent of their money in the fund and when they lose money they've lost any day more than fifty times what the next largest client has in the fund in proportion of the net worth it's the same thing of why and now skin in the game is very important because for example helicopter pilots the two-dimensional skin again the first I mention is what I call the crooks of randomness is the agency problem and that's sort of known in economics but not given that economics does know my fat tails there was a fat tail twitch dough but helicopter pilots for example in Brazil requested that helicopter maintenance people take random rice on helicopters and sure enough okay the thing to prove okay the but the dimension is skin in the game that I'm investigating now is completely different is evolutionary in the following sense a lot of people engaged in this there's a fool of the randomness they had the crooks of randomness you have the fools of randomness the full lot of people believe their own you see the idea like if you look at economists they believe their own stuff it's not like they're gaming the system and when a measure is wrong you know their objective function isn't you know that they're not penalized by their own mistakes so but in nature you don't have that in nature anyone who in dangers others all right okay you don't have evolution unless those one date or others are themselves the same bear the same risk so let's say if you have you been on highway there highways here to get here okay go on a highway any participant okay anybody could do it on the highway you don't need to be on a plane you can kill 30 people you can go wild and kill 30 people on the highway no you go against traffic because 30 people why aren't they that many of these and someone tell me sorry no it's not because they have skin in the game they could be crazy and not have skin in the game I'm not at the crooks of randomness agency problem ever the fools of randomness why because they're dead like this heli the guy who killed the hundred and some people he's dead so you can't have that higher ratio in the population of these people because they end up killing themselves you know if they won they kill other they come themselves as well you see the filter dollar system you see so so it's like entrepreneurs who make mistakes they're dead you see they're dead so this is pretty much what people don't realize as a filtering tool is that you you're because in an opacity in a very opaque environment sorry yeah but okay there are exceptions but it helps that they're dying as well and then the operation you see it does help but I mean of course you have exceptions Isis but typically traditionally the ratio of people who are bellicose and a population has stayed lower you know then then the level you need to blow up the planet because the nuts like alexander napoleon and these people would go to battle with stable he had frontline and battle Hannibal the nut complete nut if you look at it there's people say make him a hero is completely irrational fellow all right Hannibal was the frontline it was first in battle and that was the oldest so that was that's traditionally what has happened - I mean suicide bomber killed themselves so they filtered out they kill other people but I don't think that we can talk about suicide bombers as a real danger to the system as a whole given that the number of casualties coming from them is still very very minut every day in America 7,000 people die and they multiply by 20 to get the number of people with eye on the planet and count how many of these comes from suicide bombers and you realize that we're still talking about low-risk much bigger risk okay this is okay we're talking about skin in the game and as a filtering tool for people who endanger others okay well so I know we're not judging whether it is terrorism as dangerous or not I believe it is dangerous but there's another I believe it's very very dangerous psychologically and stuff like that I'm talking about the mechanism of filtering people in the pool you see as a mechanism of filtering bacteria but I was a bit trader and the magnums of filtering there was a mechanism filtering and and and and and and people would go bust with their own money they wouldn't survive hiding risk and effectively traders don't like to hide risk what happened is they they they closed this okay now all right so now I've given two points let me move into fragility this is what I was telling you why you cannot measure risk and effectively you do a very simple test and it's a very very trivial test you know when Greenspan went to Congress to tell them the crisis didn't happen sorry the crisis he apologized for why the crisis happened the tone of these events never happened before he didn't have nobody had the presence of mind among Congress people could tell him but you never died before and effectively there's a very simple test to do that the test is you take an axis of events tale events between time say t minus 1 and t t t plus 1 ok if the past predicted the future you would see tale events like this in fact what you see is a bunch of tale events here without successors and a bunch of tale events here without predecessors which tells us if you do a very simple in purses or which takes about 10 minutes on Excel for those who use Excel ok and about half a minute on Mathematica so if you can run that and then tell mr. Greenspan that it's a silly thing to say that the other events have to have predecessors it's just that the native journals typically they don't otherwise people don't don't get harmed by them so you can't use that as excuse you have to now look at the system and think like engineers I happen to be a professor of engineering which is the you know the only the job I ever wanted all right why because there's no in engineering all right so now let's move into engineering ok so I have just wrote a paper at a paper I wrote something for the RAND Corporation ok on and then it was a version of it dumped down diluted transformed journalistic fide how do you say destroyed edited debased debased what else disfigured of course well in appearance Foreign Affairs all right I know if you've heard of that place I had the misfortune of publishing that version there and let me tell you what we did with the thing so we start with the fact that people want to see what country is gonna blow up next okay so fragility theory is vastly more effective you look at a bridge you know what bridge is fragile or not okay and we have metrics to figure out what's fragile was not so the whole idea that was my DNA black swans people the nose is I'm not telling predicting these events I'm just describing the phenomenon and saying hey you know how do you get in trouble okay how do you build system that can resist any black swan and effectively that's my last book and everything I'm working on its fragility metrics of fragility because there's universality fragility so we define now fragility fragility as something that does not like disorder okay that's simple that's the finish and fragility we can we still don't have them to find the solar yet we can define this order okay how did you find this order you want to a physicist oh it's at the entropy I say oh I'm sorry you know it's confusing okay let's go you go to a mathematician who would define this order as Sigma all right or whatever scale of distribution you go to a finance person to tell you standard deviations and so okay we say all right we go to someone at say time time to makes things so you think about it think about it yes there is a metrics that's universal to the solar and there's a sensitivity to a disorder that's pretty much invariant there's beautiful thing in mathematics is that if you have a metric that's represented in some way it cannot become predicted in another discipline that approached it from a different mathematical method so they had to be unified and all you have to do is prove one you prove all the others there's a beauties mathematic cause differential equation or or integration or or or other technique so probability theory or engineer you get to the same thing definition doesn't like this order okay and looking at coffee cup doesn't like this order and what doesn't like the sort of have to have this property sharks you remember this graph alright sharks this is a coffee cup alright the health of the coffee cups you have sharks RockShox alright it doesn't benefit from shark doesn't have upside it has bigger downside than upside ok that's simple that's definition frigid so we came up with this definition for a local fragility right after the coffee cup is broken it's no longer fragile okay and there's another representation we can have of a coffee cup doesn't like the solver is this is a coffee cup there's an intensity of shock you tap you tap and then boom it breaks and now you doesn't care anymore all right so this is known as concave second-order effect that's accelerating okay complicated but it's sort of it's a necessary relation it's not like it's something you thought of it has to be like this necessary relation from the fact I'm not liking disorder you don't like the sort of the rule is as follows a shock of 10% harms you more than twice a shock of 5% okay I don't know how many of you have driven cars when you go home drive your car into the wall 50 times at 1 mile per hour and then try it once at 50 miles per hour and compare right and then we can talk about it again and then you'll see that we're fragile because sharks right are disproportionately harmful to us okay so we just said that with that that notion of fragility which didn't exist before because you have to be vicious or an option trader or obsessed with coffee cups to figure it out it so happened that all option traders know about it we call it negative convexity or concavity right and as if you have reactions as accelerating so there's a an anti fragile I show the graph of a stone I don't know how much more time we have but this is sorry this is a stone and this is a harm function okay a stone of in kilogram if someone hits me on a head with a stone of 100 pounds I'm harmed a lot more than 100 times a stone one pound you see says that acceleration and that acceleration determines it okay when you have an exertion like that you much rather have and much rather have one stone at twenty-five pound okay rather than one at ten and one at 40 how much more time do I have before the penny more time okay so this is so during the conversation I'll continue but I'll give you two more minutes all right so this is acceleration okay and the rule is as follows this is concave this is convex you know the difference this is bad in other words if you had the second-order effect your reaction to anything to death you increase that you're much more fragile if you have a billion dollar debt and a country it was just horrible about that you're not a hundred times more fragile hyung didn't know that you're ten million times more fragile you say so with this we can do matrix so if you want to get bored read my paper about high V as I discard this figure in the foreign affairs okay and but I was very lucky because nothing can be as bad at something called Harvard Business Review where they transform the rule is you should not have more than one idea per paper with an average of zero alright so that is so that's it was a target average of zero so good foreign affairs they may tolerate one an average right but I so dumb it down you know to not explain these things alright so this is sort of the gist of it and I know that I haven't explained enough for you to be bored with it but at the same time you know this is a baguette I mean you can look it up and continue it with the idea that there are two kind of portfolios a portfolio where bad news you lose money portfolio is bad news you make money one is convex one is concave when I call anti fragile when I call fragile one Lexus order one dislike this order so now look at countries you see hey size makes you more fragile to this order size of unit not necessarily you know size of country size of decision-making unit corporation same thing that makes you vulnerable to disorder what else the bunch of things makes you just focusing one commodity or I'll have more than one commodity follow the two disorder so the idea is vulnerable to disorder which we can measure you can explain is pretty much the universal definition of fragility I'd like to apply to countries and it's not predictive you see I can I can tell that this is if there's an earthquake here this will do better than a computer great thank you for listening to me I'll keep this because I will continue the lecturers work thank you very much the same and there's one thing I wanted to add is that on the way here I discovered that machine has a rating by taxi drivers by uber drivers of 4.9 so we know he's nice to taxi driver yeah I don't know but there's the same product because a lot of people have reading a five with one ride or two rides and I've been on three years worldwide on uber so this is a difference so that's my yeah maybe the only piece on my resume that I care about now we're operating we have about 15 minutes also I have a rating of negative one with policy people in Washington and I want to decrease it you know we have about 20 questions and 15 minutes and and I want to slip my comment and as a follow-up to the gentleman who spoke about Isis there is a there's a panic in the West about returning about Isis jihadists who returned from Syria or from the field I have no clue about that this is that's not honestly this is so minor currently as a source of risk the bigger source of risk is that let's talk about real things like airplanes crashing make the frontline you know of the newspaper Isis makes a frontline newspaper unless you live in the area or like me or from Lebanon all right it's not a big people talk about it a lot but if you look at the risks we face on a daily basis the real risks are not Isis that's motion the thing they think about the real risk was Ebola because Ebola can spread you see Ebola can accelerate can spread these kind of thing and the next a bola will be facing the first time in history with an epidemic that flies on British air you say and of course Delta and gets bad food on now United all right treatment by flight attendants so so that's that's the problem and we never face it because of so epidemics follow this mode so when people tell me let's talk about risk the first thing I talk about is if Isis or someone else maybe can accelerate it but it could be naturally coming is an epidemic okay so I have some questions here from from the audience so I think one of them you've already sort of answered because it says can you discuss some examples in the world that are fragile and that may be in the five days of coming of coming events okay so that's the let me tell you one thing about fragility I can explain it can tell us that the Soviet Union did not collapse because a political regime but because of size because socialization it can tell us because you can compare China to Singapore and you realize the size effect is monstrous okay then again a lot of people don't quite realize history and when you discuss the history of Italy three in a three hundred years before the unification it was a lot of turmoil but low-grade turmoil which makes things stable you see when it unified is one starts saying bad things start happening when they had the big to war participation big to Wars and stuff like that so effectively the the there is this myth about where the risks are that are when I wrote fooled by randomness I was talking about the regular people not understanding risk in later on start talking about psychologist at risk not on sunny risk when academics not understanding risk and that this is where things started you know I started getting hate mail and and and you know but from academic because they start hating me because I'm saying they don't send risk you see so but we talk about spate of risk facing us today in this room the first thing you got to think about is the real thing that can harm us and spread uncontrollably which is epidemics or biological agents spread by people not the Isis was a bunch of guys on YouTube you know so this is in a way also related to the next question which is should we still be thinking about risks on a country level or only on a country level and how does frigidity apply to transnational phenomenon like al-qaeda cyber security I mean I'm not again I mean I think I'm a very simple cybersecurity is a big problem cybersecurity luckily we have things going for us is that people are so paranoid about it banks spent five percent of their money on cybersecurity and you have hackers all the time and let me tell you what the interesting thing about about the computer industry they're smart people all right they're very very smart people because they're not over educated ph d--'s and stuff like that they're and and they know for example let me give you an example when i was finished anti fragile a lot of people wrote to me and then start engaging with them and I discovered that effectively they were doing things that I thought I discovered like for example there's something called the cows monkey on a part bye-bye Netflix Netflix every day engineers failures so the systems are it's sort of like the Saudi Arabia's don't try every day to invade Saudi Arabia alright ok pay people to invade hey let's see but netflix every day has an army of agents I mean the things trying to destroy a system and they know if they stop the first so they discover other ability so they understand very well this notion an anti fragile that is the time here that things that experience a lot of variability are a lot more stable so I'm not were that worried about cyber as much as the breathing you see when when you get an uber taxi and the guy had the passenger before you these kind of things are more worrisome but we looked closely at cyber security and and effectively there are these are the risk we identify but we're doing so much about cyber security that we don't have to worry too much the only thing I think may happen in the future is you may have it may be actually to bird it may the cost of controlling you know these risks may go up quite a bit you see like people spending more and more their budget and on cyber security and there's indirect cost as well right there's there's I mean cyber security can slow things down and there's it is that for example I went to London last week the first time I use the credit card and use it in foreign country Puft things they stopped right so they're they have a type one they have a very very very very paranoid attitude toward that okay so the next question is how does your fragility framework compare with systems thinking I don't know if Stan systems I didn't I was trying to read it as though I understood it no III tried no no there is a complex system Institute and there's a member here and we work together on on taxonomies and stuff on scaling but I don't know if you're thinking we could have bought a few books the non-standard work I don't understand things unless they're explained to me mathematically you see so as I'm a trader so unless I see a contract you know you get paid here a dollar you lose a dollar of this happens that kind of thing and then like the resilience all these things so I don't understand so I'm not saying anything bad I don't understand it right so it much rather I like the rigor of mathematical framework for things and and may fit contradict I have no idea now this one we need to know who asked it because it's a it's a it's a trick question yeah so what's your blind spot or your black sports I would like to know them I mean I have this whole idea is so quiet if you know your blind spot no longer blind spot yeah are you did that's a that's sort of like a you know the the liars paradox or someone if you know your blind spot it wouldn't be a blind spot but the other words do you make an effort to look at what my point is I'm developing something that probably improves like an the stress testing a little more acceleration detection of acceleration and proof star surfing a bit so a little improvement here and there all right I'm not trying to solve the problems of the world you don't get there with a very simple framework the good thing about fragility theory is is you can touch a lot of things and work with a lot of scientists and it's mostly sighted mode and physics for some weird reason with so it's very narrow and I want to prove from here to here I don't want to try to find you know the flaws in the world and stuff like that right now it statistics a useless science or are there some redeemable principles in statistics I mean everything now the more there's a big mistake Hilbert said it said any science that develops ok becomes applied mathematics and if it's not applied mathematics yet it's not yet a science or that was Hilbert and I believe him that I'm gonna change it and say any signs that develop properly becomes probability theory right statistics is one branch of applied statistics is the probability theory is what cooking is to chemistry you see the idea is it's cooking and the problem with statistics is the way it's used now mechanically a lot of people used statistic mechanistic so my major work is not the insert' though the assert of the philosophical thing is the parallel work that I have is about a thousand pages 500 unreadable outside finance outside the riveters and 500 pages now in something called silent risk which I made available an other way for anybody to buy publish to whatever you want sell to you know put the nice cover okay so salmon risk well-formatted it's on the web for free for free and in silent risk you go through deep down deep what what's happening we discovered one thing well gee I'm always like you know when a scientist when someone and does experiments and stuff like that they go to a statistician puts a stamp like you remember there is it used to be ad for Hanes they say they don't say it's Hanes unless I say sayings so the statistician puts a stamp but typically a statistician doesn't really care about biology looks at the numbers say okay this is converge right so it's not enough because statistics has never made claimed to be dealing with tail events a statistician tells you we are studying thing with a 98% confidence interval the problem is if you're doing airplanes with 98 constants intervals it's a plane is statistically safe without intervals you'd have between 900 and 5,000 the airplane crashes a day you see so the risk management is not completely in a tail way beyond what a statistician have been touching traditionally where the tool so it's complete different set of tools and these tools are not refined yet they don't exist but there's a group of people in 80 Asian Zurich working on it they're a bunch of people work in a lot of areas on on on formalizing a method and approach a is just like what statistics is a probability theory the risk as an application of to probability theory and a lot of the people are come from insurance so is it useful it's imperative and will control by it because a lot of problem is the verbal istic cannot reveal the the risk you have to have to format it in a way and you realize how people make a lot of mistakes so how does bad data affect your thesis and this is a question from someone who's saying that they're using data to understand tails and distribution and what about the question in the data itself it's not important what would you do when they work with the tails what happens in the body of the distribution is of no relevance to you like like people the tip is that you look at what we say peak of a threshold and stuff like that pop method of stuff you take the max amount of data so if I for example Steve Pinker wrote eight hundred page book was paid data every page and you look at it you say entertainment because if you're really doing your job on fat tails you'd have half a page right was one thing taking the Maxima I've seen the behavior of the Maxima so the there's something called large deviation theory another branch called the the the distribution of extrema ok extreme value theory or different names there's sort of few branches that deal with tails and was one deal with tails with yellow state let me give you an idea when LG Samsung he killed his wife no all right so you don't say oh the other day he ate breakfast he didn't kill anyone all right you only look at one piece of data when he killed his wife you don't look at the other don't judge the character based on a regular alright you judge a character based on the tail events okay so therefore I'm not interested in what he eats for breakfast or what he does I'm only interested in when he kills his wife what does he do it you know what he kills his wife right that's sort of that's the story was all Jason likewise with when you look at events you look at the tail the large deviations a lot more information and there's an error people make and there's something that serum actually say we show that a million piece of data all right can be anecdote and of one if it's in a tail okay is information most people think that data is information one can be information or a million can be no information okay depending on how far you are in the tails Wow so is there any Big Data Platform that you worked with that again like the whole idea where they can see offends me because well there's a problem big data let me explain it to you so yeah let me explain the Lord oh don't hit me let me explain the large D problem and big data okay if you generate a coherence matrix okay you generate the convention matrix okay a hundred with the hundred units okay so it's ad equal a hundred and you have n equal a thousand all right randomly you generate randomly data you're gonna see we just did it with the eigenvalues how many spurious correlation would you see when you say draw normal yeah you're likely to have fifty pieces of whatever no exactly so you will have the random data so what happened is that and this is discovered was discovered a long time ago by people in finance and then later on in medicine what epi data we look at big data in medicine there's a guy called you I needed to figure it out he said if I have the more the larger the cover the the they look at it for any piece of data I have n times n minus 1/2 this is every time I add the dimension okay every time I add the dimension I have more spirits correlation likely to happen so the numbers pairs correlation are convex this is bad correlation convex to data right convex 2d the dimensionality is called curse dimensionality to mitigate it you need to increase the data points per variable and unfortunately we don't have enough of these you see so we have a large what is called a large the small end problem so you're likely to see spirits correlation like and and this was discovered in epidemiology rediscovered in the terminology I got call your Andy these five years after we've been finance discovered that after fooled by randomness that when you have what I said fubar enemies the more data I have the more likely I am like likely to find a stock that correlates perfectly with your blood pressure okay so that's a problem of now how do you clean this up by having a large end moldy butt and you can't manufacture and so we're these increasing too fast that's the problem of data so the prop this is why you got a data in the hands of people who think hey more data is good and stuff like that is very dangerous this is why the the it is very very day let me repeat the point you have variables I'm adding variables and variables are swelling every day and every variable I add okay creates a billion various correlations you say and inference is the thing with Gaussian would go root end which is concave you see so you have root and information grows at root end so slowly was N and very quick and bad information grows very quickly was D so this is what if I looking at big data I can find two correlation tomorrow that look very plausible or something is plain yes right okay there are two problems there are two problems the first one is under fat tails data grows much more slowly than n all right then it would end that's the first thing you see you need a lot more data not more n okay under fat there and then we just verified it with the eigenvalues explode under fat tails that's the first one that's technical the second one is I discussed it an anti fragile to answer that question I said there's something called the bonferroni upper bound that tells you the more variables you have you put high threshold on faux fur fur on and there's something called the saenko past or you know bound for stuff but the problem that researchers and this is I put it in my chapter on ethics really of all things I said the the problem of ethics is that I can if I want to write a paper you say I test test test till I find something that fits and then I give you a statistic the significant thing you say I'm not going to tell you it's like hiding risk I'm not going to reveal to you how many times I've tried you say so the two problems are fat tails and ethics so I'll let you I would love to introduce you some data scientists who are very good at with ok are they're rich they're getting there my definition of someone who says that work a lot was data and I can predict that tell them do you work for a living or no the minute they say yes I discounted because someone who really has if we have the hang they claim to have on the data all right they'd be owning a piece of real estate of the master state of Massachusetts and stuff like that and that was something we noticed that was a rule that we had been financed early on do they work for a living yes no they will work for a living then it's suspicious because they're giving you equations and smoke and data this is skin in the game parts can discuss and scan the game even if they work for a living and they have gold taps of the Jacuzzi of the private airplane they can fail in the end no no in other words you can tell if someone has the the person who is a most sophisticated by far but data is a Renaissance guy and they only use it for short-term trading you see for inventory there's a finance fellow here yes wait for the microphone because it's being recorded so I thank you and it's such a wonderful formalization of this idea and I'm and I'm right with you it intuitively you know but do you worry that there you're gonna give them this great model and they're gonna miss use it I mean is the problem not in the math but in like saikal you know hubris and self delusion the know that big data either answer them there to problem the first one is in a mess the mass is not in the the make methods to control they say for multiple testing all right is doesn't take into account a fat tail and the second one is the problem of ethics okay that's the problem big data is ethics the guy comes in writes a paper you know that the FDA since 2005 all right if you if you tell them I want to introduce this drug this is a epidemiological study done on the computer they take in the short and a garbage okay so even if if you did all the things but just more broadly it on the trading floor it so yes he's a dictator no I think well I think it's time to to wrap up yeah I've been getting looks looks from our organizer yeah maybe he has a quote maybe maybe he has been angry so so it's good you seem not angry but he seems the but you got the Edition it's passionate about know what I'm thinking about not not just related to big data but on the trading floor and in the banks and in life I mean it seems like the the problems if it's garbage in and garbage out even into a terrific quantitative model if it's you know how do we I mean as a pot you you consult on policy and I mean we live in Boston starlet restraining floor people are interested in Tori market-making so the lot of the profits come from the fact that they have the franchise but we live in Boston or the a fidelity investment and you have the other guys all right and how much data big data fidelity investment has and how much data and we know that random fellow you know works as operated on a subway his odds of outperforming fidelity of these guys is very high is it odds are very high okay so this tells you that that we know this from finance and say hey if they they of course you hear people who are perform it over all the finance industry is a big scam to cash in fees from people by telling you we're using these computers for fancy stuff but they don't want to perform the market okay cumulative leave of course you're gonna have that you're gonna show you the ones that outperform and not the others so this is where I mean we have enough evidence that data has not helped people in socio-economic domain but if you happen to be in bird and bird the theory bird flying theory data is great all right what or if you're into hydrology data is perfect or what else if you're in UC but if things are nonlinear data is hard and this is and you can see it how much money was thrown and how many Russian scientists were flown in after Soviet Union how much money to try to predict my land overall only here you here you have a biased version of the other story but you look at it okay I'm getting I'm getting signals from from the leadership that we should that we should finish and there's a question but it looks like it's by an economist asked yeah yeah go ahead go ahead and also say yeah but but it says that that's the last one okay so the question is it if we stop using statistics yeah as the question thinks you suggest how can we make decisions okay that's great that's good and again and no and also don't we have to make assumptions okay this is the kind of thing that only cars would say I'm glad you know the name because they can keep supplying you with bad models and yet survive all right that's good in a game that be dead all right other people becoming the other the problem is in real life we don't use let me tell you one thing I discovered but only recently it was recently that five years ago I'm slow you see it took me 20 years to ride the Black Swan so the the and silent risk now its tenth year all right so the took me late I discovered that said hey you know what what are the fields that are the smart field where okay physics okay do they have pages and pages statistics no why there's a mobile trip they don't need to put a lot right so you start looking at people the more people use statistics the more than back it up with statistics the more guilty they are in fact and and and there's another thing that that I mean we can look at effectively that the predictability but there's this one thing that people in economics and logics that they lose one of the economics if you are let me tell you if you are lost in in Washington DC and you don't have a map and and you run into an economists here have no map you could ever have a map of Moscow you know all right would you take that map of Moscow no so why do you take back what do you take a random analysis which is playing random all right to make decisions rather than okay so don't make these decisions so what do you do okay so it's like what I tell people if you're on a plane and and and the pilot says we have a map of the Himalayas and we're going to the Alps but good enough right you get off the plane and you go by boat so you cheat you calibrate so my the first axiom I have is only use statistics for decision all right if the statistics is reliable all right if not then do something don't take that decision try to be robust and that's the whole idea of river that's the whole idea fragility theory and a suggested economists download silent risk and if he goes past the insults every five pages all right on the I said here I said well wrote the Black Swan hundred percent of firemen understood it all right and 95 percent of mathematicians 82 percent of physicists this is this and then 0.01% of economists understood what I was talking about right so because the logic that you have in the real life in risk management I don't have a map on Moscow in DC I'd rather have nothing right and look they lose it the minute we go into the classroom okay and I've named these things I've named it the Stigler effect as a sailor's dude mystical it's strictly SIA and then a lot of things and there's like you know Stiglitz said the Fannie Mae is safe right and Fannie Mae blew up and now he's lecturing us on what to do next I mean in a real world you blow up you're out of the pool or the only exist anymore other people are coming to the pool and that's evolution he this blood blocking evolution lets him Tyler whatever whatever made you thank you very much was it whatever made you take the decision to come and disturb us we're very grateful

Channel: The Fletcher School

Views: 20,519

Keywords: The Fletcher School, International Relations, Global Affairs, Graduate School, Dean Stavridis, Fares Center, Nadim Shehadi, Nassim Nicholas Taleb (Author)

Id: QCGtfz6TtyU

Length: 64min 29sec (3869 seconds)

Published: Fri Aug 28 2015