First, notice that the definition of the nth term is just a weighted average of the two previous terms. That is, the nth term will always be between the (n-1)st term and the (n-2)nd term, and will be twice as far from the (n-2)nd term as it is from the (n-1)st term. In other words, if a, b, c are three consecutive terms in the sequence, we will always have:
|c-a| = 2|c-b| (in other words, the distances will be in a 2 to 1 ratio)
and either a < c < b or b < c < a
So if b_1 = 10, b_2 = 20, then:
b_3 = 10 + (2/3)(20 - 10) [b_3 is 10 plus 2/3rds of the distance between 10 and 20]
b_4 = 10 + (2/3)(10) + 1/9(10) [the distance from b_2 to b_3 is 1/3 of the distance from b_1 to b_2]
b_5 = 10 + (2/3)(10) + (1/9)(10) - (1/27)(10)
b_6 = 10 + (2/3)(10) + 1/9(10) - (1/27)(10) + (1/81)(10)
and so on.
In general, for i > 2,
b_i = 50/3 + SUM_(k from 2 to i-2) [(-1)^k * 10 * (1/3^k)]
This sum converges very quickly. If we work out the sum over an infinite number of terms, we'll get an excellent approximation of b_15. Rewriting:
b_i = 50/3 + 10*SUM_(k from 2 to i-2) [(-1)/3]^k
Let S = SUM_(k from 2 to i-2) [(-1)/3]^k
S = (1/3^2 - 1/3^3) + (1/3^4 - 1/3^5) + (1/3^6 - 1/3^7) + ...
S = 2/3^3 + 2/3^5 + 2/3^7 + ...
S = 2(1/3^3 + 1/3^5 + 1/3^7 + ...)
S = 2[1/3^3 + (1/3^2)(1/3^3 + 1/3^5 + 1/3^7 + ...)]
S = 2(1/3^3 + (1/3^2)(S/2))
S = 2/27 + S/9
8S/9 = 2/27
S = 1/12
(or you could just use a formula for the sum of an alternating series here).
So limit(i --> infinity) b_i = 50/3 + 10*(1/12) = 100/6 + 5/6 = 17.5
b_15 will be *very* close to 17.5. Working out the exact value of b_15 is a bit annoying, but you can do it following the method above; I find that it's equal to:
(82*3^6 + 10)(20/3^13)
Still, I'd use the answer choices in a question like this - and would never expect to see such a question on the GMAT!