I agree with Levi Page's comment below. User identified start and end point is important. I believe the old WSOD now MarkitonDemand have something similar. Thanks

As an addition to this idea, implementing the 95-percent rule would be great.

Neat feature. Additionally, if possible, a neat feature for us is to have both positive and negative 1 and 2* standard deviation lines along the linear trend lines. This helps traditional data miners understand predictions better. Please let me know if this needs to be a separate new request.

Mokumax - I calculate it server side and the offer it up as a separate data series. In perl I used Math::NumberCruncher::BestFit

@Torstein - that was my first thought as well. It seems like complex formula however. Anybody have the algorithm?

This would be extremely useful. If we could also calculate the trend based on the visible canvas... So if you zoom in on a chart, the trend line is calculated based on the visible values, rather than the entire dataset.

I would make it independent of the series... Just allow the user to specify the start and end value. There are instances where the trend line might be the result of multiple series, so pointing to the parent in this case would not work. Also the user might want to use their own algorithm, so how would high charts know what the start and end points should be. But, a new series type called trendline would be great. It wouldn't hurt though to have built in algorthms for linear regression and so forth... in this case pointing it to a series or multiple series would be awesome (as long as the option is still there to specify your own start and end point).

that would be good, what I think is necessary, is to automate the process of creating the trend line: I've added one on a test graph, but I had to compute the line formula by myself.

This is probably best added as a separate series type, possibly with reference to a parent series that holds the input data.