software for data presentation, statistical analysis, marketing and prediction.
Free download:
LeoStatistic.zip
or
LeoStatistic.exe
(selfextracting winzip file)
Registration
Is market predictable?
Most generalized answer is Yes. Definitely so. Just from the principle of determinism, every course has its reason. If somebody has all information about present situation he can in principle calculate next situation. Market prediction is analogy to weather forecast but in a scale of whole economy. Yes, it's possible and most probable already fulfilled for very “big dogs”. Not for you and me with our personal computers. In the spite of limited amazing progress of calculating productivity for last twenty years they not yet there and what is much more critical we have no access to demographic and other economic information. Major players have.
Here we will discuss specific problem of day trading. Can one looking on the time - price chart make a prediction about the next move? Buy low, sell high. Is such information in the curve? I personally worked for the Chemistry professor who religiously believed that it’s possible. An image of very smart and generally old man marking tops and bottoms of quasi-cycles on long print of currency chart will be with me forever. At one time this exercise was even profitable. Just until the moment of blackout of conscience and luring by story of Silver speculation. The oil was that time real cheap. Silver went nowhere. Technology (digital photography) and geology are influence market fundamentals much stronger then short time speculations. Alchemistrymetrics (numerical alchemistry) is much more adequate name for such exercise then chemometrics or econometrics as they wanted to name it.
In any way is attempt to predict market price just looking at one chart have any statistical reasons? Let's use LeoStatistic for analysis of DOW chart.
In any way is attempt to predict market price just looking at one chart have any statistical reasons? Let's use LeoStatistic for analysis of DOW chart.
Import DOW data.
Go to for example to moneycentral of msn.com. There is an option to download a file of the chart to the computer and open with MS Excel. I did it and here is the file for ten years with one day step. Really it's not very easy task specially taking into account that there is a limit on size of downloaded file, it in wrong order - not descending. In any case you can find already fixed for usage with LeoStatistic MS Excel file here. If select two columns area from row names "t" and "P" down; copy; open LeoStatistic; click on "Add" button; select "t" parameter click in "Modify" do by formula 0.001*t. Scale "P" in the same way 0.001*P. Scaling will just help accuracy of calculations. Do some fashion job to help presentation. If to all of this you will have this picture in LeoStatistic. You can just open this file. It is our initial point of investigation.
Does market has memory?
Question of predictability of market can be reformulate. Is there dependence between move of price before and after so named historical now - current record during analysis of historical curve. To do this lets insert two parameters by formulas:
- dF=P[1]-P, as value - changing price in one day after historical now
- dP=P-P[-1], as argument - changing price one day before historical now
and check their mutual dependence. Here the picture:
No dependence whatsoever. None.
Probably we need more arguments? Not just first derivative of price but several more? Lets create three additional parameters there are derivatives of changing price in historical past of second, third, forth order :
- dP2=dP-dP[-1]
- dP3=dP2-dP2[-1]
- dP4=dP3-dP3[-1]
and check correlations between all of existing parameters:
No correlations between moving prices in future and any of these arguments. Some correlations between arguments themselves... No idea what it could mean.
No correlations for multivariate regression of function dF(dP,dP2,dP3,P4) in any statistical method and number or combination of arguments. Try it yourself.
Interpolation of the fitting curve.
There is no short time correlation between past and future of price changing on the short time interval. Probably we can find a prolonging trend? Of course you always fit any such data with the curve. And even very precise. But how could we could be sure that found trend will prolong in future. LeoStatistic permits to do fitting only for the part of the data and then insert parameter on the base theoretical curve. Then we can compare how real prices behave versus theoretical interpolation.
Just as example arbitrary chose fitting interval till begin of 2004. Then will do fitting with several methods, insert this fitting curves as parameters and then show data in all interval.
- f1 = -9.2e+5 +9.2e+5*t -2.3e+5*t^2 +0.14*cos((2*pi/0.001)*(t+0.0007))
- f2 = +9 +2*cos((2*pi/0.00804)*(t+0.00088)) +0.96*cos((2*pi/0.00536)*(t+0.00025)) +0.14*cos((2*pi/0.000516)*(t+7.7e-5)) +0.56*cos((2*pi/0.002128)*(t+0.00061)) +0.24*cos((2*pi/0.000936)*(t+0.00085))
- f3 = -9.2e+5*1.0 +9.2e+5*t -2.3e+5*t^2
- ln(f4) = -1.1e+2 +56*t
Not impressive at all. Of course (probably) you always can find some areas that will fit and interpolate your data just fine. But nothing can tell you where established trend ends. Usual assumption is that trends are broken with events. Very reasonable. But what event is more important then other? Election of President? Terrorist attack? Last drop of oil burned in limo? You need outside knowledge to sort out events by importance that sent us to initial thesis - market is predictable but not with the day trader who knows nothing except technical analysis.
click on picture to enlarge
Distribution of two variables.
Approximation
(constructor style interface).
