Percent Increase in time series

Question

I want to compute percent increases on time series data, but I also want to take into account the absolute volume of the data. For e.g. suppose I have 2 sets of data (a) [10, 15, 50, 100] and (b) [1, 5, 15, 35]. The percent increase vector for each of them is [50.0, 233.34, 100.0] and [400.0, 200.0, 133.33] respectively. From my application's perspective, the former case is more favorable as the computation is being occurring on larger values, which is what I want. However, I my module is detecting lots of cases like latter which I want to avoid in some way. Is there some sort of weighted average that I can compute (or some other metric) to account for the absolute values?

Answer 1

You can do lots of things. You can just average the values and rank the series on that basis. It sounds like you also want to reward high amounts of increase. You need to define how to balance magnitude with rate of increase. Which is more important to you? How much more important? Does it matter if the increases are steady or can it be one big jump? This is a question of figuring out what you want to reward, not of mathematics.