This is indeed a bit confusing. I think it boils down to how Matplotlib handles the secondary axes. Pandas probably calls ax.twinx() somewhere which superimposes a secondary axes on the first one, but this is actually a separate axes. Therefore also with separate lines & labels and a separate legend. Calling plt.legend() only applies to one of the axes (the active one) which in your example is the second axes.
Pandas fortunately does store both axes, so you can grab all line objects from both of them and pass them to the .legend() command yourself. Given your example data:
You can plot exactly as you did:
ax = var.total.plot(label="Variance")
ax = shares.average.plot(secondary_y=True, label="Average Age")
ax.set_ylabel('Variance of log wages')
ax.right_ax.set_ylabel('Average age')
Both axes objects are available with ax (left axe) and ax.right_ax, so you can grab the line objects from them. Matplotlib’s .get_lines() return a list so you can merge them by simple addition.
lines = ax.get_lines() + ax.right_ax.get_lines()
The line objects have a label property which can be used to read and pass the label to the .legend() command.
ax.legend(lines, [l.get_label() for l in lines], loc="upper center")
And the rest of the plotting:
ax.set_title('Wage Variance and Mean Age')
plt.show()
edit:
It might be less confusing if you separate the Pandas (data) and the Matplotlib (plotting) parts more strictly, so avoid using the Pandas build-in plotting (which only wraps Matplotlib anyway):
fig, ax = plt.subplots()
ax.plot(var.index.to_datetime(), var.total, 'b', label="Variance")
ax.set_ylabel('Variance of log wages')
ax2 = ax.twinx()
ax2.plot(shares.index.to_datetime(), shares.average, 'g' , label="Average Age")
ax2.set_ylabel('Average age')
lines = ax.get_lines() + ax2.get_lines()
ax.legend(lines, [line.get_label() for line in lines], loc="upper center")
ax.set_title('Wage Variance and Mean Age')
plt.show()