Calling all Quants and Statistical Analysts

Hi Jam,

"quantify how far up or down it will move FOLLOWING that week GIVEN some condition (i.e. it was UP or DN in the prior week/n periods)"

Yes, that is a good idea ... I'll see if I can figure out how to do that in R.

I was also going to look at if its an up week, what is the (1) distribution and avg # of pips move below the prevoius week's high, and (2) distribution and avg # of pips move below the current week's open. This would be useful to konw if its likely to move X pips below ... if it moves much more than that, then historically it won't be an up week and will instead be a down week.

Overall, I think having the stats are helpful. It gives mathematical proof of closing near extremes for the week (rather than in the middle), and if I'm long and it looks to be an up week ... to hold. Look at EURUSD the past week (I know ... its just one data point ...), but I was quite long at the start of the week ... normally I would have punched out much earlier, but after running these stats I forced myself to hold ... and it close at the high for the week. Since EUR data is coming out Monday am, I will unload a bunch at the open this evening, and hold just a small portion long with SL at b/e.

Overall the stats will lend insight ... and help put things in perspective and if we use them correctly, help them give us an edge.

I'll post later here a summary of EURUSD (and other pairs) ... I'm currently working on this now.

Regards,
EZ

Horse racing has a lot more in common with markets than casino games do, if you think about it.

Casino games have fixed odds (except for blackjack, where counting the cards can make small, some insist, important, differences).

Pari mutual payoffs are determined by the collective sentiment of the "crowd" - the amount of money going to a horse influences the payoff of that and all others in the race.

This is a much closer state to the auction markets we have in forex than is the spin of a roulette wheel.

FWIW, a book titled, 'Trading Realities' is on my current reading list. It's a bit more options/equity oriented, but he covers some of these conditional studies and how they evolved over different eras in the chapter on 'a new era'. You might have a look if you get a chance. Just be careful about look ahead bias when you set up the study.

If not already, you also should get familiar with the various tables commands in R (prop.table, etc), which are useful for your study.

Here's a simple, but in fact exact, formula for the determining the probability distribution of your study, under the assumption that the prices are normally distributed. It's easily derived from the joint distribution of the maximum, final position, and time of maximum, which was published by Larry Shepp (you can find it via Google and JStor).

Say 0 < u <= 1, and U = -ln(u). Then, conditional on the week ending value, p, the Max corresponding to a given value of U is:

M(U) = 1/2 * (p + sqrt( p^2 + 2 * s^2 * U)), where s is the standard deviation for the week.

Note that if u = 1, ln(u) = 0, and M(0) = p; the price ends the week at its max. As u -> 0, then -ln(u) "explodes", and the Max rises at a rate of about the root of "explodes."

If your results have more frequent higher range values than those indicated by this formula, then you're seeing one reason why trend following is profitable: too many extreme observations. But that's another story entirely.

Thanks ...