I'm working on a project that looks to harness the power of the GPU to, via brute force, identify trending pairs. It will look to find the correlation value between the two tickers and then monitor the price ratio to find a good entry/exit point. I will be looping through a list of 70 ticker symbols to find the most closely correlated pairs for the last, say, 15-100 ticks. Does anyone have any metrics on how fast current automated correlation identification schemes work? Do people even try and automate it at this level of detail? Is it done on a per tick basis? Is it even worth going subsecond to find trending pairs out of a list of many symbols. Cheers.