puppeye posted this over on the historical tick data thread (:
(the download is over there ()
One of the first thoughts that popped into my mind was how to use this utility to thwart curve fitting.
My idea is to have a c# program create a bunch of random data and output in the tick data format. This utility then reads it and passes it to Ninja to build bars out of.
NT should store whatever data you feed it through this utility in its database. So you can just feed it a bunch of data once, then it will remember it later -- no need for utility again and again.
Then you can backtest and optimize against this random data, as well as against real instrument data. In fact, you can create multiple sets of fake data (just use different instrument names or expiration dates for the fake data sets) and test against them individually. The results should clue you in to whether or not your strategy is curve fitted.
I don't have any extra time right now, but maybe someone else is willing to write a c# program (or any language) to create us some random tick-level data .txt files. I'd like to test against a few million ticks per run, at least.