How many have you worked in a finance world?

[twofish-quant]

This is true, but this requires the development to reach a critical stage. Before this stage, there is not enough hard evidence to convince the average funder.

Depends. The people that I have to convince are my supervisors, and if I go up to my boss and say "I think that you should let me study algebraic topology for a week because ..." he is usually pretty receptive.

And rightfully so. I would not fund fuzzy research either if I had the money to speculate with.

But the people I work for - will. And that's one reason I work for them. One reason that my company is willing to fund fuzzy research is that we can point to specific examples in which fuzzy research has led to mega-$$$$$, so if I go "I think we can make $$$ if we study algebraic topology" they listen to me because that's one reason they hired me.

The problem that I have isn't being able to do research. The big problem which I'm trying to figure out is how to "publish" the research. The problem is that if we discover that studying X will lead to mega-$$$$, this isn't something that we want our competitors to know immediately.

They will find out. In three months, they'll figure out what we are studying, and in one year, they will have copied everything. But in this business being able to do something that your competitors can't for three to six months is $$$$$.

If someone wants to "give me money" fine, but I would not accept a deal where they expect sometthing at all in return as that would disturb the creative process. I think this process works best when freely floating. Money just screws this up.

For me expecting something in return is part of the process. What the people that employ me expect is something that will make them $$$$. I write stuff that makes them millions, they keep most of that, give me some table scraps, but since I'm not hugely motivated by money, that works out.

Also it's not that different from the academic world.

I have a good life and have nothing to complain about.

Curiously, I've got a lot to complain about.

There is an interesting thing about bureaucratic systems in that they want people that are "loyal complainers." If you are just a "yes boss, everything that you say is right" then what ends up happening is that the system breaks down. What ends up happening is that when the cameras are rolling it's all smiles, but then people go into the back room and the conversation becomes "we are SOOOOO screwed."

 

[twofish-quant]

What kind of undergrad courses do employers look at, if any, for the more technical jobs? Or is hiring of undergrads rare? How well-regarded is the MFE degree?

Unfortunately, I don't know. The jobs that I know about require at least a masters degree. Most undergraduate physics majors into finance go into management consulting and that's a different world.

As far as an MFE degree, I'd rather avoid it. It's better if you take a masters in something technical or a MBA, or masters in finance. The trouble with the MFE is that if you get an MBA or a CS masters or anything else, then you have options other than quantitative finance.

 

[autre]

The trouble with the MFE is that if you get an MBA or a CS masters or anything else, then you have options other than quantitative finance.

But if you want to go into quant finance, an MBA, MS in CS or even MS in Finance isn't sufficient preparation, right?

 

[Fra]

The problem with intellectual satisfaction as a goal is that
1) how do you know if you are right?
2) sometimes the truth is deeply unsatisfying.

1) I'm right when my models increase my predictive power and I learn faster than those that are unaware of my model. This is testable.

The problem that I have isn't being able to do research. The big problem which I'm trying to figure out is how to "publish" the research. The problem is that if we discover that studying X will lead to mega-$$$$, this isn't something that we want our competitors to know immediately.

They will find out. In three months, they'll figure out what we are studying, and in one year, they will have copied everything. But in this business being able to do something that your competitors can't for three to six months is $$$$$.

Yes exactly, this is all in line with what I try to convey. This is exactly the meaning of that a theory is not a objective description of the world (be it nature or market); it is an interacting tool, and every "player" does have a dissimilar copy but which are constantly equilibrating and evolving.

So your financial models are not corroborated by letting every player test it in infinite trials. Because the model itself will get useless while doing it.

Your financial models are "tested" by actually generating money.

What I tried to say far back is that while this is gut feeling for finance feeling, this view on "theory" is quite obscure for a typical physicists.

/Fredrik

 

[Fra]

Finally, one reason I post as much as I do is that I'm trying to give some context for how I see the world working.

This is great, thanks for the your patience. I think after all that was the original topic here, so hopefully the OP can extract some information from the discussion.

My argumentation serves a different goal, which subtly connects to the OT: The common aspects of ANY modelling, and ANY evolving theories. This touches with what my own pet research is about. My point is that there are aspects of in modelling and understanding what a theory is, where looking at the abstractions (ie shaving out the money, shaving out the physics) can lead to insights about generic inference models, that can be useful both in finance and in physics as well as in AI research.

This provides a conceptual picture where one can appreciate looking for common denominators.

I see a theory (wether it's a physical theory or financial theory is irrelevant) as an interaction tool, not a static description. This also forces revision of the scientific method.

An interaction tool can not be tested in quite the same way. The only way to test it, is to throw the dice and play. But in the process of doing so, the theory will change. To further test it you just keep playing. So the ultimate "test" is that the theory that stays is business "is right" and the one that gets out of business is wrong.

Now this thinking can be applied, abstractly to both economy and physics as well as general inference/learning.

In finance, the measure of "existence" is acces to money (it doesn't need to be Your money). Ultimately a player than runs completely out of money faces game over - his strategy did not keep him fat.

Nash Equilibrium configurations are the player/strategy configuration where everyone maintains status quo and has nothing to gain from evolving.

In physics, the measure is confined energy or inertial mass. Ultimately a physical system that fails to equilibrate and negotiate with it's environment, will destabilize and finally loose energy and inertia - the system was not stable.

Physical equilibrium is where you have a system that is somewhat stable. For example elementary particles. In contrast one can construct "theoretical" weird particles that would instantly fall apart.

In inference, the measure us just confidence and inertia, if you ar persistently wrong and failing to learn you will loose confidence in your priors that just keeps loosing inertia. Eventually your knowledge is ripped apart and your identity lost.

Equilibrium is where many inference systems, exist making inferences about each other in a way that they have reached mutual agreemtns about their disagreements. This can even contain encapsulated stable "lies" etc. They are stable before no one benefits from revealing them.

What I'm looking for is trying to find the abstraction here, and apply it to physics. But the intuition is more obvious in finance, where it's obvious that it's a game! In physics, we normally don't think in terms of two interacting atoms as playing a game.

In particular the problem of the origin of mass, one can get some nice insights by looking at the abstraction from a gaming picture. The concept of borring money and borrowing energy and borrowing information (ie GUESSING) have some strikiing similarites.

I also think the standard model (particle zoo and actions) can be understood as a stable configuration of theories interacting. So the standard model is more like an equilibirum point. Nash equilibrium fails to be a perfect analogy of course, but it loosely speaking gives intuitive hints.

/Fredrik

 

[twofish-quant]

But if you want to go into quant finance, an MBA, MS in CS or even MS in Finance isn't sufficient preparation, right?

It doesn't quite work that way...

What happens is that you end up working on a team with people with very different skills. I know a lot about crunching PDE's. I'd make a miserable salesman or trader, but I work with people that are excellent at sales and trading, but know practically nothing about crunching PDE's.

So what people are looking for are people that are really, really, really good at one or two things, and can work with people that are really, really, really good at other things.

In any case, nothing is sufficient preparation for quant finance. It's not like Newtonian mechanics, where you can take a textbook and learn the material. If you give me a standard quantitative finance textbook (say Hull, Wilmott, or Joshi), I can point out that 75% of the material is either wrong or irrelevant. It's not that the authors are incompetent. It's that things change quickly enough so that things that are correct in Q1 2011 become wildly incorrect in Q4 2011.

This is why a strong general mathematical background is better the learning the details. For example, you learn some PDE's in Hull. They are mostly wrong. But once you get the job, someone hands you what people think are the right PDE's and then away you go. If you don't have a strong math background, you can't do that.

And then you have to have a parachute ready in case it all blows up. The big problem with an MFE degree is that they teach skills that are relevant for a particular job, and if that job disappears, then you are sunk. If you have an masters in statistics, you can still get a quant job if they are available, but if things fall apart, you may be able to get something in biotech.

 

[twofish-quant]

So your financial models are not corroborated by letting every player test it in infinite trials. Because the model itself will get useless while doing it.

It's a worse problem. The trouble is that you *can't* have infinite trials even if you wanted to. For example, I have a model for how bonds work in Q1 2005. I *know* that bonds work differently in Q1 2010, so how do I even test a model?

This is different from the situation in say particle physics or biology. In particle physics, you can be reasonably sure that the laws of physics are the same today as they were yesterday, so you can repeat an experiment. Also, you can be sure that one electron is pretty much the same as another electron.

For biology, you can be reasonably sure that the human body hasn't changed radically in the last week, so that you can repeat experiments. Also, you can do statistical tests to see how much human beings change from each other, and you can be pretty sure that someone won't walk into your experiment with the ability to bend steel or who is two inches high.

For finance you can't do that. Curiously the *philosophical* problems that you run into finance are similar to the once that you run into in astrophysics. In astrophysics, you really can't repeat an experiment. You see supernova 1987A or the big bang, you can't make another supernova or big bang go off, so how do you know that your statements about 1987A or the BB are true.

Your financial models are "tested" by actually generating money.

But even that is questionable. I could generate a ton of money in 2005-2006 by assuming that the price of real estate would go up forever. If you have a lot of people sitting in around a roulette wheel each with their "lucky theory" one of them is going to end up making a ton of money.

This is something that causes a lot of paranoia. Models work until they don't, and if you aren't careful, once it's obvious that they don't work, it's too late. You *think* that your lucky rabbit's foot caused the roulette wheel to show number 23, so you bet a thousand on 15. By some stroke of magic, you win, so you bet a million on 12. You lose. You figure out you are wrong, but it's too late.

Conversely, if you lose money, that doesn't mean that you were wrong. You could have just been unlucky. If you say that there is a 95% chance that you will make money, and you lose money. Well... Hello 5%. Something that one trader mentioned is that if you are, really, really, really good, then you will be wrong about the direction of the market 40% of the time.

What I tried to say far back is that while this is gut feeling for finance feeling, this view on "theory" is quite obscure for a typical physicists.

I don't think it is.

One reason that investment banks hire physicists rather than economists to create financial models is that physicists tend to ask deeper questions than economics or finance Ph.D's. What economics and finance Ph.D.'s tend to do is to take a pre-built model of the markets and then plug numbers into spreadsheets and come up with results, and about 98% of the work of banks involve doing this sort of thing.

For example, there is the Markowitz model of portfolio allocation, and a lot of what finance and econ majors do is to type numbers in spreadsheets and come up with weights. If you want to use that model or tweak it, then you want an econ major to do that. However, if you are in a situation in which Markowitz is no where close and you have to create something else totally from scratch, that's what astrophysicists get hired for.

The fact that things in finance change in both space and time is why you need relatively large numbers of physicists. If some one comes up with a "theory of everything" about electrons, we are done. They write the paper, collect the Nobel, and no one has to do it again. That's horrible for employment. I call it the "second Einstein problem." It's great to be Albert Einstein, but once you've figured out general relativity, there is no job for a second Einstein to figure it out again.

What happens in finance is that you create a model in Q1 2005. It seems to work, but by Q2 2007, it looks shaky, and by Q4 2007, it's obvious that there are some huge new things that make it wrong. At that point you have to pay some physicist money to come up with a new/revised model, which works for another years until the markets change, and you have to pay to get it fixed again.

Lots of jobs for astrophysicists...

 

[Fra]

I think we probalby agree to a reasonable extent, some of our arguments seems to agree even though we are superficially argumenting. But maybe the entire discussion can still be enlightening as an elaboration.

So your financial models are not corroborated by letting every player test it in infinite trials. Because the model itself will get useless while doing it.

It's a worse problem. The trouble is that you *can't* have infinite trials even if you wanted to.

Yes, we fully agree. You *can't* test it in the way we normally "test" theories. This was my point too! So the question is then, then what? Because the game nevertheless goes on with or without tested strategies.

In particle physics, you can be reasonably sure that the laws of physics are the same today as they were yesterday, so you can repeat an experiment. Also, you can be sure that one electron is pretty much the same as another electron.

Yes, but my point is that when you look at the open problem in physics, in particular for unification and comoslogical models, AND required that a theory as an inside inference (just like you desires local observables that is inferrable by a local observer). Then this fails too! In analogy with finance, it's just that this is indeed LESS obvious in physics, while it is obvious in finance (or SHOULD be, maybe I overestimate the average finance guy, I just know that at least some finance people are WELL aware of this).

This is the problem I am talking about. For my the finance is just another application I use for analogy.

What fails is the physical realisation of "ensembles" and infinite trials, when you have a information bounded observer that is NOT looking at a small subsystem. It means that a cosmological model, can not be tested in the same sense a scattering can. You count evidence from the inside, rather than repeat trials and collect statistics at infinite. These two ways doesn't quite merge at the moment.

For finance you can't do that. Curiously the *philosophical* problems that you run into finance are similar to the once that you run into in astrophysics. In astrophysics, you really can't repeat an experiment.

Yes exactly Indeed we fully agree so far. You are trying to make the same point I was trying to make. Maybe it just looked like we disagree.

But there is in my picture an additional big problem, that I'm not sure how you'd characterize.

It's how to combined a "cosmological perspective" with a "measurement theory". This is of course an open problen, but this also has analogies in finance as far as I can see. In particular in the sense that the bets and feedbacks of a players qualifies as measurements, but this measurement theory can't like you mentioned be abstracted with ensembles like in QM. We are lead to a subjective odds interpretation wuth the additional twist that the odds are constructed by combining information from non-commuting counter states. This latter thing is what leaves astrophysics since it's not just classical GR anymore.

My original assertion was that an evolving theory of expectations and rational actions, would be the economic equivalent to this open issue.

So you can't test objectively a theory, all you can do is let two systems with different theories interact and see what happens. If they do not destroy each other they are in equilibrium. In the similar way we can define emergent symmetries by populations of interacting systems or players.

But even that is questionable
...
Models work until they don't, and if you aren't careful, once it's obvious that they don't work, it's too late.

Yes exactly, but this is part of the game. This is not a "problem", it's an insight, a clue :)

The interesting thing is howto use feedback to always improve the theory. Ie. a theory should generally always be in motion; unless you hit an equilibrium point.

I think the apparently "timeless" theory of physics, SM etc, are equilibrium points in a bigger picture. This bigger picture is what we need to understand better unification, and why the relations between masses are they way they are. And to understand this we needto understand how (very abstract I agree) the laws of physics are encoded in every subsystem of the universe. In General at dissimilar copies, the case where they agree is the equilibirum. But to understand the equilibrium we need to perturb it.

/Fredrik

 

[twofish-quant]

Yes, we fully agree. You *can't* test it in the way we normally "test" theories.

Normally? We are getting into some deep issues about the philosophy of science. If you go into any textbook you'll probably read something that "this is how science handles truth." But one thing that I've found is that this is often an incomplete and inaccurate description of what goes on.

This was my point too! So the question is then, then what?

For me, the solution is not to focus on "truth" but to focus on "usefulness." Can I get predictions from this model that help improve the world. To use an example, under most circumstances, Newtonian physics is "better" than general relativity or string theory, because you can actually use Newtonian physics to build bridges.

I suspect that my way of looking at things has pretty deep Chinese roots, which means that you have a lot of different things that go on versus people that's philosophy is based on Plato and his theory of forms. Also the Chinese philosophers that I think that I'm very strongly influenced by were profoundly anti-Buddhist and anti-Taoist, so the philosophical tradition that I think I come from is quite different then the one that people talk about when they think "Eastern philosophy."

In analogy with finance, it's just that this is indeed LESS obvious in physics, while it is obvious in finance (or SHOULD be, maybe I overestimate the average finance guy, I just know that at least some finance people are WELL aware of this).

Not every physicist works in string theory. In fact, very, very few physicists work in string theory. Also most people in finance have no particular need (and often no desire) to think about what is going on. A lot of finance work involves plugging numbers A into spreadsheet B, and then doing something based on what comes out.

One problem is that one of the jobs of finance and economic professors is to be "ideological salesmen." For example, you have this political idea that "less regulation is good" and then you have people from the University of Chicago that end up selling that idea to the public in much the same way that people sell vacuum cleaners. The fact that said professor isn't going to suffer in any way if he turns out to be wrong, causes problems.

It means that a cosmological model, can not be tested in the same sense a scattering can. You count evidence from the inside, rather than repeat trials and collect statistics at infinite. These two ways doesn't quite merge at the moment.

They do, but it's not obvious how.

So you can't test objectively a theory, all you can do is let two systems with different theories interact and see what happens. If they do not destroy each other they are in equilibrium. In the similar way we can define emergent symmetries by populations of interacting systems or players.

The problem is that in finance there are some experiments that you just can't and shouldn't do. What I'd love to be able to do is to an experiment in which Congress passes different budgets, see what happens, and then press a "reset button" and repeat the experiment. The trouble is that the world has no reset buttons, and if you blow up the world, you can figure out that you are right, but you end up destroying the planet in the process.

You have one global financial system. So you can't have two systems interact.

Also to step back and explain a lot of what I do...

Suppose you are willing to pay US$100 for a banana. How much should you pay for two bananas. The answer is more or less US$200. What I do is to basically do these sorts of calculations. There is a lot more math, but this is the basic principle.

Now you may point out that you are freaking insane to be willing to pay US$100 for a banana, but that's not part of the calculation. In some ways, the willingness for someone to pay US$100 for a banana tells you something about their mental state, and if someone is willing to pay US$100 for a banana, I can tell you that they are crazy, but that craziness is such that the are willing to pay US$200 for two bananas.

The big mistake that I think people made was to confuse statements about mental state with statements about physical reality. For example, in 2005, I can tell based what you are willing to buy and sell that you think that the chances of a housing crash are close to zero. I can mathematically take your actions and calculate an "implied probability." The trouble is that this has nothing to do with a "real probability" and one can argue that a "real probability" doesn't even exist. The "implied probability" of a housing crash could be near zero, but that is just a reflection of the fact that you are insane.

I think the apparently "timeless" theory of physics, SM etc, are equilibrium points in a bigger picture. This bigger picture is what we need to understand better unification, and why the relations between masses are they way they are. And to understand this we needto understand how (very abstract I agree) the laws of physics are encoded in every subsystem of the universe. In General at dissimilar copies, the case where they agree is the equilibirum. But to understand the equilibrium we need to perturb it.

Again I think we are coming from different philosophical traditions, and the philosophical tradition that I come from regards that sort of thinking as "useless wastes of time." Part of the problem is that if you spend too much time thinking about a "theory of everything" you end up with a "theory of nothing" and that's not good enough for the boss/supervisor/CEO/emperor that wants a report by close of business today.

I think that you are coming from a Platoist background in which you have these universal truths, and that the physical world is a "dirty" reflection of those truths which means that that you need to keep things separate.

My background is different. The emperor/CEO/supervisor is paying me money and giving me social status so that I can look at the stars and figure out what the impact is on the empire. One thing that was sort of funny when I took a tour of the Forbidden City was that there were sundials all over the place. The reason that you need a lot of sundials so that when the emperor tells you that he wants a report on his desk by 5:00 p.m. you can see that.

 

[Fra]

Normally? We are getting into some deep issues about the philosophy of science. If you go into any textbook you'll probably read something that "this is how science handles truth." But one thing that I've found is that this is often an incomplete and inaccurate description of what goes on.

We actually agree on this.

For me, the solution is not to focus on "truth" but to focus on "usefulness."

I admit I am very ignorant about chineese philosophy but I think we talk past each other. If you read what I tried to say (in they way I want you to;) then you will see that this is exactly my perspective as well. We do not disagree on this.

They do, but it's not obvious how.

Not sure what you mean. I agree they have to merge, but there is not yet a consensus on it, only speculations. Ie. they obviously ARE merged in nature, but what I mean is that we humans do not yet understand it.

But I was referring to combining measurement theory as in QM, and cosmological perspective as in GR. One of the KEY problems is how to define observables, that make sense. Clearly scattering matrix is not good for cosmomodels. OTOH, local observables seem to be hard to contrsturct and still respect unitarity.

I think that you are coming from a Platoist background in which you have these universal truths, and that the physical world is a "dirty" reflection of those truths which means that that you need to keep things separate.

I thought I was able to convey more than that :( No I've tried to argue against he opposite. I do not adhere to universal truths. My point is that this view is flawed, and instead the focus is on improvement. But this doesn't mean we can have theories. It just means we need a new understanding of theories, as an interaction tool, not a static description.

/Fredrik