The short answer is: peaks of positive components do not have to have positive values, and of negative components - negative. You're dealing with continuous signals, so earlier components "drive" the later ones in their amplitudes. For example, if your N2 is particularly large, it may cause the following positivity not to reach positive values, even though the values will indeed increase. Early components (N1, P1) often show the respective intuitive negative / positive values, because they usually directly follow baseline correction, but later ones depend a lot on what happens before them. What you care for is the relative change.

Many thanks for your insight Magdalena Matyjek !

Could you refer me to studies on effect of early component of ERP on the later one?

You're welcome!

I'm not familiar with any studies that would focus on this specifically, but an example of a correction for such influence can be found in Bayer et al., 2016 (doi: 10.1093/scan/nsw162). From there: "In order to account for differences in the preceding C1 component, P1 amplitudes were analyzed as absolute base-topeak

values by subtracting C1 amplitudes."

Agreed with Magdalena's well articulated response. I would also suggest that you can simplify this by perhaps just looking at absolute power base to peak.