Hi Fred,
Doing a few stocks might not be a problem but trying to track 1000 of them would sure be a task for a PC. Maybe my SW guys are full of crap too?
Let's see. Let's say two billion shares traded on the NYSE, NASDAQ and AMEX -- I think that's more than the record. And that average trade was 500 shares. That's 4 million transactions. The market is open 6.5 hours or 23400 seconds. That leaves 6 milliseconds to process each transaction on average -- trivial with any modern computer. More than trivial.
To calculate money flow for each stock, the program would have to keep track of last price, ticker, block and non-block money flow dynamically (4 bytes for each should do). For 10,000 stocks that's only 160K in data structure. Each additional period you wanted to track, (7day, 14 day etc) you add an additional 80K in data requirenment.
The conditionals would be easy. Which stock are you? Were you an up, down or zero tick? Block or non-block trade? Ok, price*transaction gets added/subtracted to the block or non-block cumulative sum or thrown out for zero tick.
As far as data transfer, lets say transactions are coded really inefficiently from your data feed at 100 bytes per. That's an average transfer rate of 17K baud. Even an analog modem should be able to handle that.
Let's double the complexity of the coding to handle cancelled transactions. Seems simple to code and something a 486 could run while sipping on a pina colada. Run that by your SW guys and tell me what I've missed.
The thing I can't get is the data. Tick by tick data feeds are 1/6 the cost of a bloomberg at the cheapest I've found.
Rich |
