Well, since you have the average, you can compare it quite directly.

You can't add confidence intervals (error bars), because you have no date on the variation. However, confidence intervals are kind of unnecessary anyway, since you have sampled the entire population of houses sold each year (we could argue there is still a hypothetical population). You could estimate them if you have data on typical variance.

Note that the type of houses sold might differ (between each year and each city) so what the differences actually mean is up for interpreation.

What I would do is make a simple line graph with indexed house prices for each city (100=2012). As long as you have a decent number of transactions each year (say, >20), that should give you a relatively stable measure.