Yahoo data sources:
http://finance.yahoo.com/exchanges
So Telekurs (http://www.telekurs-financial.com) or Interactive Data Rts (http://www.interactivedata-rts.com).
You're "one of the leading providers of financial data" and you won't say who you are?
I sent you an email, but I do agree with lrm242 that it would be great if you could publicly give us some more information here. Openness is awesome and stealthiness is overrated.
How "fresh" are stock markets data? Does those include EU market too? Italy, in particular? thank you
Personally I think this is pretty cool and appreciate the posting, although I would prefer some sort of link.
Do you have historical stock market data (going more than thirty years back)? Or just recent data?
Do you provide real time data? Last sale? Bid-offer? Full book? Quotes or streams?
Sounds interesting, at the same time, mighty sketchy. Are you going to cut us off from the data once we develop something interesting, and run off with the idea and sell if yourself?
No thanks..
you looking for any particular type of user (company), or are hacker types ok?
"Show us what you can build."
Okay:
(1) As we know well from J. Doob's work in martingale theory and there the Doob decomposition, every stochastic process is the sum of a martingale and a predictable process.
So, using the data, estimate both processes. Then use the predictable part of make money?
(2) Also, from the martingale convergence theorem, every martingale either converges almost surely to some random variable or runs off to infinity. The interesting case is the convergence in which case the predictable part is more interesting. Look at the real data and test for and try to find convergence.
(3) For i = 1, 2, ..., let X(i) be the change in the price each time 'tick' i. So, each X(i) is a real valued random variable. Assume that it has an expectation and that E[X(i)^2] is finite.
If the set of all X(i) is independent and if all the X(i) have the same distribution, then, by an elementary version of the central limit theorem, for any i and for not very large n,
Y = X(i) + X(i + 1) + ... X(i + n)
will have a distribution that is quite accurately Gaussian. Since it is accepted, e.g., in the Jim Simons talk at
http://paul.kedrosky.com/archives/2011/01/james_simons_sp.ht...
that the tails of the distribution are 'fat', then at least one of independent or identically distributed has to fail. I vote for independence fails. Then there should be some predictability. Look for it.
(4) If convert dollars to yen to marks to francs to pounds to dollars, then should come out even. Similarly for any such currency trades. But without trading to enforce these relationships, they need not hold in which case there will be arbitrage opportunities. Look at the currency data and see if there are some significant arbitrage opportunities.
(5) Imagine a simulation of currency values: Someone sells $1 billion dollars for pounds. Now other currency values have to change. In this simulation, the rule is, to get the currency values back for no arbitrage, can only make trades of currency pairs, one at a time. So, are doing crude two dimensional iterations to get back to a multi-dimensional 'equilibrium', and that can't be very fast. Or as we know well from non-linear minimization, such iterations tend to oscillate; we try to damp the oscillations with conjugate gradients or quasi-Newton iteration, but, still, there is a lot of oscillation. Also, we do not have accurate global definitions of all the 'supply-demand' curves. So, there should be some 'dynamics' getting back to equilibrium and some predictability. Maybe there will be some multidimensional 'ringing' that would result in some predictability. Look for it.
(6) Take some possibly related real valued stochastic processes and do a real time, distribution-free, multidimensional hypothesis test where the null hypothesis is that the processes are acting as usual. When can soundly reject that hypothesis, take a position that will make money when the process becomes 'usual' again. Where to get a family of distribution-free, multidimensional hypothesis tests? Glad you asked. Use the data to test this idea.
For "hacking", this is applied math, especially about stochastic processes, and not 'hacking'.
Flagged. If you want the feedback of the community you have to at least have the backbone to post a damn URL. Advertising spam like this with some unbounded claim to email so we can "work something out" is an abuse of this system, IMO. Post a link or don't ask for help.