I come from an econometrics background and recently I am seeing the presence of a chasm between the way econometricians analyse data which tends to differ in the way that the more applied science/physics analyst may analyse market data. I am specifically interested in the comments that physics people can add to this thread as to their take on the weaknesses of econometric analysis of market data.In econometrics, the tools tend to work only when you transform the original time series via first differences from price data into return data so that you can operate under the assumptions of normality and stationarity (caveats here). The exception is here is co-integration. Then you apply OLS or a Max Likelihood function to develop a regression model of some sort. There is ARCH, GARCH, Kalman and others and the more recent work on non-linear time series analysis but that seems to be leading edge stuff at the moment.On the other hand, people from a physics background tend to not know nor learn things like Jarque-Bera tests, GARCH and Kalmans. What seems to be the fundamental difference with the way the physics world looks at time series versus the supporters of econometrics ??? I am genuinely interested in the reasons why some people from a physics trained background tend to frown upon some of the methods used in the econometrics discipline???I would appreciate any comments. Thanks.

'Frown' is perhaps to strong a word. As for my 2c:Generally econometrics tend to work with a fairly limited class of models (usually linear(ish)) and are by far too hung up about testing. Aas someone with a science and mathematical stats background, testing usually is far down the line - after I have looked at how to model the process under study.Econometric analysis appears very normative to me, lots of strong assumptions, tests for them, etc.

You don't need to be a physicist to frown on econometrics. I come from an engineering background and the first time I came across econometrics and specifically how it is used to present and/or develop so called "theories" or "models", my jaw literally dropped. Understandably, physics, and many of the other areas that sprouted out of it, have been around for a few centuries. So, thankfully, they are developed to the point that they don't have to rely on regression-style models.The problem with econometrics, I think, is that people tend to become too involved with the underlying statistical methodologies and, therefore, the focus shifts on to developing and implementing tests rather than to furthering the area of economics/finance itself. It's good Black-Scholes came out when it did. Otherwise, I wonder where the field of option pricing would be today without it.

Last edited by rcohen on June 29th, 2004, 10:00 pm, edited 1 time in total.

- DogonMatrix
**Posts:**242**Joined:**

QuoteOriginally posted by: rcohenYou don't need to be a physicist to frown on econometrics. I come from an engineering background and the first time I came across econometrics and specifically how it is used to present and/or develop so called "theories" or "models", my jaw literally dropped. Understandably, physics, and many of the other areas that sprouted out of it, have been around for a few centuries. So, thankfully, they are developed to the point that they don't have to rely on regression-style models.The problem with econometrics, I think, is that people tend to become too involved with the underlying statistical methodologies and, therefore, the focus shifts on to developing and implementing tests rather than to furthering the area of economics/finance itself. It's good Black-Scholes came out when it did. Otherwise, I wonder where the field of option pricing would be today without it.Mostly misinformed point of view.First thing, option pricing belongs to the field of financial theory or mathematical finance, not econometrics. You can have to solve some econometrics problems related to option pricing like estimating the parameters of a stochastic vol models, but the two are distinct.Second thing, I can not think of one methodology developed in engineering that is useful today in finance. The only example that may come to mind is kalman filter or frequency domain type of estimation methodology, but they become useful only once one has translated them in econometrics terms.Third thing, econometrics is not hung up on testing, but rather on the significance of the relationships that are estimated. Without property and results related to statistical distributions I really don't see how one can choose one parameterization of a relationship over another. Fourth thing, linear models are usually favored by practitioners over non linear ones, just because practitioners favor parsimonious models. But academia has moved to and studied more complex non-linear models for more than 20 years ( where they perform well, and can be estimated in a robust fashion).

I really do not think any particualr methodoly is frown upon in any particualr field; what might be frown upon is the way you use it. For example, most "econometrics methods" can be found in standard book for electrical engineering (communications, basic stocahstic process, etc); I think that is you talk to an EE working for a communication firm and you mention "stochatics process" he will, more likely than not, think of time series analysis; Go ahead test this assumption.If you say the same phrase to an industrial engineer he might think of "queuueing models"; For the EE the assumptions that someone is try to convey information is very realistic, tehfore, time series analysis when looking to "get a model" of what information is being transmitted is quite useful; for the IE any idea that the "line is trying to convey information" will look real weird; the line is not "intelligent" so ther is no way is trying to ciommunicate anything.My point; econometric assumptions and methodolgies wil mean something only is what you are try to model is in close agreement with the model assumptuion 9here my main objection to "astrology methodology. A.KA. "technical analysis" (what the heck you mena with "the market is telling me anything??? The market is not an intelligent being, it is not concoius; thus it cannot "tell me" anything; would you accept the idea that "a chair moving around is doing so because: it is "smart" or it trying to tell yo where it is going? or you you accept the idea that several hundreds guys are trying to move the chair to one place (more oor less each one of them to a particualr place of their liking) the result has to be that the chair is not moving with any particualr path.If you use econometric methods careless the result might be elegant (if you use any kind of math there will be an elegant result) but they will tell you anything about what the heck is actually going on. I think that is why is might look as if physics" or whoever) look frown upon (That is the reason I do frown upon on certain mathods badly applied).Nevertheless I do use time series analysys, mostly for EE purposes and, depending they might be applicable in finance as well (others problems do arise; nonlinearities for eaxmpel; etc. etc. time dependent parameters, etc . etc

QuoteOriginally posted by: DogonMatrixThird thing, econometrics is not hung up on testing, but rather on the significance of the relationships that are estimated. Without property and results related to statistical distributions I really don't see how one can choose one parameterization of a relationship over another. I guess that is a good description of what I meant: models in science are not about chosing a parameterization but a good model. A better model might well have a worse fit/distribution etc. but explains relationships better (case in point: ptolemaic v keplerian model). Without an 'organic' reason, statistical significance can (and often is) meaningless.

QuoteOriginally posted by: DogonMatrixMostly misinformed point of view.I strongly disagree. I have graduate degrees in both engineering (mechanical & chemical) and economics and I have also worked extensively in both areas. So I could confidently claim that I have been on "both sides of the fence." Can you say the same about yourself?QuoteOriginally posted by: DogonMatrixFirst thing, option pricing belongs to the field of financial theory or mathematical finance, not econometrics. You can have to solve some econometrics problems related to option pricing like estimating the parameters of a stochastic vol models, but the two are distinct.Option pricing is NOW a part of mathematical finance, after the development of Black-Scholes. Look at how equities and the risk premium are priced via APT. For this you can refer to many of Fama's well-cited works, among many others. From this, also, it's not difficult to conclude that, in the absence of some form of underlying rule or law (i.e. which may be represented by, let's say, some sort of a pde) econometrics is still the norm in many areas of finance.QuoteOriginally posted by: DogonMatrixSecond thing, I can not think of one methodology developed in engineering that is useful today in finance. The only example that may come to mind is kalman filter or frequency domain type of estimation methodology, but they become useful only once one has translated them in econometrics terms.Engineering grew out of physics and, admittedly, it could be classified as applied physics. For example, while physicists are concerned about statistical thermodynamics, engineers are more involved in its continuum form, which is classical thermodynamics. Do you know what the heat, the viscous diffusion or the soil consolidation equations are? They are all the diffusion equation, which forms the very logic behind Black Scholes. These equations, along with neural networks, Laplace transforms, Fourier transforms, etc., are all part of the everyday language in engineering. If you claim that, except the Kalman Filter and frequency domain estimation, engineering has had little impact on finance, I suggest you go read some books outside your area of expertise.QuoteOriginally posted by: DogonMatrixThird thing, econometrics is not hung up on testing, but rather on the significance of the relationships that are estimated. Without property and results related to statistical distributions I really don't see how one can choose one parameterization of a relationship over another. To me, a tool that allows you to choose one parameter over another (i.e. which one is more significant) is nothing but a testing tool. QuoteOriginally posted by: DogonMatrixFourth thing, linear models are usually favored by practitioners over non linear ones, just because practitioners favor parsimonious models. But academia has moved to and studied more complex non-linear models for more than 20 years ( where they perform well, and can be estimated in a robust fashion). It doesn't matter how complicated things can become in econometrics. You could make the equations nonlinear and increase the number of independent parameters from 6 to 26. At the end what counts is the lack of a governing law. For example, in dynamics you have F=ma; in solid mechanics you have Hook's law, which relates the applied force to deformation; in thermodynamics you have conservation of mass and energy; Brownian motion is governed by the diffusion equation, etc. I just fail to see any underlying rule that is applied in econometrics, except for choosing as many parameters as possible and massaging the equations to the point of getting a good fit.

- DogonMatrix
**Posts:**242**Joined:**

QuoteOriginally posted by: ZedQuoteOriginally posted by: DogonMatrixThird thing, econometrics is not hung up on testing, but rather on the significance of the relationships that are estimated. Without property and results related to statistical distributions I really don't see how one can choose one parameterization of a relationship over another. I guess that is a good description of what I meant: models in science are not about chosing a parameterization but a good model. A better model might well have a worse fit/distribution etc. but explains relationships better (case in point: ptolemaic v keplerian model). Without an 'organic' reason, statistical significance can (and often is) meaningless.Could you describe to me what will be a trading strategy based on " ptolemaic model" ? we are not talking philosophy here, are we ? the ultimate goal is to make money trading, and underlying my argumenrt is that, for that purpose (trading), econometrics methods have proven to be much more useful than engeneering***A better model might well have a worse fit/distribution etc. but explains relationships better *** that's really weird to me to have a model that explain none of the actual variance of the data but still will claim to capture a meaningfull relationship. What rigorous criteria will you use to define meaningfull in that case ?

Last edited by DogonMatrix on June 30th, 2004, 10:00 pm, edited 1 time in total.

- DogonMatrix
**Posts:**242**Joined:**

QuoteOriginally posted by: rcohenQuoteOriginally posted by: DogonMatrixMostly misinformed point of view.I strongly disagree. I have graduate degrees in both engineering (mechanical & chemical) and economics and I have also worked extensively in both areas. So I could confidently claim that I have been on "both sides of the fence." Can you say the same about yourself?QuoteOriginally posted by: DogonMatrixFirst thing, option pricing belongs to the field of financial theory or mathematical finance, not econometrics. You can have to solve some econometrics problems related to option pricing like estimating the parameters of a stochastic vol models, but the two are distinct.Option pricing is NOW a part of mathematical finance, after the development of Black-Scholes. Look at how equities and the risk premium are priced via APT. For this you can refer to many of Fama's well-cited works, among many others. From this, also, it's not difficult to conclude that, in the absence of some form of underlying rule or law (i.e. which may be represented by, let's say, some sort of a pde) econometrics is still the norm in many areas of finance.QuoteOriginally posted by: DogonMatrixSecond thing, I can not think of one methodology developed in engineering that is useful today in finance. The only example that may come to mind is kalman filter or frequency domain type of estimation methodology, but they become useful only once one has translated them in econometrics terms.Engineering grew out of physics and, admittedly, it could be classified as applied physics. For example, while physicists are concerned about statistical thermodynamics, engineers are more involved in its continuum form, which is classical thermodynamics. Do you know what the heat, the viscous diffusion or the soil consolidation equations are? They are all the diffusion equation, which forms the very logic behind Black Scholes. These equations, along with neural networks, Laplace transforms, Fourier transforms, etc., are all part of the everyday language in engineering. If you claim that, except the Kalman Filter and frequency domain estimation, engineering has had little impact on finance, I suggest you go read some books outside your area of expertise.QuoteOriginally posted by: DogonMatrixThird thing, econometrics is not hung up on testing, but rather on the significance of the relationships that are estimated. Without property and results related to statistical distributions I really don't see how one can choose one parameterization of a relationship over another. To me, a tool that allows you to choose one parameter over another (i.e. which one is more significant) is nothing but a testing tool. QuoteOriginally posted by: DogonMatrixFourth thing, linear models are usually favored by practitioners over non linear ones, just because practitioners favor parsimonious models. But academia has moved to and studied more complex non-linear models for more than 20 years ( where they perform well, and can be estimated in a robust fashion). It doesn't matter how complicated things can become in econometrics. You could make the equations nonlinear and increase the number of independent parameters from 6 to 26. At the end what counts is the lack of a governing law. For example, in dynamics you have F=ma; in solid mechanics you have Hook's law, which relates the applied force to deformation; in thermodynamics you have conservation of mass and energy; Brownian motion is governed by the diffusion equation, etc. I just fail to see any underlying rule that is applied in econometrics, except for choosing as many parameters as possible and massaging the equations to the point of getting a good fit.Once again a lot of confusion here. First thing about Laplace, Fourier transform, stochastic processes, etc, these are principles and methods out of theoretical physics not specifically engineering methodologies. So it's little bit of a short cut to claim that there are(theoretical physics and engeneering are distinct). But that's probably a sterile debate...Second thing, on the " absence of governing laws" in econometrics. I almost want to say, " who cares about governing laws!". What matters is trading performance, and that comes along with models that will capture most of the stylized facts out of the market. But what's more, econometrics in many aspects is a methodology to tie a model to reality, it is not in iteself a model. Example I can have a stochastic volatility model which in my mind is distinct from the maximum entropy or kalman filter or generalized method of moments that I will use to estimate the parameters of that model. If you are looking for what will resemble to most the governing laws that you find in Physics , then you probably need to go upstream an look at economics: things like no-arbitrage argument ( which is probably as important to the Black Scholes argument as the GBM process), or Say's law, etc.But I want to emphasize that for me the ultimate criteria is trading performance: if you are " massaging the equations to the point of getting a good fit", first, you are probably a very bad econometrician, second , chances are you will not get any trading performance out of that. Period.You are right I can't claim to know anything about engeneering, but never had any interest because never appeared to me as being the best way to understand the market.

Last edited by DogonMatrix on June 30th, 2004, 10:00 pm, edited 1 time in total.

QuoteOriginally posted by: DogonMatrixBut I want to emphasize that for me the ultimate criteria is trading performance: if you are " massaging the equations to the point of getting a good fit", first, you are probably a very bad econometrician, second , chances are you will not get any trading performance out of that. Period.You are right I can't claim to know anything about engeneering, but never had any interest because never appeared to me as being the best way to understand the market.I guess we can all agree that actual P/L delivery trumps over epistemological and technical reasons for or against certain methods, in the end trading is not a natural science. I for myself are happy to make money without using econometric methods to any great extend (but using - amongst others - statistical and mathematical tools). If others do make money using them, fine by me. As for understanding the market, the tools we need are probably not yet developed...

My experience with applied physicists and econometricians is the former are “doggy-dog” about the efficacy of their results (pricing models), and the later are too cautious to be useful contributors in a trading environment. On the average, it is easier to “pick-the-pockets” of applied physicists in a trade than econometricians, who by designs are extracting “pound-of-fresh” for their models' errors.

Last edited by pobazee on July 1st, 2004, 10:00 pm, edited 1 time in total.

If you are acting as a mathematician, the key question is, "Is it right?"If you are acting as a scientist, the key question is, "Is it interesting?"If you are acting as an engineer, the key question is, "Is it workable?"I think that mathematical finance, especially as practiced by the sell side, is firmly in the field of engineering.I think that econometrics and physics (and mathematical finance as practiced at universities) is usually science, where the key criteria for judging a work is how interesting are the results. In fact, a lot of the field of economics is firmly in the area of mathematics (proving correctness within an assumed set of preconditions is the key result, not showing applicability to real world situations).Using econometric or applied physics or engineering tools to analyze a phenomenom in the market is usually just the first step in trying to understanding which forces are currently balanced off in the marketplace; it is this understanding which is valuable, not the means of achieving it. If one discovers, say, a statistical relationship between two market instruments, , one would be advised to understand what's causing the relationship before trading on it; otherwise you can get hung out to dry.An example: some years back we noticed that whenever the Nikkei took big hit, the US 30y treasury soared for the next day or so.The next time the Nikkei tanked, we immediately put X million into the US long bond. We did well. The next time the Nikkei tanked we had more confidence in the strategy, and we put 2X million to work, and did twice as well. And then 4X mllion. But the next time, we put 8X million to work, and the trade moved against us. Because of the size of the trade, we wiped out all previous trading profits created by strategy. This was a very instructive incident, worth every penny of the "tuition" we paid. First, clearly there was some investor class that was getting hurt in Japan, and each time the Nikkei took a hit, they were running to cover into the safest security they could find, the US 30 year bond. After this happened several times, the investors were probably full up on the US 30Y bond (or the hot money had left the Nikkei or else some other security was viewed as more attractive than the overpriced 30Y bond). We never bothered to look into who was selling the Nikkei and who was buying the bonds, and asking whether their appetites were the same. But our big mistake was continually increasing the size of our bet because it had paid off for us before. The market couldn't care less if we made or lost money on this strategy N times in the past, so rationally this should not pick the size of the bet. Rational would be to pick the size based on the perceived risks versus our appetite for risk

Last edited by Pat on July 1st, 2004, 10:00 pm, edited 1 time in total.

Pat:Very, very insightful rejoinder…and I thank you for that. Your rejoinder shred some lights on my poser in the technical forum, which I recall below:"Thu Jun 17, 04 09:41 AM A prevalent verbiage among market professional is the notion of Richness/Cheapness, and I have always wondered if the perceived richness/cheapness of a given security is based on temporal equilibrium configuration, core equilibria, or risk-neutral value (a la arbitrage-free pricing model). If it is based on arbitrage-free pricing model, how does one insure that playing richness/cheapness in the market would not lead to “doubly ruin” strategy in a time continuum? Your thought is appreciated."

- DogonMatrix
**Posts:**242**Joined:**

I think the field of quantitative trading is by essence multi-disciplinary: there are some tools useful to model certain behaviors and other to model other types of behaviors. Being hung up to use only one methodology is usually at the source of trader's undoing. Moreover I think the goal of the game is nothing else than capturing inefficiencies, and that means nothing else than , doing something that nobody else's is doing. When Jim Simon was hiring the whole speech recognition department of Bell lab, I am sure nobody else ( and still today most likely) was thinking about using these types of approaches ( and most likely whether or not there are "governing laws" in speech recognition was totally a moot point). So may be bashing econometrics is a very good thing--all the attention from the Nobel prize committee on Engle and Granger is probably unwarranred--..... Please don't use it, it's useless

Last edited by DogonMatrix on July 1st, 2004, 10:00 pm, edited 1 time in total.

The phy people I have met turn to belittle econometrics in general. They will make some assumptions to begin with, working with some SDE's, then using whatever optimization to get the parameters. No tests of any kind.... They think finance is some thing you can learn in a couple of weeks...Also I notice a few replies from Phy background people here I guess , only mention econometrics has assumptions, never talk about the assumptions made by the phyisists...

GZIP: On