The short and simple answer is that when AncestryDNA gives you an estimate of your ethnicity, they're just guessing. Each company has different ways of making the estimate, and over time, they try to refine the estimate -- but it's still an educated guess.
In her Apr 16, 2017 post "Still not soup" on The Legal Genealogist, Judy G. Russell explains:
[A]dmixture tests ... take the DNA of living people — us, the test
takers — and they compare it to the DNA of other living people —
people whose parents and grandparents and, sometimes, even great
grandparents all come from one geographic area. Then they try to
extrapolate backwards into time.
and she goes on to say:
In other words, these percentages are:
• estimates based on comparisons not to actual historical populations
but rather to small groups of people living today, and
• estimates based purely on the statistical odds that those small
groups tell us something meaningful about past populations.
The companies do refine their methods of making estimates, and sometimes publish detailed explanations about how they arrived at the estimates. Ancestry has published a whitepaper on their Genetic Communities™, and they've made their 2013 Ethnicity Estimate White Paper available, so if you want to plunge into population genetics and read all about it, you can.
But as Russell points out, the fundamental difficulty is the assumption that looking at a population that has been in the same place for a limited number of generations can predict what that population was thousands of years ago. Your DNA is not being compared to old bones which have been dug up -- it is being compared with a (possibly) moving target.
For more on AncestryDNA's ethnicity estimates, see the article about AncestryDNA at the ISOGG Wiki. (Thanks to Debbie Kennett for the link to the Genetic Communities white paper.)