It is obviously not possible to create spread-bars (OHLC) using daily data. However, as several of you have pointed out, this is not a very big limitation for the average trendfollowing trader since a) a breakout from the close is more or less as powerful as a breakout from the high/low, b) ATR isn't the only measure of risk (stdev comes to mind
)
We do spreads, even if it's not to the extent I would like, and they can definitely add to diversification. However, which someone pointed out, it's a very different beast than outright positions. Very often you end up with a more or less mean reverting time-serie when you cross two trending instruments. Hence, crossing every instrument with each other and trade them all is not a viable approach. Consequently, trading spreads increases the level of datamining since we have to pick only certain spreads to remain profitable in backtests (and live). This is the tricky part. We try to solve this by finding some kind of economic relationship between the instruments in the spread, which may explain a potential trend behavior (if we want to trendfollow, that is). Still, this is dangerous stuff.
In addition to the above, spreads come with a tricky decision how to risk adjust the legs, if at all. And depending on what you decide, the outcome will be different. Thus, spreads introduce an almost endless number of possibilities to overfit the system. Very interesting area, but also very dangerous.
A slightly less scary road to take (since the possibilities are much fewer) is to do "relative value" trades, which essentially is the same as portfolio wide, or sub-portfolio wide, spreads.
Just my 2 cents...