...I don't do it myself, but all the books recommend it. Suppose you are trading intraday on a one minute chart and you have 2 years of data to work with. The rule of thumb is to divide that into ten parts. Optimize on the earliest tenth of the data. Then run the optimized system on the next tenth only and see how it works. Then reoptimize on both tenths, and see how that works on the third tenth. Repeat until finished. The idea, which I think is stupid, is to get a feel for how it would have traded, and for how robust it is over changing conditions. I prefer to get a wild assed idea, BT it for 60 days, watch its performance for a month, trade it if it's good, and reoptimize it weekly to keep up with current conditions.
...I apologize for killing your promising and intelligent thread. I had hoped that by saying that walkforward was stupid that an avalanche of replies would roll in, making you huge, like I have done for TheStudent. Go figure. Maybe nobody else walks forward, either.
Walk forward testing is just out-of-sample data tested sequentially. It's used to see drawdowns month by month going forward. Nowadays most developers use Monte Carlo testing to determine pessimistic drawdown estimates. Here's a link for some info. on the whole out-of-sample testing process. http://www.customtradingsolutions.com/presentations/out_of_sample.pdf