Soft robotic arms have gained popularity in the recent years because of their dexterity, robustness and safe interaction with humans. However, since these arms are subject to non-linear mechanics and are intrinsically under-actuated, their control still present many challenges. Octopus arms are one of the most popular biological models for soft robotics. It is known that the octopus reaching movement consists in two steps: (1) the rotation of the arm's base towards the target, and (2) the extension of the arm to reach the target. From a robotics point of view, the rotation of the base adds one additional degree of freedom to an already hyper-redundant system. Therefore, its role in the effectiveness of the control is ambiguous. In this work, we investigate the role of the base rotation for learning an effective reaching strategy. We conduct numerical experiments based on a mathematical model of the mechanics of the octopus arm in water and a simple neural network enabling to encode the control strategy through optimization learning. The network node corresponding to the base rotation is switched on or off for comparison. We test the reaching success rate with and without base rotation with targets in various positions. The results show that the addition of the base rotation can be highly beneficial or even detrimental, based on the position of the target. Nonetheless, globally the addition of base rotation affects the control strategy and expand the reachable regions.