Reproducing annealing results with classical simulations

I am trying to compare the annealing results of the Pegasus machine with exact diagonalization. For that I am using a simple QUBO problem described below. There is an unexpected, large discrepancy between the quantum annealing results and the exact classical simulation results as shown in the histogram below.

Is there a way to determine the reason for that? Is there a way to precisely determine the initial state (or the temperature) and the exact annealing schedule for the specific machine used? 

I used the maximum annealing time of 20 microseconds and the provided annealing schedule for the classical simulation. I set all h_i to zero and use the following nonzero J_ij couplings

(4, 3): -0.6, (7, 3): 0.2, (7, 4): 0.9, (6, 1): -0.8, (14, 7): -0.2, (14, 0): -0.8, (11, 1): 0.3, (11, 0): -0.1, (10, 6): -0.4, (10, 14): -0.4, (15, 3): 0.1, (15, 10): 0.1, (8, 4): 1.0, (8, 11): 0.1, (5, 15): -0.1, (5, 8): -0.7, (12, 5): 0.1, (2, 1): 1.0, (2, 12): 0.9, (9, 0): -0.4, (9, 12): -0.1, (13, 6): -0.6, (13, 2): 0.6, (13, 9): 0.8 

Thanks for help.
Bojan