QuoteOriginally posted by: abumazenQuoteOriginally posted by: chiral3a large propensity of people that bought bug spray also bought bananas.A few days ago, someone was using one of my computers to order from FreshDirect.com. If I remember correctly, when she ordered bananas, they asked her if she also wanted peanut butter.But the real question is, do customers who buy S&P futures also buy dollars? Perhaps customers who buy five-lots at-the-market more likley to than people who buy 20-lots. Perhaps customers who buy S&P futures at 2:00 are more likely to. Perhaps customers whose orders are routed through major brokers, rather than having their own Globex market-data connection, are more likely to.Perhaps girls are less likely to.MPThe only thing that I don't like about that idea is that the supermarket model does not attampt to explain why people want bananas in July when they buy bug spay. It just picks up on numerous weighted relationships. The drivers are not quantifiable and are behavioral. Buying SnP futures is model driven, and therefore is driven differently. Someone is buying something, for instance, to set up a hedge. Now, ideally the models will not even pick up on this or, if they do, their is no significance. If the models do pick up on this, and hopefully they don't, they give some insane R^2 like 95%, and you act on it. But is never works again, because the position depends on something with no relationship, some hedge that you model says will move against the position. The supermarkets have relationships that, while you can't predict or see the cause, you can predict the outcome.

QuoteOriginally posted by: chiral3The only thing that I don't like about that idea is that the supermarket model does not attampt to explain why people want bananas in July when they buy bug spay. It just picks up on numerous weighted relationships.Where on Earth did I attempt to explain any more complex relationship between buying US stock, and buying dollars?Some physical connection is implied by the empirical correlation in either case. If we discovered enough correlations, we could draw a picture of the space through which they were filtered. But we have already drawn the picture in the statistical relations.Admittedly, we can use out-of-sample data to fine-tune confidence in a given discovery - as Monroe Trout claimed to do. But you have not shown that Amazon.com doesn't do the same thing.QuoteOriginally posted by: chiral3Buying SnP futures is model driven, and therefore is driven differently. Someone is buying something, for instance, to set up a hedge.Nonsense. There is no difference. You need to think about this more deeply. It is not a closed system. The very fact that your hedge has another leg implies this. What, was the guy who took the other side of that other leg also hedging? The simplest example of how it is not "model-driven" is that if you have no money, you can't buy futures, not matter what your model says to do. Models are demand driven. You could similarly say the grocer who takes an inventory of bananas from the wholesalers is "model driven."MP

If you think about it, Chiral, you could run it in reverse, like a particle filter. Suppose you wanted to track the positions of two people walking through the store, but you could only do it using inventory-control sensors on the shelves.You could separate the two people when one bought rice cakes, and the other one bought hot dogs. But then when they both picked up toilet paper, you'd get them confused. Then when one bought mustard, and the other bought flowers, you'd realize the one was a woman buying supplies for her family cookout, while the other was a single guy picking up some stuff for his girlfriend.Then, when one picks up a copy of Vogue, and the other one picks up a package of ground beef, you have to use your prior knowledge of the types of people who come into the store to sort it out...MP

I used the following approximation to get the parameters of the distributions. I was going to assume that the data I would get follow exponentail distri. So this new approximation is ok I think. Using those parameters, I found the all the outputs that are needed. There is a software comes with Frederick and Gerald book on "Introduction to Stichastic Models in O.R". That'd solve for the outputs if u give the parameters. I compared my calculated values with it, and it was right. I'm almost done with my project. I just have to do the formal write up now. Thank u all for u'r inputs. lambda = number of arrivals/observation timemie = N/ Twhere N= number of customers served T= Time taken to serve those N customres.