11
votes

A relative of mine works for a big bank (of bailout fame) in the investment division. We had a discussion where he claimed that what he does is actually a science. Me being an engineer called baloney on this. IMHO there is no science that can predict the future without a decent mathematical model, and I think the world of investing lacks said models.

Some of the arguments he presented have to do with the sophisticated software models that are used for automated trading. The argument I presented is that all models are wrong (see 2008) and they are just delusional in their science.

To settle this, I need to know if there is any falsifiable and repeatable propositions that form the basis of the science for big time, multi-billion dollar bank investing.

Update: I found an older article here with mathematical modeling for macroeconomics for the Bank of New Zealand, and it shows (below) the forecast (wide gray areas) nowhere near the actual data (dashed lines).

Model Predictions for Bank of NZ

  • 3
    How do you define the difference between "art" and "science" in this context? – ESultanik May 17 '11 at 21:07
  • 3
    @ESultanik: The question he's really asking is "Do mathematical models for investments help maximizing profits, or are they unreliable?" If you see a way to make his question clearer, feel free to propose an edit. – Borror0 May 18 '11 at 0:59
  • 4
    xkcd.com/435 – ChrisW May 18 '11 at 1:14
  • 1
    There might be models that work well during the normal days, but fall apart when the market conditions change. In the days preceding the change, there would be "indicators" falling off the charts, signalling that what is going to happen is unpredictable. – user2547 May 18 '11 at 3:45
  • 2
    just because they models produced aren't accurate doesn't make the process of creating them unscientific. It just means that the science isn't yet able to yield accurate enough results for the purpose those results are being employed to do. It's similar in nuclear physics. I worked as a grad student on R&D relating to nuclear waste disposal, trying to reduce the margin of error of our technique to analyse waste composition from 5000% (which was acceptable if not optimal) to 500%. – jwenting Nov 30 '11 at 12:51
9
votes

A famous example of the efficacy (and also problems inherent in) mathematical modeling is the concept of Dynamic Hedging ← a transcript of an excellent documentary on the subject. The model was created by Black and Scholes, for which they were awarded the Nobel prize in Economics. Under certain reasonable assumptions, their model is guaranteed to reduce risk in proportion to the amount of investment one has made (which seems a bit counter-intuitive at first). The best analogy up with which I can come is that Dynamic Hedging is a lot like opening a casino. The model ensures that the odds will always be in the house's favor, so in the long run you will, in expectation, make a profit. The problem is that every once in a while someone will likely "win big", and your casino will have to make a payout to them. That's fine, as long as you have enough liquid money to make that payout; you'll eventually recoup the loss in the long run. The problem is that, unlike a casino, the instantaneous risks of the market are unbounded. For example, there is unbounded risk associated with shorting a stock. Therefore, as long as you have an unlimited supply of liquid cash (or credit) to deliver those payouts along the way, you are guaranteed to at least break even. Black and Scholes' model was implemented in the form of a number of hedge funds, most notably Long-Term Capital Management, which enjoyed spectacularly consistent returns in the 40% range throughout the 1980s and most of the 1990s. The problem (I'm greatly simplifying things here) was that eventually the fund got so big that it was no longer able to borrow/acquire enough liquid capital to sustain itself.

So what's happening these days? The financial concepts of arbitrage and hedging are theoretically guaranteed to be risk-free. The idea is quite simple:

  1. A widget is being sold by party X in London for $5.
  2. At the exact same point in time, due to market fluctuations, the same type of widget is being bought by party Y in New York for $6.
  3. If we can find out about #1 and #2 quicker than Y can find that same information, then we can buy the widget from X and sell it to Y at a $1 profit.

That's why we currently see the "big banks" investing millions (and some speculate billions) of dollars into high-tech computer centers very close to the big stock exchanges. (Sorry, that video is in Nederlands, but a lot of it is just dubbed English, and many Dutch words are mutually intelligible with English.) Since almost all trading is done electronically these days, the quicker one can get information from and make trades on the markets, the greater chance one has at exploiting arbitrage. In practice, though, stochasticity of the market does induce risk, which has led to widespread use of statistical arbitrage.

Edit:

I feel I should elaborate on the "science" vs. "art" component of this question, and why I brought up the question of their differentiation in the original set of comments. In my mind, the classifications of science and art are not mutually exclusive. The primary defining characteristic of science is a method by which predictions about a previously mysterious process can be made. Art is the display, application, or expression of a method, almost always with the intention of affecting others' emotions or intellect. By these definitions, science and art are not mutually exclusive, e.g., I believe one could make an argument that cooking is both a science and an art.

Anyway, without getting sidetracked too much, I also want to mention models. I think Newtonian mechanics and Newton's law of universal gravitation are great examples. Newton's models for the way the universe works are extremely good at "predicting the future", at least on the scale of ordinary interactions here on Earth. Unfortunately, Newton's models break down if we change the scale. For example, if we look at a planetary scale, they don't accurately predict the perihelion of Mercury; it wasn't until the turn of the 20th century that models were created that could both accurately predict mechanics on the scale of humans and also mechanics on the scale of planets (most notably Einstein's general relativity). Given that we know there are phenomena in the universe that Newton's model is incapable of predicting, does that mean Newton's model isn't science or is "delusional in [its] science"? Of course not. In fact, we still use Newton's model for many tasks (e.g., engineering) down on the human scale, because it's a lot easier to use than the more robust models.

(I apologize for all of the Wikipedia references; I just think they're a good way to get a high-level understanding of some of these concepts.)

  • 2
    If the Black and Scholes model would be true Long-Term Capital Management would have to borrow more money and wouldn't have become broke. The fact that the fund crashed showed that the basic assumptions of the model don't hold. – Christian May 18 '11 at 13:33
  • 3
    While I agree with the premise of what you're saying, I don't buy the exact argument. Does the fact that the fact that the Fukushima Daiichi Nuclear Power Plant functioned well for ~40 years and then had a meltdown as a result of an unprecedented, catastrophic series of events show that the basic assumptions of the model of safe nuclear power do not hold? – ESultanik May 18 '11 at 13:59
  • 1
    @Christian - no. The fact that it crashed showed that the model has limitation of applicability (e.g. amount of available liquidity/credit). – user5341 May 18 '11 at 14:59
  • 2
    @ESultanik: The model predicted that the Nuclear Power Plant will be safe as long as there aren't any earthquakes that are stronger than a certain threshold. If the Fukushima would have crashed with an earthquake under the threshold we would have to rethink the basic assumptions of nuclear power. We weren't willing to pay the price to guard the nuclear plants against stronger earthquakes. The Black and Scholes model doesn't tell you about cases where it doesn't hold. In Science a single observation is enough to disprove a theory. – Christian May 18 '11 at 15:22
  • 3
    @DVK: There no parameter in the model that allows it to predict events that don't follow the Gaussian distribution. The Black Scholes approximates financial processes that follow an extreme value distribution with a Gaussian distribution. In you want to know more about the problem Benoit Mandelbrot wrote a full book about it titled: "The Misbehavior of the Markets". – – Christian May 18 '11 at 22:49
1
vote

The argument I presented is that all models are wrong (see 2008) and they are just delusional in their science.

Is medicine a science? Medicine has test tubes, and statistics, and machines with blinky lights. And, models: trials, hypotheses, etc. So it looks like a science.

But sometimes people die. Unexpectedly. Does that mean that "all models are wrong", to quote your argument?

Similarly I think that investing is a science, with various caveats which apply to medicine too, for example:

  • Don't know everything
  • Don't always apply the model correctly (the right model at the right time, given that different models are appropriate for different times)
  • The part of medicine that uses mathematical models (the field of bio-engineering) is definitely a science, but the stochastic trial and error methods of today are not necessarily. In diagnosis, if 99% of people with flu like symptoms have the flu, that means there is a 1% probability it is not the flu, but something else. If you follow the model (symptomps)=>(diagnosis) then you will be wrong. – John Alexiou May 19 '11 at 11:58
1
vote

There a two questions mixed in together here.

The first is whether the mathematical modelling they do is a science. That question can't be answered until we have an agreed definition of science (or at least an approximate one).

  • If science is about producing and testing theories that model the real-world, then yes, it is a science.
  • If having issues with reflexivity (see separate answer) means it can't be called a science, then no, it is not a science.
  • If the best theories are not perfect means it is not a science, then no it isn't a science and your definition of science is pretty poor.
  • If peer-review is a key requirement for science, then many of the proprietary models that individual organisations are using are not science, but other elements discussed in universities are.

My point here is that the question is futile, if we can't agree on the definition. Once we agree on the definition, the answer will be trivial and uninteresting.

The second is whether the modelling is successful.

Yes, it is moderately successful. Not all of it is perfect - some people may have private models that are plain wrong. Even the higher-quality models may not be perfect, and have limitations.

The goal, though, is not to be always right, but to be right often enough to make a living.

Note: In the example fan-chart given, the blue lines do not represent all of the possibilities, but just the 95% confidence range (or 99% confidence range or the 99.997 confidence range - alas the reference did not define it.) Coming up with one example of it being outside this range counts as anecdote in a world where risk is an intrinsic part of life.

(No references, because this is all my basic understanding from working in a related field of mathematical modelling, where being right 55% of the time would be very profitable.)

  • 3
    +1 and I'd just like to add that 'it uses "the scientific method" might be a definition of whether it's science' ... which, I was told, consisted of a) make observations b) create a hypothesis that fits (isn't contradicted by) the observations c) devise experiments to test ('prove') the hypothesis d) synthesis (modify the hypothesis as necessary to account for experimental results). – ChrisW May 21 '11 at 11:49
  • it is not necessary to be right in investing but it is essential to know when you are wrong. I read sometimes a study about advisors and they were most helpful during times when clients did not notice false premises or were diverted to dangerous waters. Good advisors let their clients "to stay on course" and forget the non-sonse about "truths". Perhaps, you mean the same thing but your sentence is highly dangerous: "The goal, though, is not to be always right, but to be right often enough to make a living.". "55%", non-sense. Only marketers and "news"-papers can make bucks with being 55% right – user4312 Aug 1 '11 at 14:55
  • +1 despite the shortcomings I liked your answer. Throwing trivial numbers however at the end is not convincing, 55%, where did you get that? – user4312 Aug 1 '11 at 15:09
  • @hh, sorry, perhaps the 55% number is misleading. I am not claiming that investment bank modellers are only 55% right. I am claiming in my industry that is the case, and that is why the concepts of risk taking fall under "basic understanding". Whether you should accept that as a justifications of having no references is a whole different question :-( – Oddthinking Aug 1 '11 at 18:18
0
votes

I don't know if you would regard this as 'science' or 'art' but there are definitely mathematical models for stockprizes and investing. The theory of stochastic differential equations and stochastic processes deal with modeling these kinds of situations. See here and here, for example

Edit

Actually, yes, this is a science. Since it's not about what we feel mathematics to be (a science or an artform) but it's about how these models are used to describe real life phenomenon. Any selfrespecting mathematicians will use scientific rigor in applying these models (fitting a model to real life data, reformulating a theory if there seems to be a difference between real life and the model describing it, etcetera). So it's definitely a science.

  • Models are reliable only for as long future events reflect previous events. While that's often the case, sometimes the future is simply unpredictable. So, the question is: are the models profitable over time, or does unpredictability get the best of them? You have not address that in your answer. You can't call it a science if it's harmful to rely on it. – Borror0 May 18 '11 at 1:04
  • 2
    Do any of the models actually fare significantly better than chance? At least for predicting the stock market, they didn’t, last time I looked. Of course, investment is slightly different from stock market. – Konrad Rudolph May 18 '11 at 8:55
0
votes

Economics can not be a science because the study itself effects the markets they're studying. Expectations can be self-validating, therefore removing independence between observer and subject that is necessary to be considered a science. George Soros figured it out decades ago but, because it negates economics all together, it never really caught on. The concept is called reflexivity and it demotes economics to a fancy ouija board.

http://en.wikipedia.org/wiki/Reflexivity_(social_theory)#Reflexivity_in_Economics

  • 2
    Can you please provide a reference (or explanation) for the "independence between observer and subject" requirement for science? It strikes me that many sciences may be troubled by what Popper called the "Oedipal Effect", but that doesn't preclude them from being sciences. – Oddthinking May 21 '11 at 5:46
  • This may be correct answer but there is no guarantee about truth in it. Similarly, investing/medicine/etc models can be correct but there is no guarantee about truth. If you define science only as "something than is correct (models correspond to conclusions) and true (models correspond to experiences)", there is some sense thrown in but I am totally lost with your conclusion about reflexivity and its relation to this question. Reflexivity is a totally unrelated topic or am I missing something? – user4312 Aug 1 '11 at 15:16
  • There has been an experimental laboratory branch to Economics (esp. microeconomics) for quite a while now. Vernon Smith shared a Nobel Prize for his early work in this specialty. – Paul Sep 5 '11 at 23:13
  • "Economics can not be a science because the study itself effects the markets they're studying. " Observing a subatomic particle affects its state. Does that mean quantum physics isn't a science? – ESultanik Oct 22 '15 at 16:42

Not the answer you're looking for? Browse other questions tagged .