Anybody simulate their own prices?

Hi

Just a general qry on this, does anybody go to the bother of simulating their own prices (from the characteristics of their chosen market) when backtesting a potential system?

Im thinking along the lines of say something like an Ito process/ brownian motion/ etc. Ultimately you should end up with a recursive time dependent function like this;
future_price = (now_price * mean_return) + (now_price * random~N[0,1] * volatility_return)

Any opinions on whether this might be a better way to tweak a potential system rather than ultimately data-fitting on one data-set (of actual prices)? That is; set the parameters of your system for the best-fit over thousands of runs rather than set them so that they fit the actual data that happened.

Thanks
John5000

In my opinion you cannot simulate market data. Market data is market data. I believe it is impossible to replicate it. Yes you can come up with a stream of numbers that make it look like market data but it will have lost some of the characteristics of market data. The only characteristics it will have are the ones you've chosen it to have. I think it is very important how real market data is used when building a system in that you want enough of it in order to properly test a system on unseen data. I tend to use a minimum of curve fit to 1 part of market data to an unseen walk forward of 6 parts.

Agree with Tufty

The "best" you can realistically do without a Phd in hyper mathematics is to take a set of prices, do a natural log of the %changes and randomise them.

However, this fails to take account of the "clustered" nature of extreme price movements.