In their article (Context-dependent optimal substitution matrices for exposed residues (Koshi-Goldstein, 1995)), more precisely in the substitution matrix displayed in Figure 2, one one hand, the legend says: "Graphical representation of the probability of a mutation from the amino acid in a given row to that in a given column during an evolutionary time of 40 point-accepted mutations per 100 residues"
And on the other hand, I've noticed it is a lot more likely for an amino acid to be deleted than to be inserted (cf. screenshot, blue is low probability, orange/red is high probability).
Taken from: context-dependent optimal substitution matrices for exposed residues (Koshi-Goldstein, 1995)
How could one explain that? If that was true, all proteins sequence would slowly become smaller and smaller, right? Is it because it doesn't consider when proteins sequences merge? (I'm no biology expert, I'm sorry if the vocabulary I'm using is not exact. You are encouraged to correct me).