Predicting Randomness - great research

Discussion in 'Automated Trading' started by syswizard, Feb 27, 2006.

    I'm just showing another article I stumbled upon on the 'net....with my commentary natch.
    It appears as though our Russian friends are interested in trading the markets with neural nets....and they are connected to this large brokerage house in Switzerland. Interesting stuff.
  2. Hmm, interesting, are there any more detailed results? I don't know if they are accounting for survivorship bias accurately.. also.. I don't know how accurate their 'games' are. They are making an awful lot of assumptions about the exact nature of randomness.
  3. Note: I am trying to contact the authors and get more information.
    Regarding the above: you saw the number of randomness "tests" they performed....I think they have a good handle on the nature of randomness. In fact, this A. Duka, is supposedly an expert in that area. I will post his/her article shortly.
  4. I meant a handle on the exact nature of the randomness in the markets. Wasn't commenting about their knowledge of tests, random statistics, etc.

    I sure don't see what trading random number generators, or radiation, or randomly corrupted EUR/USD data has to do with actually trading real markets.

  5. nitro


    You guys keep talking about randomness in a totally uneducated way. The correct term is Deterministic Chaos, and most likely high dimensional chaos.

    By the mathematical definition or randomness, it is impossible to predict randomness. That paper is written by hacks.

  6. bitrend


    I think what is possible is to predict within a range, not the exact number that will come. Generally they use Statistics (Variance, Standard Deviation) to predict the outcome random variable based on past values.

    For example, we can build a software to calculate the variance/std_deviation of the prices of a given security and reject any new price that is unexpected (out of range). Ex. stock ABC is trading 5.10, 5.12, 5.09, 5.15 and then suddenly 5.50. The last value 5.50 will be rejected. We can also put the 5.50 in the calculation but we will not make trade decision on it, not for now maybe later. At one moment if the stock continue to trade long time enough around 5.50 then the new price of 5.60 will be a good value since the new variance/stddev is recalculated.

    More detail on Statistics:
  7. The paper definately isn't useful, exept as maybe mental masturbation. But, the definition of a "random variable" is such that it is not deterministic, but can be described as a probability distribution. Just because a variable is 'random' doesn't mean its completely unpredictable, but I'm sure you knew that already.

  8. nitro


    You cannot predict anything that is truly random. Period.

    A distribution describes a random variable through its moments, but that is not the same thing as saying that there is anything predictable about it. Random means without cause - therefore if something is predictable there must be cause. There is no causation in a truly random process. All predictable sequences AFAIK afford "compression." Randomness affords no compression.

    However, many processes that appear to be random are in fact not, e.g., Deterministic Chaos. That means they can be predicted under some constraints.

  9. You are absolutely right... also, just because you calculate the mean and standard deviation of some series, that does not mean that it is the actual moments of the generating process. That just means that you calculated the sample moments of the past... not that useful except for describing the past. Random walks by definition have no mean and infinite variance. There are many things that appear random at first inspection and one might be fooled to think that it is completely random, it also goes the other way, one might see patterns in causation where in reality there is none.

  10. bitrend


    Exact. Trying to predict a random is like trying to validate Einstein Theory God doesn't play with dices which is disqualified by the Quantum Theory.
    #10     Feb 27, 2006